EPA May Significantly Underestimate PFAS in U.S. Drinking Water, Study Suggests
Need to Know
- Researchers found that EPA's drinking water contaminant monitoring data interpretation method, in which PFAS concentrations that fall below laboratory reporting limits are treated as "zero," may significantly underestimate "forever chemical" occurrence in U.S. drinking water
- Based on EPA UCMR5 data, researchers found that zero substitution underestimated aggregate PFAS occurrence by approximately 56 percent compared to their primary statistical model
- The researchers suggested that censoring-aware statistical methods and improved laboratory sensitivity could provide more accurate PFAS occurrence estimates to inform drinking water monitoring and prioritization.
A new study has found that commonly used methods for interpreting per- and polyfluoroalkyl substances (PFAS) monitoring data may significantly underestimate the occurrence of these “forever chemicals” in U.S. drinking water. Specifically, researchers demonstrated that treating PFAS concentrations below laboratory minimum reporting levels as “zero” can substantially reduce estimates of aggregate PFAS occurrence.
Published in Cell Reports Sustainability, the study was conducted by researchers from Texas Christian University, the University of Florida, and Oklahoma State University. The researchers analyzed approximately 1.86 million PFAS measurement records from 10,297 public water systems included in the U.S. Environmental Protection Agency's (EPA's) fifth Unregulated Contaminant Monitoring Rule (UCMR 5) dataset, released in January 2026.
Only 2 percent of the records contained PFAS detections above laboratory minimum reporting levels, meaning that 98 percent of measurements were reported as below those thresholds. The researchers explained that concentrations falling below the laboratory's reporting limit do not necessarily indicate the total absence of PFAS.
To assess how this data interpretation affects national occurrence estimates, the researchers developed a statistical framework using censoring-aware multiple imputation (MI), which estimates possible concentrations below reporting limits while accounting for uncertainty. The approach was evaluated against other statistical methods, including censored maximum likelihood estimation and alternative distribution models.
Zero Substitution Underestimated PFAS Occurrence by Approximately 56 Percent
Using the researchers' primary statistical model, treating concentrations below reporting limits as zero underestimated the population-weighted aggregate occurrence of 25 PFAS analytes by 55.9 percent, relative to the censoring-aware estimate.
Specifically, zero substitution produced an estimated aggregate PFAS concentration of 0.00663 micrograms per liter (μg/L), compared with 0.01504 μg/L using the censoring-aware approach.
For the six PFAS included in EPA's April 2024 National Primary Drinking Water Regulation framework (which set the first national, legally enforceable maximum contaminant levels for PFAS in drinking water, but were later rolled back by the Trump Administration), zero substitution underestimated the corresponding occurrence estimate by 49.9 percent.
Looking for quick answers on food safety topics?
Try Ask FSM, our new smart AI search tool.
Ask FSM →
The researchers found that the direction of underestimation persisted across different statistical models, water sources, and water system sizes, although its magnitude varied depending on the modeling assumptions. Separately fitting models by source-water category and system size changed pooled national estimates by less than 5 percent.
Perfluorooctanoic acid (PFOA) and perfluorooctanesulfonic acid (PFOS) accounted for approximately 55–60 percent of the estimated occurrence gap among the six PFAS included in the 2024 regulatory framework. However, the researchers found that other PFAS contributed substantially to the broader aggregate occurrence gap.
Implications, Suggestions for PFAS Monitoring and Regulatory Screening
Despite the substantial differences in aggregate PFAS occurrence estimates, the researchers found that censoring-aware models produced relatively modest changes in the number of drinking water sampling locations approaching EPA's maximum contaminant levels (MCLs) for PFOA and PFOS.
Depending on the statistical model, the censoring-aware approach identified an additional 16–33 sampling locations for PFOA and 4–25 locations for PFOS at or above the study's 4.05-nanogram-per-liter (ng/L) comparison threshold. No locations with exclusively below-reporting-limit results crossed the threshold.
The researchers emphasized that their findings represented model-based occurrence estimates, not direct measurements of PFAS concentrations below laboratory reporting limits. The analysis also did not establish regulatory compliance, consumer exposure, or associated health risks.
Additionally, the researchers demonstrated that lowering laboratory reporting limits could substantially reduce the occurrence gap associated with zero substitution. In hypothetical scenarios involving PFOA and PFOS, reducing reporting limits from 4 ng/L to 0.5 ng/L increased the proportion of modeled occurrence captured by zero substitution to more than 92 percent.
Overall, the researchers concluded that censoring-aware statistical approaches could complement improvements in laboratory analytical sensitivity by providing more complete, uncertainty-aware interpretations of PFAS monitoring data. They suggested that such methods could support drinking water monitoring, screening, and prioritization efforts, while acknowledging that additional information would be necessary for regulatory compliance and health risk determinations.
