Common Reporting Method Underestimates PFAS in US Tap Water by Half

A study published October 23, 2026, in the Cell Press journal Cell Reports Sustainability estimates that common reporting methods may underestimate average PFAS (“forever chemicals”) concentrations in U.S. drinking water by roughly half (or more).

Researchers led by Gehendra Kharel (Texas Christian University), with coauthors from TCU, the University of Florida, and Oklahoma State University, reanalyzed more than 1.8 million PFAS measurement records from the EPA’s national drinking- water monitoring data (January 2026 release). The data covered 25 PFAS detected at least once across 9,991 community water systems. About 98% of individual test results fell below the EPA’s minimum reporting levels (MRLs), the concentrations labs can reliably measure and report and are typically treated as zero when calculating averages.

Using statistical models to estimate possible concentrations for the “<MRL” results (including generating many possible values per result), the team found that treating below- limit results as zero captured only about 44% of the modeled average concentration of those 25 PFAS. In other words, conventional summaries left out more than half of the estimated total.

Key modeled figures include:

• Combined average for the 25 PFAS: roughly 15.0 ng/L (parts per trillion) under the primary model (95% uncertainty interval ~13.8– 16.3 ng/L), versus ~6.6 ng/L when zeros are used.
• For the six PFAS covered by the EPA’s enforceable drinking- water limits (including PFOA and PFOS): conventional average ~2.6 ng/L, modeled estimate ~5.2 ng/L.

The study focuses on how measurements are summarized statistically; it does not claim the EPA’s lab testing itself is flawed, collect new samples, assess individual water systems’ safety, or quantify personal exposure and health effects.

The authors note that “below the reporting limit does not mean zero”, and that this distinction matters for understanding overall occurrence. Modeling also suggested some additional sampling locations could approach or exceed comparison thresholds near the EPA’s 4 ppt limits for certain compounds, though results varied by model.

PFAS are persistent synthetic chemicals linked to health concerns (cancers, cholesterol issues, reproductive effects, etc.) and found in products that can enter soil and water.

The EPA has set enforceable limits for a subset of them. The findings highlight a known statistical challenge with left- censored environmental data rather than new contamination discoveries.

The 4 ppt limit refers to the U.S. Environmental Protection Agency’s (EPA) legally enforceable Maximum Contaminant Level (MCL) of 4.0 parts per trillion (ppt), also expressed as nanograms per liter (ng/L), for two of the most studied PFAS compounds: PFOA (perfluorooctanoic acid) and PFOS (perfluorooctanesulfonic acid).

Background on the Limit

In April 2024, the EPA finalized the first National Primary Drinking Water Regulation (NPDWR) for six PFAS under the Safe Drinking Water Act. Key details include:

• PFOA and PFOS: Individual MCLs of 4.0 ppt each.
• The non- enforceable Maximum Contaminant Level Goals (MCLGs) for both are set at zero, because the EPA determined they are likely carcinogenic to humans and there is no known safe threshold for cancer risk.
• Other regulated PFAS (as of the 2024 rule): PFHxS, PFNA, and HFPO- DA (GenX chemicals) each have MCLs of 10 ppt; mixtures of certain PFAS are regulated via a Hazard Index of 1.

Public water systems must complete initial monitoring by 2027 and achieve compliance with the MCLs by 2029 (with some later proposals discussed for possible deadline extensions to 2031 for PFOA/PFOS while keeping the 4 ppt levels).

Compliance is based on running annual averages. The 4 ppt level was chosen as the lowest concentration that is both protective of public health and technologically and economically feasible for water systems to achieve and reliably measure.

One ppt is an extremely low concentration, equivalent to about one drop of water in 20 Olympic- sized swimming pools.

Health Risks Associated with PFAS Exposure

PFAS are called “forever chemicals” because they are highly persistent in the environment and the human body (they bioaccumulate). They have been detected in the blood of nearly all Americans tested. Exposure can occur through drinking water, food, consumer products, and other pathways. The EPA and other scientific reviews link certain levels of exposure (especially to PFOA and PFOS) to multiple adverse health effects. Research continues, particularly on low-level, long-term exposure.⁠

Documented or strongly associated risks include:

• Cancer: Increased risk of kidney and testicular cancers (and evidence for others in some studies). The EPA classifies PFOA as “likely to be carcinogenic to humans.”
• Immune system effects: Reduced ability to fight infections and decreased vaccine response (particularly in children).
• Developmental and reproductive effects: Low birth weight, developmental delays in children, decreased fertility, and high blood pressure or preeclampsia in pregnant women.
• Cardiovascular and metabolic effects: Elevated cholesterol levels and increased risk of obesity.
• Other effects: Interference with hormones, liver effects, and potential impacts on thyroid and other systems.

These associations come from human epidemiological studies and supporting animal research. Risks can occur at very low doses, which is why the MCLG is zero for PFOA and PFOS and why the enforceable MCL was set so low.

Individual risk depends on the specific PFAS, concentration, duration of exposure, and personal factors (age, health status, etc.). Not every exposure at or near 4 ppt will cause harm, but the standards aim to reduce population- level risk.

In the context of the recent modeling study discussed earlier, the national average for the six regulated PFAS was estimated around 5.2 ng/L when accounting for below- reporting- limit results (versus ~2.6 ng/L when treating them as zero).

This is close to or slightly above the 4 ppt individual limits for PFOA/PFOS, illustrating why accurate characterization of low- level contamination matters for understanding overall exposure.

Censoring- aware models quantify systematic underestimation of PFAS occurrence in US drinking- water monitoring

Core Finding

Environmental monitoring data for PFAS (and many other chemicals) frequently report results only as “below the laboratory reporting limit” (minimum reporting level, or MRL). The common practice of treating these non- detects as zero systematically underestimates average concentrations.

Using the EPA’s fifth Unregulated Contaminant Monitoring Rule (UCMR 5) January 2026 data release (~1.86 million PFAS records from 9,991 community water systems), the authors show that zero- substitution captures only about 44% of the model- estimated population- weighted average concentration for 25 data- identified PFAS analytes. For the six PFAS covered by the 2024 National Primary Drinking Water Regulation (NPDWR), zero-substitution captures about 50%.

Methods (High- Level)

• The team treated results below MRLs as left- censored observations rather than zeros.
• They developed and validated a censoring- aware multiple- imputation (MI) framework.
• This was benchmarked against direct censored maximum likelihood estimation (Cens- MLE), regression on order statistics (ROS), and alternative distributional assumptions (lognormal, gamma, Weibull).
• The primary model used a regularized lognormal distribution.
• They performed extensive validation (held- out observations, multi- analyte masking, semi- synthetic simulations with known truth) and robustness checks (by water source, system size, marginal family, etc.).
• Population- weighted national averages were calculated, with full uncertainty propagation.

Key Quantitative Results

• Under the primary regularized lognormal model:
  • Zero substitution captured 44% of the estimated average for 25 analytes.
  • Zero substitution captured 50% for the 6 NPDWR PFAS.
• PFOA and PFOS together accounted for only 27- 31% of the main censoring gap; other PFAS contributed substantially.
• Results were robust: refitting by source- water type or system size changed national estimates by less than 5%.
• Near- MCL screening effects (around the 4 ppt / 4.0- 4.05 ng/L threshold for PFOA/PFOS) were modest, the models added only a limited number of additional sampling locations near that threshold.

Important Caveats Stated by the Authors

• The analysis examines how measurements are summarized, not regulatory compliance of any specific water system.
• It does not quantify actual human exposure, intake dose, toxicity- weighted risk, or health outcomes.
• “Sub- MRL concentration contribution” is a model- estimated statistical increment, not a measured physical mass delivered to consumers.
• The work supports more transparent, uncertainty- aware interpretation of heavily censored data and can help prioritize where additional sampling or clearer communication of uncertainty would be most useful.

Science for Society Statement (from the paper)

Treating below- limit results as zero can make low- level occurrence appear smaller than the data support. The framework does not determine compliance or health risk, but it can help agencies and utilities decide where additional sampling and clearer communication of uncertainty are needed.

In short, the paper provides rigorous statistical evidence that the widespread convention of zero- substituting non- detects understates national PFAS occurrence estimates in U.S. drinking- water monitoring data, and it offers a reproducible, validated alternative approach for handling such heavily censored datasets.

Title: Censoring-aware models quantify systematic underestimation of PFAS occurrence in US drinking-water monitoring

Journal: Cell Reports Sustainability, Open access

DOI: 10.1016/j.crsus.2026.100893

Provided: Cell Press

Lead author and corresponding author: Gehendra Kharel (Texas Christian University)

Co- authors: Andrei P. Kirilenko (University of Florida), Omar R. Harvey (Texas Christian University), Chris B. Zou (Oklahoma State University)