Ecotoxicity Data Systematic Reviews: Updated Methods, Applications, and Integration for Biomedical and Regulatory Science

Abstract

This article provides a comprehensive guide to modern systematic review (SR) methods for ecotoxicity data, tailored for researchers and professionals in drug development and environmental health. It addresses the urgent need to update statistical practices in ecotoxicology, highlighted by the ongoing revision of key guidelines like OECD No. 54 [citation:1][citation:2]. The scope covers foundational concepts, advanced methodologies like PRISMA-guided reviews and dose-response modeling using Generalized Linear Models (GLMs) [citation:1][citation:5], strategies for troubleshooting data heterogeneity and quality assessment, and frameworks for validating and integrating evidence into regulatory decision-making [citation:3]. The article synthesizes current initiatives and offers a practical roadmap for implementing robust, transparent, and scientifically defensible SRs to support chemical risk assessment and biomedical research.

The Evolving Landscape of Ecotoxicology: Why Systematic Reviews and Modern Statistics Are Now Essential

The field of ecotoxicology stands at a critical juncture. Regulatory hazard assessment, which relies heavily on extensive animal testing, faces escalating ethical, financial, and practical pressures [1]. With over 350,000 chemicals and mixtures registered globally, the traditional paradigm of toxicity testing is unsustainable [1]. Concurrently, the statistical and methodological foundations of the discipline have been identified as fragmented, inconsistent, and outdated [2]. For decades, regulatory risk assessments have been based on statistical principles that are no longer considered state-of-the-art, creating a significant gap between available scientific innovation and applied regulatory practice [2].

This whitepaper frames this challenge within the context of systematic review methods for ecotoxicity data research. Systematic review provides a framework for identifying, evaluating, and synthesizing evidence with transparency, objectivity, and consistency [3]. The evolution of curated databases like the ECOTOXicology Knowledgebase (ECOTOX) demonstrates a long history of applying systematic principles to data curation [3]. However, the full potential of systematic review is hindered by persistent use of outdated analytical practices and barriers to integrating contemporary academic research into regulatory decision-making [4] [2].

The push for change is driven by a confluence of factors: the urgent need for alternatives to animal testing, the advent of powerful computational and machine learning tools, and a growing consensus within the scientific community that methodological overhaul is essential for robust chemical safety assessments [2] [1]. This document provides an in-depth technical guide to the core outdated practices, the contemporary methods poised to replace them, and the integrated systems—encompassing data curation, statistical analysis, and computational modeling—required to realize a modernized ecotoxicology.

The Legacy Framework: Outdated Practices and Their Limitations

The historical framework of ecotoxicology is characterized by methodological conventions that have become outmoded. A central and protracted debate exemplifies this stagnation: the use of the No-Observed Effect Concentration (NOEC), which has been contested for over 30 years [2]. The NOEC approach, which relies on statistical hypothesis testing to identify the highest test concentration showing no significant difference from controls, is fundamentally flawed. It is critically dependent on study design (e.g., the chosen test concentrations and sample size), lacks statistical power, and does not provide information on the concentration-response relationship or the magnitude of effects [2].

The regulatory guidance anchoring many of these practices is the Organisation for Economic Co-operation and Development (OECD) document No. 54, "Current approaches in the statistical analysis of ecotoxicity data: A guidance to application," published in 2006 [2]. This document is now widely recognized as no longer reflective of contemporary statistical methods or the computational platforms available to scientists [2]. Its proposed dichotomy between "hypothesis testing" (e.g., ANOVA-type models) and "dose-response modeling" (regression models) neglects the underlying unity of these approaches as variants of linear models [2]. This artificial separation has hindered the adoption of more sophisticated and informative analysis techniques.

Beyond statistics, the evaluation of data quality for use in assessments relies on various reliability evaluation methods. These methods are inconsistently applied and vary in their criteria, as summarized in Table 1 [5].

Table 1: Comparison of Reliability Evaluation Methods for Ecotoxicity Data [5]

Method (Source)	Klimisch et al.	Durda & Preziosi	Hobbs et al.	Schneider et al. (ToxRTool)
Primary Data Types	Toxicity & ecotoxicity	Ecotoxicity	Ecotoxicity (acute & chronic)	Toxicity (in vivo & in vitro)
Evaluation Categories	Reliable without/with restrictions, Not reliable, Not assignable	High, Moderate, Low quality, Not reliable, Not assignable	High, Acceptable, Unacceptable quality	Reliable without/with restrictions, Not reliable, Not assignable
Number of Criteria	12 (acute) / 14 (chronic)	40	20	21
Guidance & Summarization	Limited guidance; summarization not stated	Detailed guidance; summarization stated	Limited guidance; summarization stated	Detailed guidance; automated calculation
Regulatory Association	Recommended in REACH guidance	Based on US EPA, OECD, ASTM standards	For Australasian database	-

Furthermore, a systems-based analysis of European chemical assessment reveals deep-seated barriers to the uptake of modern academic research. Stakeholders are divided on the extent to which available evidence is used, citing issues of reliability, transparency, and a fundamental misalignment between the goals of academic knowledge production and the demands of regulatory decision-making [4]. These technical and social factors are interconnected, requiring a systems-level approach to overcome [4].

The Contemporary Statistical Toolbox: Moving Beyond OECD No. 54

The revision of OECD No. 54, planned for 2026, represents a pivotal opportunity to align regulatory guidance with modern statistical science [2]. Contemporary practice moves beyond the simple ANOVA vs. regression dichotomy to a unified framework based on Generalized Linear Models (GLMs) and their extensions [2].

GLMs handle non-normal data (e.g., binomial mortality counts, Poisson counts) through link functions, eliminating the need for problematic data transformations. Mixed-effect GLMs (hierarchical models) can account for nested data structures and variability, such as individuals within test chambers or multiple studies on the same chemical [2]. For more complex, nonlinear response patterns, Generalized Additive Models (GAMs) allow the relationship between concentration and response to be described by smooth, data-driven curves [2].

The benchmark for analysis is now continuous dose-response modeling to derive point estimates like the ECx (the concentration causing an x% effect). Modern software, particularly the open-source R environment and packages like drc, has made fitting multi-parameter models (e.g., 2 to 5 parameter logistic models) routine [2]. Furthermore, advanced metrics are gaining traction:

• Benchmark Dose (BMD) Modeling: Promoted by the European Food Safety Authority, this method fits a dose-response model to identify a lower confidence limit for a predetermined benchmark response (e.g., a 10% effect), offering advantages in statistical stability and biological relevance [2].
• No-Significant Effect Concentration (NSEC): A recently proposed metric designed to offer a more statistically robust alternative to the NOEC [2].

Table 2: Evolution of Key Statistical Practices in Ecotoxicology

Aspect	Legacy/Outdated Practice	Contemporary/Recommended Practice	Key Advantage of Contemporary Practice
Core Analysis	Dichotomy: Hypothesis Testing (ANOVA) vs. Dose-Response	Unified framework of Generalized Linear Models (GLMs)	Unified theory, handles non-normal data correctly via link functions.
Primary Metric	No-Observed Effect Concentration (NOEC)	Effective Concentration (ECx) or Benchmark Dose (BMD)	Quantifies effect magnitude, accounts for full dose-response curve, more statistically robust.
Model Flexibility	Fixed-parameter, pre-defined curves (e.g., standard logistic)	Flexible models (GAMs), multi-parameter models	Can describe complex, non-standard response patterns observed in real data.
Data Structure	Ignores hierarchical data structure	Mixed-effects models (Hierarchical GLMs)	Accounts for variability at multiple levels (e.g., within study, between labs), more accurate uncertainty estimates.
Inferential Framework	Almost exclusively Frequentist	Inclusion of Bayesian methods	Allows for incorporation of prior knowledge, direct probability statements about parameters.
Software & Accessibility	Proprietary software, limited methods	Open-source platforms (e.g., R with extensive packages)	Democratizes access to cutting-edge methods, ensures reproducibility.

A critical "wish list" for the OECD No. 54 revision includes clarifying the connection between hypothesis testing and dose-response, making continuous regression the default, incorporating GLMs and Bayesian methods, and advocating for better training in statistical design for ecotoxicologists [2].

Foundational Data Curation: Systematic Review and the ECOTOX Model

Robust contemporary analysis is impossible without high-quality, consistently curated data. The ECOTOXicology Knowledgebase (ECOTOX) is the world's largest curated ecotoxicity database and serves as a premier model for systematic data curation in the field [3]. Its methodology aligns closely with systematic review and PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [3].

The ECOTOX workflow is a rigorous, multi-stage pipeline, as illustrated in the following diagram.

The process begins with comprehensive literature searches for a chemical of interest, spanning both open and "grey" literature (e.g., government reports) [3]. References are screened in two phases: first for applicability (ecologically relevant species, correct chemical, reported exposure details), then for acceptability (documented controls, reported endpoints) [3]. Accepted studies undergo detailed data extraction using well-established controlled vocabularies to ensure consistency. Key extracted information includes chemical identity (CAS, DTXSID, structure), species taxonomy, detailed experimental design (exposure duration, media, endpoints like LC50/EC50), and test conditions [3]. Following quality assurance, the data is added to the publicly accessible database, which is updated quarterly [3].

This systematic, transparent process transforms disparate primary literature into FAIR data (Findable, Accessible, Interoperable, and Reusable), creating the essential feedstock for modern computational toxicology, model development, and integrated risk assessment [3].

Driving Innovation: Machine Learning and Benchmark Datasets

The curated data from systematic repositories like ECOTOX enables the next frontier: predictive computational toxicology. Machine Learning (ML) offers a powerful paradigm for predicting ecotoxicological outcomes, potentially reducing reliance on animal testing [1]. However, its adoption has been hampered by the lack of standardized, high-quality benchmark datasets, which are essential for fairly comparing model performance and advancing the field [1].

The ADORE (Aquatic Toxicity) dataset has been introduced to meet this need [1]. It is a curated benchmark dataset for acute aquatic toxicity focusing on fish, crustaceans, and algae. The dataset is constructed from ECOTOX data and enriched with chemical descriptors (e.g., molecular fingerprints), phylogenetic data, and species-specific traits [1].

Table 3: The ADORE Benchmark Dataset for Machine Learning [1]

Aspect	Description	Purpose & Significance
Core Source	Curated data from the US EPA ECOTOX Knowledgebase.	Provides a high-quality, toxicity-focused foundation.
Taxonomic Groups	Fish, Crustaceans, Algae.	Covers key aquatic trophic levels with substantial available data.
Key Endpoints	Acute mortality & comparable effects (LC50, EC50).	Aligns with standard regulatory endpoints (OECD Test Guidelines 203, 202, 201).
Data Enrichment	Chemical features (e.g., molecular representations, properties), phylogenetic features, species-specific traits.	Provides the multidimensional "feature space" necessary for ML algorithms to learn complex relationships.
Defined Splits	Provides predefined training/test splits based on chemical occurrence and molecular scaffolds.	Prevents data leakage; enables direct, fair comparison of different ML models on identical data partitions.
Primary Challenge	To predict toxicity values (regression) or classes (classification) for new chemicals or across taxonomic groups.	Drives community progress toward robust, generalizable predictive models for chemical hazard.

The ADORE dataset embodies the transition to contemporary methods by addressing a major barrier to ML adoption. It provides the community with a common benchmark to develop models for Quantitative Structure-Activity Relationships (QSARs) and beyond, facilitating exploration of complex questions like extrapolation across taxonomic groups [1].

Transitioning to contemporary methods requires familiarity with a new set of core resources and tools. The following table details key solutions spanning data, software, and methodological guidance.

Table 4: Research Reagent Solutions for Contemporary Ecotoxicology

Item / Resource	Type	Function / Purpose	Key Features / Relevance
ECOTOX Knowledgebase	Curated Database	The world's largest source of curated single-chemical ecotoxicity data [3].	Provides systematically reviewed data for ~12,000 chemicals and ~14,000 species; essential for assessment, modeling, and systematic review [3].
ADORE Dataset	Benchmark ML Dataset	A curated, feature-enriched dataset for developing and benchmarking ML models in aquatic toxicology [1].	Includes defined train/test splits; enables fair model comparison and focuses research on prediction challenges [1].
R Statistical Environment	Software Platform	An open-source language and environment for statistical computing and graphics [2].	The de facto standard for implementing contemporary statistical methods (GLMs, mixed-models, dose-response analysis, Bayesian inference) in ecotoxicology [2].
ToxRTool	Evaluation Tool	Toxicological data Reliability assessment Tool for evaluating study reliability [5].	Provides a structured, transparent method with 21 criteria to score and categorize study reliability for use in assessments [5].
OECD No. 54 (Future Revision)	Guidance Document	The forthcoming updated guidance on the statistical analysis of ecotoxicity data (revision planned for 2026) [2].	Will be the central regulatory guidance promoting modern statistical methods (GLMs, BMD, Bayesian) as standard practice [2].
BMD Methodology	Statistical Approach	Benchmark Dose modeling for identifying a lower confidence limit on a predefined effect size [2].	Supported by EFSA; offers a more stable and biologically informative alternative to NOEC/LOEC and point estimates like EC10 [2].

Synthesis and Future Directions: Implementing a Systems-Level Change

The historical context of ecotoxicology reveals a field constrained by outdated statistical paradigms, inconsistent data evaluation, and a socio-technical system that limits the flow of academic innovation into regulatory practice [4] [2]. The push for change is now unequivocal, driven by the confluence of ethical imperatives, regulatory needs, and scientific advancement.

The path forward requires a systems-level approach that simultaneously addresses technical, social, and training dimensions [4]. The following diagram synthesizes the key barriers and intervention points within the socio-technical system of regulatory toxicology [4].

Technical modernization is underway through the statistical toolbox update, the systematic curation of FAIR data, and the development of ML-ready benchmarks. However, as the diagram illustrates, these technical factors are interdependent with social factors [4]. Lasting change requires:

• Finalizing and adopting modernized guidance, specifically the OECD No. 54 revision, to align regulatory standards with scientific best practices [2].
• Embracing open science and benchmarking through resources like the ADORE dataset to accelerate and standardize computational model development [1].
• Investing in targeted training to elevate statistical and data science literacy within the ecotoxicology community, moving beyond sporadic courses to embedded expertise [2].
• Fostering sustained collaboration between academic statisticians, ecotoxicologists, and regulatory scientists to align goals, clarify assessment objectives, and co-develop solutions [4] [2].

By integrating robust systematic review for data curation, contemporary statistical and computational methods for analysis, and a collaborative systems mindset for implementation, the field of ecotoxicology can fully transition from its outdated practices to a more predictive, efficient, and evidence-based future.

Within the context of developing rigorous systematic review methods for ecotoxicity data research, three core challenges persistently hinder progress: fragmented and outdated statistical guidelines, the long‑standing debate over the use of the no‑observed‑effect concentration (NOEC), and the limited integration of academic research into regulatory decision‑making. This whitepaper examines each challenge in detail, presents quantitative data from recent studies, outlines key experimental methodologies, and provides visual workflows to guide researchers and risk assessors.

Fragmented Statistical Guidelines

The statistical methods used in regulatory ecotoxicology have evolved in a fragmented, inconsistent, and often outdated manner, creating a landscape that is difficult to navigate[reference:0]. A prominent illustration of this fragmentation is the decades‑long debate over whether NOECs should be banned from or used in regulatory contexts[reference:1]. The key reference document, OECD No. 54 (“Current approaches in the statistical analysis of ecotoxicity data”), published in 2006, is no longer considered reflective of contemporary statistical methods or computational platforms[reference:2]. A revision is planned for 2026, coordinated by the German Environment Agency (UBA)[reference:3].

This fragmentation is not limited to hypothesis‑testing versus dose‑response modeling. The evaluation of study reliability itself relies on inconsistent criteria. The widely used Klimisch method (1997) has been criticized for lacking detail and guidance, leading to evaluations that depend heavily on expert judgment and yield inconsistent results among assessors[reference:4][reference:5]. In response, the Criteria for Reporting and Evaluating ecotoxicity Data (CRED) method was developed to provide a more transparent, consistent, and science‑based framework for evaluating both reliability and relevance[reference:6].

The NOEC Debate

The NOEC, derived from hypothesis‑testing (e.g., ANOVA), identifies the highest tested concentration at which no statistically significant effect is observed. Its critics argue that the NOEC is inherently dependent on study design (e.g., chosen concentration spacing, sample size) and does not estimate a true threshold of effect. Alternatives such as the effect concentration for x% of the population (ECx), the benchmark dose (BMD), and the no‑significant‑effect concentration (NSEC) have been proposed as more robust, regression‑based metrics[reference:7].

A recent meta‑analysis of chronic freshwater toxicity tests quantified the actual effect levels associated with traditional hypothesis‑testing endpoints[reference:8]. The findings are summarized in Table 1.

Table 1. Median percent effect and adjustment factors for hypothesis‑based toxicity endpoints (Justice et al., 2025)

Endpoint	Median percent effect occurring at the endpoint	Median adjustment factor to approximate EC5
NOEC	8.5%	1.2
LOEC	46.5%	2.5
MATC	23.5%	1.8
EC20	–	1.7
EC10	–	1.3

Source: [reference:9]

The data show that a “no‑observed” effect can correspond to a median effect of 8.5%, highlighting the disconnect between the statistical significance and the biological effect size. The derived adjustment factors allow hypothesis‑based results to be converted to a common scale (e.g., EC5), facilitating their use in screening‑level risk assessments[reference:10].

Integrating Academic Research

Academic peer‑reviewed studies are a vital source of evidence for chemical assessment, yet they face significant barriers to use in regulatory decision‑making[reference:11]. A systems‑based analysis of the European regulatory toxicology system identified deep divisions among stakeholders regarding the extent to which available academic evidence is used[reference:12]. Barriers are not merely technical (e.g., perceptions of variable reliability and transparency) but are often interdependent with social factors, such as misaligned goals between academic and regulatory knowledge production[reference:13]. Overcoming these barriers requires a coordinated, systems‑level approach that considers both technical and social dimensions[reference:14].

Experimental Protocols

Protocol 1: Meta‑Analysis of Hypothesis‑Based vs. Point‑Estimate Toxicity Values

Based on Justice et al. (2025)[reference:15]

• Literature search: Systematic search of chronic freshwater toxicity tests reporting NOEC, LOEC, MATC, EC10, and EC20 values.
• Inclusion criteria: Studies must report both hypothesis‑based endpoints (NOEC/LOEC/MATC) and point‑estimate endpoints (EC10/EC20) for the same test.
• Data extraction: For each test, record the reported endpoint values, test organism, chemical, and experimental design.
• Statistical analysis:
  • Calculate the percent effect corresponding to each NOEC, LOEC, and MATC by comparing the response at that concentration to the control response.
  • Derive adjustment factors for each endpoint by determining the multiplier needed to equate the endpoint to an EC5 value (e.g., NOEC/EC5).
  • Analyze the influence of chemical class and taxon (invertebrate vs. vertebrate) on the percent effect and adjustment factors using robust statistical tests (e.g., Kruskal‑Wallis).

Protocol 2: Ring Test for Evaluating Ecotoxicity Study Reliability (CRED Method)

Based on Kase et al. (2016)[reference:16]

• Participant recruitment: Assemble a diverse group of risk assessors (e.g., 75 assessors from 12 countries) representing regulatory agencies, industry, and academia.
• Study selection: Provide a set of aquatic ecotoxicity study reports covering a range of qualities and test types.
• Evaluation procedure:
  • Phase 1: Each assessor evaluates the same set of studies using the traditional Klimisch method.
  • Phase 2: Assessors evaluate the same studies using the new CRED method, which includes detailed criteria for reliability (e.g., test validity, reporting completeness) and relevance (e.g., ecological realism, endpoint appropriateness).
• Consistency assessment: Compare the evaluation outcomes (reliable/not reliable, relevance score) within and between the two methods. Measure agreement using statistical metrics (e.g., Cohen’s kappa).
• Perception survey: Collect feedback from participants on the practicality, transparency, and perceived accuracy of each method.

Visualization of Workflows and Relationships

Diagram 1: Systematic Review Workflow for Ecotoxicity Data

Diagram 2: Decision Tree for Selecting Ecotoxicity Statistical Endpoints

Diagram 3: Interrelationship of Core Challenges

The Scientist’s Toolkit

Table 2. Essential materials and tools for ecotoxicity research and systematic review

Item	Function/Description
Test organisms (e.g., Daphnia magna, Danio rerio)	Standardized model species for acute/chronic toxicity testing; provide reproducible biological responses.
Exposure systems (static, renewal, flow‑through)	Control and maintain precise chemical concentrations during tests; flow‑through systems best mimic environmental conditions.
Chemical analysis tools (LC‑MS, GC‑MS, spectrophotometry)	Verify exposure concentrations, measure chemical degradation, and assess metabolite formation.
Statistical software (R) with packages (drc, bmd, ggplot2)	Fit dose‑response models (ECx, BMD), perform meta‑analyses, and create publication‑quality graphics.
Reliability assessment framework (CRED checklist)	Standardized tool to evaluate the reliability and relevance of individual ecotoxicity studies for inclusion in reviews.
Curated ecotoxicity database (e.g., US EPA ECOTOX Knowledgebase)	Centralized repository of published toxicity data; essential for systematic literature searches and data extraction.

In ecotoxicity research, the volume and complexity of data—from standardized laboratory assays to complex field studies—present a significant challenge for risk assessment and regulatory decision-making. The critical need to distinguish signal from noise, to reconcile conflicting studies on a chemical's environmental impact, and to build a reliable evidence base for policy underscores the importance of rigorous evidence synthesis methodologies [6].

This whitepaper defines and contrasts the core methodologies of evidence synthesis: the narrative review, the systematic review, and the quantitative meta-analysis, with a specific focus on their application within ecotoxicity data research. Systematic reviews, which minimize bias through explicit, pre-defined methods, have become the standard for providing confident evidence, increasingly supplanting traditional narrative reviews [6] [7]. Furthermore, we explore the principles of evidence integration, which governs how synthesized findings are incorporated into broader scientific narratives and decision-making frameworks. For researchers and drug development professionals, mastering these concepts is not academic; it is essential for validating ecotoxicological models, supporting chemical safety assessments, and ensuring that environmental health protections are built on a foundation of transparent, reproducible, and robust science.

Defining the Core Methodologies

A narrative review (or traditional literature review) provides a broad, qualitative summary of the literature on a topic by an expert in the field. It is characterized by its flexibility in scope and methodology, which is not pre-specified or systematic. The author selectively cites literature to provide an experiential perspective, track the development of a concept, or explore existing debates and knowledge gaps [6] [7]. While valuable for offering a foundational overview or generating novel hypotheses, its major limitation is the absence of objective, systematic selection criteria. This can introduce selection bias, where the review reflects the author's predispositions, leading to different narrative reviews on the same topic reaching conflicting conclusions [6]. As such, while they may be evidence-informed, narrative reviews are not considered rigorous scientific evidence on their own.

Systematic Review: The Structured, Bias-Minimizing Synthesis

A systematic review is a high-level research project that answers a specific, focused research question by systematically identifying, selecting, appraising, and synthesizing all relevant studies [7]. Its paramount objective is to minimize bias and error at every stage, ensuring transparency and reproducibility. This is achieved through a pre-published protocol that details the research question (often formulated using frameworks like PICO—Population, Intervention, Comparator, Outcome), explicit inclusion/exclusion criteria, a comprehensive search strategy across multiple databases, a critical appraisal of study quality, and a structured data synthesis [7]. The synthesis can be qualitative (descriptive) or quantitative (meta-analysis). Systematic reviews are the cornerstone of evidence-based practice, designed to provide the most valid evidence to guide decision-making in fields like clinical medicine and, increasingly, environmental policy [6] [7].

Meta-Analysis: The Quantitative Statistical Synthesis

Meta-analysis is a statistical methodology for quantitatively combining the results of multiple independent studies that address a common research question [8]. It is often, but not always, a core component of a systematic review. The process involves extracting a common effect size (e.g., standardized mean difference, risk ratio) and its variance from each study, then calculating a pooled (combined) effect estimate across studies [8] [9].

• Primary Purpose: To improve the precision of an effect estimate, resolve uncertainties from conflicting studies, and investigate sources of variation (heterogeneity) across studies [9].
• Key Statistical Models:
  • Fixed-Effect Model: Assumes all studies estimate a single, common underlying effect. The inverse of each study's variance is used as its weight [9].
  • Random-Effects Model: Acknowledges and models that the true effect may vary across studies (heterogeneity). It incorporates this between-study variation into the weighting, which redistributes weight more equally among studies, especially when heterogeneity is high [8] [9].

The results are typically displayed in a forest plot, which visually presents the effect size and confidence interval for each study alongside the pooled estimate [9].

Table 1: Comparative Analysis of Narrative and Systematic Review Methodologies [6] [7].

Feature	Narrative (Traditional) Review	Systematic Review
Research Question	Broad, exploratory, often not explicitly stated.	Narrow, specific, and pre-defined (e.g., using PICO).
Search Strategy	Not systematic or comprehensive; selection may be subjective.	Comprehensive, explicit, and reproducible across multiple databases.
Study Selection	No pre-defined criteria; prone to selection bias.	Based on explicit, pre-specified inclusion/exclusion criteria.
Quality Appraisal	Variable, often not formalized.	Rigorous critical appraisal of study validity (risk of bias).
Data Synthesis	Qualitative, narrative summary.	Structured synthesis, which can be qualitative or quantitative (meta-analysis).
Conclusions	Based on author's interpretation, potentially subjective.	Evidence-based, derived directly from the analyzed data.
Reproducibility	Low, due to lack of methodological transparency.	High, due to documented protocol and methods.
Primary Utility	Background, hypothesis generation, expert perspective.	Answering focused questions, supporting evidence-based decisions and policy.

The Systematic Review Workflow: A Step-by-Step Protocol for Ecotoxicity Data

The following protocol, adapted from guidelines like Cochrane and PRISMA, provides a detailed roadmap for conducting a systematic review in ecotoxicity.

Pre-Protocol Phase: Scoping & Team Assembly

• Define the Rationale: Clearly articulate the environmental or regulatory problem (e.g., "Uncertainty in the chronic aquatic toxicity of Chemical X").
• Assemble a Team: Include subject matter experts (ecotoxicologists), a librarian/information specialist for search design, and a statistician if a meta-analysis is anticipated.

Phase 1: Protocol Development & Registration

• Formulate the Research Question: Use a structured framework. For ecotoxicity: Population (e.g., Daphnia magna), Exposure (Chemical X concentration/duration), Comparator (Control/solvent control), Outcome (e.g., 48-hr LC50, reproduction NOEC) [7].
• Develop & Register the Protocol: Document the methods before starting. Key elements include:
  • Eligibility Criteria: Define included species, exposure regimes, study designs (e.g., OECD guideline tests, microcosm studies), language, and publication status (including "gray literature" like regulatory reports) [8].
  • Search Strategy: Design with an information specialist. Use controlled vocabulary (e.g., MeSH, Emtree) and keywords. Pre-specify databases (e.g., PubMed, TOXLINE, Web of Science, Scopus) and other sources [8].
  • Study Selection Process: Outline how titles/abstracts and full texts will be screened, typically by two reviewers independently, with conflicts resolved by consensus or a third reviewer.
  • Data Extraction Plan: Design a standardized form to capture study characteristics (test organism, exposure conditions, endpoint, effect size data, study funding) and risk of bias items.
  • Risk of Bias Assessment Plan: Specify the tool (e.g., adapted from Cochrane RoB, SYRCLE's RoB tool for animal studies) to evaluate study reliability.
  • Synthesis Strategy: State plans for qualitative synthesis and the conditions under which a meta-analysis will be performed (e.g., sufficient data, comparable outcomes).

Phase 2: Study Identification & Selection

• Execute the Search: Run the search across all planned sources. Record the number of records identified from each source.
• Manage Records & Screen: Use reference management software (e.g., DistillerSR, Rayyan). Perform deduplication. Screen titles/abstracts, then full texts, against eligibility criteria. Document reasons for exclusion at the full-text stage.
• Diagram the Flow: Create a PRISMA flow diagram to transparently report the study selection process.

Diagram 1: Systematic Review & Meta-Analysis Workflow

Phase 3: Data Collection & Critical Appraisal

• Data Extraction: Use the pre-designed form. Extract numeric data needed for effect size calculation (means, SDs, sample sizes) and descriptive data. Contact study authors for missing data if necessary.
• Assess Risk of Bias (RoB): Have at least two reviewers independently apply the chosen RoB tool to each study. This evaluates internal validity (e.g., random allocation, blinding in effect assessment, completeness of outcome data).

Phase 4: Synthesis & Reporting

• Qualitative Synthesis: Systematically describe and summarize the findings from the included studies, often grouped by key characteristics (e.g., species, exposure pathway).
• Meta-Analysis (if appropriate):
  • Calculate Effect Sizes: Transform extracted data into a common, comparable effect size (e.g., log-transformed ratio of treated to control mean for continuous outcomes).
  • Choose a Statistical Model: Assess statistical heterogeneity (e.g., using I² statistic). A random-effects model is often appropriate for ecotoxicity data due to expected methodological and biological variation [8] [9].
  • Conduct Analysis & Assess Heterogeneity: Perform the meta-analysis to generate a pooled effect estimate with a confidence interval. Investigate sources of high heterogeneity via subgroup analysis (e.g., by test type, species class) or meta-regression.
• Report Findings: Follow reporting guidelines (PRISMA). Present results with tables, forest plots, and a clear discussion of the strength of evidence, limitations, and implications for research and regulation.

Meta-Analysis: Statistical Foundations and Application

Meta-analysis translates the qualitative question "What does the body of evidence show?" into a quantitative answer. Its validity hinges on the quality of the preceding systematic review steps [9].

Core Statistical Models and Selection

Table 2: Key Statistical Models in Meta-Analysis [8] [9].

Model	Underlying Assumption	Weighting of Studies	Interpretation of Result	When to Use in Ecotoxicity
Fixed-Effect	All studies estimate one single, true effect size. Differences are due to sampling error alone.	Weight is inverse of study's within-study variance (1/SE²). Larger studies get more weight.	The pooled estimate is the best estimate of that single common effect.	Rarely appropriate. Only if studies are near-identical (e.g., same species, same lab protocol).
Random-Effects	The true effect size varies across studies (heterogeneity). Studies estimate different, related effects.	Weight is inverse of (within-study variance + between-study variance (τ²)). Weight is distributed more equally.	The pooled estimate is the mean of the distribution of true effects.	Almost always appropriate. Accounts for real variation due to species sensitivity, test conditions, etc.

Diagram 2: Meta-Analysis Statistical Model Selection Process

Investigating Heterogeneity and Sensitivity Analysis

In ecotoxicity, heterogeneity is the rule, not the exception. Key sources include:

• Biological: Differences in species, life stage, genetic strain.
• Methodological: Differences in test duration, exposure medium, endpoint measurement.
• Statistical: Differences in study quality and design.

Subgroup analysis and meta-regression are used to explore if these factors explain heterogeneity. For example, a meta-regression might test if log-transformed toxicity values are predicted by the octanol-water partition coefficient (log Kow) of the test chemicals. Sensitivity analyses test the robustness of results, for instance, by repeating the meta-analysis excluding studies with a high risk of bias or those published only as abstracts.

Evidence Integration: From Synthesis to Application

Evidence integration is the process of critically evaluating and incorporating synthesized evidence into a broader context, such as a research paper, a risk assessment report, or a regulatory guideline [10]. It moves beyond simply reporting a pooled effect size to interpreting its meaning, credibility, and relevance.

Principles of Effective Evidence Integration

• Credibility Assessment: Evaluate the reliability of the synthesized evidence itself. Consider the rigor of the systematic review (Was the protocol followed? Was the search comprehensive?), the consistency of findings across studies, and the risk of bias in the included studies [10].
• Contextualization: Relate the findings to the specific problem at hand. Does the pooled EC50 for a chemical from laboratory species predict effects in a specific local ecosystem? Discuss biological plausibility and relevance.
• Transparency in Uncertainty: Clearly communicate the limitations of the evidence. This includes statistical uncertainty (confidence intervals), methodological limitations (risk of bias, publication bias), and applicability concerns (e.g., laboratory-to-field extrapolation).
• Balanced Interpretation: Avoid "cherry-picking" only evidence that supports a pre-determined conclusion. Acknowledge and interpret conflicting evidence or indeterminate results [10].

Types of Evidence for Integration

Table 3: Hierarchy and Integration of Evidence Types in Ecotoxicological Risk Assessment [10].

Evidence Type	Description	Role in Integration	Example in Ecotoxicity
Primary Research Data	Original data from individual experiments or studies (e.g., a toxicity test).	The raw material for systematic review and meta-analysis.	A published paper reporting the LC50 of Chemical Y to fathead minnow.
Synthesized Quantitative Evidence	The output of meta-analysis (pooled effect estimates, confidence intervals).	Provides the most precise and statistically powerful summary of the available primary data.	A pooled no-observed-effect concentration (NOEC) for invertebrates exposed to Chemical Z.
Synthesized Qualitative Evidence	Systematic descriptive summaries of study findings, often where meta-analysis is not feasible.	Identifies patterns, knowledge gaps, and contextual factors not captured by numbers.	A systematic review categorizing the types of sub-lethal behavioral effects observed across fish studies.
Expert Judgment/Testimony	Informed opinions from recognized authorities in the field.	Provides interpretation, identifies plausible mechanisms, and helps bridge gaps where direct evidence is lacking. Must be used transparently.	A panel interpreting mechanistic toxicology data to assess the relevance of a tumor finding in rodents for aquatic species.

Table 4: Essential Toolkit for Conducting Systematic Reviews & Meta-Analyses in Ecotoxicity.

Tool / Resource Category	Specific Examples	Primary Function
Protocol & Reporting Guidelines	PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses); ROSES (Reporting standards for Systematic Evidence Syntheses in environmental research).	Ensure completeness, transparency, and quality of reporting.
Review Management Software	DistillerSR, Rayyan, Covidence.	Streamline and manage the collaborative process of screening references, data extraction, and quality appraisal.
Specialized Databases	PubMed, TOXLINE, Web of Science, Scopus, ECOTOX (US EPA).	Enable comprehensive, structured literature searches for primary ecotoxicity studies.
Statistical Software for Meta-Analysis	R (packages: metafor, meta), RevMan (Cochrane), Comprehensive Meta-Analysis.	Perform complex meta-analyses, create forest and funnel plots, conduct subgroup and meta-regression analyses.
Risk of Bias / Quality Assessment Tools	SYRCLE's RoB tool (for animal studies), NIH Quality Assessment Tool, custom tools based on OECD test guideline criteria.	Systematically evaluate the internal validity and reliability of included primary studies.
Gray Literature Sources	Government and regulatory agency reports (e.g., US EPA, ECHA), university theses, conference proceedings.	Reduce publication bias by including studies not published in commercial journals.

In the data-rich and high-stakes field of ecotoxicity research, the disciplined application of systematic review and meta-analysis methodologies is paramount. Moving from traditional, selective narrative summaries to transparent, protocol-driven syntheses represents a fundamental shift toward greater scientific rigor and reliability. As shown, a systematic review provides the essential structural framework to minimize bias, while meta-analysis offers powerful statistical tools to quantify overall effects and explore variability. Finally, the principled integration of this synthesized evidence into the broader scientific and regulatory context ensures that conclusions are not only statistically sound but also relevant and actionable.

For researchers and drug development professionals, mastery of these concepts enables the creation of defensible, high-quality environmental safety assessments. It empowers them to build a compelling evidence base that can effectively inform substance regulation, guide the development of safer chemicals, and ultimately contribute to more robust protection of environmental health.

The robust assessment of chemical hazards to ecosystems hinges on the transparent, reproducible, and statistically sound analysis of ecotoxicity data. In recent years, the scientific community has increasingly adopted systematic review (SR) methodologies to minimize bias, ensure comprehensive evidence synthesis, and support regulatory decision-making[reference:0]. This whitepaper frames the ongoing revision of OECD Document No. 54 and concurrent global initiatives within this broader thesis: the modernization of statistical and methodological frameworks is a regulatory imperative for advancing ecotoxicity data research. These coordinated efforts aim to replace outdated practices with contemporary, evidence-based approaches, ultimately enhancing the reliability and global harmonization of environmental risk assessments.

The Pivotal Revision: OECD Document No. 54

OECD Document No. 54, “Current approaches in the statistical analysis of ecotoxicity data: a guidance to application” (2006), has been the cornerstone for statistical analysis in regulatory ecotoxicology. However, it is no longer considered reflective of modern statistical methods or computational platforms[reference:1].

Key Drivers and Timeline

A consensus has emerged that the document requires a substantial overhaul. The revision is formally planned for 2026, with the German Environment Agency (UBA) coordinating scientific contributions[reference:2]. The process was launched with the “1st UBA expert workshop on the OECD No. 54 revision” in September 2024, which focused on updating chapters concerning hypothesis testing, dose-response models, and biological effect models[reference:3].

Core Statistical Upgrades

The revision addresses critical limitations in the current guidance:

• Moving Beyond the Hypothesis Testing vs. Dose-Response Dichotomy: The existing distinction between ANOVA-type "hypothesis testing" and regression-based "dose-response modeling" is artificial, as both are variants of linear models[reference:4]. The revision advocates for continuous regression-based models as the default.
• Incorporating Modern Statistical Tools: The update will integrate contemporary methods such as Generalized Linear Models (GLMs), mixed-effect models, Generalized Additive Models (GAMs), and Bayesian frameworks into the ecotoxicologist's toolbox[reference:5][reference:6].
• Embracing Open-Source Software: The guidance will promote the use of open-source platforms like R, which have made advanced statistical methodologies widely accessible[reference:7].
• Evaluating Modern Metrics: The revision will consider alternative effect metrics like the Benchmark Dose (BMD) and the No-Significant Effect Concentration (NSEC), alongside traditional ECx values, evaluating their properties for use in higher-tier assessments[reference:8].

Concurrent Global Regulatory Initiatives

The push for modernization extends beyond the OECD, manifesting in parallel initiatives across major regulatory bodies.

European Food Safety Authority (EFSA)

In June 2024, EFSA received a mandate to review its Guidance Document on Terrestrial Ecotoxicology and to develop guidance for assessing indirect effects on biodiversity[reference:9]. This revision aims to strengthen the methodological foundation for pesticide and chemical risk assessment in the EU.

U.S. Environmental Protection Agency (EPA)

The EPA is advancing the systematic review of ecotoxicity data through two key pillars:

• The ECOTOX Knowledgebase: This is the world's largest curated database of single-chemical toxicity data for aquatic and terrestrial species, providing foundational data for SR processes[reference:10]. As of 2023, it contained data from over 54,000 references, comprising 1.1 million test records on nearly 14,000 species and 13,000 chemicals[reference:11].
• Standard Operating Procedures (SOPs): The EPA has developed formal SOPs for the Systematic Review of Ecological Toxicity Data to support the development of Ambient Water Quality Criteria, ensuring consistency and transparency in data evaluation[reference:12].

World Health Organization (WHO)

In February 2024, WHO launched a Repository of Systematic Reviews on interventions in environment, climate change, and health[reference:13]. While broader in scope, this repository underscores the global institutional shift towards evidence-based assessment and directly intersects with ecotoxicity through its coverage of chemicals and environmental health.

International Standardization (ISO/OECD)

A vast battery of standardized microbial toxicity tests (e.g., ISO 8192, OECD 209) are globally accepted for assessing chemical impacts on microbial communities, particularly in wastewater treatment and biodegradation testing[reference:14]. These standardized methods provide the reliable, reproducible experimental data that feed into higher-level systematic reviews and risk assessments.

Initiative	Leading Organization	Primary Focus	Current Status/Timeline
Statistical Analysis Guidance	OECD	Modernizing statistical methods for ecotoxicity data analysis.	Revision ongoing; planned publication in 2026[reference:15].
Terrestrial Ecotoxicology Guidance	EFSA	Updating risk assessment guidance for terrestrial organisms and indirect effects.	Revision mandated in June 2024; outline published[reference:16].
Systematic Review SOPs	U.S. EPA	Standardizing procedures for reviewing ecological toxicity data.	SOPs published in September 2024[reference:17].
Evidence Repository	WHO	Cataloging systematic reviews on environmental health interventions.	Repository launched in February 2024[reference:18].
Microbial Toxicity Tests	ISO/OECD	Standardizing tests for bacterial and microbial toxicity.	Multiple guidelines actively maintained and applied[reference:19].

Table 2: Scale of the U.S. EPA ECOTOX Knowledgebase (2023)

Data Category	Volume	Description
Scientific References	>54,000	Curated sources from the peer-reviewed literature.
Test Records	>1.1 million	Individual toxicity test results.
Species Covered	~14,000	Aquatic and terrestrial test organisms.
Chemicals Covered	~13,000	Unique chemical substances.

Detailed Experimental Protocols

Protocol for a Standard Aquatic Ecotoxicity Test (Algal Growth Inhibition)

This protocol is based on the principles of OECD Test Guideline 201.

1. Test Organism and Pre-culture:

• Organism: Use a validated freshwater algal species (e.g., Pseudokirchneriella subcapitata).
• Culture Conditions: Maintain stock cultures in sterile OECD TG 201 standard growth medium (e.g., AAP medium) under continuous cool-white fluorescent light (60-120 µE/m²/s) at 21±2°C with shaking.

2. Test Chemical Preparation:

• Prepare a stock solution of the test chemical in sterile medium, using a solvent if necessary (e.g., acetone, DMSO ≤ 0.01% v/v). Prepare a series of at least 5 geometrically spaced concentrations (e.g., dilution factor of 1.8-2.2).

3. Test Initiation and Design:

• Inoculate fresh medium with algae from an exponentially growing pre-culture to an initial cell density of ~10⁴ cells/mL.
• Dispense 50-100 mL of inoculated medium into sterile Erlenmeyer flasks. Add the appropriate volume of test solution or solvent control.
• Include at least three replicates per concentration and controls (solvent and negative).

4. Incubation and Monitoring:

• Incubate flasks under the same conditions as the pre-culture for 72 hours.
• Measure algal growth (cell density or fluorescence) at 0, 24, 48, and 72 hours.

5. Data Analysis and Endpoint Calculation:

• Calculate the average specific growth rate for each replicate.
• Fit a dose-response model (e.g., a 4-parameter log-logistic model) to the mean growth rate inhibition data.
• Calculate the EC₅₀ (concentration causing 50% growth inhibition) or EC₁₀/EC₂₀ with confidence intervals.

Protocol for Systematic Review of Ecotoxicity Data

This protocol aligns with the U.S. EPA's SOPs for systematic review[reference:20].

1. Problem Formulation & Protocol Registration:

• Define a clear, structured review question (PECO: Population, Exposure, Comparator, Outcome).
• Develop and register a detailed review protocol outlining search strategy, inclusion/exclusion criteria, and analysis plan.

2. Comprehensive Literature Search:

• Search multiple electronic databases (e.g., Web of Science, PubMed, Scopus).
• Use chemical names, CAS numbers, and relevant ecotoxicity keywords.
• Supplement with searches of regulatory dossiers, grey literature, and backward/forward citation chasing.

3. Study Screening & Selection:

• Screen titles/abstracts and then full texts against pre-defined eligibility criteria.
• Perform screening in duplicate, with conflicts resolved by a third reviewer.
• Document the flow of studies (PRISMA diagram).

4. Data Extraction & Quality Assessment:

• Extract relevant data (test organism, exposure conditions, endpoints, results) into a standardized form.
• Assess the reliability/risk of bias of each study using a validated tool (e.g., adapted from Klimisch scores or the EPA's Data Evaluation Record system).

5. Evidence Synthesis & Integration:

• For quantitative synthesis (meta-analysis), calculate summary effect estimates (e.g., pooled EC₅₀) where appropriate.
• For narrative synthesis, organize findings by taxon, endpoint, or exposure scenario.
• Assess the overall strength and confidence in the body of evidence.

Visualizing Workflows and Relationships

Diagram 1: Systematic Review Workflow for Ecotoxicity Data

Diagram 2: Interplay of Global Regulatory Initiatives

The Scientist's Toolkit: Essential Research Reagents & Materials

Table 3: Key Reagent Solutions for Standard Ecotoxicity Testing

Item	Function & Explanation	Example/Standard
Standard Growth Medium	Provides essential nutrients for test organisms in a reproducible, defined matrix. Critical for controlling test conditions.	OECD TG 201 AAP medium for algae; OECD TG 202 Daphnia medium.
Reference Toxicant	A chemical with known, stable toxicity used to validate the health and sensitivity of test organisms across different test runs.	Potassium dichromate (for Daphnia), Copper sulfate (for algae).
Solvent Control	A vehicle (e.g., acetone, DMSO) used to dissolve poorly water-soluble test chemicals. Ensures any observed effects are due to the chemical, not the solvent.	Concentration typically ≤ 0.01% v/v to avoid solvent toxicity.
Culture Organisms	Standardized, genetically defined strains of test species that ensure reproducibility and inter-laboratory comparison.	Pseudokirchneriella subcapitata (algae), Daphnia magna (cladocera).
Biomass Indicator	A dye or assay used to quantify organism growth or viability as the test endpoint.	Fluorescence probes (e.g., chlorophyll a for algae), cell counting.
Quality Control Buffers & Standards	Used to calibrate and verify the performance of analytical equipment (pH meters, dissolved oxygen probes, spectrophotometers).	pH buffer solutions (4.0, 7.0, 10.0), DO calibration standards.
Data Analysis Software	Open-source statistical platforms essential for applying modern analysis methods recommended in revised guidelines.	R with packages (e.g., drc for dose-response, metafor for SR).

The concurrent revision of OECD No. 54 and the advancement of complementary initiatives by the EFSA, U.S. EPA, WHO, and ISO represent a cohesive global regulatory imperative. This movement is fundamentally rooted in the principles of systematic review, aiming to replace fragmented and outdated practices with transparent, statistically robust, and evidence-based methodologies. For researchers and drug development professionals, engagement with these evolving frameworks is not merely compliance but an opportunity to contribute to and leverage higher-quality, globally harmonized ecotoxicity assessments that better protect environmental and public health.

Conducting Rigorous Systematic Reviews: A Step-by-Step Guide from Protocol to Synthesis

In the evolving landscape of ecotoxicology, the shift towards evidence-based environmental decision-making necessitates robust, transparent, and reproducible systematic review methodologies [3]. The foundation of any high-quality systematic review is a precisely formulated research question, which dictates the entire process—from study identification and selection to evidence synthesis and risk characterization [11]. Within the context of systematic review methods for ecotoxicity data research, the Population, Exposure, Comparator, Outcome (PECO) and Population, Stressor, Outcome (PSO) frameworks provide this critical scaffolding [11].

These frameworks are not merely academic exercises; they address a tangible challenge in environmental science. The proliferation of contaminants, including contaminants of emerging concern (CECs), has created a data-rich but often fragmented landscape [12]. Without a clearly defined question, systematic reviews risk being inefficient, non-transparent, or misaligned with regulatory and research needs [11]. The PECO framework, an adaptation of the clinical PICO (Population, Intervention, Comparator, Outcome) model, is increasingly accepted for structuring questions about the association between environmental exposures and ecological outcomes [11]. It is specifically designed to address the complexities of unintentional exposure scenarios, which are fundamental to environmental and occupational health [11]. Concurrently, frameworks like PSO serve in contexts where defining a precise comparator is less critical than characterizing the stressor-effect relationship itself. This technical guide details the operationalization of these frameworks, their integration into systematic review workflows, and their application in advanced ecotoxicological research, positioning them as the cornerstone of a rigorous, defensible, and decision-relevant evidence assessment process.

Core Frameworks: PECO and PSO in Detail

The PECO Framework: Operationalizing Exposure-Based Questions

The PECO framework deconstructs a research question into four interdependent pillars. A well-constructed PECO question defines the review's objectives and directly informs study design, inclusion/exclusion criteria, and the interpretation of findings [11].

• Population (P): The ecological entity of interest. This must be specified with taxonomic, life-stage, and contextual precision (e.g., "adult fathead minnows (Pimephales promelas) in freshwater lotic systems," "the earthworm Eisenia fetida in agricultural soils") [12] [13]. The population defines the biological domain of the review.
• Exposure (E): The chemical, physical, or biological agent under investigation, with detailed characterization of its form, magnitude, duration, and route. This is a critical differentiator from clinical PICO, as environmental exposures are often passive and complex [11]. Examples include "waterborne exposure to ≥ 80 dB anthropogenic noise" or "dietary exposure to a mixture of acetamiprid and chlorpyrifos at a 1:1 concentration ratio" [11] [13].
• Comparator (C): The reference against which the exposure is evaluated. Defining an appropriate comparator is a noted challenge in environmental health [11]. It may be a different exposure level (e.g., "exposure to < 80 dB"), a background condition, or a different substance. The comparator is essential for estimating effect size and understanding dose-response relationships.
• Outcome (O): The measurable ecological or toxicological endpoint. Outcomes should be biologically relevant, measurable, and tiered where possible (e.g., from molecular initiating events like "CYP1A1 gene expression" to apical endpoints like "96-hour mortality" or "reproductive success") [12].

Research and regulatory contexts dictate how these elements are combined. A seminal framework outlines five paradigmatic PECO scenarios, moving from exploratory association to intervention-based questions [11].

Table 1: PECO Framework Scenarios for Systematic Review [11]

Scenario & Context	Approach	Example PECO Question
1. Explore dose-effect relationship	Explore the shape of the exposure-outcome relationship.	Among freshwater amphipods, what is the effect of a 10 μg/L incremental increase in waterborne triclosan on 48-hour mortality?
2. Evaluate data-driven exposure cut-offs	Use cut-offs (e.g., tertiles) identified from the distribution within retrieved studies.	Among soil nematodes, what is the effect of exposure to the highest versus lowest quartile of cadmium concentration in soil on reproductive rate?
3. Evaluate externally defined cut-offs	Use cut-offs (e.g., regulatory limits) known from other populations or standards.	Among coastal marine fish, what is the effect of sediment PAH concentrations exceeding EPA sediment quality guidelines compared to those below guidelines on liver somatic index?
4. Identify a protective exposure cut-off	Use an existing health-based exposure limit as the comparator.	Among honeybee colonies, what is the effect of exposure to neonicotinoid levels below the LD₅₀ compared to levels at or above the LD₅₀ on colony collapse?
5. Evaluate an intervention	Select a comparator based on an achievable exposure reduction.	In an urban watershed, what is the effect of implementing a constructed wetland (reducing effluent pyrene by 50%) compared to no intervention on the genetic diversity of benthic macroinvertebrate populations?

The PSO and Related Frameworks

The Population, Stressor, Outcome (PSO) framework is a streamlined variant used when a formal comparator is either implicit or not the focus of the review. It is particularly useful for hazard identification and screening-level assessments. For instance, a question may be: "In juvenile zebrafish (Danio rerio), what is the effect of microplastic exposure (1-10 μm) on larval deformity rate?" The comparator here is implicit (no exposure). PSO is often applied in prioritization frameworks that screen chemicals based on their intrinsic Persistence (P), Bioaccumulation (B), and Toxicity (T) potential [14].

Complementary frameworks like qAOP (Quantitative Adverse Outcome Pathway) provide a mechanistic structure for organizing evidence [15]. While not a direct substitute for PECO/PSO, a qAOP framework quantifies the relationships along a pathway from a Molecular Initiating Event (MIE) to an Adverse Outcome (AO), directly informing the 'O' in PECO and enabling prediction of outcomes for novel exposure scenarios [15].

Table 2: Comparison of Question Framing Frameworks in Ecotoxicology

Framework	Primary Purpose	Key Components	Typical Application Context
PECO	To structure questions about the association or effect of an exposure relative to a comparator.	Population, Exposure, Comparator, Outcome	Systematic reviews for risk assessment, dose-response analysis, intervention evaluation [11].
PSO	To structure questions for hazard identification and stressor-effect characterization.	Population, Stressor, Outcome	Screening-level reviews, hazard ranking, PBT assessments [14].
qAOP	To organize mechanistic, quantitative evidence across biological scales for predictive toxicology.	Molecular Initiating Event, Key Events, Adverse Outcome, Quantitative Relationships	Developing predictive models, integrating NAMs (e.g., in vitro, in silico), supporting read-across [15].

Integrating PECO/PSO into Systematic Review Workflows

Framing the question is the inaugural and defining step in a systematic review. Its integration into the subsequent workflow is critical for maintaining focus and rigor. The following diagram illustrates this integration, highlighting decision points informed by the PECO/PSO structure.

Diagram 1: Systematic Review Workflow with Integrated PECO/PSO Framing

The PECO question directly shapes the review protocol. Inclusion and exclusion criteria are explicit translations of the PECO elements. For example, a PECO specifying "chronic toxicity in freshwater fish" excludes acute toxicity studies and studies on terrestrial species. The search strategy is built using controlled vocabularies (e.g., MeSH, Emtree) and keywords derived from each PECO component [3].

During study screening and data extraction, PECO acts as a consistent benchmark. Tools like the ECOTOXicology Knowledgebase (ECOTOX), which houses over one million curated test results, exemplify the application of systematic, PECO-informed review principles for data curation [3] [16]. Furthermore, the rise of New Approach Methodologies (NAMs)—including high-throughput in vitro assays, toxicogenomics, and in silico models—generates data that must be integrated within this structured framework [12] [3]. A PECO-defined question ensures that NAM data relevant to specific key events or outcomes are appropriately incorporated into a Weight-of-Evidence (WoE) assessment [12].

Advanced Applications and Quantitative Methodologies

Quantitative Analysis Informing PECO Comparators

Quantitative toxicological methods provide the data needed to define meaningful 'E' and 'C' elements, especially in Scenarios 2-5 of the PECO framework [11]. Two key methodologies are:

• Benchmark Dose (BMD) Modeling: This approach is superior to traditional NOAEC/LOEC methods as it models the full dose-response curve to estimate a predetermined level of change (e.g., BMD₁₀, the dose associated with a 10% effect) [13] [2]. BMD is invaluable for quantifying the toxicity of chemical mixtures. A study on pesticide mixtures (acetamiprid, carbendazim, chlorpyrifos, cyhalothrin) found that mixture toxicity to earthworms could be up to 40 times higher than that of individual components, a finding critical for defining relevant exposure levels in a PECO question about mixture effects [13].
• Species Sensitivity Distributions (SSDs): SSDs model the variation in sensitivity of multiple species to a single stressor. They are used to derive Predicted No-Effect Concentrations (PNECs), which can serve as scientifically robust comparators (the 'C') in PECO questions focused on protective thresholds for ecosystems [3].

Statistical Best Practices for Ecotoxicity Data Analysis

The statistical analysis of data underlying PECO-based reviews is undergoing significant modernization. There is a strong movement away from hypothesis-testing methods based on NOEC/LOEC towards continuous regression-based modeling [2]. Contemporary best practices recommend:

• Using generalized linear models (GLMs) and non-linear dose-response models (e.g., 2-5 parameter models) to estimate ECₓ or BMD values [2].
• Employing generalized additive models (GAMs) to explore non-linear response patterns without pre-specified shapes [2].
• Considering Bayesian methods for incorporating prior knowledge and quantifying uncertainty [2].

These advanced statistical practices yield more reliable and informative toxicity estimates, which are the fundamental data points for answering PECO questions and conducting robust evidence synthesis [2].

The qAOP Modeling Workflow: From Mechanistic Data to Quantitative Predictions

The development of a Quantitative Adverse Outcome Pathway (qAOP) is a paradigm for integrating diverse data streams to support predictive toxicology. It directly operationalizes mechanistic evidence within a quantifiable framework [15]. The workflow, depicted below, shows how data curated under a PECO/PSO structure can feed into predictive models.

Diagram 2: Quantitative AOP (qAOP) Development Workflow

This process requires identifying and extracting reliable quantitative data for key events, which is facilitated by a clear initial question framing the relationship of interest [15]. Resources like the ECOTOX Knowledgebase are essential for sourcing curated in vivo data for model calibration and validation [3].

Detailed Experimental Protocols

Protocol 1: Aquatic Ecotoxicity Screening for Polymers (Adapted from OECD TG) [17] This protocol assesses the intrinsic toxicity of polymer filtrates and suspensions.

• Test Organisms: Algae (Raphidocelis subcapitata), daphnids (Daphnia magna), and zebrafish embryos (Danio rerio).
• Test Substance Preparation: Polymers (e.g., alginate, chitosan) are prepared as both filtrates (0.45 μm filtered to assess intrinsic toxicity) and suspensions (to assess physical effects) in reconstituted standard test waters.
• Exposure Design: A concentration range of 1–100 mg/L is tested. For algae, a 72-hour growth inhibition test (OECD TG 201) is conducted. For daphnids, a 48-hour immobility test (OECD TG 202, miniaturized) is used. For zebrafish embryos, a 96-hour exposure (adapted from OECD TG 236) is performed, followed by transcriptomic analysis.
• Endpoint Measurement: Algal growth (cell density/biomass), daphnid immobility, zebrafish embryo mortality/morphology, and differential gene expression via RNA sequencing.
• Data Analysis: Determine LOEC/NOEC. For transcriptomics, identify significantly differentially expressed genes (DEGs) and perform pathway enrichment analysis to infer mode of action [17].

Protocol 2: Benchmark Dose Analysis for Pesticide Mixtures in Earthworms [13] This protocol quantifies mixture toxicity using the BMD approach.

• Test Organism: Adult earthworms (Eisenia fetida) with developed clitellum, cultured in artificial soil.
• Test Chemicals: Individual pesticides (acetamiprid, carbendazim, chlorpyrifos, cyhalothrin) and their mixtures at environmentally relevant ratios.
• Exposure Design: Two tests: (i) Acute toxicity test: Exposure to a range of concentrations in artificial soil for 14 days; endpoint is mortality. (ii) Avoidance response test: Dual-chamber test with contaminated vs. control soil for 48 hours; endpoint is the effective avoidance rate.
• Data Analysis: Fit dose-response models to mortality and avoidance data for each chemical and mixture. Calculate the Benchmark Dose (BMD) for a specified Benchmark Response (BMR), e.g., 10% extra risk (BMD₁₀). Compute the mixture toxicity index by comparing the BMD of the mixture to the BMDs of individual components [13].

Research Reagent Solutions & Essential Materials

Table 3: Key Research Reagents and Materials for Featured Ecotoxicity Studies

Item	Function/Description	Example Study/Use Case
Reconstituted Standard Test Water	Provides a consistent, defined ionic composition and hardness for aquatic toxicity tests, minimizing confounding water quality variables.	Used in OECD TG tests for algae, daphnids, and fish [17].
Artificial Soil	A standardized matrix of peat, kaolin clay, and sand for terrestrial toxicity tests, ensuring reproducibility.	Used in earthworm acute and avoidance tests per OECD guidelines [13].
Polymer Test Substances (e.g., Chitosan, Alginate)	Biopolymers of interest as "green" substitutes; tested as both filtrates (intrinsic toxicity) and suspensions (physical effects).	Central test materials in aquatic ecotoxicity screening of biopolymers [17].
Pesticide Active Ingredients (e.g., Chlorpyrifos, Carbendazim)	High-purity analytical standards used to spike soils or waters at precise concentrations for dose-response testing.	Used to generate dose-response data for BMD modeling of single and mixed pesticides [13].
RNA Sequencing Kits	For library preparation and next-generation sequencing of transcriptomes from exposed organisms (e.g., zebrafish embryos).	Enables transcriptomic analysis to identify differentially expressed genes and infer toxic modes-of-action [17].
T47D-kBluc or Attagene Factorial Assay Kits	In vitro cell-based bioreporter assays used in effects-based monitoring. They detect activation of specific biological pathways (e.g., estrogen receptor, stress response).	Used in WoE prioritization frameworks to screen environmental water extracts for biological activity [12].

In the field of ecotoxicology, researchers and regulators are confronted with a complex and ever-expanding body of data on the effects of chemicals, pharmaceuticals, and novel materials like nanoplastics on ecosystems[reference:0]. Systematic reviews are essential for distilling this evidence into reliable conclusions that can inform risk assessment and policy. However, the credibility of any systematic review hinges on the transparency and a priori planning of its methods, which mitigates bias and ensures reproducibility. This is where the development and registration of a detailed protocol becomes paramount. Framed within a broader thesis on advancing systematic review methods for ecotoxicity data, this technical guide elucidates the process of creating and registering a robust protocol using the PRISMA 2020 framework as its cornerstone.

The PRISMA 2020 Framework and Protocol Registration

The Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) 2020 statement is the contemporary guideline for reporting systematic reviews, designed to facilitate complete and transparent reporting[reference:1]. It comprises a 27-item checklist. While PRISMA 2020 guides the reporting of a completed review, its Item 24 explicitly addresses the protocol: "Provide the register name and registration number" or "Indicate where the review protocol can be accessed"[reference:2]. This requirement underscores that a publicly available protocol is not an optional extra but an "essential element" of a trustworthy systematic review[reference:3].

A protocol acts as the project's roadmap, detailing the rationale, question (often structured as PICO), inclusion/exclusion criteria, search strategy, data management, risk-of-bias assessment, and synthesis plans before the review begins[reference:4]. Registering this protocol on a public platform locks in these plans, reducing opportunities for selective reporting and outcome switching, thereby enhancing the review's validity.

Developing a Detailed Protocol: A PRISMA-P Guided Methodology

The dedicated guideline for protocol development is the PRISMA extension for Protocols (PRISMA-P) 2015 statement[reference:5]. The following steps outline a robust methodology for ecotoxicity reviews:

• Problem Formulation & Rationale: Clearly define the ecotoxicological problem, the populations (e.g., specific species, ecosystems), interventions/exposures (e.g., chemical contaminants), comparators, and outcomes (e.g., mortality, sublethal biomarkers like oxidative stress)[reference:6]. Justify the review in the context of existing knowledge and regulatory needs.

• Eligibility Criteria: Pre-specify detailed inclusion and exclusion criteria. For ecotoxicity, this includes study designs (in vivo, in vitro, field studies), organism models (e.g., Danio rerio, earthworms), exposure durations, and measured endpoints.

• Information Sources & Search Strategy: Identify all databases to be searched (e.g., Web of Science, Scopus, PubMed, specialized databases like ECOTOX). Draft a full, reproducible search strategy for each, including keywords, controlled vocabulary (e.g., MeSH terms), and filters[reference:7]. Specify dates of search and plans for identifying grey literature.

• Study Selection Process: Describe the process for deduplication and a two-stage screening (title/abstract, then full-text) using tools like Rayyan or Covidence. Specify that screening will be performed independently by at least two reviewers, with conflicts resolved by consensus or a third reviewer[reference:8].

• Data Extraction & Management: Design a standardized data extraction form to capture study characteristics, exposure details, outcome data, and key conclusions. Define the process for independent extraction and management of data.

• Risk of Bias Assessment: Select and justify appropriate tools for assessing the methodological quality or risk of bias of individual ecotoxicity studies (e.g., SYRCLE's risk of bias tool for animal studies, adapted tools for in vitro research).

• Data Synthesis Plan: Outline the approach for synthesis. If a meta-analysis is plausible, describe the effect measures, heterogeneity assessment (I² statistic), and model (fixed/random effects). For narrative synthesis, describe the planned grouping and comparison of studies.

• Protocol Amendments: Establish a process for documenting and reporting any deviations from the registered protocol, with justifications.

Registering the Protocol: Platforms and Procedures

Registration makes the protocol publicly accessible and discoverable. The two primary platforms are:

• PROSPERO: An international, free, open-access register for systematic review protocols, maintained by the University of York. It requires completion of a minimum dataset of key methodological details[reference:9]. Registration is prospective (before data extraction begins) and provides a unique registration number (e.g., CRD420XXXXXX).
• Open Science Framework (OSF) Registries: A flexible, open-source platform that allows for the registration and archiving of systematic review protocols, including the upload of full protocol documents and associated materials[reference:10].

Table 1: Comparison of Protocol Registration Platforms

Feature	PROSPERO	OSF Registries
Primary Purpose	Dedicated registry for systematic review protocols	General registry for any research study, including protocols
Cost	Free	Free
Required Fields	Pre-defined minimum dataset for systematic reviews	Flexible; user-defined structure
Peer Review	Administrative check only, no methodological review[reference:11]	No editorial review before acceptance[reference:12]
Best For	Standardized registration for health & ecotoxicity reviews requiring a recognized registration number.	Reviews requiring flexibility, or when planning to share full protocol documents and data.

Quantitative Insights: The State of Protocol Reporting

The importance of detailed protocol registration is highlighted by empirical research. A 2020 study assessed a random sample of 439 PROSPERO records against the 19 PRISMA-P checklist items[reference:13]. The results, summarized below, reveal a significant gap between recommended and reported information, underscoring the need for stricter adherence to PRISMA-P during registration.

Table 2: Summary of PROSPERO Record Compliance with PRISMA-P Items (2020 Assessment)[reference:14]

Metric	Result
Mean score per record (out of 19 PRISMA-P items)	4.8 (SD 1.8)
Median score	4
Range of scores	2 to 11
Overall percentage of PRISMA-P items reported	25% (2081/8227 items)
Conclusion	Key methodological details are often missing from PROSPERO registrations, limiting their utility for transparency and bias reduction.

Table 3: Key Digital Tools and Resources for Protocol Development

Tool/Resource	Function in Protocol Development
PRISMA-P 2015 Checklist	Guideline specifying essential items to include in a systematic review protocol.
Rayyan / Covidence	Web-based tools for collaborative management of the study screening and selection process.
Reference Management Software (e.g., EndNote, Zotero, Mendeley)	For deduplication of search results and organizing included studies[reference:15].
Database Search Interfaces (e.g., PubMed, Web of Science, Scopus)	Primary sources for comprehensive literature searching.
PROSPERO Registry	The primary platform for prospectively registering the protocol and obtaining a unique ID.
Open Science Framework (OSF)	A platform for registering the protocol, sharing the full document, and hosting supplementary materials.
VOSviewer / Bibliometric Software	For planning and executing bibliometric analyses if part of the review scope[reference:16].

Visualizing the Workflow: From Protocol to Registered Review

The following diagram maps the logical workflow for developing and registering a systematic review protocol within the PRISMA 2020 framework, highlighting key decision points and outputs.

Diagram 1: Protocol Development and Registration Workflow (Max width: 760px)

For researchers synthesizing ecotoxicity evidence, a meticulously developed and publicly registered protocol is the foundation of a rigorous, transparent, and credible systematic review. The PRISMA 2020 framework mandates this practice, and guidelines like PRISMA-P provide the necessary structure. While current registration practices, as evidenced by PROSPERO records, often fall short of ideal reporting standards, adherence to these guidelines is a critical step toward improving the reliability of ecological risk assessments. By following the detailed methodologies, utilizing the essential toolkit, and navigating the registration pathways outlined in this guide, scientists can ensure their work meets the highest standards of evidence-based environmental science.

Comprehensive Search Strategies for Academic and Grey Literature in Toxicology

The foundation of robust ecotoxicity research and chemical risk assessment is a comprehensive, transparent, and reproducible literature search. Systematic review methodology, adapted from clinical medicine, provides the necessary framework to minimize bias and ensure all relevant evidence is considered[reference:0]. This guide details advanced strategies for searching both academic (peer-reviewed) and grey literature within toxicology, framed within the context of a broader thesis on systematic review methods for ecotoxicity data. The integration of grey literature—encompassing government reports, regulatory documents, theses, and conference proceedings—is critical to counteract publication bias and provide a complete evidence base for environmental decision-making[reference:1].

Core Methodology for Comprehensive Searching

A systematic search strategy is built on pre-defined protocols and transparent reporting. The following methodology, exemplified by major projects like the U.S. EPA's ECOTOX Knowledgebase and PFAS systematic reviews, outlines the essential steps.

Defining the Scope: The PECO Framework

The Population, Exposure, Comparator, Outcome (PECO) framework precisely scopes the review question and determines inclusion/exclusion criteria. For ecotoxicity reviews, criteria are tailored to capture environmentally relevant data[reference:2].

Developing and Executing Search Strategies

Search strategies involve multiple, complementary approaches:

• Academic Database Searches: Systematic searches across major multidisciplinary (e.g., Web of Science, Scopus) and subject-specific (e.g., PubMed/Medline, Embase, BIOSIS Previews) databases using controlled vocabularies (e.g., MeSH, Emtree) and tailored keyword strings[reference:3][reference:4].
• Grey Literature Searches: Targeted searches in specialized repositories, government agency websites, regulatory databases (e.g., EPA's HERO, NTP database), and thesis portals to capture unpublished or hard-to-find studies[reference:5][reference:6].
• Supplemental Searches: Scanning reference lists of included studies and relevant reviews, and consulting with subject matter experts to identify additional sources[reference:7].

Screening and Selection Pipeline

Identified records undergo a multi-stage screening process to select relevant studies, often visualized via a PRISMA flow diagram[reference:8].

• Deduplication: Using automated tools (e.g., ICF's Deduper) and manual checks to remove duplicate citations[reference:9].
• Title/Abstract Screening: Two independent reviewers assess records against PECO criteria. Machine-learning assisted tools (e.g., SWIFT-Active Screener) can prioritize records to improve efficiency[reference:10].
• Full-Text Review: The full documents of potentially relevant studies are retrieved and assessed for final inclusion by independent reviewers, with conflicts resolved by consensus or a third reviewer[reference:11].

Data Extraction and Curation

Data from included studies are extracted into structured formats using standardized forms. Key details include chemical identity, species, test conditions, endpoints, and results. This process is often supported by specialized software (e.g., DistillerSR, HAWC) to ensure consistency and facilitate quality control[reference:12].

Quantitative Data Summaries

The scale and scope of comprehensive searching in toxicology are illustrated by curated databases and large-scale review projects. The following tables summarize key quantitative metrics and data structures.

Table 1: Scale of a Curated Ecotoxicity Database (ECOTOX Knowledgebase)

Metric	Quantity	Source
Number of unique chemicals	>12,000	[reference:13]
Number of ecological species	>12,000	[reference:14]
Total test results	>1,000,000	[reference:15]
Number of source references	>50,000	[reference:16]
Data update frequency	Quarterly	[reference:17]

Table 2: Core Data Fields for Ecotoxicity Study Curation (Adapted from ECOTOX)

Category	Example Data Fields	Purpose in Evidence Assessment
Chemical	Name, CAS RN, purity, formulation, concentration type (nominal/measured)	Verify test substance identity and exposure relevance.
Species	Scientific name, taxonomy, life stage, source	Define ecological population and assess organism relevance.
Study Design	Test method (e.g., OECD guideline), media, exposure duration, control type	Evaluate methodological reliability and appropriateness.
Test Results	Effect, endpoint (e.g., LC50, NOEC), statistical significance, units	Extract quantitative toxicity data for synthesis and analysis.

Table 3: Key Databases and Grey Literature Sources for Toxicology Systematic Reviews

Source Type	Example Resources	Primary Function in Search Strategy
Multidisciplinary Academic Databases	Web of Science, Scopus, Google Scholar	Broad coverage of peer-reviewed journal literature; cited reference searching.
Biomedical/Toxicology Databases	PubMed/MEDLINE, Embase, ToxLine (historical), BIOSIS Previews	Subject-specific coverage with strong toxicology indexing; includes conference abstracts.
Grey Literature Repositories	EPA HERO, NTP Database, OpenGrey, ProQuest Dissertations & Theses	Access to regulatory reports, unpublished studies, theses, and conference papers.
Chemical/Regulatory Databases	EPA CompTox Chemicals Dashboard (ToxValDB), ECHA website	Identify toxicity values and assessments from government sources.

Detailed Experimental Protocols for Cited Systematic Reviews

Protocol: ECOTOX Knowledgebase Literature Curation Pipeline

The ECOTOX team employs a standardized, quarterly pipeline to curate ecotoxicity data[reference:18].

• Search Execution: Comprehensive searches are conducted across open and grey literature sources using chemical-specific terms. No language or date restrictions are initially applied[reference:19].
• Screening: References undergo title/abstract screening followed by full-text review against pre-defined applicability (e.g., relevant species, chemical) and acceptability (documented controls, reported endpoints) criteria[reference:20].
• Data Extraction: For included studies, trained curators extract detailed information on chemical, species, study design, and results into a structured relational database using controlled vocabularies[reference:21].
• Quality Assurance: Data extractions undergo peer review, and standard operating procedures (SOPs) for all steps are maintained and updated regularly[reference:22].

Protocol: EPA PFAS Systematic Review Literature Search & Screening

This protocol for ~150 per- and polyfluoroalkyl substances (PFAS) demonstrates a large-scale, machine-learning assisted approach[reference:23].

• Database Searching: An information specialist executed complex search strategies in PubMed, Web of Science, and (historically) ToxLine in 2019, with updates in 2020 and 2021. All results were stored in the HERO database[reference:24].
• Deduplication: A two-phase process used automated logic followed by a machine-learning model (Python's Dedupe package) to identify and verify duplicate records[reference:25].
• Relevance Filtering: The SWIFT-Review software was used to filter unique references into "evidence streams" (e.g., human, animal, in vitro) based on preset search strategies applied to title/abstract fields[reference:26].
• Machine-Learning Assisted Screening: The PFAS review team used SWIFT-Active Screener for title/abstract screening. The software used active learning to prioritize records, and screening continued until a 95% likelihood of capturing all relevant studies was reached[reference:27].
• Full-Text Review & Data Extraction: Studies passing the initial screen underwent full-text review in DistillerSR. Data from prioritized animal studies were then extracted into the Health Assessment Workspace Collaborative (HAWC) by one reviewer and verified by a second[reference:28][reference:29].

Visualizing Search Strategies and Workflows

Diagram: Systematic Review Workflow for Ecotoxicity Data

Diagram: Literature Search & Screening Process

Table 4: Key Research Reagent Solutions for Toxicology Literature Review

Tool/Resource Category	Specific Example	Primary Function in Search Strategy
Bibliographic Databases	PubMed/MEDLINE, Embase, Web of Science, Scopus	Core sources for peer-reviewed journal articles and conference abstracts. Embase offers exceptional coverage of drug/toxicology topics[reference:30].
Grey Literature Databases	EPA HERO, NTP Database, OpenGrey, ProQuest Dissertations & Theses	Essential for finding unpublished studies, government reports, regulatory data, and academic theses.
Chemical Information Hubs	EPA CompTox Chemicals Dashboard (ToxValDB), PubChem	Provide curated chemical identifiers, properties, and aggregated toxicity values from multiple sources, including grey literature[reference:31].
Screening & Deduplication Software	SWIFT-Active Screener, DistillerSR, Rayyan, CADIMA	Facilitate collaborative title/abstract and full-text screening, often incorporating machine learning to prioritize records[reference:32].
Data Extraction & Management Platforms	Health Assessment Workspace Collaborative (HAWC), DistillerSR, Systematic Review Data Repository (SRDR+)	Structured environments for extracting, managing, and quality-checking data from included studies[reference:33].
Reference Management	EndNote, Zotero, Mendeley	Store, organize, and deduplicate search results; generate bibliographies.

Implementing comprehensive search strategies that seamlessly integrate academic and grey literature is a cornerstone of rigorous systematic reviews in toxicology. By adhering to pre-defined protocols like PECO, utilizing a wide array of specialized databases and repositories, and employing modern screening and data management tools, researchers can construct a complete and unbiased evidence base. This thorough approach is critical for supporting robust ecological risk assessments, informing regulatory decisions, and advancing the field of evidence-based toxicology.

In the field of ecotoxicity data research, systematic reviews are fundamental for synthesizing evidence to inform regulatory decisions, risk assessments, and the development of toxicity factors [18]. The complexity and volume of scientific literature on contaminants—from emerging compounds like PFAS (per- and polyfluoroalkyl substances) to established toxicants—demand a structured, transparent, and bias-minimizing approach to evidence synthesis [19] [20]. The dual-independent review methodology stands as a critical procedural safeguard within this process. It involves at least two researchers independently executing key stages of the review—screening studies against eligibility criteria, assessing their quality, and extracting data—with a formal process for reconciling discrepancies. This guide details the technical implementation of this methodology, framing it within the established systematic review framework for toxicology as outlined by regulatory bodies and contemporary best practices [21] [18]. The goal is to equip researchers and drug development professionals with a replicable protocol to enhance the credibility, reproducibility, and regulatory acceptance of their systematic reviews in ecotoxicology.

Foundational Framework and Process Workflow

Systematic reviews in toxicology extend beyond mere literature summaries; they are structured scientific investigations that aim to answer a specific question by identifying, selecting, appraising, and synthesizing all relevant studies. The Texas Commission on Environmental Quality (TCEQ) guidelines formalize this into a six-step process, within which dual-independent review is a cross-cutting principle applied to several stages [18].

The following workflow diagram illustrates this integrated process, highlighting the stages where dual-independent review is implemented and how they feed into final evidence integration.

Figure 1: Integrated Systematic Review Workflow with Dual-Independent Review Stages. The process, adapted from TCEQ guidelines [18], shows key stages where dual review is critical for minimizing error and bias.

The process begins with Problem Formulation, which establishes the review's scope and protocol. This is followed by a comprehensive Systematic Search. The core of the dual-independent methodology is then applied to Screening (title/abstract and full-text), Data Extraction, and Study Quality/Risk of Bias Assessment. Discrepancies between independent reviewers are reconciled at each stage before proceeding. Finally, the extracted and appraised evidence is integrated to answer the review question [18]. A recent review on global emerging contaminants effectively followed this model, implementing dual-independent screening protocols aligned with PRISMA 2020 guidelines [19].

Implementing Dual-Independent Screening and Selection

The screening phase filters the initially identified records to those truly relevant. Dual-independent review at this stage guards against the accidental exclusion of pertinent studies due to human error or individual reviewer bias.

Developing Inclusion/Exclusion Criteria (PECO Framework)

Criteria must be objective, unambiguous, and established a priori in the review protocol. The PECO framework is recommended for ecotoxicity reviews:

• Population (P): The test system (e.g., Daphnia magna, zebrafish embryos, soil microbial communities, specific human cell lines).
• Exposure (E): The contaminant(s) of interest, including specific forms, concentrations, and durations (e.g., chronic exposure to PFOS at environmental concentrations).
• Comparator (C): The control condition (e.g., vehicle control, clean medium, unexposed population).
• Outcome (O): The measured toxicological endpoints (e.g., mortality, reproduction, genotoxicity, gene expression changes, apical endpoints relevant to risk assessment) [21] [18].

A recent five-year review on emerging contaminants (ECs) employed stringent criteria, including: publication in English; focus on specific EC categories (e.g., PFAS, endocrine-disrupting chemicals); provision of concentration data and sampling details in aquatic environments; and clear analytical methods [19].

Table 1: Quantitative Summary of Screening Outcomes from a Recent Ecotoxicology Review [19]

Screening Stage	Records Identified	Action	Resulting Records	Key Criteria Applied
Initial Search	3,241	Removal of duplicates	3,018	Duplicate detection via reference manager.
Title/Abstract Screen	3,018	Exclusion based on title/abstract	416	Focus on five EC categories in aquatic environments; exclude non-research articles.
Full-Text Screen	416	Eligibility assessment	327 (Included)	Must provide concentration data, sampling info, and QA/QC methods.

The Dual-Independent Screening Protocol

• Calibration: Reviewers independently screen a common subset of 50-100 records. Agreement is measured using Cohen's kappa (κ). A κ ≥ 0.6 indicates acceptable agreement; if lower, criteria are clarified.
• Independent Screening: Using systematic review software (e.g., Rayyan, Covidence) or shared spreadsheets, each reviewer classifies every record as "Include," "Exclude," or "Maybe," citing the specific criterion for exclusion.
• Reconciliation: The software or lead reviewer identifies conflicts. Reviewers meet to discuss each conflict and reach consensus. Unresolved conflicts are arbitrated by a third senior researcher.
• Documentation: The final included list and reasons for all exclusions are recorded, forming the PRISMA flow diagram.

Implementing Dual-Independent Data Extraction

Accurate and consistent data extraction transforms study findings into analyzable data. Dual-independent extraction is vital for minimizing transcription errors and subjective interpretation.

Designing the Extraction Form

The form, piloted on a few studies, should capture:

• Bibliographic & Administrative Data: Author, year, DOI, funding source.
• Study Characteristics: Test organism (species, life stage), experimental design (control type, replication), exposure regime (concentration, duration, medium).
• Outcome Data: Quantitative results for pre-specified endpoints (means, standard deviations, sample sizes, p-values, EC/LC50 values). For graphical data, tools like GetData Graph Digitizer can be used [19].
• Methodological Quality Indicators: Details pertinent to risk of bias assessment (e.g., randomization, blinding, test substance verification, solvent controls, adherence to test guidelines) [21].

Table 2: Key Data Extraction Fields for Ecotoxicity Studies

Category	Specific Data Field	Example from PFAS Research [20]	Notes for Extractors
Study ID	Citation, Digital Object Identifier (DOI)	-	Essential for tracking.
Test System	Species, strain, life stage, cell type	Fathead minnow (Pimephales promelas), larval stage.	Record as reported.
Exposure	PFAS compound(s), measured concentration, duration, medium	PFOS, 10 µg/L (measured), 28-day chronic, freshwater.	Note if concentration is nominal or measured.
Outcomes	Endpoint, result value, variance, statistical significance	Survival: 85% mean, SD=5%; Reproduction (# offspring): 50 mean, SD=8.	Extract all relevant numeric data. Use graph digitizer if needed.
Methods	Test guideline (e.g., OECD), solvent, control type, renewal frequency	OECD 210, water-only exposure, solvent control (0.1% DMSO), static renewal.	Critical for quality assessment.

The Dual-Independent Extraction Protocol

• Piloting and Training: Reviewers independently extract data from 2-3 studies using the draft form. The team compares results to refine field definitions and instructions.
• Independent Extraction: Reviewers extract data from all included studies into separate forms.
• Verification and Reconciliation: Extracted data is compared, typically by a third party or using software features. All discrepancies are highlighted. Reviewers discuss each discrepancy, refer to the original source document, and agree on the correct entry.
• Resolution and Locking: Consensus data is entered into a master database. The process ensures the final dataset is error-checked.

Dual-Independent Study Quality and Risk of Bias Assessment

In ecotoxicology, assessing a study's "quality" involves more than just risk of bias; it includes evaluating technical reliability and relevance to the review question [21]. Dual-independent assessment is crucial here due to its inherent judgment.

Selecting an Appropriate Tool

The tool should match the study type:

• In Vivo / Whole Organism Studies: Tools like the TCEQ's internal criteria or SYRCLE's Risk of Bias tool evaluate sequence generation, blinding, selective reporting, and other biases. Key toxicology-specific items include verification of test substance identity and concentration [21] [18].
• In Vitro Studies: Specific guidance is less common but should address cell line authentication, mycoplasma testing, and appropriate controls [21].
• Human Observational Studies: Tools like NIH Quality Assessment Tool or ROBINS-I are applicable [21].

The Dual-Independent Assessment Protocol

• Tool Calibration: Reviewers independently assess the same set of studies. Agreement is measured (e.g., percentage agreement, weighted kappa for ordinal scales).
• Independent Assessment: Each reviewer judges each study against the tool's domains or questions.
• Consensus Meeting: Reviewers meet to discuss ratings, focusing on discrepancies. The goal is to reach a consensus rating for each domain, supported by explicit text from the study.
• Sensitivity Analysis: The review protocol should plan to analyze the impact of including or excluding studies judged to have "high" risk of bias or "low" reliability on the overall conclusions.

The following diagram illustrates how evidence from studies, after undergoing rigorous dual-independent screening, extraction, and quality assessment, is integrated to form a final review conclusion with an associated confidence rating.

Figure 2: Evidence Integration and Confidence Rating Process. Following dual-independent appraisal, evidence is stratified by quality. Confidence in the final conclusion is rated based on the weight, coherence, consistency, and relevance of the integrated evidence [18].

Implementing a rigorous dual-independent review requires both conceptual tools and practical software solutions.

Table 3: Research Reagent Solutions for Systematic Review Execution

Item / Resource	Category	Function in the Review Process	Example / Note
PRISMA 2020 Statement	Reporting Guideline	Provides a 27-item checklist and flow diagram template to ensure transparent and complete reporting of the review.	Mandatory for high-quality reviews; endorsed by major journals [19].
PECO Framework	Protocol Tool	Structures the formulation of the review question and eligibility criteria to ensure they are focused and answerable.	Population, Exposure, Comparator, Outcome [18].
Rayyan / Covidence	Software Platform	Web-based tools designed to manage the systematic review process, facilitating blind dual screening, conflict resolution, and data extraction.	Greatly improves efficiency and auditability over manual spreadsheet methods.
GetData Graph Digitizer	Data Extraction Tool	A software application that extracts numerical data from points on published graphs or images when tabular data is unavailable.	Essential for recovering data from older or poorly reported studies [19].
TCEQ Systematic Review Guidelines	Methodology Guide	Provides a six-step framework specifically for developing toxicity factors, emphasizing dual review and evidence integration [18].	A key regulatory-informed resource.
Risk of Bias / Quality Assessment Tools	Critical Appraisal Tool	Structured checklists to evaluate the methodological strengths and limitations of individual studies (e.g., SYRCLE for animal studies).	Choice depends on study design; piloting is essential [21].
Reference Management Software	Organizational Tool	Manages bibliographic records, removes duplicates, and formats citations (e.g., EndNote, Zotero, Mendeley).	Integrated searching and deduplication features are valuable.

The scientific and regulatory evaluation of chemicals depends on the systematic review and synthesis of ecotoxicity data. Traditional methods, which often rely on standard ANOVA-based comparisons of treatment groups to a control, are increasingly inadequate for modern risk assessment [22]. These methods fail to model the full dose-response continuum, struggle with non-normal data typical of toxicological endpoints (like counts, proportions, and binary responses), and do not provide a probabilistic estimate of low-effect concentrations. This creates a critical gap in systematic reviews, where diverse studies must be quantitatively integrated to derive robust safety thresholds.

This whitepines a paradigm shift toward Generalized Linear Models (GLMs) and Benchmark Dose (BMD) modeling within a framework of rigorous data evaluation. GLMs provide the flexibility to handle diverse data types and directly model the dose-response relationship. The BMD approach, a model-derived estimate of the dose corresponding to a specified low level of adverse effect (e.g., a 5% or 10% increase), offers a more scientifically consistent and statistically stable alternative to the No Observed Adverse Effect Level (NOAEL) [23]. The integration of these advanced statistical techniques with transparent study evaluation frameworks, such as the Criteria for Reporting and Evaluating Ecotoxicity Data (CRED), is essential for strengthening the objectivity, reproducibility, and predictive power of systematic reviews in ecotoxicity research [24].

Limitations of Traditional ANOVA-Based Methods in Dose-Response Analysis

The reliance on ANOVA with post-hoc tests to identify a NOAEL is a foundational weakness in conventional ecotoxicity data analysis. This approach discards the information contained in the dose-response pattern by treating dose as a categorical rather than a continuous variable. Its outcomes are heavily dependent on the arbitrary spacing and selection of test doses, sample size, and statistical power. A study with fewer animals or wider dose intervals may yield a higher, less protective NOAEL, demonstrating a lack of statistical robustness [23].

Furthermore, ANOVA assumes normally distributed, homoscedastic data, an assumption frequently violated by proportional data (e.g., percent affected) or count data (e.g., number of offspring). Applying transformations to meet these assumptions complicates interpretation and may not fully address the underlying data structure. Crucially, the NOAEL is constrained to be one of the experimental doses tested and provides no information on the expected response at doses below or between test levels, limiting its utility for extrapolation in risk assessment.

Foundational Statistical Models: GLMs and the Benchmark Dose (BMD)

Generalized Linear Models (GLMs) form the robust statistical backbone for modern dose-response analysis. A GLM relaxes the strict assumptions of ordinary linear regression by allowing the response variable to follow any distribution from the exponential family (e.g., Binomial for quantal data, Poisson for count data) and by using a link function to relate the linear predictor to the mean of the distribution [25]. For binary response data, common link functions include the logit (for logistic regression) and probit models. This framework directly estimates the probability of response as a smooth function of dose, enabling prediction across the dose continuum.

Benchmark Dose (BMD) Modeling utilizes the fitted dose-response curve to estimate the dose associated with a predetermined benchmark response (BMR), such as a 5% or 10% increase in the incidence of an adverse effect over the background rate. The lower confidence limit on this dose, the BMDL, is typically used as a point of departure for risk assessment [23]. This method uses all the experimental data, is less sensitive to dose spacing, and provides a consistent statistical confidence measure. For developmental toxicity data with intralitter correlations, specialized models (e.g., the beta-binomial LOG, RVR, and NCTR models) have been developed that account for litter effects as covariates, significantly improving fit and accuracy [23].

Table: Comparison of Traditional NOAEL and Modern BMD Approaches

Feature	NOAEL Approach	BMD Approach
Statistical Basis	Inferential (ANOVA); identifies highest dose with no statistically significant difference from control.	Model-based; fits a curve to all dose-response data.
Use of Data	Uses only data from the NOAEL and control doses; ignores shape of dose-response.	Uses all dose-response data to characterize the entire curve.
Dose Dependency	Highly dependent on the selected dose spacing and sample size.	Relatively stable across different experimental designs.
Result	A single experimental dose.	An estimated dose (BMD) and its confidence limit (BMDL).
Extrapolation	Cannot predict response at untested doses.	Allows for prediction of response at any dose within the model range.
Handling Covariates	Difficult to incorporate.	Can directly include covariates (e.g., litter size) in the model [23].

The Systematic Review Context: Integrating CRED for Data Reliability and Relevance

Advanced statistical synthesis is only as valid as the underlying data. Systematic reviews must therefore incorporate a critical evaluation of study reliability and relevance. The widely used Klimisch method has been criticized for its lack of detail, inconsistency between assessors, and over-reliance on Good Laboratory Practice (GLP) status rather than scientific merit [24] [22].

The Criteria for Reporting and Evaluating Ecotoxicity Data (CRED) framework was developed to address these shortcomings. It provides a transparent, criteria-based system for evaluating both the reliability (internal validity) and relevance (external validity, ecological applicability) of aquatic ecotoxicity studies [24]. A major ring test demonstrated that CRED produces more consistent and transparent evaluations than the Klimisch method [24]. Integrating CRED into the systematic review workflow ensures that only high-quality, relevant data are funneled into advanced dose-response modeling, thereby strengthening the overall synthesis.

Table: Key Differences Between Klimisch and CRED Evaluation Methods [24]

Characteristic	Klimisch Method	CRED Method
Primary Focus	Reliability only.	Reliability and relevance.
Number of Criteria	12-14 (ecotoxicity).	20 reliability criteria, 13 relevance criteria.
Guidance Provided	Limited, leading to high expert judgement dependence.	Detailed guidance for each criterion to improve consistency.
Basis for Reliability	Heavily favors GLP and standardized protocols.	Scientifically detailed criteria, independent of GLP status.
Outcome Transparency	Simple categorization (e.g., "reliable without restrictions").	Qualitative summary based on explicit criteria.

Detailed Experimental and Analytical Protocols

Protocol for Dose-Response Analysis Using GLMs and BMD

This protocol outlines the steps for fitting a dose-response model and deriving a BMD from quantal (binary) data, accounting for potential over-dispersion.

• Data Preparation: Organize data with columns for dose (continuous), number of subjects tested (n), and number of subjects affected (x). Include a group identifier if multiple replicates exist per dose [26].
• Exploratory Analysis: Plot the observed response proportion (x/n) against dose (often log-transformed). Examine variability between replicates at the same dose.
• Model Selection and Fitting: Fit a generalized linear model with a Binomial distribution and an appropriate link function (e.g., logit, probit). The model is: g(μ) = β₀ + β₁×Dose, where g is the link function and μ is the expected proportion responding.
• Model Diagnostics: Check for over-dispersion (variance > mean variance assumed by binomial model). This is common when subjects are not independent (e.g., organisms in the same container or litter) [23] [26].
• Addressing Over-Dispersion:
  • Quasi-Likelihood: Adjust standard errors using a dispersion parameter.
  • Bootstrap Method: The preferred method per OECD guidance for calculating confidence intervals for effect concentrations like the EC50 or BMD [26].
    • For each dose, sample the experimental replicates (e.g., test containers) with replacement.
    • Within each resampled replicate, sample the individual organisms with replacement.
    • Calculate a new response rate for each dose from the resampled data.
    • Fit the GLM to this new bootstrap dataset and estimate the BMD.
    • Repeat this process thousands of times (e.g., 5,000) to build an empirical distribution of the BMD.
    • The 2.5th and 97.5th percentiles of this distribution form the 95% confidence interval for the BMD. The BMDL is the 2.5th percentile [26].
• BMD Calculation: Using the original fitted model, calculate the dose at which the estimated additional risk (over background) equals the predetermined BMR (e.g., 5%). Use the bootstrap distribution to derive the BMDL.

Protocol for Applying the CRED Evaluation Method

The following steps are based on the CRED framework for evaluating a single ecotoxicity study [24].

• Phase 1: Reliability Assessment

  • Use the 20 reliability criteria, which cover all sections of a study report: Test Substance, Test Organism, Test Design and Conditions, Results, and Reporting.
  • For each criterion (e.g., "Concentration of test substance verified analytically"), assign a score based on the provided guidance: "Fully" (2), "Partially" (1), "Not at all" (0), or "Not reported" (NR).
  • A summary score is calculated, leading to an overall reliability category (e.g., "Reliable without restrictions," "Reliable with restrictions").

• Phase 2: Relevance Assessment

  • Use the 13 relevance criteria addressing ecological, biological, and endpoint relevance (e.g., "Is the test organism representative for the assessed ecosystem?").
  • Score each criterion similarly ("Fully," "Partially," etc.).
  • The relevance assessment is summarized qualitatively, identifying key strengths and limitations regarding the study's applicability to the specific risk assessment question.

• Integration for Systematic Review: Only studies deemed at least "Reliable with restrictions" and of sufficient relevance are carried forward for data extraction and quantitative synthesis via GLM/BMD modeling.

The Modern Toolkit: Integrating New Approach Methods (NAMs) and AI

The future of statistical synthesis in ecotoxicology lies in integrating traditional in vivo data with New Approach Methods (NAMs) and artificial intelligence (AI). The U.S. EPA and other organizations are actively developing training and tools in this area [27].

• High-Throughput Screening Data: Programs like ToxCast generate vast in vitro bioactivity data across hundreds of pathways. AI models are increasingly used to predict in vivo toxicity endpoints from this data, helping to prioritize chemicals for further testing and fill data gaps [28].
• Computational Toxicology Tools: Resources like the CompTox Chemicals Dashboard, ECOTOX Knowledgebase, and SeqAPASS (for cross-species extrapolation) provide essential data for modeling and assessment [27].
• Advanced Training: Hands-on courses and webinars, such as those offered by the Society of Toxicology's Risk Assessment Specialty Section and EU-PARC, are critical for training researchers in these advanced data analysis and modeling techniques [29] [30].

Table: Essential Research Reagent Solutions and Computational Tools

Tool/Reagent Category	Specific Example(s)	Primary Function in Synthesis/Modeling
Statistical Software & Packages	R (drc, benchmarkd, boot packages), Stata (doseresponse2) [25]	Fitting GLMs, performing bootstrap resampling, and calculating BMDs and confidence intervals.
Data Reliability Evaluation Framework	CRED (Criteria for Reporting and Evaluating Ecotoxicity Data) [24]	Providing a structured, transparent checklist to assess the reliability and relevance of primary studies for inclusion in a systematic review.
High-Throughput Bioactivity Database	EPA ToxCast Database (via invitroDB) [27] [28]	Serves as a data source for AI/QSAR models to predict toxicity and inform mechanisms, supplementing traditional ecotoxicity data.
Chemical & Toxicity Databases	EPA CompTox Dashboard, ECOTOX Knowledgebase [27]	Providing curated chemical identifiers, properties, and aggregated toxicity data necessary for data harmonization and modeling.
Cross-Species Extrapolation Tool	SeqAPASS (Sequence Alignment to Predict Across Species Susceptibility) [27]	Informs relevance assessment by evaluating the conservation of molecular targets across species, supporting read-across.

Moving beyond ANOVA to a synthesis paradigm centered on GLMs and BMD modeling, underpinned by rigorous study evaluation via frameworks like CRED, represents a necessary evolution in ecotoxicity research. This approach yields more robust, reproducible, and scientifically defensible estimates of low-effect concentrations for risk assessment.

Future advancements will be driven by the integration of diverse data streams: standardized in vivo studies evaluated via CRED, high-throughput in vitro NAMs data, and AI-driven predictive models [27] [28]. Successfully leveraging this integrated evidence base requires ongoing training for researchers in advanced statistics, data science, and systematic review methodology [30]. By adopting these advanced synthesis and modeling techniques, the field can improve the efficiency and accuracy of chemical safety evaluations, ultimately leading to better protection of ecological health.

The evaluation of Endocrine Active Chemicals (EACs) presents a unique and persistent challenge in toxicological risk assessment due to their potential to elicit biological effects at low exposure levels. These substances, capable of interacting with and modulating hormonal systems, may produce effects that are not readily detected by traditional, high-dose toxicological study designs [31]. The "low-dose hypothesis" posits that EACs can cause adverse outcomes at doses below those typically tested in standardized guideline studies and potentially below the established No Observed Adverse Effect Level (NOAEL) [32]. This concept is contentious, with debates centering on methodological reproducibility, toxicological relevance, and the implications for established testing paradigms [32].

The necessity for robust, transparent methods to evaluate this evidence is paramount. Systematic review methodologies offer a framework to objectively assemble, appraise, and synthesize the often complex and contradictory scientific literature on low-dose effects. This technical guide details the application of systematic review principles within a broader strategy for assessing low-dose toxicity of EACs, providing researchers and assessors with a structured approach to this critical issue [31].

Systematic Review Framework for Ecotoxicity Data

A systematic review minimizes bias through an explicit, pre-defined protocol for searching, selecting, appraising, and synthesizing evidence. In ecotoxicity, this process is essential for managing vast data and resolving contradictory findings.

The overarching strategy for evaluating low-dose evidence, as outlined by the National Academies, consists of three iterative phases: Surveillance, Investigation and Analysis, and Action [31].

Table: Core Phases in a Strategy for Evaluating Low-Dose Toxicity Evidence

Phase	Primary Objective	Key Activities	Outcome
Surveillance	Signal Detection	Actively monitor scientific literature, databases (e.g., ECOTOX, ToxValDB), and stakeholder input for new data on EACs and low-dose effects [31] [33] [34].	Identification of chemicals, effects, or methodological advances requiring further investigation.
Investigation & Analysis	Evidence Evaluation	Conduct systematic reviews, apply study quality criteria (e.g., Klimisch scores), perform dose-response analysis, and integrate mechanistic data [31] [33] [35].	A synthesized, weight-of-evidence conclusion regarding low-dose hazard.
Action	Risk Management & Knowledge Integration	Update toxicity values, refine testing guidelines, prioritize new data generation, or continue surveillance [31].	Regulatory decisions, revised assessments, or improved testing strategies.

Central to this process is the use of curated toxicological databases. Resources like the ECOTOXicology Knowledgebase (ECOTOX) and the Toxicity Values Database (ToxValDB) exemplify the application of systematic review principles at scale. ECOTOX, the world's largest curated ecotoxicity database, employs a structured pipeline for literature search, relevance screening, data extraction, and quality control to provide over one million test results for environmental assessments [33]. Similarly, ToxValDB curates and standardizes human health-relevant toxicity values (e.g., NOAELs, benchmark doses) from multiple sources into a consistent format, enabling large-scale analysis and modeling [34]. These databases are foundational for efficient evidence surveillance and synthesis.

Experimental Protocols for Low-Dose Endocrine Toxicity Assessment

In Vivo Testing Guidelines for Endocrine-Sensitive Outcomes

Traditional toxicity testing has been modified to enhance sensitivity to endocrine-mediated effects. Key OECD and EPA guidelines include:

• Extended One-Generation Reproductive Toxicity Study (OECD TG 443): This is a pivotal study that assesses endocrine-sensitive endpoints across generations. It includes cohorts for expanded reproductive/developmental toxicity, developmental neurotoxicity, and developmental immunotoxicity. Dosing begins in parental (P) generation animals before mating and continues through to offspring (F1) weaning and sometimes into adulthood [31].
• Rodent Two-Generation Reproduction Study (OECD TG 416): The historical standard for reproductive toxicity, which evaluates effects on fertility, gestation, lactation, and offspring development over two generations. Modifications for endocrine assessment include adding specific endpoints like anogenital distance, nipple retention, and detailed histological examination of endocrine tissues [31].
• Enhanced 28-Day Toxicity Study (OECD TG 407): Includes enhanced endocrine-sensitive endpoints such as thyroid histology, estrous cyclicity, and additional hormone measurements to screen for endocrine activity in a subacute exposure scenario [31].

Core Protocol Elements for Low-Dose Evaluation:

• Dose Selection: Critical for low-dose assessment. Doses should bracket anticipated human exposure levels and include a very low dose to probe for non-monotonic responses. A wide dose range is necessary to characterize the full dose-response curve [31] [32].
• Critical Windows of Exposure: Dosing must cover sensitive developmental periods (e.g., in utero, prenatal, neonatal, pubertal), as EAC effects are often latent and manifest later in life after early-life exposure [31].
• Endpoint Expansion: Beyond standard apical endpoints, studies should incorporate:
  • Morphological: Anogenital distance, nipple/areola count, puberty onset (vaginal opening, preputial separation).
  • Histopathological: Comprehensive evaluation of hormone-sensitive tissues (mammary gland, prostate, uterus, ovaries, testes, thyroid).
  • Functional: Sperm analysis, estrous cyclicity, hormone assays (estradiol, testosterone, TSH, T3/T4), behavioral assessments.
  • Multi-Generational Follow-up: To capture heritable effects or those that only become apparent in subsequent generations [31].

In Vitro and New Approach Methodologies (NAMs)

NAMs provide mechanistic data and help reduce animal testing. Core protocols include:

• Cell Viability & Proliferation Assays: Used to determine cytotoxic concentrations. Protocols must account for assay interference from EACs. A 2024 study demonstrated that metabolic assays like XTT can overestimate viability for certain drug classes (e.g., CDK4/6 inhibitors) due to drug-induced changes in cell metabolism and size. Quantitative nuclei imaging (e.g., using H2B-fluorescent protein or Hoechst staining) provides a more direct and reliable count of cell number, especially for low-dose, growth-inhibitory effects [36].
• Hormone Receptor Transactivation Assays: Standardized (e.g., OECD TG 455, 457) to detect agonist/antagonist activity on estrogen, androgen, and thyroid hormone receptors.
• High-Throughput Transcriptomics: Short-term in vivo or in vitro exposure followed by genome-wide expression analysis to derive transcriptomic points of departure (tPODs), which can be compared to apical endpoint PODs to assess sensitivity [34].
• Statistical Design & Dose-Response Analysis: A critical protocol component. Guidance recommends:
  • Using sufficient biological replicates and dose groups (typically ≥5 plus control) to model low-dose curves [35].
  • Selecting appropriate statistical models (e.g., nonlinear regression for curve fitting) over simple pairwise comparisons to the control (e.g., Dunnett's test) to interpolate effect concentrations between tested doses [35].
  • Using model-derived benchmark doses (BMD) over NOAELs/LOAELs where possible, as BMD modeling uses all dose-response data and is less dependent on dose spacing [35].

Table: Key Databases Curating Toxicity Data for Systematic Review

Database	Primary Focus	Key Data Types	Record Count (Approx.)	Role in Low-Dose Review
ECOTOX [33]	Ecotoxicology	Single-chemical toxicity tests for aquatic and terrestrial species.	>1 million test results	Provides ecological hazard data, identifies sensitive species and endpoints.
ToxValDB v9.6.1 [34]	Human Health	In vivo study results (NOAEL, LOAEL), derived toxicity values, exposure guidelines.	242,149 records	Source of standardized mammalian toxicity values for dose-response analysis and modeling.
ToxRefDB [34]	Mammalian Toxicology	Detailed in vivo guideline study data from EPA pesticide programs.	Not specified in search results	Provides rich study data for in-depth evaluation of study quality and endpoint selection.

Visualizing Workflows and Pathways

Systematic Review Workflow for EAC Low-Dose Toxicity [31] [33]

Dose-Response Analysis Pathway for Low-Dose Effects [35]

Toxicity Database Curation and Standardization Pipeline [33] [34]

The Scientist's Toolkit: Research Reagent & Resource Solutions

Table: Essential Research Tools for Low-Dose EAC Investigation

Tool/Resource Category	Specific Item/Example	Function in Low-Dose EAC Research
Curated Toxicity Databases	ECOTOX Knowledgebase [33]	Provides systematically curated ecotoxicity data for ecological risk assessment and hypothesis generation.
	ToxValDB [34]	Supplies standardized mammalian toxicity values (NOAEL, BMD) for dose-response modeling and chemical prioritization.
	ToxRefDB [34]	Offers detailed in vivo guideline study data for deep-dive evaluation of study quality and endocrine-sensitive endpoints.
In Vivo Test Guidelines	OECD TG 443 (Extended One-Generation) [31]	The primary guideline study for detecting endocrine-sensitive developmental and reproductive effects across life stages.
	OECD TG 416/421 (Reproductive Screening) [31]	Foundational studies for reproductive toxicity, often modified to include endocrine endpoints.
In Vitro & NAM Assays	Quantitative Nuclei Imaging (H2B-mRuby/Hoechst) [36]	Provides accurate, direct cell counting for viability/proliferation, mitigating artifacts from metabolic assays at low doses.
	Metabolic Assays (XTT, MTT, CellTiter-Glo) [36]	Measures cellular metabolic activity; requires careful interpretation for EACs that may alter metabolism independently of viability.
	Receptor Transactivation Assays (OECD TG 455, 457) [31]	Screens for direct agonist/antagonist activity on nuclear hormone receptors.
Statistical & Analytical Software	Benchmark Dose (BMD) Software (e.g., EPA BMDS, PROAST)	Fits dose-response models to calculate BMDs, which are more robust for low-dose extrapolation than NOAELs [35].
	R/Python Packages for Dose-Response (e.g., drc, doseR)	Facilitates advanced nonlinear regression and statistical comparison of dose-response curves [35].
Reference Compounds	Diethylstilbestrol (DES) [31]	Classic positive control EAC for estrogenic effects and developmental toxicity studies.
	Bisphenol A (BPA) [32]	Widely studied reference EAC for low-dose and non-monotonic dose-response research.
Systematic Review Tools	PECO/PSMO Frameworks	Structures the review question (Population, Exposure, Comparator, Outcome / Problem, Search, Manage, Output).
	Study Quality Checklists (e.g., Klimisch score) [33] [34]	Provides criteria for evaluating the reliability and relevance of individual toxicological studies.

Navigating Complexities in Ecotoxicity Reviews: Solutions for Data Heterogeneity, Quality, and Bias

Ecotoxicity research is foundational to environmental risk assessment, aiming to understand the harmful effects of chemicals on ecosystems. However, the field is inherently challenged by extensive data heterogeneity arising from three primary sources: variability in test species, the diversity of measured endpoints, and differences in experimental designs. This heterogeneity complicates the synthesis of evidence, a core task of systematic reviews (SRs) and meta-analyses (MAs) which seek to provide definitive conclusions for regulatory and scientific guidance [37] [38].

High-quality systematic reviews in ecotoxicology are critical for moving beyond narrative summaries to quantitative, transparent, and reproducible evidence synthesis. They are essential for validating new approach methodologies (NAMs), prioritizing chemicals for regulatory action, and setting protective environmental standards [39] [1]. Yet, as seen in recent large-scale assessments, significant methodological flaws and high between-study heterogeneity can undermine the validity and utility of such reviews [37]. A primary contributor to this heterogeneity is intertest variability—the differences in results for the same chemical-species combination across different studies. Quantifying this is crucial; one analysis of acute aquatic toxicity data found the standard deviation of intertest variability to be approximately a factor of 3 [40]. Ignoring this variability, or aggregating data via unjustified harmonization methods like the geometric mean without proper statistical modeling, weakens the defensibility of derived safety thresholds such as the Predicted No-Effect Concentration (PNEC) [40].

This whitepaper provides an in-depth technical guide for researchers conducting systematic reviews of ecotoxicity data. It deconstructs the sources of heterogeneity, presents quantitative frameworks for its assessment, and offers detailed methodologies for designing robust synthesis protocols that account for variability in test species, endpoints, and experimental designs.

Systematic Review Methodology for Heterogeneous Ecotoxicity Data

Conducting a systematic review on ecotoxicity data requires a protocol that is both rigorous in its standards and flexible enough to account for expected heterogeneity. The process must be transparent, reproducible, and aligned with guidelines such as PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) [41] [19].

1. Protocol Development and Registration A pre-defined, publicly registered protocol is essential. It should detail the research question using PECO/PICO elements (Population, Exposure, Comparator, Outcome), and explicitly define eligibility criteria for studies. For ecotoxicity, this means specifying:

• Population: Accepted taxonomic groups (e.g., freshwater invertebrates, fish families), life stages, and source (lab-cultured vs. field-collected).
• Exposure: Chemical of interest, acceptable exposure media (e.g., water, sediment), and exposure routes.
• Comparator: Requirement for an appropriate control group (concurrent negative control is mandatory) [42].
• Outcome: Relevant ecotoxicity endpoints (e.g., LC50, EC50, NOEC, LOEC).

2. Systematic Search and Screening Searches should be performed across multiple bibliographic databases (e.g., Web of Science, Scopus, PubMed) and specialized resources like the U.S. EPA ECOTOX database [42] [1]. The ECOTOX database alone contains over 1.1 million entries and serves as a primary curated source for experimental ecotoxicity data [1]. Screening should be performed independently by two reviewers against the eligibility criteria, with good inter-rater reliability (e.g., Kappa > 0.6) [38]. A flow diagram documenting the process from initial search to final inclusion is mandatory.

3. Data Extraction and Critical Appraisal A standardized form should extract quantitative data (mean effect, variance, sample size), experimental conditions (species, exposure duration, temperature, endpoint type), and study validity indicators. Each study must be critically appraised. Regulatory agencies like the U.S. EPA Office of Pesticide Programs provide explicit criteria for accepting open literature data, which can serve as a robust appraisal tool [42]. Key acceptance criteria include:

• The study investigates single-chemical exposure on whole, live organisms.
• A concurrent control is used for comparison.
• An explicit exposure duration and measured concentration/dose are reported.
• The tested species is verified [42].

4. Quantitative Synthesis and Heterogeneity Analysis Where appropriate, meta-analysis is conducted. Given the expected variability, a random-effects model is typically more appropriate than a fixed-effect model, as it assumes the true effect size varies between studies. Heterogeneity must be statistically quantified using metrics like I² and τ². Exploring sources of heterogeneity through subgroup analysis (e.g., by species class, endpoint type, or test guideline) or meta-regression (using continuous variables like exposure time or temperature) is a core objective [41]. The workflow for managing this process is outlined below.

Systematic review workflow for ecotoxicity data

Quantifying and Managing Variability Across Test Species

Species sensitivity is a major source of variability. A core tool for modeling this in risk assessment is the Species Sensitivity Distribution (SSD). SSDs are statistical models that aggregate toxicity data across multiple species to estimate a hazardous concentration (e.g., HC₅, affecting 5% of species) [39]. Building robust SSDs requires addressing interspecies variability and the related issue of intertest variability.

1. Data Aggregation for Species-Specific Toxicity For a given chemical, multiple toxicity values (e.g., LC50) for the same species may exist from different studies. The REACH guidance suggests aggregating these via the geometric mean [40]. However, a more statistically rigorous approach is to model the data using a Bayesian hierarchical model that explicitly estimates both the central tendency for that species (μ) and the intertest variability (σintertest). Research indicates σintertest has a standard deviation of approximately a factor of 3 (fold-change) [40]. Ignoring this can lead to underestimating uncertainty in the SSD and the derived HC₅.

2. Building and Interpreting Species Sensitivity Distributions Modern SSD modeling uses curated datasets from sources like ECOTOX, spanning multiple taxonomic groups and trophic levels (producers, primary consumers, secondary consumers, decomposers) [39]. Advanced models integrate both acute (e.g., LC50) and chronic (e.g., NOEC) endpoints to predict hazard concentrations for data-poor chemicals [39]. The output is a cumulative distribution function of species sensitivities, from which protective benchmarks are derived.

Table 1: Representative Sensitivity of Taxonomic Groups to General Chemical Stressors

Taxonomic Group	Trophic Level	Typical Acute Endpoint (LC50/EC50)	Relative Sensitivity (Generalized)	Key Endpoint Variability Notes
Green Algae (e.g., Raphidocelis subcapitata)	Producer	Growth inhibition (72-96 hr EC50)	High	Population-level endpoint; highly sensitive to photosynthesis inhibitors.
Crustaceans (e.g., Daphnia magna)	Primary Consumer	Immobilization (48 hr EC50)	High-Very High	Sensitive indicator; endpoint often immobilization, not direct mortality [1].
Insects (e.g., Chironomus dilutus)	Primary Consumer	Mortality or growth (96 hr LC50/EC50)	Moderate-High	Larval stages tested; sediment-dwelling species important for benthic assessment.
Fish (e.g., Oncorhynchus mykiss)	Secondary Consumer	Mortality (96 hr LC50)	Moderate	Standard vertebrate model; inter-species variability significant [1].
Amphibians (e.g., Xenopus laevis)	Secondary Consumer	Mortality or malformation (96 hr LC50/EC50)	Variable	High sensitivity during larval stages; data often less abundant.

Harmonizing Diverse Ecotoxicity Endpoints

Ecotoxicity studies measure a wide array of endpoints, from acute mortality to sublethal effects on reproduction, growth, and behavior. This diversity is necessary to capture full hazard but creates integration challenges.

1. Endpoint Types and Derivation Endpoints fall into three main classes [43]:

• Effective Concentrations (ECx): The exposure concentration causing a x% effect (e.g., 50%, 20%, 10%). Derived from continuous dose-response data.
• No Observed Effect Concentration (NOEC): The highest tested concentration showing no statistically significant difference from the control. Heavily dependent on experimental design (e.g., concentration spacing, statistical power).
• Lowest Observed Effect Concentration (LOEC): The lowest tested concentration showing a statistically significant effect.

The derivation of these endpoints from raw experimental data is a critical step that influences their comparability.

Derivation pathways for ecotoxicity endpoints

2. Endpoint Selection and Conversion for Meta-Analysis For synthesis, endpoint selection must be dictated by the review question. A common approach is to use the most sensitive endpoint from each study or to separate analyses by endpoint type (e.g., acute vs. chronic). Converting between endpoints (e.g., using Acute-to-Chronic Ratios) introduces uncertainty and should be done cautiously with clear justification. The use of benchmark dose (BMD) modeling, which provides a more robust estimate of a predefined effect level (e.g., BMD10) than NOEC, is increasingly advocated [43].

Table 2: Common Ecotoxicity Endpoints and Meta-Analysis Considerations

Endpoint	Description	Typical Exposure Duration	Key Advantage	Key Limitation for Synthesis
LC50	Lethal Concentration for 50% of population.	Acute (24-96 hr)	Standardized, widely available, allows direct comparison.	Only measures mortality, may not be most ecologically relevant.
EC50	Effective Concentration for 50% effect (e.g., immobilization, growth inhibition).	Acute or Chronic	Can capture sublethal effects; defined for various outcomes.	Effect type must be clearly defined and consistent across studies.
NOEC	No Observed Effect Concentration.	Usually Chronic (e.g., 21-28 d)	Simple concept for regulatory thresholds.	Statistically weak, depends on test design/power; not an estimate of threshold [43].
LOEC	Lowest Observed Effect Concentration.	Usually Chronic	Identifies a clear effect level.	Dependent on concentration spacing; value is a tested concentration, not a true threshold.

Evaluating and Accounting for Experimental Design Variability

Differences in how tests are conducted introduce significant variability. Key design factors include exposure characteristics, environmental parameters, and control data.

1. Exposure and Environmental Parameters A meta-analysis on microplastics and temperature illustrates how design factors are analyzed. It required studies to report quantitative data on both stressors and measured endpoints like growth, mortality, reproduction, or physiological stress [41]. Such factors become covariates in meta-regression to explain heterogeneity. For instance, the effect size of a toxicant may be significantly moderated by water temperature or exposure duration [41].

2. The Critical Role of Control Data Distinguishing treatment effects from background biological variability is paramount. The use of Historical Control Data (HCD)—compiled from control groups of previous studies using the same species and methods—is a powerful but underutilized tool in ecotoxicology [43] [44]. HCD provides a reference range for "normal" variability, helping to interpret whether an effect in a concurrent control is atypical or if a small statistical effect in a treatment group is biologically meaningful. While routine in mammalian toxicology, its use in ecotoxicology is hampered by lack of standardization and guidance [43] [44].

Table 3: Key Experimental Design Variables and Their Impact on Data Heterogeneity

Design Variable	Source of Variability	Potential Impact on Endpoint	Method for Harmonization/Account in SR
Exposure Duration	Acute (hrs/days) vs. Chronic (weeks/months) tests.	Fundamental difference in endpoint sensitivity (Chronic > Acute).	Perform separate meta-analyses for acute and chronic data [1].
Temperature	Metabolic rate, chemical toxicity, and solubility changes.	Can amplify or mitigate toxicity (e.g., with microplastics [41]).	Include as a continuous covariate in meta-regression [41].
Test Medium/Water Chemistry	pH, hardness, dissolved organic matter affect bioavailability.	Alters the freely dissolved concentration of toxicant.	Record as study characteristic; subgroup analysis if extremes exist.
Chemical Concentration Verification	Nominal vs. measured concentration.	Measured is more reliable; nominal overestimates exposure if loss occurs.	Prefer studies with measured concentrations; note method as quality item [42].
Control Group Performance	Natural biological variability in control organisms.	Affects statistical power to detect true effects; atypical control can skew results.	Use Historical Control Data (HCD) to contextualize concurrent control results [43] [44].

Impact of experimental design variability on data synthesis

The Scientist's Toolkit: Essential Reagents and Materials

Conducting and synthesizing ecotoxicity research requires both physical materials and conceptual tools. The following table details key solutions and resources.

Table 4: Research Reagent Solutions for Ecotoxicity Testing and Synthesis

Item/Tool Name	Category	Function in Ecotoxicity Research
U.S. EPA ECOTOX Database	Data Repository	A publicly accessible database compiling over 1.1 million toxicity test results for single chemicals on aquatic and terrestrial species. Serves as the primary source for curated data for systematic reviews, SSD modeling, and machine learning [42] [39] [1].
OECD Test Guidelines	Methodological Standard	Internationally agreed test methods (e.g., OECD TG 203 for fish acute toxicity, TG 211 for Daphnia reproduction) that standardize procedures for species culturing, exposure, and endpoint measurement, reducing inter-laboratory variability.
Historical Control Data (HCD) Compendium	Reference Data	A laboratory-specific or consortium-wide collection of control group results from past studies. Used to establish the normal range of background variability for key endpoints, aiding in the interpretation of individual study results [43] [44].
Culture Media for Standard Test Species	Biological Reagent	Standardized recipes (e.g., ASTM hard water, OECD algal medium) for culturing and testing organisms like Daphnia magna, Raphidocelis subcapitata, and fathead minnows. Ensures organism health and reproducible baseline responses.
Reference Toxicants	Quality Control Chemical	Standard chemicals (e.g., potassium dichromate for Daphnia, sodium chloride for algae) used in periodic tests to confirm the sensitivity of biological cultures falls within an accepted historical range, ensuring data reliability.
Statistical Software for Meta-Analysis & SSD Modeling	Analytical Tool	Software packages (e.g., R with metafor, ssdtools, OpenTox SSDM platform [39]) capable of performing random-effects meta-analysis, meta-regression, and fitting species sensitivity distributions, which are essential for quantitative evidence synthesis.

Addressing heterogeneity in test species, endpoints, and experimental designs is not an obstacle to be eliminated but a fundamental characteristic of ecotoxicology that must be rigorously managed. Systematic reviews and meta-analyses that proactively model these sources of variability—using hierarchical models for intertest variability, subgroup analysis for species and endpoint types, and meta-regression for experimental conditions—produce more reliable, transparent, and actionable evidence for decision-making.

Future progress depends on several key developments:

• Adoption of HCD: Regulatory encouragement and guidance for the systematic collection and use of Historical Control Data would significantly improve study interpretation and reduce false positives [43] [44].
• Standardized Reporting: Journals and data repositories should mandate the comprehensive reporting of all experimental design metadata to facilitate future synthesis.
• Embracing NAMs and Computational Tools: The integration of machine learning-ready benchmark datasets [1] and robust SSD modeling platforms [39] will accelerate risk assessment while reducing reliance on animal testing.
• Complex Stressor Integration: Methodologies must evolve to synthesize data on combined stressors (e.g., chemicals × temperature [41]), reflecting real-world ecological challenges.

By implementing the detailed methodologies and frameworks presented here, researchers can enhance the scientific rigor and regulatory relevance of systematic reviews in ecotoxicology, ultimately leading to more robust protections for ecosystem health.

Within the systematic review of ecotoxicity data, the critical appraisal of individual studies is a foundational step that determines the validity and strength of the resulting evidence synthesis. This process involves assessing the risk of bias (RoB)—the potential for systematic error due to flaws in study design or conduct—and the overall reliability of the data [45] [46]. Unlike narrative reviews, a systematic review mandates a transparent and standardized evaluation of these factors to judge the trustworthiness of evidence before its synthesis [47] [46]. In ecotoxicology, this task is uniquely complex due to the diversity of test organisms, exposure systems, and endpoints, alongside the frequent use of both standardized guideline tests and higher-tier, environmentally realistic studies [48] [49].

Historically, the field has relied on tools like the Klimisch method, which categorizes studies as "reliable without restrictions" to "not reliable" [50]. However, this approach has been criticized for its lack of detail, over-reliance on Good Laboratory Practice (GLP) status, and insufficient guidance, leading to inconsistent evaluations between assessors [50]. The evolution toward more transparent and structured frameworks, such as the Criteria for Reporting and Evaluating ecotoxicity Data (CRED) and the newer Ecotoxicological Study Reliability (EcoSR) framework, marks a significant advancement toward harmonized and scientifically robust systematic reviews in environmental science [45] [50].

Core Critical Appraisal Tools and Frameworks

Several dedicated tools have been developed to address the specific needs of ecotoxicity study evaluation. The table below summarizes the key characteristics of prominent frameworks.

Table 1: Comparison of Key Critical Appraisal Frameworks for Ecotoxicity Studies

Framework/Tool	Primary Developer/Context	Core Purpose	Key Features & Structure
Ecotoxicological Study Reliability (EcoSR) Framework [45]	Kennedy et al. (2025); For toxicity value development.	A comprehensive, two-tiered system to assess internal validity (risk of bias) and reliability.	Tier 1: Optional rapid screening. Tier 2: Full assessment across 5 bias domains (Confounding, Selection, Exposure/Intervention, Outcome, Reporting). Emphasizes a priori customization.
Criteria for Reporting & Evaluating Ecotoxicity Data (CRED) [50]	Moermond et al. (2016); Improvement over Klimisch method.	Evaluate reliability and relevance of aquatic ecotoxicity studies for regulatory hazard/risk assessment.	20 reliability criteria (e.g., test organism, exposure, measurement) and 13 relevance criteria (e.g., endpoint, exposure). Detailed guidance for each criterion.
EFSA Critical Appraisal Tools (CATs) [48]	European Food Safety Authority (EFSA); For non-standard higher-tier studies.	Support evaluation of reliability and relevance of higher-tier ecotoxicity studies for pesticide risk assessment.	Based on CRED. Consists of Excel tools and handbooks for seven study types (e.g., aquatic, bees, birds). Includes a (semi-)quantitative scoring system.
Chemical-Specific Information (CSI-CAT) [51]	Systematic review community; For exposure measurement bias.	Supplement general CATs with chemical-specific data to appraise exposure measurement in human observational studies.	Four categories: Overarching Considerations, Exposure Setting, Sampling/Lab Methods, Biological/Physiological Considerations.
FEAT Principles [46]	Collaboration for Environmental Evidence (CEE); For environmental systematic reviews.	Core principles to guide fit-for-purpose RoB assessment in quantitative environmental reviews.	Principles: Focused, Extensive, Applied, Transparent. Supports a Plan-Conduct-Apply-Report framework.
CEE Critical Appraisal Tool (Prototype) [52]	Collaboration for Environmental Evidence (CEE).	A prototype tool for assessing RoB in primary studies on effectiveness or impact in environmental management.	Under development (v0.3 as of 2023). Currently covers seven risk-of-bias criteria for internal validity.

The EcoSR framework represents a modern synthesis, explicitly building on the classic RoB approach from human health and integrating ecotoxicity-specific criteria from tools like CRED [45]. Its two-tiered process is designed for efficiency and depth.

Diagram: Two-Tiered Workflow of the EcoSR Framework [45]

The CRED method provides granular criteria, moving beyond a single score to a transparent evaluation of specific study elements. A major ring test demonstrated that CRED yielded more consistent and transparent evaluations than the Klimisch method [50]. EFSA's CATs operationalize CRED for regulatory use, providing tailored tools for specific study types like bee field studies or aquatic mesocosms [48].

Foundational Principles: Risk of Bias and the FEAT Approach

Assessing risk of bias is distinct from general "quality assessment." It specifically evaluates a study's internal validity—the degree to which its design and conduct prevent systematic error (bias) that would lead to under- or over-estimation of the true effect [46]. This is separate from external validity (relevance or applicability) and precision (random error) [46].

To ensure RoB assessments are robust, the FEAT principles have been proposed as a universal guide [46]:

• Focused on internal validity.
• Extensive in covering all key sources of bias.
• Applied to the synthesis and conclusions.
• Transparent in process and reporting.

A survey revealed that 64% of environmental systematic reviews published between 2018-2020 lacked any RoB assessment, and those that did often omitted key bias sources [46]. Adhering to FEAT addresses these shortcomings. The process integrates into the systematic review as follows:

Diagram: Integration of Risk of Bias Assessment in Systematic Review

Methodological Pillars: Experimental Protocols and Statistical Analysis

Critical appraisal requires understanding standard ecotoxicity test methodologies. Common guidelines include OECD Test Guidelines (e.g., OECD 201 for algae, 210 for fish) and analogous EPA and ISO standards [50]. The reliability criteria in tools like CRED directly reflect the key methodological components of these tests.

Table 2: Key Experimental Protocols and Methodological Criteria in Ecotoxicity Studies

Protocol Domain	Standard Guideline Example	Key Appraisal Criteria (e.g., from CRED) [50]	Common Sources of Bias
Test Organism	OECD 211: Daphnia magna reproduction test.	Species identification, life stage, health status, acclimation.	Selection bias if organisms are not representative or healthy.
Exposure System	OECD 210: Fish early-life stage test.	Exposure concentration verification (analytical measurements), stability, renewal, environmental conditions (pH, O₂, temp).	Exposure misclassification bias if nominal concentrations are used without verification for poorly soluble/toxic substances [51].
Endpoint Measurement	OECD 201: Freshwater algal growth inhibition test.	Definition of endpoint, measurement method, timing, replication.	Detection bias if measurement is not blinded or is inconsistent.
Study Design	OECD 203: Fish acute toxicity test.	Use of controls, randomization, concentration levels, replication.	Performance bias if groups are not treated equally aside from exposure.
Data Reporting & Statistics	OECD Document No. 54 (under revision) [2].	Raw data availability, appropriate statistical tests, results clarity.	Reporting bias if only significant or favorable results are presented.

A critical and evolving aspect of methodology is statistical analysis. The long-standing use of hypothesis-testing methods like ANOVA to derive No-Observed-Effect Concentrations (NOECs) is increasingly criticized as outdated and less informative [2]. Modern appraisal emphasizes the use of dose-response modeling (e.g., using generalized linear models - GLMs) to derive point estimates like the EC_x or the Benchmark Dose (BMD), which make more efficient use of data and provide confidence intervals [2]. The planned 2026 revision of OECD Document No. 54 is expected to recommend continuous regression models as the default over categorical ANOVA-type analyses [2].

The Scientist's Toolkit: Essential Reagents and Materials

Critical appraisal involves evaluating the suitability and handling of materials used in the primary studies. The following table details key research reagent solutions and their role in ecotoxicity testing.

Table 3: Research Reagent Solutions and Essential Materials in Ecotoxicity Testing

Item/Category	Function in Ecotoxicity Studies	Appraisal Considerations
Standard Test Organisms	Daphnia magna (crustacean), Pseudokirchneriella subcapitata (alga), Danio rerio (fish). Serve as biological models for response.	Is the species/strains standard? Are source, health, and life stage documented? Culture conditions? [50]
Reference Toxicants	E.g., Potassium dichromate (K₂Cr₂O₇) for Daphnia. Used to assess organism sensitivity and test system health.	Was a reference toxicity test performed? Did results fall within acceptable historical ranges? [50]
Stock & Test Solutions	Prepared solutions of the test chemical in solvent (e.g., acetone, DMSO) or dilution water.	Was solvent control used? Was concentration verified analytically (crucial for nanomaterials, poorly soluble compounds)? [51] [49]
Culture/Dilution Media	Reconstituted hard/soft water (e.g., ISO or EPA media), algal growth media. Provides standardized hydration/nutrition.	Was media composition appropriate and consistent? Were key parameters (hardness, pH, nutrients) measured and reported? [50]
Analytical Standards	High-purity chemical standards for analytical verification (e.g., via HPLC, GC-MS, ICP-MS).	For tests not following GLP, was analytical chemistry used to confirm exposure concentrations? This is a key reliability criterion [50].
Positive Control Substances	Substances with known mechanism/effect (e.g., atrazine for photosynthesis inhibition). Used in higher-tier or mechanistic studies.	If used, was the control response as expected, validating the test system for the specific endpoint?

The development and adoption of structured critical appraisal tools like the EcoSR framework and CRED mark significant progress toward robust, transparent systematic reviews in ecotoxicology. The integration of risk of bias assessment grounded in FEAT principles ensures that evidence synthesis accounts for internal validity, strengthening the credibility of conclusions for regulatory and scientific decision-making [45] [46].

Future directions include the wider adoption of these tools, the completion and validation of the CEE prototype tool, and the crucial revision of statistical guidance (OECD No. 54) to promote modern dose-response modeling [2] [52]. Furthermore, addressing emerging challenges—such as appraising studies on complex materials like graphene-family nanomaterials (GFMs), where characterization in environmental media is difficult, or evaluating data-poor chemicals in life-cycle assessment—will require ongoing refinement of appraisal criteria [53] [49]. Ultimately, consistent application of rigorous critical appraisal tools is indispensable for building a reliable evidence base to inform environmental protection policies.

Handling Missing Data and Non-Standard Statistical Reporting (e.g., NOEC/LOEC)

Systematic reviews of ecotoxicity data aim to synthesize scientific evidence to inform environmental risk assessments and regulatory decisions for chemicals and pharmaceuticals. The integrity of these reviews is fundamentally challenged by two pervasive issues: missing data and the use of non-standard statistical endpoints, such as the No Observed Effect Concentration (NOEC) and Lowest Observed Effect Concentration (LOEC). Missing data can reduce statistical power, introduce bias, and threaten the validity of meta-analyses [54]. Concurrently, reliance on NOEC/LOEC values, which depend on arbitrarily selected test concentrations and lack measures of variability, can lead to misleading conclusions about a substance's hazard [55]. Within the framework of a thesis on systematic review methods, this guide provides a technical examination of these challenges. It outlines robust methodologies for identifying, handling, and analyzing incomplete datasets while critically evaluating legacy statistical reporting practices. The goal is to advance transparent, reproducible, and statistically sound synthesis methods in ecotoxicology.

Handling Missing Data in Ecotoxicity Datasets

Missing data is a common occurrence in almost all research, and ecotoxicity studies within systematic reviews are no exception [54]. The approach to handling missing data must be deliberate and justified, as improper methods can bias effect estimates and reduce the reliability of the review's conclusions.

Mechanisms and Types of Missing Data

The strategy for handling missing data is determined by its underlying mechanism, classified as follows [54] [56]:

• Missing Completely at Random (MCAR): The probability of data being missing is unrelated to any observed or unobserved variable. An example is a sample lost due to equipment failure. Analyses remain unbiased but lose statistical power.
• Missing at Random (MAR): The probability of missingness depends on observed data but not on the missing value itself. For instance, if younger fish in a study are more likely to have unmeasured weight data, but this propensity is fully explained by their recorded length, the data is MAR.
• Missing Not at Random (MNAR): The probability of missingness depends on the unobserved missing value itself. This is the most problematic scenario. An example is the failure to record mortality events in a tank where the observer was absent precisely because a sudden toxic effect was suspected.

Table: Mechanisms and Handling Strategies for Missing Data in Ecotoxicity

Mechanism	Definition	Ecotoxicity Example	Recommended Handling Strategy
MCAR	Missingness is independent of all data.	Sample destroyed in a freezer malfunction.	Listwise deletion may be acceptable; multiple imputation is preferred.
MAR	Missingness depends on observed data only.	Measurement of a secondary endpoint (e.g., growth) is missing for all individuals from one specific lab due to a protocol oversight.	Model-based methods (multiple imputation, maximum likelihood).
MNAR	Missingness depends on the unobserved missing value.	Organisms that died early in a chronic test were not weighed, causing missing growth data.	Sensitivity analysis (e.g., pattern-mixture models, selection models).

Prevention and Primary Handling Strategies

The optimal method is to prevent missing data through rigorous study design and data collection protocols [54]. For systematic reviewers, this emphasizes the importance of screening for data completeness as a core quality criterion. The U.S. EPA's guidelines for evaluating open literature stress the necessity of reporting key study elements, including concurrent controls, explicit exposure durations, and calculated endpoints [42].

When data is missing, the following techniques, summarized from general statistical and machine learning practices, are applicable [54] [56]:

• Deletion Methods:

  • Listwise Deletion: Removes any study or experimental unit with missing values. It is unbiased only if data is MCAR but can severely reduce sample size and power.
  • Pairwise Deletion: Uses all available data for each specific analysis. It can lead to inconsistent models and is generally not recommended for complex synthesis.

• Single Imputation Methods (Generally discouraged for primary analysis):

  • Mean/Median/Mode Substitution: Replaces missing values with the central tendency of the observed data. It distorts distributions and underestimates standard errors.
  • Last Observation Carried Forward (LOCF): Used in longitudinal data. It assumes no change after dropout, an unrealistic assumption in growing or responding organisms, and introduces bias [54].

• Model-Based Methods (Recommended):

  • Maximum Likelihood (ML): Estimates parameters directly using all available data under an assumed model (e.g., multivariate normal). It provides valid inferences under MAR conditions.
  • Multiple Imputation (MI): Creates several (m) plausible complete datasets by replacing missing values with draws from a predictive model. The m analyses are combined, yielding estimates and confidence intervals that account for imputation uncertainty. MI is the current best practice for handling MAR data in systematic reviews.

Protocol for a Multiple Imputation Workflow in a Systematic Review

• Diagnosis: Report the amount and patterns of missing data for key variables (e.g., EC50, sample size, control response) across included studies.
• Imputation Model: Specify a multivariate model (e.g., predictive mean matching, Bayesian regression) to impute missing values. Include all variables involved in the analysis and auxiliary variables correlated with missingness to strengthen the MAR assumption.
• Imputation: Generate m complete datasets (typically m=20-50). The imputation process should be performed separately for different taxa or chemical classes if the relationships between variables differ.
• Analysis: Perform the planned meta-analysis or synthesis separately on each of the m datasets.
• Pooling: Combine the m results using Rubin's rules to obtain final effect estimates, confidence intervals, and p-values that incorporate between-imputation variance.

Systematic Review Multiple Imputation Workflow

Non-Standard Reporting: The NOEC/LOEC Paradigm and Alternatives

A significant hurdle in synthesizing historical ecotoxicity literature is the prevalent use of NOEC and LOEC values. These endpoints were once standard but are now recognized as statistically flawed [55].

Definition and Critical Limitations

• NOEC (No Observed Effect Concentration): The highest tested concentration at which there is no statistically significant difference (typically p ≥ 0.05) from the control group.
• LOEC (Lowest Observed Effect Concentration): The lowest tested concentration that does show a statistically significant difference (p < 0.05) from the control [55].

Table: Critical Limitations of NOEC/LOEC versus Regression-Based ECx Values

Aspect	NOEC/LOEC	Regression-Based ECx (e.g., EC10, EC50)
Basis	Depends on selected, often arbitrary, test concentrations.	Derived from a fitted model of the entire concentration-response relationship.
Statistical Power	Highly sensitive to sample size, control variance, and number of concentrations. Low power increases NOEC.	Power is reflected in the confidence intervals around the ECx estimate.
Information Use	Ignores the shape of the concentration-response curve and data from other concentrations.	Uses all datapoints to model the biological response pattern.
Uncertainty Quantification	Provides no confidence interval or measure of uncertainty.	Confidence intervals can be calculated (e.g., bootstrap, delta method).
Interpretation	Misleading name; "No Observed Effect" is not "No Effect."	Explicit: the concentration causing an X% effect (e.g., 10% reduction in growth).
Value Extrapolation	Cannot estimate effects between or below test concentrations.	Can predict effect levels for untested concentrations within the model range.

The core problem is that the NOEC is a function of experimental design (chosen concentrations, replication) and statistical power, not a biological threshold. A large NOEC may indicate low toxicity, or it may simply indicate an underpowered test [55].

Modern Alternative: Regression-Based ECx Estimation

International guidance (OECD, EPA) now favors regression-based procedures [55]. This involves:

• Model Fitting: Fitting a suitable mathematical model (e.g., log-logistic, probit) to the dose-response data.
• Endpoint Derivation: Using the fitted model to calculate an Effective Concentration (ECx)—the concentration estimated to cause a defined percentage (x%) effect relative to the control. Common values are EC10 (low-effect level) and EC50 (median effect).
• Uncertainty Estimation: Calculating confidence intervals (e.g., 95% CI) for the ECx, often via bootstrap methods.

Regression-Based ECx Determination Workflow

Protocol for Re-analysis of Published NOEC/LOEC Data

When individual-level or mean/SD data are available from studies reporting only NOEC/LOEC, reviewers can re-analyze to derive ECx values.

• Data Extraction: Extract raw response data (means, standard deviations/errors, sample sizes) for each test concentration and control from tables, figures (using digitization software), or by contacting authors.
• Data Assumption: If individual data are unavailable, assume the mean response at each concentration represents normally distributed data with the reported variance.
• Model Selection & Fitting: Use statistical software (e.g., R with drc package) to fit 3-4 common ecotoxicity models (e.g., log-logistic, Weibull) to the extracted data.
• Model Averaging (if necessary): If no single model is clearly superior, use model averaging techniques to derive a robust ECx estimate that accounts for model uncertainty.
• Reporting: Report the derived ECx (e.g., EC10) with its confidence interval, the best-fitting model, and the original NOEC/LOEC for comparison.

Integrating Approaches for Robust Systematic Reviews

Advanced systematic reviews must integrate handling for both missing data and non-standard endpoints into a unified protocol.

Tiered Screening and Data Acceptability Criteria

Adopting criteria from the U.S. EPA's guidelines for open literature provides a robust screening framework [42]. Studies should be evaluated based on:

• Reliability: Was the study conducted according to standard guidelines (e.g., OECD, EPA)? Are control responses acceptable? Is the test substance characterization adequate?
• Relevance: Is the species, endpoint, and exposure duration relevant to the review question?
• Statistical Reporting: Does the study report data suitable for synthesis (e.g., ECx with CI, means with variance, raw data)? Studies reporting only NOEC/LOEC without supporting data may be categorized as having "limited usability" and their conclusions treated with caution.

Table: Data Quality Criteria for Ecotoxicity Studies in Systematic Reviews

Criteria Category	High Usability	Moderate/Limited Usability	Unacceptable/Excluded
Experimental Design	Guideline-compliant (OECD, EPA), clear control, reported exposure duration.	Minor deviations from guidelines, control performance marginal but acceptable.	No control, unacceptable solvent use, exposure not verified.
Data Reporting	Raw data available; or means with SD/SE and n; or ECx with CI.	Only means reported without variance; or only NOEC/LOEC with some graphical data.	Only NOEC/LOEC without supporting data; only qualitative statements.
Statistical Method	Regression-based ECx estimation.	Hypothesis testing (ANOVA) with post-hoc leading to NOEC/LOEC.	Inappropriate statistical test; no statistical analysis.

Advanced Analytical Integration: Addressing Data Scarcity with SDML

A cutting-edge approach to integrate incomplete and variably reported data is the use of Small Data Machine Learning (SDML) [57]. SDML techniques, such as Bayesian hierarchical modeling and data augmentation, can:

• Model across heterogeneous data: Integrate studies reporting ECx, NOEC, or even binary (effect/no-effect) outcomes by modeling the underlying concentration-response relationship.
• Impute missing endpoints: Predict missing EC50 values for a chemical-species pair based on chemical descriptors, taxonomic similarity, and data from other tested species.
• Quantify uncertainty: Propagate uncertainty from imputation, model selection, and between-study heterogeneity through to final risk estimates.

Table: Research Reagent Solutions for Ecotoxicity Data Synthesis

Tool/Resource	Function	Application in Systematic Review
ECOTOX Database (U.S. EPA)	Comprehensive public database of ecotoxicity test results for single chemicals [42].	Primary source for literature mining and initial data collection.
R Statistical Software	Open-source environment for statistical computing and graphics.	Core platform for data cleaning, imputation (mice package), dose-response analysis (drc package), meta-analysis (metafor), and SDML.
Digitization Software (e.g., WebPlotDigitizer)	Extracts numerical data from published graphs and figures.	Critical for recovering raw data from studies that did not publish numerical tables.
Multiple Imputation Software (e.g., mice in R)	Creates multiple plausible imputations for missing data.	Handles missing covariates or effect sizes under the MAR assumption.
Dose-Response Analysis Packages (e.g., drc in R)	Fits and analyzes a wide range of non-linear dose-response models.	Re-analyzes published data to derive ECx values from studies reporting only NOEC/LOEC.
Bayesian Modeling Software (e.g., Stan, JAGS)	Fits complex hierarchical models using Markov Chain Monte Carlo (MCMC) sampling.	Implements SDML, integrates diverse data types, and quantifies all sources of uncertainty.
Reporting Guidelines (e.g., PRISMA-EcoTox)	Checklist for transparent reporting of ecological systematic reviews.	Ensures methodological rigor and completeness in documenting the review process.

Advancing systematic review methods for ecotoxicity research requires confronting the dual challenges of missing data and legacy statistical reporting head-on. Moving beyond simplistic listwise deletion and critically re-evaluating NOEC/LOEC values are not just statistical improvements but necessities for robust evidence synthesis. By implementing a protocol that combines preventive screening based on clear data quality criteria [42], principled multiple imputation for missing data [54], and regression-based re-analysis or modeling of non-standard endpoints [55], reviewers can produce more reliable and informative syntheses. The integration of Small Data Machine Learning frameworks presents a promising frontier for maximizing the utility of sparse and heterogeneous ecotoxicity data [57]. Embracing these methods will strengthen the scientific foundation of environmental risk assessment and support more credible decision-making.

Optimizing Search Strategies to Overcome Database and Publication Bias

Within the rigorous framework of systematic review methods for ecotoxicity data research, the integrity of the final synthesis is fundamentally dependent on the completeness and representativeness of the evidence base. Database bias and publication bias represent two pervasive, interconnected threats to this integrity. Database bias arises when a search is limited to one or a few sources, failing to capture all relevant studies due to uneven coverage across disciplinary databases [58]. Publication bias, a form of dissemination bias, occurs when the publication of research findings is influenced by their nature, direction, or origin, leading to a published literature that is a non-random, often overly optimistic, subset of all conducted research [59].

The consequences are severe. In clinical medicine, publication bias has led to harmful patient outcomes and wasteful expenditures, as exemplified by cases where billions were spent on stockpiling drugs based on incomplete evidence [59]. In ecotoxicology, these biases compromise regulatory risk assessments by skewing the available data on chemical hazards. A systems-based analysis of European chemical management highlights that disagreements on the reliability and transparency of academic research act as barriers to its uptake, further complicating evidence-based decision-making [4]. This technical guide provides a detailed methodology for designing and executing search strategies that proactively mitigate these biases, ensuring a more robust and reliable foundation for systematic reviews in ecotoxicity and related fields.

Defining the Bias Landscape in Evidence Synthesis

A precise understanding of bias mechanisms is essential for developing effective countermeasures. The following table categorizes and defines the primary biases addressed by optimized search strategies.

Table: Key Biases in Evidence Synthesis and Their Impact

Bias Type	Definition	Primary Cause	Consequence for Ecotoxicity Reviews
Publication Bias	The publication or non-publication of research depends on the nature and direction of the results [59].	Selective submission/acceptance of studies with statistically significant or "positive" results.	Overestimation of a chemical's toxicity or potency; hazard thresholds (e.g., NOEC, ECx) derived from an unrepresentative subset of all tests.
Database Bias	The failure to identify relevant studies due to reliance on a single or limited number of databases with incomplete coverage [58].	Varying scope, indexing practices, and journal coverage across bibliographic databases.	Missing key studies published in specialized environmental science journals not indexed in major biomedical databases.
Outcome Reporting Bias	The selective reporting of some outcomes (e.g., statistically significant ones) but not others within a published study [59].	Authors or sponsors choosing to highlight only favorable or significant findings from a suite of measured endpoints.	Incomplete understanding of a chemical's effects profile; missing data on sub-lethal or long-term chronic effects.
Language & Geographical Bias	The preferential publication or indexing of research from certain countries/institutions, often in English [59].	Editorial and reviewer preferences, resource limitations for translation, and database focus.	Underrepresentation of locally relevant field studies from non-English speaking regions, limiting the geographical applicability of the review.

The pathway from research conception to dissemination, as adapted from Song et al. and detailed in a 2015 review, illustrates critical intervention points where these biases are introduced and where search strategies can intervene [59]. The following diagram maps this pathway within the ecotoxicology context.

Core Methodology for Optimized Search Design

An optimized search strategy is multi-pronged, extending beyond a single database query to actively seek out studies vulnerable to being missed.

Principle 1: Multi-Database Search as a Foundation

A 2022 metaresearch study provides quantitative evidence for this principle: searching two or more databases significantly decreases the risk of missing relevant studies [58]. The analysis of 60 Cochrane reviews showed that while overall coverage (indexation in at least one database) was high (96%), the recall (actual findability) varied. In reviews where searching fewer databases would not change conclusions, searching two databases achieved a median recall of 87.9% to 93.3%. Crucially, in reviews where a limited search would lead to opposite or impossible conclusions, recall from single-database searches plummeted to as low as 20.0% to 78.7% [58].

Table: Impact of Database Search Strategy on Evidence Base Integrity [58]

Search Scenario	Median Coverage (Indexation)	Median Recall (Findability)	Risk to Review Conclusions
Adequate Search (≥2 DBs) in stable reviews	95.0% - 100.0%	87.9% - 93.3%	Low. Conclusions and certainty unchanged.
Limited Search (1 DB) in stable reviews	87.9% - 96.6%	78.2% - 86.6%	Moderate. May lower confidence.
Limited Search (1 DB) in sensitive reviews	60.6% - 86.0%	20.0% - 78.7%	High. Can reverse or invalidate conclusions.

Protocol for Database Selection & Search:

• Mandatory Core: Search at least two major multidisciplinary and subject-specific databases. For ecotoxicology, a combination like PubMed/MEDLINE (biomedical) and Web of Science or Scopus (multidisciplinary) is essential.
• Subject-Specific Databases: Integrate specialized sources such as Environmental Sciences and Pollution Management (ProQuest), TOXLINE, or GreenFILE.
• Systematic Search Syntax: Develop a search string using Boolean operators (AND, OR, NOT). Structure it around core concepts: (1) Population (e.g., Daphnia magna, zebrafish), (2) Exposure (e.g., chemical name, CAS RN, "pharmaceutical"), and (3) Outcome (e.g., "ecotox*", "LC50", "chronic toxicity"). Explode database-specific thesaurus terms (MeSH, Emtree) and include free-text synonyms.
• No Date/Language Filters: Apply these only in the study selection phase, not the initial search, to avoid prematurely excluding relevant older or non-English studies.

Principle 2: Targeted Grey Literature Searching

Grey literature is critical to combat publication bias. The unfound references in the metaresearch study were more often abstractless (30% vs. 11% of found references) and older [58], characteristics common in technical reports and theses.

Protocol for Grey Literature Search:

• Search Trial and Study Registries: Consult ClinicalTrials.gov, the WHO ICTRP, and environmental data repositories like the Environmental Protection Agency's (EPA) Health and Environmental Research Online (HERO).
• Scan Conference Proceedings: Search websites of major societies (e.g., Society of Environmental Toxicology and Chemistry (SETAC), American Chemical Society (ACS)).
• Search Theses/Dissertations: Use ProQuest Dissertations & Theses Global.
• Contact Regulatory Bodies & Industry: Request unpublished study reports from agencies like the European Chemicals Agency (ECHA) or the U.S. EPA.
• Use Specialized Search Engines: Utilize Google Scholar with advanced operators and tools like OpenGrey.

Supplementary manual methods address database indexing gaps.

Protocol for Supplementary Searching:

• Forward Citation Tracking: Use databases like Scopus or Web of Science to find all papers that cite key included studies.
• Backward Citation Searching: Manually review the reference lists of all included studies and relevant review articles.
• Hand-Searching Key Journals: Identify the top 3-5 journals in your field and manually browse their tables of contents for the last 2-3 years.

Experimental Protocols for Search Strategy Validation

Protocol: Assessing Database Coverage

This protocol quantifies the risk of database bias for a specific research question.

• Define a Gold Standard Set: Assemble a final list of included studies from a completed systematic review or a pilot search.
• Check Indexation: For each study in the set, query each database under evaluation (e.g., MEDLINE, Embase, Web of Science) to determine if it is indexed there.
• Calculate Metrics:
  • Coverage (Sensitivity): (Number of gold standard studies indexed in Database A / Total gold standard studies) * 100.
  • Unique Contribution: Identify studies only found in Database A and in no other searched database.
• Interpretation: Based on the metaresearch, a unique contribution of >5% from any single database strongly justifies its inclusion in the core search strategy [58].

Protocol: Testing for Publication Bias (Post-Hoc Analysis)

While not a search strategy, statistical tests applied during data synthesis can indicate the likely presence of publication bias, informing the interpretation of results.

• Funnel Plot Analysis: Plot a measure of study precision (e.g., standard error) against its effect size (e.g., log-transformed hazard ratio). Asymmetry (a gap in smaller, less precise studies near the null effect) suggests potential publication bias.
• Statistical Tests: Apply Egger's regression test or the trim-and-fill method to quantitatively assess funnel plot asymmetry. Crucial Note: These methods have low statistical power, especially with fewer than 10 studies, and a non-significant result does not prove the absence of bias [59].
• Comparison of Registered vs. Reported Outcomes: For clinical trials, compare the pre-specified primary outcomes in the trial registry entry with those reported in the final publication to detect outcome-reporting bias [59].

Data Extraction and Management to Mitigate Bias

Consistent and rigorous data extraction is vital for handling the heterogeneous data retrieved via comprehensive searches. Best practice mandates dual independent extraction by at least two reviewers to minimize error and bias [60].

Key Protocol Steps:

• Develop a Pilot-Tested Form: Create a structured form in software like Covidence, REDCap, or a spreadsheet. Pilot it on 5-10 studies and refine for clarity [60].
• Extract Key Study Elements:
  • Bibliographic & Study Design: Author, year, source (journal, report, thesis), design (OECD test guideline, field study), sponsor/funder.
  • Population & Test System: Species, life stage, test duration.
  • Exposure: Chemical, concentration/ dose metrics, vehicle/control.
  • Outcomes & Results: All measured endpoints (even non-significant ones), raw data (means, SD, n), calculated metrics (NOEC, LOEC, EC50, BMD), and statistical tests used.
  • Indicators of Risk of Bias: Information on randomization, blinding, compliance with test guidelines, and any statements on conflict of interest (CoI) [59] [60].
• Resolve Discrepancies: Reviewers compare extractions, discuss disagreements, and reach consensus, often involving a third arbitrator [60].

The following diagram synthesizes the socio-technical factors affecting evidence use in ecotoxicology [4] and the statistical modernization needs [2] into a systems view of bias.

The Ecotoxicologist's Toolkit for Bias-Resistant Reviews

Table: Essential Tools and Resources for Optimized Systematic Reviews

Tool Category	Specific Tool / Resource	Primary Function in Mitigating Bias
Bibliographic Databases	Web of Science, Scopus, PubMed/MEDLINE, Embase, Environment Complete	Foundation for multi-database searching to overcome database bias and improve recall [58].
Specialized & Grey Lit Sources	ProQuest Dissertations, Google Scholar, agency websites (ECHA, EPA), OpenGrey, conference proceedings.	Targeted retrieval of unpublished or hard-to-find studies to combat publication and outcome reporting bias.
Study Registries	ClinicalTrials.gov, WHO ICTRP, OSF Registries, institutional repositories.	Identification of ongoing/completed but unpublished studies and comparison of planned vs. reported outcomes [59].
Reference Management	EndNote, Zotero, Mendeley.	Deduplication of search results and organization of studies from diverse sources.
Systematic Review Software	Covidence, Rayyan, EPPI-Reviewer.	Streamlines screening, dual data extraction, and quality assessment, reducing human error and enhancing reproducibility [60].
Statistical Software	R (with packages like metafor, drc, bmdb), OpenBUGS/JAGS.	Advanced data synthesis, dose-response modeling (e.g., BMD analysis), and statistical testing for publication bias [2].
Reporting Guidelines	PRISMA, ITSIE, COSTER.	Ensures transparent and complete reporting of the review methods, including search strategies, allowing for critical appraisal and replication.

Optimizing search strategies is not a peripheral task but a central, methodologically rigorous component of a valid systematic review in ecotoxicology. By implementing multi-database searches, proactively seeking grey literature, and employing supplementary manual techniques, researchers can significantly reduce database and publication biases. This requires moving beyond reliance on a single database and embracing a systematic, protocol-driven approach to evidence retrieval. When combined with rigorous data extraction, appropriate statistical analysis of the retrieved evidence, and transparent reporting, these optimized strategies form the bedrock of a reliable and unbiased evidence synthesis. This, in turn, supports sound regulatory decision-making and a more accurate understanding of chemical hazards in the environment.

Managing and Synthesizing Evidence from Diverse Evidence Streams (in vivo, in vitro, in silico)

The systematic review and synthesis of evidence from in vivo, in vitro, and in silico streams constitute a critical methodological frontier in ecotoxicity and human health risk assessment. This technical guide details rigorous frameworks and protocols for integrating these heterogeneous data types to form robust, transparent hazard conclusions. The evolution from expert-led narrative reviews to structured, protocol-driven systematic reviews minimizes bias, enhances reproducibility, and strengthens the scientific foundation for regulatory and research decisions [61] [62]. Central to this process is the application of modified systematic review methodologies, such as the Navigation Guide and the Systematic Review and Integrated Assessment (SYRINA) framework, which provide structured steps for formulating questions, retrieving evidence, appraising study quality, and synthesizing findings across evidence streams [61] [62]. A significant contemporary challenge and opportunity lie in building scientific confidence in New Approach Methodologies (NAMs), which include advanced in vitro and in silico models. Successfully incorporating NAMs requires validation against apical health outcomes and integration into a holistic "weight of evidence" assessment that considers biological plausibility, consistency, and the inherent variability of each evidence stream [63].

The Imperative for Systematic Evidence Integration in Ecotoxicology

The field of environmental health has historically relied on expert-based narrative reviews, an approach analogous to clinical medicine over four decades ago [61]. Such reviews are vulnerable to bias, lack of transparency, and delayed incorporation of new scientific evidence, potentially leading to continued exposure to harmful substances while evidence of harm accumulates [61]. The transition to systematic review methodology, well-established in clinical sciences, addresses these deficiencies by employing pre-specified protocols, comprehensive search strategies, standardized risk-of-bias assessments, and explicit methods for evidence synthesis [62].

The complexity of modern toxicology, with its diverse data streams, demands this structured approach. In vivo mammalian studies provide whole-organism, apical endpoint data but raise ethical and translational concerns. In vitro assays offer mechanistic insights and high-throughput capability but may lack physiological context. In silico models, including quantitative structure-activity relationships (QSARs) and molecular docking simulations, enable prediction and prioritization for thousands of data-poor chemicals. The core challenge is to credibly integrate these distinct lines of evidence into a coherent and actionable hazard assessment. This integration is not merely a summation of data but a rigorous evaluation of the strengths, limitations, and interdependencies of each stream to answer a specific question, such as whether a chemical meets the World Health Organization (WHO) definition of an endocrine disruptor [62].

Foundational Frameworks for Evidence Synthesis

Two prominent frameworks exemplify the systematic approach to integrating diverse evidence streams in environmental health: the Navigation Guide and the SYRINA framework.

The Navigation Guide Methodology: Developed to translate evidence-based medicine principles into environmental health, this framework involves four core steps [61]:

• Specify the Study Question: Frame a precise question relevant to decision-makers (e.g., "Does developmental exposure to chemical X impair fetal growth?").
• Select the Evidence: Execute and document a systematic search of published and unpublished literature.
• Rate Quality and Strength of Evidence: Critically appraise individual studies and rate the overall body of evidence for human and nonhuman data separately, then integrate these ratings.
• Grade Strength of Recommendations: Integrate the strength of toxicity evidence with exposure data, alternative assessments, and societal values.

A key innovation is its departure from traditional evidence-based medicine by assigning a "moderate" quality rating to well-conducted human observational studies and explicitly combining human and nonhuman evidence streams into a unified conclusion (e.g., "probably toxic") [61].

The SYRINA Framework: Explicitly designed for identifying endocrine disruptors (EDs) per the WHO/IPCS definition, SYRINA provides a seven-step process [62]:

• Formulate the problem via a PECO (Population, Exposure, Comparator, Outcome) statement.
• Develop and publish an a priori protocol.
• Conduct a systematic literature search and screening.
• Perform critical appraisal of included studies.
• Synthesize evidence within each stream (epidemiological, in vivo mammalian, in vivo non-mammalian, in vitro) for a) adverse effects and b) endocrine activity.
• Integrate evidence across streams to rate the strength of evidence for criteria (a) and (b).
• Integrate the conclusions on (a) and (b) to draw an overall conclusion on endocrine disrupting potential.

This framework forces a structured evaluation of the causal link between endocrine activity and adverse outcomes, which is central to ED identification [62].

Table 1: Comparison of Systematic Review Frameworks for Ecotoxicity Evidence Synthesis

Framework	Primary Context	Core Steps	Key Feature for Evidence Integration	Final Output Example
Navigation Guide [61]	Environmental health hazard assessment	1. Specify question2. Select evidence3. Rate quality/strength4. Grade recommendations	Explicit integration of human and nonhuman evidence streams into a single rating.	A conclusion statement such as "known to be toxic," "probably toxic," or "not classifiable."
SYRINA [62]	Identification of endocrine disruptors	1. Problem formulation2. Protocol3. Search & screen4. Appraisal5. Within-stream synthesis6. Across-stream integration7. Overall conclusion	Separate synthesis for evidence of adverse effect and endocrine activity, followed by integration to establish a causal link.	A conclusion on whether the chemical is an endocrine disruptor based on WHO/IPCS definition.

Detailed Methodological Protocols for Each Evidence Stream

Robust synthesis depends on the rigorous and standardized execution of protocols for each evidence stream. The following outlines critical methodological components.

3.1 Protocol for In Vivo (Ecotoxicity) Studies In vivo studies, particularly using mammalian or non-mammalian model organisms, provide data on apical endpoints (e.g., mortality, growth, reproduction, histopathology) under controlled exposure conditions.

• Experimental Design: Follow OECD or EPA test guidelines where applicable (e.g., OECD 203 for fish acute toxicity, OECD 416 for two-generation reproduction toxicity). Key elements include appropriate sample size calculation, randomized allocation to exposure groups, inclusion of vehicle and positive controls, and blinded outcome assessment where possible.
• Data Extraction: Extract data on species, strain, sex, age, route and duration of exposure, dosing regimen, measured endpoints, statistical results (mean, variance, significance), and no-observed-adverse-effect-level (NOAEL)/lowest-observed-adverse-effect-level (LOAEL).
• Risk of Bias (RoB) Assessment: Use domain-based tools tailored to toxicology, such as the SYRCLE's RoB tool for animal studies or the NTP/OHAT approach. Assess bias from sequence generation, baseline characteristics, blinding, random outcome assessment, incomplete outcome data, selective reporting, and other sources [62].

3.2 Protocol for In Vitro and Mechanistic Studies In vitro assays elucidate mechanisms of action, including receptor binding, gene expression changes, and cellular functional assays.

• Experimental Design: Report cell line origin and passage number, culture conditions, dosing rationale (including solvent controls), treatment duration, and replication details. Assays should be grounded in a defined toxicity pathway (e.g., estrogen receptor activation).
• Data Extraction: Extract information on the biological model, endpoint measured, assay type, concentration-response data (EC50, IC50, efficacy), positive and negative control performance, and evidence of cytotoxicity for non-specific effects.
• RoB Assessment: Adapt tools to assess in vitro study reliability. Key domains include characterization of test substance, interference from confounding properties (e.g., cytotoxicity), reproducibility of results, and appropriateness of the model system for the biological question [62].

3.3 Protocol for In Silico (Computational) Evidence In silico evidence includes QSAR predictions, read-across analyses, and molecular modeling.

• Model Application: Use models that are scientifically valid, transparent, and applicable to the chemical and endpoint of interest. Document the model's defined applicability domain and the certainty of the prediction.
• Data Extraction: Extract the predicted endpoint/activity, prediction certainty or probability, underlying algorithm or training set, and any applicability domain alerts.
• RoB/Reliability Assessment: Assess the validity of the model per OECD principles, the reliability of the input data, the chemical's fit within the model's applicability domain, and the transparency of the methodology.

Table 2: Core Data Extraction and Risk of Bias Domains by Evidence Stream

Evidence Stream	Critical Data Extraction Fields	Key Risk of Bias Assessment Domains
In Vivo	Species/strain/age/sex; Exposure route/duration; Dosing data; Apical endpoint results (NOAEL, LOAEL, statistical significance); Historical control data.	Selection bias (randomization); Performance bias (blinding); Detection bias (blinded assessment); Attrition bias (completeness of data); Selective reporting [62].
In Vitro	Cell line/organism source; Passage number/culture conditions; Test substance characterization; Concentration-response data (e.g., EC50); Control performance (positive/negative/vehicle); Cytotoxicity data.	Test substance interference (e.g., cytotoxicity, fluorescence); Biological relevance of model system; Replication and reproducibility; Appropriateness of assay for endpoint [62].
In Silico	Model name and version; Predicted endpoint and value; Applicability domain description; Prediction confidence metric; Underlying algorithm/training set.	Validity of the model (e.g., OECD principles); Reliability of input data; Relevance of the model to the specific endpoint; Fit within applicability domain.

Quantitative and Qualitative Data Synthesis Techniques

Synthesis involves combining results from individual studies to assess patterns, consistency, and overall strength of evidence.

4.1 Within-Stream Synthesis

• Narrative Synthesis: For data not amenable to meta-analysis, organize studies by outcome, species, or study design. Tabulate results and describe patterns of consistency, dose-response relationships, and the magnitude of effects.
• Meta-Analysis: When studies are sufficiently homogeneous (in PECO, exposure, outcome), quantitative meta-analysis can be performed. This involves calculating a summary effect size (e.g., standardized mean difference, relative risk) across studies, often using random-effects models to account for heterogeneity. Statistical heterogeneity (e.g., I² statistic) must be assessed and reported [64].

4.2 Across-Stream Integration (Weight of Evidence) This is the crucial step of combining conclusions from in vivo, in vitro, and in silico streams. A structured WoE approach considers several factors:

• Consistency: Are effects/activities consistent across different studies and evidence streams?
• Biological Plausibility and Coherence: Do the mechanistic data (in vitro, in silico) provide a plausible explanation for the apical effects observed in vivo? Is the evidence coherent with established biological knowledge?
• Strength and Concordance: What is the magnitude and statistical confidence of effects? Do data from different levels of biological organization (molecular, cellular, organ, organism) point to the same conclusion?
• Temporal and Dose-Response Relationships: Are effects time- and dose-dependent?

Frameworks like SYRINA formalize this by first rating the strength of evidence for an adverse effect and for endocrine activity separately, then judging the causal link between them [62]. The overall confidence is rated as high, moderate, low, or evidence of no effect.

Visualization of Integrated Evidence and Pathways

Effective visualizations are indispensable for communicating the synthesis process, biological pathways, and final conclusions.

5.1 Workflow and Evidence Integration Diagrams Diagrams should map the systematic review process and the logic of evidence integration.

Evidence Integration Workflow in Systematic Review

5.2 Signaling Pathway Disruption Maps For mechanistic toxicology, diagrams can illustrate how a chemical perturbs normal biological pathways.

Conceptual Pathway for Endocrine Disruption Leading to Adverse Outcomes

5.3 Experimental Workflow from Protocol to Synthesis A detailed diagram can show the parallel processing of different evidence types.

Parallel Processing and Synthesis of Diverse Evidence Streams

The Scientist's Toolkit: Essential Reagents and Materials

Conducting and evaluating studies across evidence streams requires specific tools. This table outlines key solutions for the integrated assessment workflow.

Table 3: Research Reagent Solutions for Integrated Ecotoxicity Assessment

Tool/Reagent Category	Specific Examples & Functions	Primary Evidence Stream
Systematic Review Software	Rayyan, Covidence, DistillerSR: Platforms for managing the systematic review process, including reference deduplication, blinded screening, and data extraction form management.	All (Meta-tool)
Risk of Bias Assessment Tools	SYRCLE's RoB Tool, OHAT/NTP RoB Tool: Structured guides to evaluate internal validity of animal studies. ECHA/EFSA Guidance Checklists: For assessing reliability of in vitro and environmental studies.	In Vivo, In Vitro
Reference Chemicals & Controls	Agonists/Antagonists (e.g., 17β-Estradiol, Flutamide): Positive controls for mechanistic assays. Vehicle Controls (e.g., DMSO, Corn Oil): Ensure solvent effects are accounted for. Historical Control Data: For contextualizing in vivo study findings.	In Vivo, In Vitro
Validated In Vitro Assay Kits	CALUX (Chemically Activated LUciferase gene eXpression) Assays: Reporter gene assays for specific receptor activity (ER, AR, etc.). ToxTracker Assay: Stem cell-based assay for detecting key toxicity pathways (genotoxicity, oxidative stress).	In Vitro
Computational Toxicology Platforms	OECD QSAR Toolbox: Software for grouping chemicals, filling data gaps via read-across, and applying QSAR models. VEGA QSAR Platform: A collection of transparent and validated QSAR models for toxicity endpoints. Molecular Docking Software (AutoDock, Glide): Predicts binding affinity of chemicals to biological targets.	In Silico
Data Analysis & Visualization Suites	R/Python with meta-analysis packages (metafor, meta): For statistical synthesis. Graphical Tools (Graphviz, Cytoscape): For creating pathway and workflow diagrams.	All (Synthesis)

Case Study Application: Triphenyl Phosphate (TPP) as an Endocrine Disruptor

A 2024 case study applied a modified SYRINA framework to assess the endocrine disrupting potential of the flame retardant triphenyl phosphate (TPP), integrating 66 studies across epidemiological, in vivo, and in vitro streams [62].

• Problem Formulation: The PECO question was structured around mammalian populations exposed to TPP, comparators of lower/no exposure, and outcomes related to endocrine disruption.
• Evidence Stream Synthesis: In vivo rodent studies showed adverse effects on metabolic parameters (increased adiposity, liver steatosis) and reproductive tissues. In vitro assays consistently demonstrated TPP's activity as a peroxisome proliferator-activated receptor gamma (PPARγ) agonist—a key regulator of adipogenesis and metabolism.
• Integration and Conclusion: The reviewers synthesized strong evidence for both adverse metabolic effects (in vivo) and endocrine activity (PPARγ agonism, in vitro). The biological plausibility and coherence between the mechanistic activity and the observed apical outcomes supported a causal link. Consequently, TPP could be identified as an endocrine disruptor based on metabolic disruption [62].
• Challenges Encountered: The study highlighted the significant resource requirement for analyzing in vitro mechanistic data and the need for more harmonized methods for evidence integration and causal link assessment [62].

Future Directions and Confidence in New Evidence Streams

The future of evidence synthesis lies in building scientific confidence in New Approach Methodologies (NAMs). A 2023 NASEM report emphasizes that variability in traditional in vivo studies should not be the sole benchmark for NAM validation [63]. Instead, confidence should be built on:

• Establishing Intrinsic Performance: Assessing within- and between-laboratory repeatability, robustness, and applicability domain of the NAM itself [63].
• Establishing External Validity: Determining the NAM's relevance in predicting protective health outcomes, potentially using a battery of assays to capture broader biological space [63].
• Integration into Risk Assessment Context: Using systematic review methodologies to integrate NAM data with other evidence streams within defined contexts of use (e.g., screening, prioritization, or as part of a WoE for hazard identification) [63].

The ongoing refinement of frameworks like SYRINA and Navigation Guide, coupled with the development of NAM confidence frameworks, will enable more efficient, predictive, and humane chemical safety assessments while enhancing the rigor and transparency of decisions based on diverse evidence streams.

Ensuring Confidence and Impact: Validating Methods and Integrating Reviews into Decision-Making

Within the systematic review and synthesis of ecotoxicity data, methodological validation constitutes the critical framework for ensuring the reliability, reproducibility, and regulatory applicability of derived conclusions. This process formally tests the robustness of models, the consistency of effects across biological and chemical subgroups, and the fundamental soundness of the experimental data underpinning the entire evidence base. In ecotoxicity, systematic reviews are increasingly employed to support ecological risk assessment (ERA) and the development of New Approach Methodologies (NAMs), which aim to reduce reliance on animal testing [39] [65]. Validation is not a single step but an integrated practice encompassing sensitivity analyses to probe model stability, subgroup analyses to detect effect modifiers, and rigorous model checking to verify foundational assumptions. This guide details the technical execution of these validation pillars within the context of contemporary ecotoxicity research, drawing on current datasets, computational models, and regulatory guidelines.

Sensitivity Analyses: Probing Model Robustness and Uncertainty

Sensitivity analyses systematically evaluate how uncertainty in a model's input variables, structure, or assumptions propagates to its outputs. In ecotoxicity modeling, these analyses are essential for establishing confidence in predictions used for chemical prioritization and safety threshold derivation.

Core Concepts and Applications

Sensitivity analysis in this field addresses two primary questions: 1) How do variations in input toxicity data influence final protective benchmarks (e.g., HC-5)? 2) How stable are model predictions to changes in model structure or parameterization? A key application is in Species Sensitivity Distribution (SSD) modeling, which aggregates toxicity data across species to estimate a hazardous concentration for 5% of species (HC-5) [39]. Sensitivity analyses test how the derived HC-5 shifts when using different statistical distributions (e.g., log-normal vs. log-logistic) or when excluding certain taxonomic groups [66].

Quantitative Frameworks and Protocols

Protocol for SSD Model Sensitivity Analysis:

• Model Fitting & Benchmark Derivation: Fit multiple statistical distributions (log-normal, log-logistic, Burr type III, Weibull) to the same curated species toxicity dataset for a chemical. Derive the HC-5 from each fitted distribution [66].
• Comparison Metric: Calculate the ratio of HC-5 values from alternative distributions to the HC-5 from a log-normal distribution, which is often used as a baseline. Document the range of these ratios [66].
• Goodness-of-Fit Assessment: Use information criteria like the corrected Akaike Information Criterion (AICc) to compare the statistical support for each distribution. Visually inspect the fit, particularly in the lower tail (e.g., 5th percentile), which is critical for risk assessment [66].
• Data Perturbation: Conduct bootstrapping or jackknife resampling of the input species toxicity data. Re-fit the preferred model to each resampled dataset to generate a confidence interval for the HC-5, directly quantifying sensitivity to sample composition.

Protocol for ICE Model Extrapolation Uncertainty: Interspecies Correlation Estimation (ICE) models predict a species' sensitivity from a surrogate species' known toxicity. Sensitivity analysis is crucial when predictions require extrapolation.

• Truncated Model Construction: For a given ICE model (e.g., surrogate-predicted species pair), create a truncated version using only the lower 75th percentile of surrogate toxicity data [67].
• Extrapolation Prediction: Use the truncated model to predict toxicity for test data points in the upper 25th percentile. Perform this prediction twice: a) by inputting the high surrogate value in µg/L, allowing standard extrapolation, and b) by "scaling" the input—converting the surrogate value to mg/L before prediction [67].
• Accuracy Evaluation: Compare both types of predictions to the actual measured toxicity value. Accuracy can be evaluated based on whether predictions fall within a pre-defined threshold (e.g., 5-fold difference from measured). This protocol assesses the sensitivity of model accuracy to input scaling and domain extrapolation [67].

Table 1: Key Datasets for Sensitivity Analysis in Ecotoxicity Modeling

Dataset/Model Name	Scope and Size	Primary Use in Sensitivity Analysis	Source/Reference
ECOTOX Database (Curated Subset)	3,250 toxicity entries across 14 taxonomic groups [39]	Input data for SSD modeling; resampling assesses HC-5 sensitivity to data composition.	U.S. EPA; Used in [39]
Web-ICE v4.0 Database	10,645 acute toxicity values for 476 species, 1,708 chemicals [67]	Underpins ICE models; analysis of inter-test variability provides a baseline for prediction accuracy.	U.S. EPA [67]
EPA CDR Chemical List	~8,449 industrial chemicals [39]	Application set for validated models; sensitivity of chemical prioritization lists can be tested.	U.S. EPA [39]
EnviroTox Database	Acute/chronic data for 191/31 chemicals [66]	Comparative testing of different statistical distributions for SSD derivation.	HESI [66]

Diagram 1: Generalized workflow for sensitivity analysis of Species Sensitivity Distribution (SSD) models.

Subgroup Analyses: Investigating Heterogeneity and Effect Modification

Subgroup analyses examine whether the effect of a stressor (e.g., a chemical's toxicity) differs across predefined subsets of the data. In ecotoxicity systematic reviews, this is vital for identifying moderator variables such as taxonomic class, trophic level, or chemical mode of action, which can explain heterogeneity in reported effects and refine risk assessments [68].

Rationale and Typology

The primary goal is to distinguish between a consistent effect across all organisms and one that is modified by specific biological or experimental factors. Analyses can be exploratory (hypothesis-generating) or confirmatory (hypothesis-testing) [68]. For example, an exploratory analysis may screen multiple taxonomic groups for differential sensitivity, while a confirmatory analysis might specifically test whether arthropods are more sensitive than fish to a particular insecticide class.

Statistical Methodology

The correct approach to detect effect modification is a test for interaction within a statistical model, not separate comparisons within each subgroup [68].

Protocol for Subgroup Analysis via Interaction Testing:

• Define Subgroup Variable (Moderator): Specify a categorical (e.g., trophic level: producer, primary consumer) or continuous (e.g., species mean body mass) baseline factor.
• Select and Specify Model: Choose a model appropriate for the endpoint. For continuous data (e.g., LC50 values), use a linear model. For binary outcomes (e.g., significant effect yes/no in a study), use a logistic model. For time-to-event data (e.g., time to mortality), use a Cox proportional hazards model [68].
• Model Formulation: Include main effect terms for the treatment/stressor and the subgroup variable, plus their interaction term.
  • Linear Model: Y = α + β₁X + β₂Z + β₃(X*Z) + ε, where X is treatment, Z is subgroup, and X*Z is the interaction. A significant β₃ indicates effect modification [68].
  • Logistic Model: logit(p) = α + β₁X + β₂Z + β₃(X*Z), where p is the probability of an event. A significant β₃ indicates the odds ratio of the treatment effect differs by subgroup [68].
• Interpretation: If the interaction term is statistically significant, present effect estimates (e.g., HC-5 differences, odds ratios) for each subgroup separately. Account for multiple testing (e.g., using Bonferroni correction) when evaluating several subgroup hypotheses to control the false discovery rate [68].

Table 2: Common Subgroup Variables and Analysis Methods in Ecotoxicity

Subgroup Variable (Moderator)	Typical Categories/Scale	Analysis Goal	Suggested Statistical Model
Taxonomic Group	Fish, Crustacea, Algae, Insects, etc.	Identify taxonomic groups with differential sensitivity.	Linear model (log-transformed EC/LC50) with interaction term.
Trophic Level	Producer, Primary Consumer, Secondary Consumer, Decomposer [39]	Determine if toxicity potency varies by ecological function.	Linear or logistic model with interaction term.
Exposure Duration	Acute (short-term), Chronic (long-term)	Evaluate if toxicity thresholds shift with exposure time.	TKTD model (e.g., GUTS) or separate meta-regression for acute vs. chronic data.
Chemical Class	Pesticides, Metals, PFAS, Personal Care Products [39]	Assess consistency of a model's performance across chemical types.	Stratified model fitting or meta-regression with chemical class as covariate.
Endpoint Type	Mortality (LC50), Growth (NOEC), Reproduction (LOEC)	Understand how the choice of endpoint influences hazard rankings.	Multivariate model or comparative analysis of SSDs built on different endpoints.

Diagram 2: Decision logic for conducting subgroup analyses in ecotoxicity reviews.

Model Checking: Validating Assumptions and Diagnostic Testing

Model checking involves verifying the fundamental assumptions of both the primary toxicity tests that generate data and the secondary statistical/computational models used to synthesize them. This step ensures that results represent consistent and comparable measures of relative toxicity [69].

Checking Primary Test Validity: The LC50 Example

The acute aquatic toxicity test (e.g., 96-hr LC50) is a model experiment relying on key assumptions. Protocol for Quality Control (QC) Review of LC50 Test Validity [69]:

• Assumption 1 - Steady-State Attainment: Check if a time-independent (incipient) LC50 can be estimated from the data. Tests where mortality does not plateau with exposure time fail this assumption. A QC study found approximately 8% of tests in a database failed this check [69].
• Assumption 2 - Equivalent Exposure Duration: Confirm that the LC50 is estimated at equivalent time points for all concentrations and tests being compared.
• Assumption 3 - Control of Toxicity Modifying Factors (TMFs): Evaluate whether factors like water chemistry (pH, hardness), organism life stage, and feeding status are reported and controlled. Unexplained variance often indicates TMFs are not fully controlled, compromising comparability [69].

Checking Secondary Model Validity

Toxicokinetic-Toxicodynamic (TKTD) Model Checking: Models like the General Unified Threshold model of Survival (GUTS) require checking fit and assumptions.

• Calibration Fit: Visually and statistically assess the model's fit to time-course survival data across multiple concentrations.
• Posterior Predictive Checks (in Bayesian framework): Simulate new data using the calibrated model's posterior parameter distributions. Compare simulated datasets to observed data to detect systematic misfit [70].
• Residual Analysis: Examine residuals for patterns over time or concentration to identify violations of model assumptions.

Protocol for SSD Model Diagnostic Checking [66]:

• Goodness-of-Fit Tests: Use Kolmogorov-Smirnov or Anderson-Darling tests to assess the fit of the chosen distribution to the empirical species sensitivity data.
• Q-Q Plot Inspection: Visually assess the quantile-quantile plot for deviations from linearity, particularly in the lower tail.
• Influence Analysis: Identify if the HC-5 is unduly influenced by a single or few sensitive species by recalculating the model after removing them.

Table 3: Model Checking Diagnostics and Interpretation

Model Type	Key Assumption	Diagnostic Check	Interpretation of Problem
Primary LC50 Test [69]	Time-independent steady-state LC50 is reached.	Plot % mortality vs. time for each concentration.	If mortality curves do not plateau, LC50 is time-dependent, invalidating comparison.
ICE Models [67]	Linear relationship of log-sensitivity is conserved across chemicals.	Examine residuals of the log-log regression for patterns.	Non-random residuals may indicate the model is not applicable for certain chemical classes.
SSD Models [66]	Species sensitivities are a random sample from a statistical distribution.	Goodness-of-fit test (e.g., K-S test); Q-Q plot.	Poor fit suggests the chosen distribution is inappropriate, casting doubt on HC-5.
TKTD/GUTS Models [70]	The dominant rate constant (k_d) is constant across individuals and concentrations.	Check parameter identifiability; perform posterior predictive checks.	Poor predictive checks indicate a structural model flaw or violated assumption.

Table 4: Research Reagent Solutions for Ecotoxicity Method Validation

Tool/Resource	Function in Validation	Key Features / Notes
OpenTox SSDM Platform [39]	Provides interactive tools for building, fitting, and performing sensitivity analyses on Species Sensitivity Distribution models.	Hosts datasets and model architectures from published studies; enables transparent HC-5 derivation and scenario testing.
U.S. EPA Web-ICE Application [67]	Facilitates Interspecies Correlation Estimation and includes functionality for uncertainty analysis of predictions, especially for extrapolations.	Contains v4.0 database; allows users to assess confidence intervals and apply updated guidance for low-toxicity compounds.
EnviroTox Database [66]	Supplies curated, quality-controlled acute and chronic aquatic toxicity data for building and validating SSDs.	Critical for model checking (comparing distributions) and subgroup analysis (data stratified by taxonomy/chemical).
OECD Test Guidelines (e.g., TG 203, 236, 254) [71]	Define standardized experimental protocols for generating primary toxicity data, forming the basis for assumption checking.	2025 updates incorporate omics sampling and refine endpoints; adherence is a prerequisite for data validity in reviews.
R package morse [70]	Implements Toxicokinetic-Toxicodynamic (TKTD) models, including GUTS, within a Bayesian framework for survival data analysis.	Enables model calibration, posterior predictive checks, and propagation of parameter uncertainty to predictions like LC(x,t).
ICCVAM Validation Guidelines [65]	Outline a modern framework for the validation, qualification, and regulatory acceptance of New Approach Methodologies (NAMs).	Provides authoritative criteria for establishing the scientific credibility and reliability of alternative methods and models.

Within the domain of systematic review methods for ecotoxicity data research, evidence synthesis represents a critical analytical frontier. Researchers and regulators are tasked with integrating disparate, heterogeneous studies—ranging from laboratory dose-response experiments to field observations—to form coherent conclusions about chemical hazards and environmental risks. This technical guide provides an in-depth analysis and comparison of three sophisticated synthesis methodologies: Meta-Regression, Bayesian Approaches, and Weight-of-Evidence (WoE) frameworks. The evolution of these methods is particularly timely, coinciding with a concerted push to modernize statistical practices in ecotoxicology, as highlighted by ongoing revisions to key guidance documents like OECD No. 54 [2].

The central thesis is that the choice of synthesis method must be guided by the structure of the available evidence, the nature of the research question, and the required output for decision-making. Traditional pairwise meta-analysis often falters in ecotoxicology due to few direct comparisons between similar treatments or stressors. Conversely, advanced methods like Network Meta-Analysis (NMA) and Bayesian Hierarchical Models can leverage indirect evidence and formally incorporate uncertainty, which is paramount for ecological risk assessment where data is often sparse and variable [72] [73]. Concurrently, the WoE approach provides a structured qualitative-quantitative framework essential for integrating different lines of evidence (e.g., in vitro, in vivo, in silico) and dealing with entrenched challenges in using academic research in regulatory settings [4].

Foundational Principles and Mathematical Frameworks

Meta-Regression

Meta-Regression extends standard meta-analysis by modeling heterogeneity in effect sizes (e.g., Hedges' g, Log Odds Ratio) as a function of study-level covariates (e.g., particle size, exposure duration, organism class). Its goal is to explain why effect sizes vary across studies.

A basic random-effects meta-regression model is specified as: yᵢ = θᵢ + εᵢ, where θᵢ = β₀ + β₁xᵢ₁ + … + βₚxᵢₚ + uᵢ. Here, yᵢ is the observed effect size in study i, θᵢ is the true effect size, and εᵢ is the within-study sampling error (~N(0, sᵢ²)). The true effect is modeled with a linear combination of covariates xᵢ with coefficients β, and residual heterogeneity uᵢ (~N(0, τ²)) [74]. A Hierarchical Meta-Regression (HMR) further allows the relationship between covariates and effect size to vary across study subgroups, offering a flexible way to assess bias or integrate disparate evidence types [75].

Bayesian Approaches

Bayesian synthesis methods treat all unknown parameters (e.g., true effect sizes, heterogeneity) as probability distributions. They combine prior knowledge (the prior) with the observed data (the likelihood) to form updated knowledge (the posterior distribution).

A foundational model for Bayesian random-effects meta-analysis is: yᵢ | θᵢ, σᵢ² ~ N(θᵢ, σᵢ²) θᵢ | μ, τ² ~ N(μ, τ²) μ ~ N(0, 1000), τ ~ Half-Cauchy(0, 5) The key output is the posterior distribution for the pooled effect μ and heterogeneity τ, which allows for direct probability statements (e.g., "the probability that the true hazard quotient exceeds 1 is 95%") [73].

Bayesian Network Meta-Analysis (NMA) is a powerful extension for comparing multiple treatments (A, B, C, D) simultaneously, even when not all have been directly compared in head-to-head trials. It synthesizes direct (A vs. B) and indirect (A vs. C, B vs. C → A vs. B) evidence within a single coherent model. Consistency between direct and indirect evidence is a critical assumption that must be checked, for instance, via node-splitting methods [72].

Weight-of-Evidence (WoE)

WoE is a structured, transparent framework for integrating and interpreting multiple, sometimes conflicting, lines of evidence to answer a specific question (e.g., "Is chemical X a persistent, bioaccumulative, and toxic substance?"). Unlike purely quantitative meta-methods, WoE incorporates both quantitative results and qualitative assessments of study reliability, relevance, and consistency.

A standard WoE assessment involves:

• Evidence Collection: Assembling all relevant data (e.g., experimental ecotoxicity, QSAR predictions, field monitoring).
• Evidence Weighting: Critically evaluating each piece of evidence based on pre-defined criteria (e.g., following Klimisch scores or similar reliability assessments).
• Evidence Integration: Combining the weighted evidence to determine the overall direction, strength, and uncertainty of the conclusion. This often employs logic or causal pathway diagrams to visualize relationships [4].

Methodological Protocols and Experimental Workflows

Protocol for Hierarchical Meta-Regression in Ecotoxicity

This protocol is designed to analyze drivers of heterogeneity in ecotoxicity endpoints, such as the effect of biodegradable microplastics (BMPs) on aquatic organism growth [74].

• Effect Size Calculation: For each study i, calculate a standardized effect size (e.g., Hedges' g) and its variance (sᵢ²) for the chosen endpoint (e.g., growth inhibition).
• Covariate Extraction: Systematically extract potential moderators from each study: Polymer type (PLA, PHA, PBS), Particle size (nm), Exposure concentration (mg/L), Organism taxon (Fish, Crustacean, Mollusk), and Test duration.
• Model Specification & Fitting: Fit a random-effects meta-regression model in a statistical environment like R using the metafor or brms package. For example: rma.mv(yi = Hedges_g, vi = Variance, mods = ~ Polymer_Type + Particle_Size + Taxon, random = ~ 1 | Study_ID, data = dataset).
• Heterogeneity Assessment: Quantify total heterogeneity (I² statistic), and partition it into variance explained by the covariates (R²) and residual unexplained variance (τ²).
• Sensitivity & Subgroup Analysis: Conduct leave-one-out analyses and subgroup analyses by major taxonomic groups to check robustness.

Protocol for Bayesian Network Meta-Analysis

This protocol, adapted from an analysis of alcoholism treatments, is applicable for comparing the relative toxicity of multiple nanomaterials or chemical alternatives where a connected network of studies exists [72].

• Network Geometry Definition: Map all direct comparisons from included studies into a network diagram. Nodes represent stressors/treatments (e.g., AgNPs, TiO₂ NPs, ZnO NPs, Control). Edges represent direct comparisons with the number of studies.
• Model Implementation in WinBUGS/Stan: Code the Bayesian NMA model. For a binary outcome (e.g., mortality), a logistic model is used: logit(pᵢₖ) = μᵢ + δᵢₖ * I(k≠b), where pᵢₖ is the event probability in study i, arm k; μᵢ is the study-specific baseline (for reference treatment b); and δᵢₖ is the log-odds ratio for treatment k vs. b in study i, with δᵢₖ ~ N(d_{bk}, σ²). The d_{bk} are the pooled treatment effects of interest.
• Consistency Check: Perform node-splitting to assess inconsistency between direct and indirect evidence for each comparison.
• Prior Specification & MCMC Simulation: Use vague priors (e.g., d_{bk} ~ N(0, 10000), σ ~ Unif(0, 5)) and run Markov Chain Monte Carlo (MCMC) simulations (e.g., 50,000 iterations after 50,000 burn-in) [72] [73].
• Ranking & Output: Calculate the posterior probability and surface under the cumulative ranking curve (SUCRA) for each treatment being the most toxic/least toxic.

Protocol for Weight-of-Evidence Assessment

This protocol is based on frameworks for integrating academic research into regulatory chemical assessment [4].

• Problem Formulation & Pathway Development: Define the assessment question. Develop a conceptual model (e.g., an adverse outcome pathway for genotoxicity) identifying key events and measurable endpoints.
• Evidence Gathering & Triaging: Collect evidence from standardized tests, academic literature, (Q)SAR models, and high-throughput screening. Triage evidence based on pre-defined criteria for reliability (e.g., test guideline compliance, reporting quality) and relevance (e.g., taxonomic, mechanistic).
• Evidence Evaluation & Weighting: Assign a weight (e.g., high, medium, low, unassignable) to each study or data point. Document the rationale transparently.
• Evidence Integration: Synthesize findings across weighted evidence streams. Assess the coherence (do different lines of evidence tell a biologically plausible story?), consistency (are findings across studies in agreement?), and adequacy (is the body of evidence sufficient for a conclusion?).
• Conclusion and Uncertainty Characterization: State the overall conclusion regarding hazard or risk. Explicitly characterize the confidence in the conclusion and the major sources of uncertainty (e.g., data gaps, conflicting results, applicability domain issues).

Table 1: Comparative Summary of Synthesis Method Protocols

Aspect	Meta-Regression	Bayesian NMA	Weight-of-Evidence
Primary Objective	Explain heterogeneity; estimate effect of continuous/categorical moderators.	Estimate relative effects/rankings across multiple treatments using direct + indirect evidence.	Integrate diverse evidence types for a holistic, transparent hazard assessment.
Core Input	Effect sizes + study-level covariate data.	Relative effect measures (OR, RR, MD) from a network of comparisons.	All relevant data, quantitative and qualitative, with quality scores.
Key Analytical Step	Modeling effect size as a function of covariates.	Checking consistency assumption; running MCMC sampling.	Critical appraisal of reliability/relevance; assessing coherence/consistency.
Typical Software	R (metafor, brms), STATA.	WinBUGS/OpenBUGS, JAGS, Stan (via gemtc, brms).	Systematic review software (e.g., HAWC, DistillerSR); graphical tools.
Main Output	Estimated coefficients (β) for covariates; proportion of heterogeneity explained.	Posterior distributions for all pairwise comparisons & treatment rankings (SUCRA).	A narrative conclusion with an assigned confidence level and identified uncertainties.

Workflow and Logical Pathway Visualization

Decision Workflow for Selecting an Evidence Synthesis Method

Hierarchical Structure of a Bayesian Network Meta-Analysis Model [72]

Table 2: Quantitative Performance Comparison Based on Case Studies

Method	Case Study Context	Key Quantitative Finding	Handling of Heterogeneity/Uncertainty
Meta-Regression	Ecotoxicity of Biodegradable Microplastics (BMPs) [74]	BMPs significantly impaired behavior (Hedges' g = -2.358), reproduction (g = -1.821), and growth (g = -0.864). Polymer type (PBS, PHB) and particle size were significant moderators.	Quantified high heterogeneity (I²); used subgroup analysis and meta-regression to explain variance from polymer type and particle size.
Bayesian NMA	Pharmacological Treatments for Alcohol Dependence [72]	Combination therapy had the highest posterior probability of being best. For poorly compared treatments, NMA provided more precise estimates (narrower CrIs) than pairwise meta-analysis.	Model incorporated between-study variance (τ²). Bayesian framework provided full posterior credible intervals (CrIs) for all estimates, directly quantifying uncertainty.
Bayesian Meta-Analysis	PTSD and Suicide Risk [73]	Pooled relative risk (RR) of suicide for individuals with PTSD was 1.68 (95% CrI: 1.21, 2.32). The probability that RR > 1 was 99.8%.	Used weakly informative priors; accurately estimated heterogeneity from a small number of studies (k=6); presented uncertainty as a probability.
Weight-of-Evidence	Use of Academic Research in Regulation [4]	Survey identified technical (reliability, transparency) and social (misaligned goals) factors as interdependent barriers to evidence uptake.	Frameworks used to characterize and integrate evidence of different reliability levels, explicitly addressing uncertainty from evidence conflicts and gaps.

Application in Ecotoxicity Research: Synthesis and Recommendations

The integration of these advanced synthesis methods is pivotal for modern ecotoxicity research. For instance, a Bayesian approach is exceptionally suited for dose-response meta-analysis, allowing the incorporation of prior knowledge about model parameters and providing full probability distributions for benchmark doses (BMDs) or ECx values, aligning with calls for more sophisticated statistical practices [2]. Meta-regression is indispensable for disentangling the effects of nanomaterial properties (e.g., size, coating, zeta potential) on observed toxicity, as seen in studies of multicomponent nanomaterials (MCNMs) [76]. WoE is the cornerstone for safety assessment of emerging materials like green-synthesized nanoparticles, where traditional toxicity data may be limited, but evidence from genotoxicity, eco-corona formation, and environmental fate studies can be integrated [77] [78].

Table 3: The Scientist's Toolkit for Advanced Evidence Synthesis

Tool/Resource	Primary Function	Application Notes
R Statistical Environment [2]	Open-source platform for statistical computing and graphics.	Core hub for analysis. Essential packages include metafor (meta-analysis/regression), brms/rstanarm (Bayesian modeling), and dose-response analysis packages [2].
Stan (via brms or rstan) [73]	Probabilistic programming language for Bayesian inference.	Preferred for complex Bayesian models (NMA, hierarchical models) due to its powerful sampling algorithms (Hamiltonian Monte Carlo).
WinBUGS/OpenBUGS [72]	Software for Bayesian analysis using MCMC methods.	Historically used for Bayesian NMA; code examples available in literature [72]. Still relevant, though Stan is often preferred for newer projects.
PRISMA Guidelines [74]	Evidence-based minimum set of items for reporting systematic reviews and meta-analyses.	Critical for ensuring transparency and reproducibility in the search, screening, and synthesis process.
OHAT / WHO WoE Framework	Structured templates for weight-of-evidence assessments.	Provides a standardized methodology for transparently evaluating and integrating evidence for hazard identification.
Graphical Tools (e.g., Graphviz)	Visualization of complex networks, workflows, and conceptual models.	Crucial for communicating network geometries in NMA, WoE pathways, and decision workflows as mandated in this analysis.

Recommendations for Practice:

• For explaining variation in ecotoxicity endpoints across studies (e.g., why toxicity of AgNPs varies), use Meta-Regression.
• For comparing relative toxicity of multiple chemicals or material types within a connected evidence network, use Bayesian NMA. It is particularly valuable when direct comparisons are scarce.
• For integrating different evidence types (experimental, computational, field) for a comprehensive hazard assessment, or when data is insufficient for quantitative pooling, use a structured WoE framework.
• Embrace Bayesian methods for their natural handling of uncertainty, ability to incorporate prior knowledge (e.g., from QSAR or read-across), and utility in small-study settings common in ecotoxicology [2] [73].
• Invest in training and collaboration with statisticians to implement these methods correctly and move beyond outdated statistical practices [2].

The landscape of evidence synthesis in ecotoxicity is evolving from simple pairwise averaging toward sophisticated, model-based integrations that respect the complexity and heterogeneity of ecological data. Meta-regression, Bayesian approaches (especially NMA), and Weight-of-Evidence frameworks are not mutually exclusive but are complementary tools in the modern systematic reviewer's arsenal. Their judicious application, guided by the specific research question and evidence structure, will lead to more robust, transparent, and informative conclusions for environmental risk assessment and the development of safe-by-design chemicals and materials.

The regulatory assessment of chemicals, pharmaceuticals, and environmental contaminants is undergoing a foundational shift. There is a strong global push to adopt New Approach Methodologies (NAMs) that enhance mechanistic understanding and reduce reliance on traditional animal studies [79]. Concurrently, regulatory frameworks such as the EU's Biocidal Products Regulation (BPR) and Plant Protection Products (PPP) regulation are evolving, with increasing focus on complex endpoints like endocrine disruption and pollinator protection [80]. This evolution creates a critical gap: a deluge of novel, often heterogeneous data from in vitro, in silico, and high-throughput sources must be reliably synthesized and evaluated for regulatory decision-making.

Systematic review (SR) methodology, long established as the pinnacle of evidence synthesis in medical research, provides the essential toolkit to bridge this gap [81] [82]. An SR uses explicit, pre-defined, and reproducible methods to identify, select, appraise, and synthesize all available evidence on a specific question [83]. Within the broader thesis on systematic review methods for ecotoxicity data research, this whitepaper posits that the rigorous application and adaptation of SR principles are non-negotiable for integrating innovative research findings into credible, transparent, and defensible regulatory frameworks. It provides a technical guide for researchers and assessors on conducting SRs for ecotoxicological data and implementing their findings within modern, tiered risk assessment paradigms.

Core Methodology: Conducting a Systematic Review for Ecotoxicity Evidence

A high-quality SR is a structured, multi-stage process designed to minimize bias and maximize reliability. The following steps, adapted for ecotoxicological and regulatory science contexts, are essential.

• Protocol Development and Registration: The process begins with a publicly accessible protocol detailing the research question, search strategy, and inclusion criteria. Registries like PROSPERO are used to prevent duplication and promote transparency [82]. A well-formulated research question is critical. While the medical field often uses PICO/PICOT frameworks, ecotoxicity questions may better suit adapted formats (e.g., Population/Test System, Exposure, Comparator, Outcome) [81] [82].

• Systematic Search and Study Selection: A comprehensive, reproducible search is conducted across multiple bibliographic databases (e.g., PubMed, Embase, Web of Science) and "grey literature" sources to mitigate publication bias [81]. Tools like Covidence or Rayyan streamline the management of references and the dual-independent screening of titles/abstracts and full texts against pre-defined eligibility criteria [81].

• Critical Appraisal and Data Extraction: This is a pivotal phase for regulatory application. The reliability of each included study must be assessed using a pre-specified critical appraisal tool (CAT). For ecotoxicity studies, the Ecotoxicological Study Reliability (EcoSR) framework is a novel two-tiered tool designed to evaluate internal validity (risk of bias) and reliability specific to ecotoxicological testing [45].

  • Tier 1 is an optional screening for manifestly unreliable studies.
  • Tier 2 is a full assessment across core domains: test substance characterization, test system/design, exposure characterization, endpoint measurement and analysis, and reporting completeness [45].

Standardized data extraction forms are then used to collect quantitative results, study characteristics, and key methodological details from studies deemed sufficiently reliable.

• Evidence Synthesis and Integration: Extracted data are synthesized qualitatively (narrative synthesis) or quantitatively (meta-analysis). In meta-analysis, statistical models combine effect sizes (e.g., log-transformed LC50 values) to generate summary estimates of toxicity, while assessing heterogeneity (I² statistic) [81]. For NAMs and mechanistic data, synthesis may involve integrative review techniques to generate new frameworks or perspectives by combining diverse study types [83]. The overall certainty of evidence across studies for a given endpoint is evaluated using structured approaches like GRADE [82].

Table 1: Key Stages of a Systematic Review for Regulatory Ecotoxicology

Stage	Core Objective	Key Tools & Outputs	Regulatory Utility
Protocol	Define scope & methods a priori	PROSPERO registration, PRISMA-P checklist [82]	Ensures transparency, prevents bias, aligns stakeholders.
Search/Selection	Identify ALL relevant evidence	Multi-database search, Grey literature, PRISMA flow diagram	Minimizes selection bias; provides auditable evidence base.
Critical Appraisal	Evaluate study reliability & bias	EcoSR Framework [45], Other CATs	Filters data for "fit-for-purpose" studies; informs weight-of-evidence.
Data Extraction	Systematically collect data	Customized extraction forms, Dual review	Creates structured, analyzable dataset from heterogeneous sources.
Synthesis	Integrate findings	Narrative synthesis, Meta-analysis, Integrative review [83]	Generates summary estimates, identifies patterns/knowledge gaps.

Frameworks for Integration: From Systematic Evidence to Regulatory Decision

The output of an SR is a synthesized body of evidence with an associated certainty rating. Integrating this into risk assessment requires structured frameworks. Two advanced, complementary models are presented here.

1. The Tiered Next-Generation Risk Assessment (NGRA) Framework: This framework integrates toxicokinetics (TK) and NAM-based toxicodynamics (TD) in a tiered, hypothesis-driven approach [84]. SR methods are crucial at each tier to gather and evaluate the underlying evidence.

Diagram 1: Tiered NGRA Framework Workflow (100 chars)

2. The Conceptual Framework for NAM Integration in Environmental Safety: This model leverages mechanistic data from NAMs within a weight-of-evidence approach [79]. It starts with collecting all relevant in vivo and in vitro effect data via an SR. The core integration step involves determining the "biological relevance" and "ecological conservation" of mechanistic pathways—identifying where targets and outcomes are conserved across species and are thus predictive of apical effects [79]. This builds confidence for using NAM data to identify sensitive species and derive Points of Departure (PODs), supplementing or replacing traditional animal data.

Table 2: Comparison of Advanced Risk Assessment Integration Frameworks

Framework Feature	Tiered NGRA Framework [84]	NAM Integration Framework [79]
Primary Application	Combined chemical exposure risk (e.g., pyrethroids).	Environmental safety assessment for chemicals.
Core Integration Strategy	Iterative tiering of TK modeling & NAM-based TD.	Weight-of-evidence on evolutionary conservation of mechanistic targets.
Role of SR	Provides synthesized bioactivity (ToxCast) & regulatory (ADI) data for each tier.	Synthesizes all available in vivo & in vitro data across species/life stages.
Key Output	Bioactivity-based Margin of Exposure (MoE).	Mechanistically informed Points of Departure (PODs).
Regulatory Goal	Refine cumulative risk assessment for related substances.	Enable safety decisions without new animal studies.

Case Study Application: Pyrethroid Cumulative Risk Assessment

A 2025 case study on pyrethroid insecticides demonstrates the practical application of an SR within the tiered NGRA framework [84]. The process mirrors the workflow in Diagram 1.

• Tiers 1 & 2 (Evidence Synthesis & Hypothesis Testing): An SR gathered in vitro bioactivity data (AC50 values) from the ToxCast database and regulatory toxicity values (NOAELs, ADIs) for six major pyrethroids. A quantitative synthesis of relative potencies was performed. The analysis rejected the hypothesis of a common single mode of action across all pyrethroids, demonstrating the need for a cumulative assessment based on specific bioactivity indicators [84].

• Tiers 3 & 4 (Integrated TK-TD Modeling): Realistic dietary exposure estimates were synthesized from EFSA reports. Physiologically Based Kinetic (PBK) modeling (using PK-Sim) simulated internal plasma and tissue concentrations. This step crucially bridged the external dose to internal bioactivity, allowing comparison between in vitro bioactivity thresholds and in vivo exposure scenarios [84].

• Tier 5 (Risk Characterization): The final integrated analysis calculated a bioactivity-based Margin of Exposure (MoE) by comparing the internal concentrations at realistic exposure levels to the bioactivity concentrations from the in vitro data. While dietary exposure alone was near thresholds of concern, the SR-informed framework clearly identified that adding non-dietary exposures (e.g., biocidal use) could lead to insufficient MoEs [84]. This provides regulators with a nuanced, mechanistically informed view of cumulative risk impossible to derive from single-substance assessments.

Implementation Protocols: Embedding SRs into Regulatory Practice

For systematic reviews to effectively bridge the research-practice gap, standardized protocols for their conduct and evaluation within agencies are essential.

Protocol 1: Conducting an EcoSR Assessment for Study Evaluation

• Objective: To consistently evaluate the reliability of individual ecotoxicity studies for use in toxicity value derivation [45].
• Procedure:
  • Pre-Customization: Define assessment goals (e.g., screening vs. high-confidence POD derivation) and customize the EcoSR Tier 2 criteria weighting if necessary [45].
  • Tier 1 Screening (Optional): Apply rapid exclusion criteria (e.g., lacking concurrent control, unacceptable test substance purity).
  • Tier 2 Full Assessment: Two independent reviewers assess the study across five domains using predefined signaling questions. Disagreements are resolved by consensus or a third reviewer.
  • Judgment and Documentation: Summarize judgments per domain and an overall reliability classification (e.g., High, Medium, Low, Unreliable). The rationale must be documented [45].

Protocol 2: Systematic Review for NAM Credibility in a Defined Context of Use (COU)

• Objective: To synthesize evidence supporting the validity of a NAM (e.g., a specific in vitro pathway assay) for predicting a defined apical in vivo outcome within a regulatory COU.
• Procedure:
  • Define COU & Review Question: Precisely specify the chemical domain, target species/life stage, and ecological endpoint the NAM is intended to predict.
  • Dual-Stream Search: Systematically search for (A) all studies on the NAM's performance (mechanistic relevance, intra-/inter-laboratory reproducibility) and (B) all relevant in vivo studies for the specified endpoint.
  • Evidence Mapping & Correlation Analysis: Extract data on NAM response levels and corresponding in vivo outcomes. Use quantitative or qualitative methods to assess the correlation strength and consistency.
  • Assessment of Biological Relevance: Apply criteria, as in the NAM Integration Framework [79], to evaluate the evolutionary conservation of the molecular initiating event and key events in the pathway.
  • Certainty of Evidence Rating: Use a modified GRADE approach to rate the certainty that the NAM output is a reliable predictor for the specified COU.

Table 3: The Ecotoxicologist's Regulatory Integration Toolkit

Tool/Reagent Category	Specific Tool/Platform	Function in Bridging the Gap	Key Consideration
Evidence Synthesis Software	Covidence, Rayyan, SysRev	Manages screening, data extraction, and critical appraisal for SR teams [81].	Cloud-based platforms enable collaborative review among regulators and researchers.
Toxicity Databases	EPA CompTox Dashboard, ECOTOX	Provide structured, curated in vivo toxicity data for extraction and synthesis.	Data quality and curation protocols vary; critical appraisal remains essential.
NAM Data Sources	ToxCast/Tox21 database, PubChem	Supply high-throughput in vitro bioactivity data for hypothesis generation and integration [84].	Requires careful interpretation of assay relevance and translation to in vivo scenarios.
Toxicokinetic Modeling Tools	PK-Sim, GastroPlus	Simulate internal dose metrics to bridge between in vitro bioactivity and in vivo exposure [84].	Model credibility must be established per FDA/EMA guidance for regulatory use.
Risk Assessment Platforms	SEURAT, OECD QSAR Toolbox	Provide workflows to integrate (Q)SAR, read-across, and in vitro data for regulatory endpoints.	SR methodology is needed to build and evaluate the underlying evidence bases for these tools.

Integrating systematic review methodologies into regulatory ecotoxicology is no longer a theoretical ideal but a practical necessity. The frameworks and protocols outlined provide a roadmap for using SRs to create a transparent, auditable evidence base, critically appraise novel data streams like NAMs, and implement integrated assessment strategies like tiered NGRA. This closes the research-practice gap by ensuring that regulatory decisions are built on the best available synthesized science, not on selective or haphazard evidence.

The future direction of this field will involve deeper integration of artificial intelligence and machine learning (AI/ML) with SR processes. The FDA's 2025 draft guidance on AI in drug development provides an initial risk-based framework for evaluating AI model credibility [85], a approach applicable to ecotoxicological models. Furthermore, lessons from the FDA's Drug Development Tools Qualification Program, which establishes qualified methods for regulatory use [86], can inform the creation of similar pathways for qualifying SR-derived evidence integration frameworks and standardized NAM batteries in environmental regulation. The ultimate goal is a responsive, evidence-driven regulatory system that efficiently incorporates scientific innovation to protect human health and the environment.

Evaluating the Impact of Systematic Reviews on Chemical Assessment and Management Policies

Within the broader thesis on advancing systematic review methods for ecotoxicity data research, this whitepaper examines the transformative role of structured evidence synthesis in chemical policy. The core premise is that systematic reviews (SRs) introduce a critical level of transparency, objectivity, and consistency to the evaluation of ecotoxicity data, thereby directly enhancing the robustness of chemical assessment and management decisions[reference:0]. While SR methodologies are well-established in clinical medicine, their adoption in environmental health and chemical risk assessment (CRA) represents a pivotal "step change" for the field[reference:1]. This guide details the quantitative impact, provides actionable experimental protocols, and outlines the essential toolkit for researchers and regulatory professionals aiming to bridge the gap between academic ecotoxicity research and evidence-based policy.

Quantitative Impact: Data on Uptake and Output

The application of SRs in chemical assessment is growing, as demonstrated by case studies and stakeholder analyses. The data below summarize key quantitative findings regarding the output of SRs and the persistent challenges in integrating academic research into regulatory frameworks.

Table 1: Output Metrics from a Recent Systematic Review Case Study (SYRINA Framework)

Metric	Value	Detail
Total Studies Included	66	Comprised in vivo, in vitro, and epidemiological data on triphenyl phosphate[reference:2].
Evidence Synthesis Outcome	Positive Identification	Concluded the chemical could be identified as an endocrine disruptor based on metabolic and reproductive effects[reference:3].
Primary Challenge Noted	High Resource Demand	Significant time and effort were required for the analysis of in vitro mechanistic data, increasing workload[reference:4].

Table 2: Perceived Barriers to Using Academic Research in Chemical Assessment (European Context)

Barrier Category	Key Findings	Source
Usage Gap	Academic peer-reviewed studies are "rarely used in regulatory and policy decision making" in practice, despite legal mandates to consider all relevant evidence[reference:5].	SETAC Survey Description (2024)
Stakeholder Divergence	Survey respondents are "deeply divided on the extent to which chemical assessment makes use of available and relevant evidence," highlighting a systemic perception issue[reference:6].	Jones et al. (2025)
Technical & Social Factors	Barriers are interdependent, linking technical issues (e.g., non-standard tests) with social factors like misaligned goals between academic and regulatory knowledge production[reference:7].	Jones et al. (2025)

Experimental Protocols: Core Methodologies for Conducting Systematic Reviews

Implementing an SR for ecotoxicity data requires a rigorous, protocol-driven approach. The following methodologies are considered gold standards.

The SYRINA Framework for Endocrine Disruptor Assessment

This framework, applied in the case study above, provides a structured 7-step process[reference:8]:

• Articulate the Research Objective: Formulate a precise question using the PECO (Population, Exposure, Comparator, Outcome) statement.
• Develop & Execute a Literature Search Protocol: Define search strings, databases (e.g., PubMed, Scopus), and inclusion/exclusion criteria a priori.
• Screen Literature: Perform title/abstract and full-text screening, typically by two independent reviewers to minimize bias.
• Evaluate Study Quality & Risk of Bias: Appraise each included study using standardized tools (e.g., OHAT, SYRCLE) for internal validity.
• Extract Data: Systematically collect relevant data on study design, exposure, outcomes, and results into a pre-defined form.
• Synthesize Evidence: Qualitatively integrate findings across studies. Perform meta-analysis if data are sufficiently homogeneous.
• Integrate & Report Evidence: Weigh the body of evidence, draw conclusions, and report transparently, acknowledging uncertainties.

The U.S. EPA TSCA Systematic Review Protocol

For regulatory assessments under the Toxic Substances Control Act (TSCA), the EPA has developed a formal protocol[reference:9]. Key stages include:

• Problem Formulation: Define the scope of the risk evaluation and the specific evidence needed.
• Literature Search & Selection: Conduct comprehensive searches, including grey literature, with documented strategies.
• Study Evaluation & Data Extraction: Apply a tiered approach to evaluate study reliability and relevance, extracting data for hazard and exposure assessment.
• Evidence Integration & Conclusion: Use weight-of-evidence approaches to integrate lines of evidence and support risk determinations.

Visualizing Workflows and Impact Pathways

Systematic Review Workflow for Ecotoxicity Data

The following diagram outlines the standard operational workflow for conducting an SR in ecotoxicology.

Diagram 1: Standard systematic review workflow for ecotoxicity evidence synthesis.

Impact Pathway from Systematic Review to Chemical Policy

This diagram illustrates how SRs function as a critical evidence-filtering mechanism within the broader chemical assessment and policy cycle.

Diagram 2: Pathway through which systematic reviews inform chemical assessment and policy.

The Scientist's Toolkit: Essential Research Reagent Solutions

Successfully executing SRs and generating the underlying ecotoxicity data requires a suite of tools and materials. The following table details key resources.

Table 3: Key Research Reagent Solutions for Systematic Reviews and Ecotoxicity Testing

Item	Category	Function/Brief Explanation
DistillerSR	SR Software	A comprehensive, web-based platform for managing the entire SR process, from screening to data extraction and reporting.
Rayyan	SR Software	A free, collaborative web tool designed to accelerate the initial screening of titles and abstracts by multiple reviewers.
PubMed / Scopus	Database	Core bibliographic databases for conducting comprehensive literature searches in biomedical and environmental sciences.
OECD Test Guidelines	Testing Standard	Internationally agreed standard methods (e.g., OECD TG 201, 211) for testing chemicals for effects on aquatic and terrestrial organisms, ensuring data reliability and regulatory acceptance.
Microtox Acute Toxicity Test	Bioassay Kit	A standardized, rapid test using luminescent bacteria (Vibrio fischeri) to assess the acute toxicity of water samples or chemical solutions.
ToxTrak Toxicity Reagent Set	Bioassay Kit	A reagent set for performing toxicity screens on water samples, based on the inhibition of dehydrogenase activity in bacteria.
LumiMARA Ecotoxicity Test	Bioassay Kit	A multi-species test using 11 luminescent bacterial strains to provide a more comprehensive toxicity profile beyond single-species assays.
OHAT / SYRCLE RoB Tool	Evaluation Tool	Structured templates for assessing the Risk of Bias in animal studies (SYRCLE) or human and animal studies (OHAT), critical for SR quality appraisal.

The integration of systematic review methodologies into chemical assessment represents a fundamental advancement towards more transparent and defensible management policies. As demonstrated, SRs can efficiently synthesize large bodies of ecotoxicity data (e.g., 66 studies in a single review) to support clear hazard identification[reference:10]. However, their full impact is currently moderated by persistent systemic barriers, including perceived divides between academic and regulatory sectors and high resource demands[reference:11][reference:12]. Overcoming these challenges requires a concerted, systems-level approach that includes the adoption of standardized protocols (e.g., SYRINA, EPA TSCA), investment in specialized software tools, and ongoing dialogue between researchers and regulators. For the field of ecotoxicity data research, embedding systematic review principles is not merely a methodological choice but an essential step for generating evidence that is robust enough to inform the critical decisions that protect human health and the environment.

The development of environmental health policy and chemical risk assessment is fundamentally grounded in the rigorous synthesis of scientific evidence. Over the past two decades, methods for assembling this evidence have evolved from opaque expert judgment towards transparent, systematic approaches for gathering, evaluating, and integrating research findings [87]. Systematic reviews now form the cornerstone of hazard assessment for agencies like the U.S. Environmental Protection Agency (EPA), bringing necessary transparency and reducing subjectivity in processes such as the Integrated Risk Information System (IRIS) assessments [87]. However, the rapid expansion of chemical diversity and the complexity of biological systems present substantial challenges. Traditional review methods, often manual and labor-intensive, struggle with the volume of data and the need to integrate heterogeneous evidence streams—from high-throughput in vitro assays and omics technologies to traditional in vivo studies and epidemiological data.

This whitepaper posits that the next evolution in systematic review methods for ecotoxicity data research lies at the confluence of three transformative paradigms: Artificial Intelligence (AI) for data mining and predictive modeling; the Adverse Outcome Pathway (AOP) framework for organizing mechanistic knowledge; and Interactive Evidence Maps for visual synthesis and gap analysis. Together, these technologies enable a more dynamic, predictive, and transparent evidence-synthesis ecosystem capable of meeting the demands of next-generation risk assessment (NGRA).

Artificial Intelligence in Data Mining and Predictive Toxicology

AI, particularly machine learning (ML) and deep learning (DL), is revolutionizing the analysis of toxicological data by extracting patterns from large-scale biological and chemical datasets that are beyond human analytical capacity [28].

The U.S. EPA’s ToxCast program, with its vast database of high-throughput screening (HTS) results for thousands of chemicals, is the most widely used data source for developing AI-driven toxicity prediction models [28]. Models increasingly move beyond classical Quantitative Structure-Activity Relationship (QSAR) approaches by using ToxCast bioactivity data as biological features to predict in vivo toxicity outcomes [28]. The field is characterized by a shift from simple molecular fingerprints to alternative representations like molecular graphs, images, and text, leveraging advances in deep learning [28].

Table 1: Performance Comparison of AI Model Types for Toxicity Prediction

Model Type	Typical Data Input	Key Strength	Representative AUC-ROC	Primary Limitation
Traditional QSAR/ML	Chemical descriptors (e.g., molecular fingerprints)	Interpretability, well-established	~0.70 - 0.85 [88]	Struggles with "activity cliffs"; limited biological context
Transcriptomics-Based	Gene expression profiles (e.g., L1000 data)	Captures cellular response; mechanistic insight	0.72 (Carcinogenicity) - 0.82 (Genotoxicity) [88]	Experimental variability; cost of data generation
Multimodal Deep Learning (e.g., GenotoxNet)	Integrated chemical graphs, transcriptomics, & HTS data	Captures cellular heterogeneity & cross-modal interactions	0.891 ± 0.017 (Genotoxicity) [88]	Model complexity; "black-box" interpretation challenges

Advanced Protocol: Multimodal Deep Learning for Genotoxicity

The GenotoxNet framework exemplifies the cutting-edge integration of multimodal data [88].

Experimental Protocol:

• Data Collection & Curation:
  • Chemicals: Obtain a curated set of substances with confirmed genotoxic (positive) and non-genotoxic (negative) labels from sources like The Carcinogenome Project [88].
  • Gene Expression: For each chemical, retrieve gene expression profiles from relevant cell lines (e.g., HepG2). Use the profile with the highest Transcriptional Activity Score (TAS). Filter to a set of genes known to be related to the endpoint (e.g., 295 genotoxicity-related genes) [88].
  • HTS Bioassays: Download binary response data from ToxCast. Filter to bioassay endpoints relevant to the mechanism (e.g., DNA damage, cell cycle, apoptosis) [88].
• Data Processing:
  • Standardize chemical structures (SMILES) using toolkits like RDKit to remove salts and generate canonical forms [88].
  • Represent each molecule as a graph G = (V, E), where V (nodes) are atoms and E (edges) are bonds. Create a node feature matrix (atom types, hybridization) and an adjacency matrix [88].
  • Normalize gene expression data using moderated z-scores (MODZ) relative to control groups [88].
  • Integrate the three modalities (graph, gene expression, bioassay) into a aligned dataset for shared compounds.
• Model Architecture & Training (GenotoxNet):
  • Chemical Graph Stream: Process the molecular graph through multiple Graph Convolutional Network (GCN) layers to learn structural features [88].
  • Biological Data Stream: Process gene expression and bioassay vectors through separate fully connected neural network layers [88].
  • Intermediate Fusion: Concatenate the high-level feature representations from each modality before the final classification layer. This allows the model to learn complex cross-modal interactions [88].
  • Train the model using a held-out test set and validate performance via metrics like AUC-ROC.

Diagram: Workflow of a Multimodal Deep Learning Model (e.g., GenotoxNet) for Toxicity Prediction.

The Adverse Outcome Pathway (AOP) Framework as an Organizing Principle

The AOP framework provides a structured, modular representation of mechanistic knowledge, linking a Molecular Initiating Event (MIE) through a series of measurable Key Events (KEs) to an Adverse Outcome (AO). This formalization is critical for integrating diverse data streams within a systematic review [89].

Integrating AI and AOPs for Mechanistic Insight

AI can both leverage and enrich the AOP framework. Graph machine learning can be directly applied to AOP networks (often represented as directed graphs) to uncover novel connections or predict vulnerabilities [89]. Conversely, AOPs provide a mechanistic scaffold for interpreting AI model predictions, moving beyond a "black box" to align chemical perturbations with known toxicity pathways.

Experimental Protocol: Exploring Genetic Influences on AOPs Using AI [89]

• AOP Network Construction: Extract disease-specific AOPs (e.g., for liver cancer) from the AOP Knowledge Base (AOP-KB). Represent this as a computational graph where nodes are Key Events and edges are key event relationships.
• Genetic Data Integration: Obtain real-world genotype and phenotype data (e.g., from UK Biobank). Select single nucleotide polymorphisms (SNPs) associated with genes in the relevant AOPs (e.g., AHR, ABCB11 for liver cancer). Use propensity score matching to create balanced case/control cohorts for model training [89].
• Graph Machine Learning Analysis: Apply Graph Neural Networks (GNNs) to the combined graph of AOPs and genetic associations. The model learns to map genetic variation patterns onto perturbations in the AOP network.
• Heuristic Simulation: Use genetic algorithms to perform heuristic searches across the AOP-genetic graph. This generative modeling can simulate how different genetic backgrounds might influence the progression and severity of the AOP, identifying putative novel risk factors [89].

Table 2: Key Research Reagent Solutions for AOP-Informed AI Research

Item / Resource	Function in Research	Example Source / Tool
AOP Knowledge Base (AOP-KB)	Central repository for curated, structured AOP information used to build mechanistic graphs.	aopkb.org
Graph Neural Network (GNN) Library	Software framework for implementing machine learning on graph-structured data (AOP networks).	PyTorch Geometric, Deep Graph Library
Genetic Dataset with Phenotypes	Provides real-world genotype (SNP) and health outcome data to link AOPs to population-level risks.	UK Biobank, All of Us [89]
Propensity Score Matching Algorithm	Statistical method to balance case and control groups in observational genetic data, reducing confounding bias.	R: MatchIt package; Python: psmpy [89]

Diagram: Integration of Genetic Data and AI Models with the AOP Framework.

Interactive Evidence Maps for Synthesis and Gap Analysis

Interactive evidence maps are visual tools that systematically catalogue and display the characteristics of available research and interventions, making the breadth of evidence and its gaps immediately apparent [90].

Extending Maps Beyond Research

Traditional evidence maps visualize research. A significant advancement is the juxtaposition of research evidence (e.g., systematic reviews on effectiveness) with practice data (e.g., interventions currently available in a specific national context) on a single map [90]. This allows stakeholders to instantly see if available interventions are underpinned by strong evidence and to identify gaps in either research or practice [90] [91].

Protocol: Creating an Interactive Evidence Map for Ecotoxicity Data [90] [91]

• Systematic Searching & Screening:
  • Conduct a systematic search of bibliographic databases for relevant primary studies or reviews.
  • Use priority screening with text-mining tools in software like EPPI-Reviewer to expedite screening. The software prioritizes references most likely to be relevant based on machine learning, allowing screening to stop once saturation is reached [90].
• Coding and Data Extraction:
  • Code each included record using a predefined framework (e.g., chemical class, exposed organism, endpoint measured, AOP relevant).
  • For systematic reviews, extract summary findings (e.g., meta-analysis results) and quality appraisals (e.g., using AMSTAR 2) [90].
  • In parallel, collect data on "practice," such as chemicals currently under regulatory review or monitored in the environment.
• Map Design and Development:
  • Develop an Organizing Framework: Choose two or more dimensions to structure the map (e.g., X-axis: Toxicological Endpoint; Y-axis: Level of Biological Organization; Color: Strength of Evidence) [91].
  • Build the Interactive Map: Use specialized software like EPPI-Mapper (which integrates with EPPI-Reviewer) to create the visualisation [90]. Iteratively design appearance and functionality based on stakeholder feedback [91].
  • Ensure Usability: Include clear labels, a glossary, filtering functions (by chemical, endpoint, study type), and detailed pop-up summaries for each map item [91].

Diagram: Conceptual Structure of an Interactive Evidence Map for Ecotoxicity.

Table 3: Quantitative Output from a Model Evidence Mapping Project [90]

Map Component	Search/Survey Input	Included Records	Key Visualized Gap
Research Evidence (Systematic Reviews)	20,961 references identified; 14,402 screened	18 systematic reviews	Lack of high-quality reviews for "relapse prevention" strategies.
Practice Data (Available Interventions)	59 survey responses; personal communications	40 digital interventions	Interventions existed for "peer support" but no rigorous reviews.
Synthesis	N/A	58 total map entries	No research or practice for "screening & brief intervention" for drugs.

Synthesis and Integrated Workflow for Systematic Reviews

The ultimate power of these technologies lies in their integration within a modernized systematic review workflow. This workflow is iterative and dynamic, contrasting with traditional linear processes.

Integrated Next-Generation Systematic Review Workflow:

• AI-Driven Evidence Discovery: Use NLP and ML to continuously scan literature and databases, identifying and prioritizing potentially relevant studies and data, feeding into a living evidence map [28] [88].
• Evidence Mapping and Gap Analysis: Populate and maintain an interactive evidence map that catalogs both research and regulatory/practice data. This visually guides the review question and highlights critical evidence gaps [90] [91].
• Mechanistic Data Synthesis via AOPs: Code and evaluate mechanistic studies (both traditional and high-throughput) within the context of relevant AOPs. The AOP framework provides a standardized vocabulary and logical structure for synthesizing heterogeneous mechanistic data [89].
• Predictive Modeling and Extrapolation: Employ validated AI models (especially multimodal models) to predict toxicity for data-poor chemicals or to extrapolate findings across species or endpoints, using AOPs to justify biological plausibility [89] [88].
• Living Review and Updating: The interactive map and associated AI tools facilitate a "living" systematic review approach, where the evidence base and predictions can be updated as new data emerges, maintaining the review's relevance [90].

This integrated approach directly addresses the core challenges noted in evolving systematic review methods: managing data volume and complexity, integrating diverse evidence streams, and maintaining transparency while incorporating advanced computational analyses [87]. It moves the field from a primarily retrospective, narrative-driven synthesis towards a prospective, predictive, and continuously updated system for evidence-based decision-making in ecotoxicology and environmental health.

Conclusion

The systematic review of ecotoxicity data is transitioning from an academic exercise to a foundational component of modern, evidence-based environmental and biomedical decision-making. This synthesis underscores that success hinges on moving beyond outdated statistical methods, fully adopting rigorous, protocol-driven review processes like PRISMA, and proactively managing the complexities inherent in toxicological evidence [citation:1][citation:5]. Crucially, the ultimate value of these reviews is realized only through their effective integration into regulatory and research pipelines, which requires addressing not just technical but also social and systemic barriers to knowledge uptake [citation:3]. For biomedical research, particularly in drug development where environmental impact and chemical safety are paramount, these methods offer a structured pathway to evaluate off-target ecotoxicological effects, integrate preclinical data, and fulfill regulatory requirements for environmental risk assessment. Future progress depends on sustained interdisciplinary collaboration, investment in methodological training, and the development of standardized, transparent tools that align scientific rigor with the practical needs of risk assessors and policymakers.

References

• 1. A benchmark dataset for machine learning in ecotoxicology [https://www.nature.com/articles/s41597-023-02612-2]
• 2. Updating statistical practice in ecotoxicology: reflections ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12047014/]
• 3. The ECOTOXicology Knowledgebase: A Curated Database of ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC9408435/]
• 4. Systems-based analysis of the factors affecting the use ... [https://www.sciencedirect.com/science/article/pii/S0160412025006105]
• 5. Comparison of four different methods for reliability evaluation ... [https://enveurope.springeropen.com/articles/10.1186/2190-4715-23-17/tables/12]
• 6. Why Systematic rather than Review ? - PMC Narrative Review [https://pmc.ncbi.nlm.nih.gov/articles/PMC4504929/]
• 7. The Difference Between Narrative and Review Systematic Review [https://www.distillersr.com/resources/systematic-literature-reviews/the-difference-between-narrative-review-and-systematic-review]
• 8. Meta-analysis [https://en.wikipedia.org/wiki/Meta-analysis]
• 9. Chapter 10: Analysing data and undertaking meta-analyses [https://www.cochrane.org/authors/handbooks-and-manuals/handbook/current/chapter-10]
• 10. Incorporating Evidence - Writing Lab [https://www.clemson.edu/centers-institutes/writing/writing-resources/writing-resources/incorporating-evidence.html]
• 11. Identifying the PECO : A framework for formulating good questions to... [https://pmc.ncbi.nlm.nih.gov/articles/PMC6908441/]
• 12. A framework for prioritizing contaminants in retrospective ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10601791/]
• 13. Quantitative ecotoxicity analysis for pesticide mixtures ... [https://www.sciencedirect.com/science/article/abs/pii/S0147651318303622]
• 14. 7 Assessment of Ecotoxicity | A Framework to Guide Selection of... [https://nap.nationalacademies.org/read/18872/chapter/9]
• 15. Towards a qAOP framework for predictive toxicology [https://www.sciencedirect.com/science/article/pii/S2468111321000414]
• 16. cfpub.epa.gov/ecotox [https://cfpub.epa.gov/ecotox/]
• 17. Natural but threatening? (I) A systematic aquatic ecotoxicity ... [https://www.sciencedirect.com/science/article/pii/S0013935125005304]
• 18. Guidelines for performing systematic reviews in the ... [https://pubmed.ncbi.nlm.nih.gov/29080853/]
• 19. Global Distribution, Ecotoxicity, and Treatment ... [https://www.mdpi.com/2305-6304/13/8/616]
• 20. State of per- and polyfluoroalkyl substances (PFAS) report [https://www.canada.ca/en/environment-climate-change/services/evaluating-existing-substances/state-per-polyfluoroalkyl-substances-report.html]
• 21. Guidance on assessing the methodological and reporting ... [https://www.sciencedirect.com/science/article/pii/S0160412016300903]
• 22. Comparison of four different methods for reliability evaluation of... [https://enveurope.springeropen.com/articles/10.1186/2190-4715-23-17]
• 23. - Dose assessment for developmental toxicity. III. response ... Statistical [https://pubmed.ncbi.nlm.nih.gov/7867901/]
• 24. Criteria for Reporting and Evaluating ecotoxicity ... Data [https://pmc.ncbi.nlm.nih.gov/articles/PMC5044958/]
• 25. Estimating the dose - response function through the GLM approach [https://ideas.repec.org/p/boc/dsug13/10.html]
• 26. Bootstrap method in dose - response - Quantics Biostatistics modelling [https://www.quantics.co.uk/blog/bootstrap-method-dose-response-modelling/]
• 27. New Approach Methods (NAMs) Training [https://www.epa.gov/chemical-research/new-approach-methods-nams-training]
• 28. AI-based toxicity prediction models using ToxCast data [https://www.sciencedirect.com/science/article/pii/S0300483X25001891]
• 29. RASS 2025–2026 Webinar Series [https://www.toxicology.org/groups/ss/rass/events.asp]
• 30. How to work with ecotoxicology data - hands on course [https://www.eu-parc.eu/index.php/events/trainings/statistics-modelling/how-work-ecotoxicology-data-hands-course-0]
• 31. Strategy for Evaluating Low-Dose Effects - NCBI - NIH [https://www.ncbi.nlm.nih.gov/books/NBK453247/]
• 32. Low-Dose Effects - Endocrine Science Matters [https://endocrinesciencematters.org/low-dose-effects-2/low-dose-effects-technical-overview/]
• 33. A Curated Database of Ecologically Relevant Toxicity Tests to ... [https://academic.oup.com/etc/article/41/6/1520/7730623]
• 34. Development of the toxicity values database, ToxValDB [https://www.sciencedirect.com/science/article/abs/pii/S2468111325000258]
• 35. Guidance for statistical design and analysis of toxicological ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10474994/]
• 36. Improving the power of drug toxicity measurements by ... [https://www.nature.com/articles/s41420-024-01950-3]
• 37. The WHO-commissioned systematic reviews on health ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12490090/]
• 38. Environmental aluminum exposure and Alzheimer's ... [https://www.sciencedirect.com/science/article/pii/S0147651325011042]
• 39. Species sensitivity distribution modeling for ecotoxicity ... [https://www.sciencedirect.com/science/article/pii/S004896972502515X]
• 40. On the quantification of intertest variability in ecotoxicity ... [https://pubmed.ncbi.nlm.nih.gov/22619109/]
• 41. Meta-analysis reveals temperature increase exacerbates ... [https://enveurope.springeropen.com/articles/10.1186/s12302-025-01227-1]
• 42. Evaluation Guidelines for Ecological Toxicity Data in the ... [https://www.epa.gov/pesticide-science-and-assessing-pesticide-risks/evaluation-guidelines-ecological-toxicity-data-open]
• 43. Historical control data for the interpretation of ecotoxicity data [https://pmc.ncbi.nlm.nih.gov/articles/PMC6872505/]
• 44. Historical control data for the interpretation of ecotoxicity data [https://pubmed.ncbi.nlm.nih.gov/31696445/]
• 45. An integrated ecotoxicological study reliability framework ... [https://pubmed.ncbi.nlm.nih.gov/39873747/]
• 46. Principles and framework for assessing the risk of bias for ... [https://environmentalevidencejournal.biomedcentral.com/articles/10.1186/s13750-022-00264-0]
• 47. Critical Appraisal and Risk of Bias Tools - Evidence ... [https://infoguides.gmu.edu/EvidenceSynthesis/Criticalappraisal]
• 48. Proposal of critical appraisal tools for the evaluation of ... - EFSA [https://www.efsa.europa.eu/en/supporting/pub/en-7787]
• 49. Towards Understanding the Factors behind the Limited... | Preprints.org [https://www.preprints.org/manuscript/202308.0483/v1]
• 50. Criteria for Reporting and Evaluating ecotoxicity ... Data [https://enveurope.springeropen.com/articles/10.1186/s12302-016-0073-x]
• 51. Integrating chemical-specific information with general ... [https://www.sciencedirect.com/science/article/pii/S0160412025003253]
• 52. CEE Critical Appraisal Tool [https://environmentalevidence.org/cee-critical-appraisal-tool/]
• 53. Impact Evaluation for Ecotoxicity Poor Chemicals Under the... Data [https://orbit.dtu.dk/en/publications/ecotoxicity-impact-evaluation-for-data-poor-chemicals-under-the-g]
• 54. The prevention and handling of the missing data - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC3668100/]
• 55. Ecotoxicology - NOEC and LOEC [https://www.quantics.co.uk/blog/ecotoxicology-noec-and-loec/]
• 56. Handling Missing Values in Machine Learning [https://www.geeksforgeeks.org/machine-learning/handling-missing-values-machine-learning/]
• 57. Addressing the Data Scarcity Problem in Ecotoxicology via ... [https://pubmed.ncbi.nlm.nih.gov/40111220/]
• 58. Searching two or more databases decreased the risk of ... [https://www.sciencedirect.com/science/article/pii/S0895435622001445]
• 59. Inadequate use and regulation of interventions against ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC4459964/]
• 60. Systematic Reviews: Step 7: Extract Data from Included Studies [https://guides.lib.unc.edu/systematic-reviews/extract-data]
• 61. The Navigation Guide Systematic Review Methodology [https://pmc.ncbi.nlm.nih.gov/articles/PMC4181919/]
• 62. Applying a modified systematic review and integrated ... [https://pubs.rsc.org/en/content/articlehtml/2024/em/d3em00353a]
• 63. Summary | Building Confidence in New Evidence Streams ... [https://nap.nationalacademies.org/read/26906/chapter/2]
• 64. Quantitative Techniques and Graphical Representations ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8894511/]
• 65. About ICCVAM - National Toxicology Program [https://ntp.niehs.nih.gov/whatwestudy/niceatm/iccvam]
• 66. Which distribution to choose for deriving a species ... [https://www.sciencedirect.com/science/article/pii/S014765132400455X]
• 67. Uncertainty analysis and updated user guidance for ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12199407/]
• 68. Detecting Moderator Effects Using Subgroup Analyses - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC3193873/]
• 69. Model validation in aquatic toxicity testing [https://www.sciencedirect.com/science/article/abs/pii/S0273230012000815]
• 70. Recommendations to address uncertainties in ... [https://www.nature.com/articles/s41598-019-47698-0]
• 71. OECD Chemical Testing Guidelines 2025 Updated [https://auxilife.com/oecd-announces-comprehensive-updates-to-chemical-testing-guidelines-strengthening-environmental-protection-and-animal-welfare/]
• 72. A Bayesian Mixed Treatment Comparisons Meta-Analysis ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC4318856/]
• 73. A Practical Guide to Random-Effects Bayesian Meta ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10021079/]
• 74. Multi-level ecotoxicity of biodegradable microplastic to ... [https://www.sciencedirect.com/science/article/abs/pii/S0013935125021462]
• 75. Two Examples of Bayesian Evidence Synthesis with the ... [https://www.intechopen.com/chapters/56655]
• 76. Assessing the ecotoxicity of multicomponent nanomaterials ... [https://pubs.rsc.org/en/content/articlehtml/2025/en/d4en01183j]
• 77. Green synthesized metal nanoparticles appear to meet ... [https://pubmed.ncbi.nlm.nih.gov/40197171/]
• 78. Ecotoxic effect of mycogenic silver nanoparticles in water ... [https://www.nature.com/articles/s41598-025-95485-x]
• 79. Unlocking the future of environmental safety: a framework ... [https://www.sciencedirect.com/science/article/pii/S3050620425000235]
• 80. Chemical Industry Regulations 2025: Key Insights from CIR ... [https://www.enviresearch.com/news/insights-from-chemical-industry-regulations-2025/]
• 81. A comprehensive guide to conduct a systematic review and ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12366998/]
• 82. Conducting a High-Quality Systematic Review [https://www.jrheum.org/content/52/7/636]
• 83. Systematic Reviews and Other Expert Reviews [https://browse.welch.jhmi.edu/sr-methods/expert-reviews]
• 84. A tiered next-generation risk assessment framework ... [https://link.springer.com/article/10.1007/s00204-025-04045-9]
• 85. FDA issues first guidance on the use of AI in drug ... [https://www.ddw-online.com/fda-issues-first-guidance-on-the-use-of-ai-in-drug-development-33107-202501/]
• 86. Has FDA's Drug Development Tools Qualification Program ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12181102/]
• 87. Invited Perspective: Systematic Review for Environmental ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10664748/]
• 88. Advancing toxicity AI-based prediction with multilevel ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12619907/]
• 89. Exploring genetic influences on adverse outcome pathways ... [https://assessments.epa.gov/risk/document/&deid%3D365128]
• 90. Advancing interactive evidence maps: Visualising service ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11795966/]
• 91. How to design interactive evidence maps: a case study from ... [https://eppi.ioe.ac.uk/cms/Default.aspx?tabid=3681&articleType=ArticleView&articleId=181]