Systematic Evidence Map for Over One Hundred and Fifty Per- and Polyfluoroalkyl Substances (PFAS)

Background: Per- and polyfluoroalkyl substances (PFAS) are a large class of synthetic (man-made) chemicals widely used in consumer products and industrial processes. Thousands of distinct PFAS exist in commerce. The 2019 U.S. Environmental Protection Agency (U.S. EPA) Per- and Polyfluoroalkyl Substances (PFAS) Action Plan outlines a multiprogram national research plan to address the challenge of PFAS. One component of this strategy involves the use of systematic evidence map (SEM) approaches to characterize the evidence base for hundreds of PFAS. Objective: SEM methods were used to summarize available epidemiological and animal bioassay evidence for a set of ∼150 PFAS that were prioritized in 2019 by the U.S. EPA’s Center for Computational Toxicology and Exposure (CCTE) for in vitro toxicity and toxicokinetic assay testing. Methods: Systematic review methods were used to identify and screen literature using manual review and machine-learning software. The Populations, Exposures, Comparators, and Outcomes (PECO) criteria were kept broad to identify mammalian animal bioassay and epidemiological studies that could inform human hazard identification. A variety of supplemental content was also tracked, including information on in vitro model systems; exposure measurement–only studies in humans; and absorption, distribution, metabolism, and excretion (ADME). Animal bioassay and epidemiology studies meeting PECO criteria were summarized with respect to study design, and health system(s) were assessed. Because animal bioassay studies with ≥21-d exposure duration (or reproductive/developmental study design) were most useful to CCTE analyses, these studies underwent study evaluation and detailed data extraction. All data extraction is publicly available online as interactive visuals with downloadable metadata. Results: More than 40,000 studies were identified from scientific databases. Screening processes identified 44 animal and 148 epidemiology studies from the peer-reviewed literature and 95 animal and 50 epidemiology studies from gray literature that met PECO criteria. Epidemiological evidence (available for 15 PFAS) mostly assessed the reproductive, endocrine, developmental, metabolic, cardiovascular, and immune systems. Animal evidence (available for 40 PFAS) commonly assessed effects in the reproductive, developmental, urinary, immunological, and hepatic systems. Overall, 45 PFAS had evidence across animal and epidemiology data streams. Discussion: Many of the ∼150 PFAS were data poor. Epidemiological and animal evidence were lacking for most of the PFAS included in our search. By disseminating this information, we hope to facilitate additional assessment work by providing the initial scoping literature survey and identifying key research needs. Future research on data-poor PFAS will help support a more complete understanding of the potential health effects from PFAS exposures. https://doi.org/10.1289/EHP10343

▪ Groups records based on editing distance across specific fields (i.e., affine gap distance). ▪ Proposes grouped records as duplicates to the user for verification and changes the relative weights of various features of each entry based on the user's response.
▪ Identifies the probability of duplication based on the distance between each record cluster after sufficient testing and allows for a more conservative or a more aggressive deduplication process based on threshold probabilities.
For rapid processing, ICF created a large training dataset from deduplication efforts on previous projects that save the user the need for training the model for each new run.
The algorithms underlying DeDuper -and its application in a case study involving a literature search related to diisononyl phthalate (DINP) -were presented at the 2019 Society of Toxicology (SOT) meeting (Magnuson, 2019}. In this case study, ICF applied DeDuper to a set of 30,000 references in which duplicates had been previously identified manually. Phase 1 achieved a precision of 100%, although recall was limited at 76%. Phase 2 achieved a recall of 99% with a 48% precision. After accounting for manual review to remove false positives from the machine-identified duplicates pile, the combined pipeline realized an 82% efficiency gain. ICF normalized these results based on maximum possible efficiency gains (which depends on the proportion of duplicate groups in the original dataset) to estimate a specificity of 85%.

Part 1: Data Extraction
Step 1: View list of studies. To view the list of studies, go to Review -> Level 3 -> Health Literature Inventory Extraction form. Alternatively, you can access the list from the assessment home page. Under Level 3, click on "Unreviewed" to access the list of studies that have not been extracted yet.
Step 2: Select a study for extraction. Select a study by clicking on it, and a new tab will open.
Once the new tab opens, search the RefID at https://heronet.epa.gov/heronet/index.cfm/search and download the PDF.
Step 3: Data extraction -Part 1 Step 3a: Enter Author information. Use the format specified in the Distiller form.
Step 3b: Select all supplemental tags that apply Step 3c: Select other PFAS chemicals that were evaluated but not included in our screening Step 3d: Select "No" for the QC question if you are doing the primary extraction Select "No" to indicate that you are doing the initial data extraction. If you find during your review that the study does not meet PECO criteria, it should not be extracted. Select "study is not PECO-relevant: update full-text screening tags" to indicate that the study needs to be retagged at the full text review level.
Step 4: Data extraction -Part 2 Step 4a: Add a subform To begin extracting data, click on 'Add' and a subform will appear. You are now ready to enter information into the form.
Step 4b: Enter evidence type. This is a dropdown menu -human, animal, or PBPK Note: For this project, select "human (abbreviated extraction)" for human.
Step 4c: Select chemical form. This is a dropdown menu of the preferred names of each chemical.
Refer to the Master List in Teams for their synonyms.
The next questions about study design will differ depending on whether the study is in humans or animals.

Respiratory Effects
• Lung weight and histopathology • Nasal cavity histopathology Cohort A group of people is examined over time to observe a health outcome. Everyone belongs to the same population (e.g., general U.S. population; an occupational group; cancer survivors). All cohort studies (prospective or retrospective) consider exposure data from before the occurrence of the health outcome.
Case-control Cases (people with the health outcome) and controls (people without the health outcome) are selected at the start of a study. Exposure is determined and compared between the two groups. A case-control study can be nested within a cohort.

Ecological
The unit of observation is at the group level (e.g., zip code; census tract), rather than the individual level. Ecological studies are often used to measure prevalence and incidence of disease. Cannot make inferences about an individual's risk based on an ecological study.

Controlled Trial
Exposure is assigned to subject and then outcome is measured.