Using the Matrix to bridge the epidemiology/risk assessment gap: a case study of 2,4-D

Abstract

Background

The Matrix is designed to facilitate discussions between practitioners of risk assessment and epidemiology and, in so doing, to enhance the utility of epidemiology research for public health decision-making. The Matrix is comprised of nine fundamental “asks” of epidemiology studies, focusing on the types of information valuable to the risk assessment process.

Objective

A 2,4-dichlorophenoxyacetic acid (2,4-D) case study highlights the extent to which existing epidemiology literature includes information generally needed for risk assessments and proffers suggestions that would assist in bridging the epidemiology/risk assessment gap.

Methods

Thirty-one publications identified in the US Environmental Protection Agency 2,4-D epidemiology review were assessed. These studies focused on associations between 2,4-D exposure and non-Hodgkin lymphoma (NHL), respiratory effects, and birth outcomes.

Results

Many of the papers met one or more specific elements of the Matrix. However, from this case study, it is clear that some aspects of risk assessment, such as evaluating source-to-intake pathways, are generally not considered in epidemiology research. Others are incorporated, but infrequently (e.g. dose-response information, harmonization of exposure categories). We indicated where additional analyses or modifications to future study design could serve to improve the translation.

Discussion

Interaction with risk assessors during the study design phase and using the Matrix “asks” to guide the conversations could shape research and provide the basis for requests for funds to support these additional activities. The use of the Matrix as a foundation for communication and education across disciplines could produce more impactful and consequential epidemiology research for robust risk assessments and decision-making.

Keywords:

• 2,4-D
• matrix
• epidemiology
• pesticides
• risk assessment
• causality
• exposure assessment

Introduction

The fields of environmental epidemiology and exposure science have provided consequential data for use in meta-analyses, systematic reviews, and ultimately, public health decision-making. Yet while activities such as the development of reference doses have been based on data from epidemiology studies, it is often the case that these data are secondary to toxicological data or are judged to be insufficient to examine exposure-outcome associations (Nachman et al. 2011; EFSA Panel on Plant Protection Products et al. 2017; Deener et al. 2018). While calls have been made for improving the utility of epidemiology for risk assessment, hurdles remain (Burns et al. 2014; Christensen et al. 2015; Birnbaum et al. 2016). In an effort to move the needle on this issue, an international, multi-sector group with expertise in risk assessment, toxicology, epidemiology, and exposure science developed the Matrix (Table 1), a structured approach to bridging the risk assessment-epidemiology gap (Burns et al. 2019).

Table 1. The Matrix: Nine risk assessor “asks” of epidemiology studies.

Asks for risk assessment
Hazard ID	Confirm outcome	Confirm exposure	Report methods fully and transparently
Dose-response	Include information on the shape of the curve	Harmonize exposure categories	Describe direction/magnitude of error
Exposure assessment	Evaluate source-to-intake pathways	Describe complete exposure data	Report on quality assurance/quality control

(From Burns et al. 2019; LaKind et al. 2020).

The disciplines of epidemiology and risk assessment share common goals of understanding and reducing human health impacts associated with exposure to environmental chemicals. But challenges arise because these disciplines have different study objectives and use different terminologies and skillsets (Burns et al. 2019). Thus, the Matrix was developed as an approach to promoting dialogue among practitioners of epidemiology and risk assessment and to improve cross-disciplinary understanding. As a communication tool, the Matrix elements are designed as “asks,” not requirements or criteria. The nine “asks” of the Matrix are fundamental aspects of risk assessment, each of which adds confidence to study observations and aids in comparisons across studies (Burns et al. 2019; LaKind et al. 2020). The “asks” are organized according to the basic risk assessment steps: (1) hazard identification, (2) dose-response assessment (aka exposure-response), and (3) exposure assessment; the fourth step – risk characterization – is a synthesis of information from the other steps. For each of these steps (shown as rows in Table 1), the Matrix includes three “asks” of an epidemiology study. The “asks” address study elements that may not be widely used in epidemiology research but at the same time can increase the value of a study for risk assessment purposes.

It is important to note that the Matrix was not developed or designed for use as a checklist to critique a body of literature or to offer insights into causality. Rather, the goal of the Matrix is to enable interested investigators, reviewers, and study sponsors to develop an improved understanding of study designs which enhance or impede their usefulness for risk assessment and to highlight opportunities for post hoc reporting and analyses of data (Burns et al. 2019; LaKind et al. 2020). In this way, the use of the Matrix can improve the translation of epidemiology research for use in policy decision-making. This is aligned with the goal of the US National Institute of Environmental Health Sciences to encourage “the translation of environmental health research into concrete strategies that protect and improve human health” (NIEHS 2019).

As an illustration, we selected as a case study the epidemiology literature on 2,4-dichlorophenoxyacetic acid (2,4-D) and associations with non-Hodgkin lymphoma (NHL), respiratory effects, and birth outcomes. This epidemiology literature has been reviewed previously (e.g. Garabrant and Philbert 2002; Burns and Swaen 2012; Goodman et al. 2015) but these reviews did not describe aspects of the studies that either enhanced or lessened their value for risk assessment purposes. This body of literature was also reviewed by the US Environmental Protection Agency (EPA) using a study quality framework for assessing causal associations (US EPA 2017a, 2017b). The intent of these reviews was not to offer an approach to informing epidemiologists about more risk assessment-conducive study design and reporting methods.

Herein, we focus on providing examples within the 2,4-D literature in which Matrix “asks” were met, as well as information on study aspects that could be enhanced with post hoc analyses, reporting, and/or modifications to future study designs. In other words, this is not a critical assessment of the literature but rather a forward-looking exercise. Through this case study, the reader may gain an understanding of the methodological and reporting practices often sought by regulatory agencies. We do not assess the study quality or overall weight of evidence, as these are not the purposes of the Matrix, nor do we delve into how risk assessment is used for regulatory decision-making.

Methods

We used the epidemiology publications cited in the US EPA 2,4-D Tier II epidemiology reviews for this case study as this literature represents a useful variety of study designs and population types to illustrate Matrix “asks” (US EPA 2017a, 2017b). As summarized in Table 2, the EPA 2,4-D Tier II epidemiology reviews included 15 noncancer publications of respiratory effects and outcomes related to in utero exposure and 16 NHL publications.

Table 2. Case study 2,4-D publication examined with the Matrix.

Reference	Study or design	Location	Outcome
Noncancer outcomes
(Arbuckle et al. 2001)	OFFHS	Canada	Spontaneous abortion
(Hoppin et al. 2002)	AHS	US (IA and NC)	Wheeze
(Hoppin et al. 2006a)	AHS	US (IA and NC)	Wheeze
(Hoppin et al. 2006b)	AHS	US (IA and NC)	Wheeze
(Hoppin et al. 2007)	AHS	US (IA and NC)	Chronic bronchitis
(Hoppin et al. 2008)	AHS	US (IA and NC)	Asthma
(Hoppin et al. 2009)	AHS	US (IA and NC)	Allergic and nonallergic asthma
(Hoppin et al. 2017)	AHS	US (IA and NC)	Wheeze
(Rull et al. 2006)	Pooled case-control	US (CA)	Neural tube birth defects
(Sathyanarayana et al. 2010)	AHS	US (IA and NC)	Birth weight
(Slager et al. 2009)	AHS	US (IA and NC)	Rhinitis
(Waller et al. 2010)	Case-control	US (WA)	Gastroschisis
(Weselak et al. 2007)	OFFHS	Canada	Respiratory effects in children
(Weselak et al. 2008)	OFFHS	Canada	Non-specific birth defects
(Valcin et al. 2007)	AHS	US (IA and NC)	Chronic bronchitis
Cancer outcomes
(Burns et al. 2011)	Retrospective cohort	US (MI)	NHL
(Cantor et al. 1992)	Case-control	US (IA/MN)	NHL
(De Roos et al. 2003)	Pooled case-control	US (IA/MN, KS, and NE)	NHL
(Hardell et al. 1994)	Case-control	Sweden	NHL
(Hartge et al. 2005)	Case-control	US (IA)	NHL
(Hoar et al. 1986)	Case-control	US (KS)	NHL
(Hohenadel et al. 2011)	Case-control	Canada	NHL
(Kogevinas et al. 1995)	Pooled nested case-control	Multi-country	NHL
(Lee et al. 2004)	Pooled case-control	US (IA/MN and NE)	NHL
(McDuffie et al. 2001)	Case-control	Canada	NHL
(McDuffie et al. 2005)	Case-control	Canada	NHL
(Miligi et al. 2006)	Case-control	Italy	NHL
(Mills et al. 2005)	Nested case-control	US (CA)	NHL
(Pahwa et al. 2012)	Case-control	Canada	NHL
(Woods and Polissar 1989)	Case-control	US (WA)	NHL
(Zahm et al. 1990)	Case-control	US (NE)	NHL

AHS: Agricultural Health Study; CA: California; IA: Iowa; KS: Kansas, MI: Michigan; MN: Minnesota; NE: Nebraska; NC: North Carolina; NHL: non-Hodgkin lymphoma; OFFHS: Ontario Farm Family Health Study; US: United States; WA: Washington.

We examined each publication to determine the extent to which the study included information related to the nine Matrix “asks.” We recognize that the definition or level of sufficiency for each “ask” will vary by reviewer and purpose (e.g. hazard identification, dose-response assessment). We, therefore, elected to extract examples from the literature to describe how “asks” can be met and opportunities for improving the utility of various aspects of the literature.

Results

The following subsections are organized according to the three “asks” for each of the three risk assessment steps (Table 1). Each subsection begins with a brief explanation of the “ask” in question form. We then provide examples from the 2,4-D literature of publications that did or did not meet the given “ask” and a general description of the extent to which the overall body of literature addressed an “ask.” We also describe whether the “ask” could be met by additional analysis/reporting of existing data or whether additional research would be needed. As a reminder, this is not a systematic assessment or quality review of each paper, but rather a collection of examples highlighting approaches to enhancing translation of epidemiology research.

Hazard identification

Confirm outcome

Is there evidence that the methods used for outcome ascertainment are sensitive and specific? Can the investigators accurately determine the presence or absence (or degree of severity) of the outcome?

The two case-control studies of birth defects confirmed cases of gastroschisis and neural tube defects using data from birth certificates, medical records, and/or genetic centers (Rull et al. 2006; Waller et al. 2010). In contrast, the remaining 2,4-D noncancer studies relied upon unconfirmed self-reported outcome information. The publications from the Ontario Farm Family Health Study (OFFHS) and the Agricultural Health Study (AHS) used self-reports on birth outcomes (spontaneous abortion, birth defects, birth weight) and respiratory outcomes (asthma, wheeze, bronchitis, and rhinitis) (e.g. Arbuckle et al. 2001; Hoppin et al. 2007; Valcin et al. 2007). Efforts to confirm other outcomes within the AHS have relied upon additional screening questions and obtaining consent to review medical records, with results varying by outcomes such as Parkinson’s Disease (84% confirmed), rheumatoid arthritis (67% confirmed), and thyroid disease (32% confirmed for hyperthyroidism, 90% confirmed for hypothyroidism) (De Roos et al. 2005; Tanner et al. 2011; Parks et al. 2016; Shrestha et al. 2018, 2020). However, the reliability and validity for respiratory outcomes in the AHS have not been evaluated.

The NHL studies largely identified cases from tumor registries or state and local programs. While some of the NHL studies relied upon the diagnoses provided by the tumor registry (Hartge et al. 2005; Mills et al. 2005; Burns et al. 2011), several investigators conducted additional histopathological verification which increased the confidence in the outcome ascertainment and refined the study population (Hoar et al. 1986; Zahm et al. 1990; Cantor et al. 1992; Hardell et al. 1994; McDuffie et al. 2001). One of these additional exercises revealed that up to 10% of the presumptive cases in the Iowa/Minnesota case-control study were not confirmed to be NHL (Cantor et al. 1992).

The confirm outcome “ask” is achieved through using, when possible, outcome ascertainment approaches that are accurate (e.g. histopathological confirmation of NHL diagnosis). For this body of NHL literature, it is of interest to note that outcome ascertainment in the NHL studies conducted prior to 2000 was complicated by the evolving diagnosis and classification system for the lymphohematopoietic cancers during that time. A uniform approach from the World Health Organization has been globally embraced since the early 2000s making registry data more comparable (Pratap and Scordino 2019). For studies that include NHL subtype analyses, variability in defining the multiple subtypes is an important consideration (Clarke et al. 2004). For investigations such as the study of United Farm Workers that evaluated NHL by subtype (Mills et al. 2005), case validation may become increasingly important.

In contrast with approaches to diagnosing NHL, self-reported data, while an important source of retrospective information in both clinical settings and etiologic research, have well-known limitations (e.g. recall validity may vary by outcome) (Baumeister et al. 2010; Liu et al. 2013; Cornish et al. 2014; Pols et al. 2016). Whether or not a self-reported outcome meets the “ask” may depend on the specific diagnosis or symptom(s), and available evidence of recall validity for the study population.

The method(s) used for outcome ascertainment is typically determined during the study design phase. However, additional post hoc information such as data on the sensitivity and specificity from registries or an add-on validation study by the investigators could increase risk assessors’ confidence in the study results.

Confirm exposure

Is the exposure specific? Is the exposure of interest defined and measured with robust methods that include frequency and duration? Was the exposure quantified?

Because this case study focused on a specific pesticide (2,4-D), all the reviewed papers evaluated a given active ingredient (in contrast with, for example, studies on less specific exposures to phenoxies or herbicides). However, most of the studies did not include measurements of 2,4-D and none of the exposure assessments was based on biomonitoring.

Two publications used concentration data from environmental sources. Waller et al. (2010) estimated 2,4-D exposures by calculating the average concentration for each site using annual surface water concentrations. This approach did not include information on actual exposures to individuals from processed drinking water. Hartge et al. (2005) included 2,4-D concentration data in carpet dust. The use of carpet dust for providing accurate information on human exposure has been examined and researchers have expressed varying degrees of confidence in this source of exposure data (Lioy et al. 2002; Deziel et al. 2015).

Most of the studies examined here obtained exposure information from questionnaires and pesticide use records. The AHS has conducted several validation and reliability studies (Blair et al. 2002; Hoppin et al. 2002; Coble et al. 2011). Recall bias notwithstanding, use information does not provide quantitative data on external or internal doses and actual exposures are known to be affected by protective equipment and personal behaviors (Arbuckle et al. 2005; Alexander et al. 2007; Burns and Swaen 2012). Without information on quantifiable estimates of specific exposure, the use of epidemiology data to support hazard identification can be difficult. A robust assessment of 2,4-D exposure for hazard identification could include, for example, environmental and/or biomonitoring-based measurements or models, with complete use information including exposure frequency and duration, product concentration, and data on use of personal protective equipment.

Meeting the confirm exposure “ask” is most readily accomplished during the study design phase, although it is possible that post hoc validation studies could provide information linking data from questionnaires and pesticide use reports to concentration information.

Report methods fully and transparently

Is methodologic information adequately reported? Are the data sources well-documented? Are the methods used to develop exposure estimates clearly described?

The publications from the AHS provide a model for method transparency via a public website that includes the questionnaires, a list of publications, and a full description of data collection methods (https://aghealth.nih.gov/). In general, more specific and detailed information can facilitate a risk assessor’s ability to utilize the information for hazard identification. Some of the investigations did not include full descriptions of exposure assessment methods, making it difficult for reviewers and risk assessors to determine the degree of confidence in the results. A few examples follow: The methods for calculating exposure intensity were not provided in the publication on the Dow Chemical manufacturing cohort (Burns et al. 2011). Waller et al. (2010) used 2,4-D concentrations in surface water to assess exposure but did not explain how these measurements correlated with the drinking water used by the study subjects (presumably the actual source of water exposure) or other potential sources. The algorithm to determine bystander exposure using residential addresses and the California Pesticide Use Report data was not clearly explained (Rull et al. 2006).

This Matrix “ask” can be met by additional reporting. The availability of online supplements and study websites can alleviate the past problem of word count restrictions for journal publications that may have prevented a full description of study methods.

Dose-Response assessment

Include information on shape of the dose-response curve

Is there a sufficient characterization of the shape of the dose (exposure)-response relationship (needed by risk assessors to quantify potential public health impacts associated with the chemical exposure)? Did the investigators evaluate both linear and nonlinear exposure-response relationships?

Despite not having concentration data, several of the NHL investigations partially met the exposure-response “ask” by reporting results categorically by days of use (Hoar et al. 1986; Zahm et al. 1990; Cantor et al. 1993; McDuffie et al. 2001), levels in household dust (Hartge et al. 2005) and occupational exposure estimates (Burns et al. 2011). Additional post hoc data were collected from the Iowa NHL study participants to obtain the number of days of pesticide use (Cantor et al. 1992, 1993). Among the noncancer studies, categorical results, by days applied, were only reported in three AHS publications (Hoppin et al. 2002; Slager et al. 2009; Hoppin et al. 2017).

While quantitative dose-response assessments require information on exposure concentrations, of the 31 studies reviewed, only Hartge et al. (2005) used measurement information (concentrations of 2,4-D in dust) to develop an exposure-response curve.

For future epidemiology studies to meet this “ask,” it is likely that both study design and reporting components will need to be enhanced. To increase the utility of future studies for translation, investigators could consider exposure methods that include measured or modeled exposure data usable for dose-response assessment and conduct analyses that offer the relevant information on the shape of the dose-response curve.

Harmonize exposure categories

Did the investigators use exposure definitions consistent with previously published studies? Is there a harmonization of exposure levels across studies? Are categories similarly defined?

This “ask” focuses on study design and reporting in comparison to that of previously published studies. The value inherent in harmonizing exposure categories with those used in the extant literature lies in the fact that inter-study comparison is an integral part of the weight of evidence assessment; inter-study consistency increases confidence in observational epidemiology study results (Goodman et al. 2010). With disparate exposure categories, inter-study comparisons may not be possible, nor recommended (Kabat et al. 2021).

In the literature examined here, harmonization of exposure definitions and/or categories was not often considered. Specifically, for the publications that included dose-response analyses, such as for days of use, exposure categories were not aligned with previously published studies. As an example, within the AHS, the investigators utilized similar categories for days of use in the past year for wheeze among farmer applicators and commercial applicators (Hoppin et al. 2002; 2006a). However, in a later wheeze analysis, the authors used different cut-points for exposure categories (Hoppin et al. 2017). The importance of harmonization of exposure categories is exemplified in a recent NHL meta-analysis by Smith et al. (2017), who compared risk estimates across publications using the category of most highly exposed within each publication, even though the definition for “highly exposed” differed appreciably across studies.

This “ask” could be met in the analysis phase for many studies by conducting additional analysis utilizing the same exposure categories as in other publications. Post hoc analyses using concordant exposure levels would contribute meaningfully to the process of inter-study comparisons (Goodman et al. 2010; LaKind et al. 2015).

Describe direction/magnitude of error

Did the investigators quantify the magnitude and direction of error in the risk estimate? Did the researchers provide information beyond standard errors and qualitative discussions of study limitations?

This element was not addressed in any of the reviewed studies. The publications at most included qualitative discussions regarding study limitations and possible errors, with some including a general “bias towards the null” statement. Conducting a quantitative assessment of error for magnitude and direction is gaining acceptance, and techniques have been developed for bias analysis (Lash et al. 2014). These analyses can be conducted post hoc.

Edwards and Keil (2017) observed that an important task of public health research is to inform decision-makers and that measurement error can “obscure or amplify uncertainty in the results.” The value of quantifying the direction and distortion of error in terms of regulatory decision-making has also been noted by US EPA. (2016): “…this ability to characterize the direction of the bias based on knowledge of the study methods and analyses can be useful to the regulatory decision-maker since it may allow the decision-maker to determine the extent to which, if any, the epidemiological effect sizes being considered (e.g. OR, RR) are likely underestimating or overestimates of the true effect size.” This ask was also described in the approach to epidemiology evaluation within the EPA Integrated Risk Information System, a program that evaluates many types of environmental chemicals (US EPA 2020). Thus, the importance of this issue extends well beyond 2,4-D research.

Exposure assessment

Evaluate source-to-intake pathway

Is there information on the source(s), fate, and transport of the chemical(s) of interest, and routes of exposure (i.e. dermal, inhalation, ingestion)? Are the relationships among the sources and fate and transport of a chemical, the conditions of use, and human and environmental receptors included?

While not commonly used in epidemiology studies, defining the source-to-intake pathway is valuable. When assumptions and conventional wisdom regarding pathways of exposure are formally evaluated, risk assessors are better informed regarding what actions will reduce exposure in a given population.

For example, it has been assumed that bisphenol A exposures are driven by diet. Yet a publication that studied exposure from multiple sources revealed that diet may make up a much smaller component of exposure than previously thought (Morgan et al. 2018). Thus, studies that assume exposure routes are from a single source may not be providing risk assessors with the complete data they need to assess general population exposure(s) and risk.

The publications on 2,4-D did not generally consider source-to-intake pathways. The different sources of exposure (e.g. occupational exposures, home use, foods) and routes were implied by the study population and design without validation of these assumptions.

Of note, a modeling study by Kauppinen et al. (1994) using the same NHL cohort as Kogevinas et al. (1995), included a source-receptor model in their exposure assessment. The model considered the occupational source, fate and transport and receptor (worker) pathways of exposure.

The process of defining and evaluating the source-to-intake pathways is not part of the epidemiology lexicon. This “ask” is especially important for epidemiology studies that rely on biomonitoring to assess exposure, as biomonitoring itself does not provide information on sources or routes of exposure. This “ask” will likely only become part of future studies after extensive outreach and communication. In addition, this “ask” would be difficult if not impossible to meet as part of a post hoc analysis. Rather, incorporation into future study design would be necessary.

Provide complete exposure data

Did the study provide a description of the distribution of chemical concentrations? Are results beyond common metrics such as ranges and means provided?

Of the publications in this case study, two studies included environmental measurements of 2,4-D in dust (Hartge et al. 2005) and water (Waller et al. 2010). Neither publication provided standard metrics such as mean or range, nor more complete distribution information (e.g. histograms). While Burns et al. (2011) offered detailed exposure information for the overall study population using intensity scores, the intensity scores and the underlying industrial hygiene data on which they were based were not provided.

Investigators designing future studies may want to explore the feasibility of collecting concentration data to increase the likelihood of translation of their research for decision-making. For those that already rely on measurement data, this “ask” could likely be readily accomplished for most – or perhaps all – of the epidemiology studies using data already collected by the researchers. Going forward, if investigators would provide fuller exposure data to risk assessors, this could facilitate a more robust understanding of population distributions of exposure and possibly a better understanding of dose-response.

Report on quality assurance and quality control (QA/QC)

Was information included on quality assurance and quality control procedures?

The National Toxicology Program has stated: “Knowledge of the quality and confidence in the evidence is essential to decision making” (NTP 2019). While scientific journals do not regularly require information on QA/QC procedures, this information is vital to decision-makers.

In general, limited information on study QA/QC was included in the publications examined here. Thus, it is not known whether QA/QC procedures were included but not reported, or simply not an aspect of the study design. The Some of the 2,4-D studies that included information on elements of QA/QC provided qualitative descriptions and in certain cases referred to assessments of the validity of study components that were not included in their study. For example, Hardell et al. (1994) noted that an earlier study validated a questionnaire that was “similar” to theirs and the result was “good.” The Italian Multi-Center Case-Control Study investigators (Miligi et al. 2003) stated that the “[a]ccuracy and standardization of the interview procedures were periodically verified” but no data on the level of accuracy was provided. Zahm et al. (1990) included blinding of the interviewers to the outcome status of study participants. Hartge et al. (2005) reported that laboratory personnel were blinded to case status and that laboratory spikes were used.

Moving forward, information related to this “ask” could be prepared and made available post hoc. Detailed guidance documents are available on approaches to describing exposure data quality for future epidemiology investigations and use of epidemiology in risk assessment (EFSA 2010; LaKind et al. 2015; US EPA 2016, 2018; LaKind et al. 2019; NTP 2019; US EPA 2020).

Discussion

In this paper, we presented a case study of the 2,4-D epidemiology literature to provide specific examples regarding how the Matrix “asks” can be used to increase the opportunity for research translation. This case study included 31 publications examining associations between 2,4-D exposure and NHL, respiratory effects, and birth outcomes. For each of the nine Matrix “asks,” we provided examples in which some or all of the papers could be considered to have met the “ask,” or indicated where additional analyses or modifications to future study design could serve to improve the translation. From this case study, as shown in Figure 1, it is clear that some elements are generally not considered in epidemiology research (e.g. source-to-intake pathways, quantitative assessments of the direction and magnitude of error) while others are incorporated, but infrequently (e.g. dose-response information, harmonization of exposure categories).

Figure 1. Inclusion of Matrix asks in 2,4-D epidemiology literature.

With the transition away from traditional animal testing as a major source of information for regulatory decision-making, other sources of exposure-outcome data will be needed (e.g. human studies, new approach methodologies (NAMs)). Thus, this is an opportune time for epidemiologists to incorporate methods that enhance the integration of epidemiology findings into decision-making. Various agencies have been developing approaches to more transparently incorporate epidemiology into weight of evidence assessments (US EPA 2016, 2018; EFSA Scientific Committee et al. 2020). These approaches include a wide array of quality elements that cover a multitude of design and reporting components. While interested epidemiologists can review these documents to gain a better understanding of what is being sought by these agencies, unfortunately, they may find the various approaches used to evaluate their work to be confusing and opaque. This is in part because the processes are evolving and also because the approaches are not consistent across agencies or even within a single agency. For those epidemiologists not well-versed in risk assessment, it may be even more difficult to understand what is wanted from their research.

The Matrix is an outgrowth of the desire for an approach to assist investigators in increasing the utility of their work for risk assessment. Its purpose is not to evaluate the quality of already-published studies or to be used for causal inference – these tasks can be accomplished by the various regulatory guidance documents and other available tools. It is also worth noting that some of the “asks” may be easier to address today than they would have been in the past. For example, the widespread availability of online supplements without word limits can facilitate more comprehensive reporting of methods and results.

The Matrix approach offers a primer on the information often sought by regulators, risk assessors, and others seeking to fully integrate epidemiology research into policy decisions. The Matrix also serves as a foundation for education for epidemiology and exposure science training. In addition, the Matrix can provide support for requesting additional funding from sponsors interested in research more suited to translation. In conclusion, the use of the Matrix as a foundation for interactions with scientists, regulators, and other stakeholders can result in more impactful and consequential epidemiology research, and improved risk assessments and regulatory decision-making.

Acknowledgments

A draft courtesy copy was shared with the Task Force (TF); only one editorial correction was received from Steve McMaster, Chairman, Technical Committee, TF.

Declaration of interest

• Competing interests. Carol J. Burns consults to the private sector. CJB is a retiree and stockholder of The Dow Chemical Company, a registrant of 2,4-D. Judy S. LaKind consults to governmental and private sectors.

• Funding. This work was financially supported by the Industry Task Force II on 2,4-D Research Data (TF). The TF is made up of companies holding technical registrations on the active ingredient in 2,4-D herbicides. They are Corteva Agriscience (U.S.), Nufarm, Ltd. (Australia) and Agro-Gor Corp (U.S.). The TF was not involved in the design, collection, management, analysis, or interpretation of the data, or in the preparation or approval of the manuscript.

• Project initiation. Burns and Lakind approached the TF with an unsolicited proposal to conduct a case study using the Matrix approach to assess the 2,4-D epidemiology literature. The outcome of this effort was a White Paper, submitted to the TF. We then proposed that the White Paper be used to develop a manuscript so that the work would have a broader readership. The current manuscript represents the results of this effort.

• The review, synthesis, and conclusions reported in this paper are the exclusive professional work product of the authors and do not necessarily represent the views of the 2,4-D TF or the member companies.