PFOS Epidemiology and Cancer Toxicity

RoseyFields evaluates the human epidemiological evidence for PFOS as follows.

1. Study Populations and Data Sources

As with PFOA, the epidemiological database for PFOS is large and draws on several distinct categories of study populations, and the consistency of findings across these different populations is central to how RoseyFields evaluates the strength of the causal case for PFOS.

Occupational cohorts consist of workers at PFOS manufacturing facilities, most notably the 3M production plant in Decatur, Alabama, which was the primary U.S. manufacturing site before 3M’s voluntary phase-out of PFOS production between 2000 and 2002. This cohort experienced markedly elevated serum concentrations (mean serum PFOS around 94 ng/mL, dramatically above general-population background) and has been followed for both mortality and self-reported medical conditions, providing high-exposure data that complements the general-population literature, though the cohort’s statistical power is limited by its size and, for cancer outcomes specifically, by follow-up periods too short for many workers to allow adequate cancer latency.

Community or contamination-site cohorts are drawn from populations living near PFOS-emitting facilities or in areas with contaminated drinking water. The West Virginia/Ohio C8 Health Project — the same large community cohort central to RoseyFields’ PFOA page — also measured PFOS and has been used in numerous PFOS analyses, since residents of these water districts had elevated exposure to both chemicals. A separate, independently studied high-exposure community, referred to in the literature as the “Isomers of C8 Health Project,” involves residents of Shenyang, China, with PFOS levels substantially above general background (medians in the range of roughly 10–24 ng/mL) and has produced several analyses of blood pressure and metabolic outcomes. A third community-based population comes from the Veneto Region of Italy, where widespread PFAS contamination of drinking water has enabled large cross-sectional studies of young adults.

General-population biomonitoring studies again rely heavily on NHANES, the CDC’s ongoing, nationally representative survey that has measured serum PFOS since 1999, permitting investigators to examine associations at background exposure levels representative of the general public rather than only at unusually high occupational or community exposure levels.

Birth cohorts and prospective pregnancy cohorts — including the Faroe Islands birth cohorts, the Danish National Birth Cohort, and numerous other national and regional birth cohorts across the United States, Europe, and Asia — have measured PFOS in maternal and/or cord serum and followed children prospectively for outcomes including vaccine antibody response, birth weight, and postnatal growth. The developmental/birth-weight literature for PFOS is unusually large, drawing on more than thirty individual cohort studies across multiple continents.

As RoseyFields’ PFOA page discusses at greater length, using multiple, diverse populations strengthens causal inference for PFOS in the same ways: replication across occupational, community, and general-population settings with different exposure magnitudes reduces the likelihood that a finding is a chance result specific to one population; populations with different co-occurring PFAS profiles (for example, the Decatur, Alabama occupational cohort, where PFOS exposure was overwhelmingly dominant) help assess whether associations are attributable specifically to PFOS rather than to correlated PFAS; and the combination of prospective birth-cohort data with cross-sectional general-population data helps address concerns about temporality and reverse causation. Because PFOS and PFOA frequently co-occurred in the same contaminated communities and consumer products historically, several of the same cohorts (the C8 Health Project chief among them) and even the same individual studies appear in both chemicals’ epidemiological literatures, which is itself a source of both strength (cross-validation using the same well-characterized populations) and complexity (the two chemicals’ effects can be difficult to fully disentangle in these shared cohorts, a point addressed further in Section 5).

2. Overview of Human Health Outcomes Investigated, by Body System

Consistent with the U.S. EPA’s 2024 systematic review of PFOS and supporting analyses in the ATSDR Toxicological Profile for Perfluoroalkyls, human epidemiological investigation of PFOS has spanned essentially every organ system, mirroring the breadth of the PFOA literature.

Outcomes with strong, consistent, quantitatively modeled evidence — those for which EPA’s systematic review concluded the evidence “indicates” that PFOS exposure is likely to cause the effect in humans, and which EPA used to derive quantitative toxicity values — are: hepatic injury (elevated ALT); immune suppression (reduced antibody response to childhood vaccination); cardiovascular/lipid effects (elevated serum total cholesterol and LDL cholesterol); and adverse developmental effects (reduced birth weight). Cancer as a general hazard category also reaches this tier, with EPA concluding PFOS is “Likely to Be Carcinogenic to Humans,” although — as discussed in Sections 3 and 6 — the epidemiological cancer evidence for PFOS is more mixed across tumor types than for PFOA, and the cancer slope factor for PFOS was ultimately derived from animal rather than human data.

Outcomes with suggestive but inconsistent evidence include: bladder, prostate, liver, and breast cancer (each showing positive associations in some studies, particularly high-exposure or genetically/hormonally susceptible subgroups, but null or inconsistent findings in others); thyroid disease and thyroid-stimulating hormone (TSH) perturbation; nervous system and neurodevelopmental outcomes in children, including attention-deficit/hyperactivity disorder, autism spectrum indicators, and hearing; blood pressure and hypertension (positive in several general-population and high-exposure-community studies, particularly among women, but not uniformly replicated); and elevated infectious disease incidence in children, a biologically coherent companion finding to the vaccine-response data.

Outcomes investigated with null or largely inconsistent findings include: HDL cholesterol and triglycerides (the available evidence does not support a consistent directional association with PFOS, in contrast to the more consistent findings for total and LDL cholesterol); most other cardiovascular endpoints beyond blood pressure and lipids, including coronary heart disease, stroke, and carotid atherosclerosis (mixed, generally non-significant findings); prostate-specific antigen as a prostate cancer biomarker; and reproductive hormone and fertility endpoints, where the available studies are comparatively few and inconsistent. As with PFOA, several of these outcomes have a comparatively small evidence base, such that “null” is often more accurately characterized as “inadequate to determine an association” than as an affirmative finding of no effect.

3. Strong-Evidence Outcomes: Key Studies, Exposure Metrics, and Effect Sizes

Hepatic injury. As with PFOA, the most extensively replicated finding for PFOS is a positive association between serum PFOS and serum ALT. Multiple NHANES-based analyses of the general U.S. population found statistically significant positive associations between PFOS and ALT, and this finding was corroborated in an independent Swedish longitudinal cohort in which within-person changes in PFOS over time tracked with changes in liver enzymes, and in a Chinese occupational/community cohort. One large NHANES analysis found the positive ALT association was concentrated in obese participants, with generally null associations in non-obese participants, suggesting metabolic status may modify the relationship. In children and adolescents, a national biomonitoring study found a positive association in girls in the highest exposure quartile, while another cohort reported an inverse association, suggesting less consistency in children than in adults. Historical occupational data from the Decatur, Alabama plant found a small but statistically significant excess of cirrhosis mortality among PFOS production workers, though this finding derives from a single small cohort. Median/geometric-mean serum PFOS concentrations in the informative general-population studies were as low as approximately 0.57 ng/mL, and the direction of the ALT association was broadly consistent from this background level up through much higher occupational and community exposure levels — a pattern EPA identified as a key basis for its “evidence indicates” hepatic hazard conclusion, though EPA also noted that confounding by co-occurring PFAS could not be ruled out, since the one general-population study that performed formal multi-pollutant modeling found the PFOS-ALT association attenuated after adjustment for other PFAS.

Immune suppression. As with PFOA, the strongest human evidence for an immune effect of PFOS is reduced antibody response to routine childhood vaccination, particularly against tetanus and diphtheria. In the Faroe Islands birth cohorts, prenatal and childhood serum PFOS concentrations were inversely associated with antibody titers against diphtheria and tetanus at multiple ages, with several associations reaching statistical significance; a follow-up multi-pollutant analysis adjusting for co-occurring PFAS found the PFOS association was not materially attenuated after accounting for PFOA, PFHxS, and PFNA, addressing a key potential confounding concern for this cohort. A cross-sectional study of Greenlandic children found elevated odds of falling below the clinically protective diphtheria antibody threshold with increasing PFOS exposure. NHANES-based analyses of U.S. adolescents found significant inverse associations between PFOS and rubella and mumps antibody response, and a study of Guinea-Bissau children found inverse associations between PFOS and measles antibody response, particularly among children who had received fewer routine vaccine doses. In adults, evidence was less consistent: a Faroese cohort followed to age 28 found associations that varied in direction across vaccines and exposure windows, suggesting the immune effects of PFOS exposure may be most relevant during early-life developmental windows rather than in adulthood. EPA’s hazard judgment for PFOS immune effects was driven primarily by the reduced antibody response findings, observed at median serum concentrations as low as approximately 0.8 ng/mL in the studies reporting an adverse effect.

Cardiovascular / serum lipid effects. A consistent, positive association between serum PFOS and total and LDL cholesterol has been observed across dozens of general-population, occupational, and high-exposure-community studies. Multiple NHANES cycle analyses of the general U.S. population found statistically significant positive associations between PFOS and total cholesterol, and a high-exposure community study in the Veneto region of Italy found a clear concentration-response relationship between PFOS and elevated total cholesterol risk across exposure quartiles and deciles. Six medium-confidence studies examining LDL cholesterol all reported positive associations with PFOS, though few individually reached statistical significance, suggesting a real but comparatively modest effect. By contrast, the evidence for HDL cholesterol and triglycerides was inconsistent in both direction and magnitude across studies and did not support a clear association in either direction. Blood pressure findings were also broadly, though not universally, positive: a large NHANES analysis found a statistically significant positive association between PFOS and systolic blood pressure, and two overlapping studies of the high-exposure Shenyang, China community found statistically significant increases in both systolic and diastolic blood pressure and in risk of hypertension, an association that appeared stronger among women in several of these studies. EPA’s cardiovascular hazard judgment was driven primarily by the serum lipid findings, observed at median PFOS levels ranging from roughly 3.7 to 36.1 ng/mL across the informative studies, and was considered further supported, though not independently sufficient, by the blood pressure findings.

Developmental effects (birth weight/fetal growth). Reduced birth weight associated with maternal PFOS exposure is supported by an unusually large literature — more than thirty individual cohort studies from multiple continents. The majority of informative studies reported at least some deficit in mean birth weight associated with higher PFOS exposure, with reported deficits generally ranging from approximately 14 to 93 grams per unit increase in exposure across most studies, though a smaller number of studies (generally lower-confidence ones) reported much larger deficits. Notably, the ten highest-confidence studies in this literature showed a narrower, more consistent range of effect sizes (approximately 14 to 18 grams per unit exposure increase) than the lower-confidence studies, a pattern consistent with less-biased studies converging on a smaller but more reliable estimate of true effect size. A large Swedish cohort study was selected by EPA as a co-critical, high-confidence study underlying its PFOS reference dose derivation. Findings were also somewhat sensitive to the timing of maternal blood sampling during pregnancy, with associations somewhat more consistently observed when serum was sampled later in pregnancy or from cord blood, a pattern the researchers attribute at least in part to pregnancy-related hemodynamic changes affecting measured PFOS concentrations rather than necessarily reflecting differences in true biological windows of susceptibility. Median maternal serum PFOS concentrations in the informative studies generally ranged from about 5.0 to 30.1 ng/mL, and the direction of effect was broadly consistent across this range and across the many independent cohorts and countries represented in this literature.

Cancer. The epidemiological cancer evidence for PFOS spans several tumor types with varying degrees of consistency, and is less uniformly concentrated on a single cancer type than the PFOA literature’s focus on kidney and testicular cancer. For kidney cancer (renal cell carcinoma), the same large case-control study nested within the National Cancer Institute’s PLCO cancer screening trial that provided key PFOA evidence also found a statistically significant increased risk associated with PFOS exposure in the highest exposure quartile and in continuous analyses; however, unlike the PFOA finding in the same study, the PFOS association was attenuated after adjustment for co-occurring PFAS, indicating the possibility that some or all of the unadjusted PFOS-kidney cancer association reflects confounding by correlated PFAS exposure (including PFOA) rather than an independent PFOS effect. For bladder cancer, the Decatur, Alabama occupational cohort found elevated bladder cancer mortality in an initial analysis and elevated, though not statistically significant, incidence in a follow-up analysis with additional years of observation, while a Danish general-population cohort found no elevated risk, leading EPA to characterize the evidence as “plausible,” particularly for high-exposure occupational settings, rather than as clearly established. For liver cancer, earlier studies found imprecise null results, but a more recent, medium-confidence nested case-control study within the Multiethnic Cohort found a statistically significant increased risk comparing high to low PFOS exposure. For prostate cancer, findings across the occupational cohort, a Danish case-control study, and a C8 Health Project biomarker study were predominantly null or non-significant, with one NHANES analysis finding a significant inverse association. For breast cancer, the majority of the epidemiological literature on PFOS cancer outcomes, findings were highly heterogeneous: several studies found associations concentrated in specific genetically or hormonally defined subgroups (for example, individuals with a specific aromatase-gene polymorphism, or estrogen/progesterone-receptor-positive tumors specifically), while other well-conducted case-control and nested case-control studies found no association, and at least one Japanese population study found a significant inverse association. EPA’s own weight-of-evidence review characterized the epidemiological cancer evidence overall as “mixed results across tumor types” providing “supporting” rather than primary evidence for its cancer classification, in contrast to the more central role epidemiology played in EPA’s PFOA cancer determination.

4. Study Quality Considerations

Confounding by co-occurring PFAS. As with PFOA, distinguishing PFOS-specific effects from those attributable to correlated PFAS is a recurring methodological challenge, and the PFOS literature provides a particularly instructive example of how this concern plays out differently across outcomes. For the immune/vaccine-response findings, a multi-pollutant re-analysis of the Faroese cohort data found the PFOS association was not materially attenuated after adjusting for PFOA, PFHxS, and PFNA, supporting an independent PFOS effect. For the kidney cancer finding in the PLCO-nested case-control study, by contrast, the PFOS association was attenuated after adjustment for co-occurring PFAS (while the parallel PFOA association in the same study was not), directly illustrating that confounding by correlated PFAS can materially affect conclusions for some outcomes and chemicals but not others, and underscoring why single-pollutant findings should be interpreted cautiously until multi-pollutant sensitivity analyses are available. For the hepatic/ALT finding, the single general-population study performing multi-pollutant modeling likewise found some attenuation of the PFOS association, adding uncertainty to an otherwise fairly consistent single-pollutant literature.

Reverse causation. As with PFOA, reverse causation is a recognized concern for outcomes involving renal function or other physiological processes that could plausibly affect PFOS elimination rather than only be affected by PFOS exposure; because PFOS, like PFOA, is eliminated primarily via the kidney, altered renal function could in principle affect measured serum PFOS independent of any causal effect of PFOS on the outcome being studied. This concern is compounded for cancer outcomes specifically by the general possibility that a developing or undiagnosed malignancy could alter a patient’s metabolism, body composition, or renal function in ways that affect measured PFOS concentrations at or near the time of diagnosis; several of the cancer studies discussed in Section 3 measured PFOS in serum collected close to the time of cancer diagnosis, which the EPA review specifically flagged as a source of potential reverse-causation bias given PFOS’s multi-year elimination half-life and the resulting difficulty of distinguishing pre-disease exposure from disease-altered biomarker levels using a single measurement.

Exposure misclassification. PFOS’s long serum half-life means a single serum measurement reasonably reflects several years of cumulative exposure, which mitigates concern about misclassification from one-time sampling relative to chemicals with shorter half-lives. However, misclassification concerns remain more significant for occupational cohort studies relying on job-title- or era-based exposure categories rather than direct biomarker measurement for the full employment history, and for outcomes with long latency such as cancer, where a measurement taken decades after initial exposure, or shortly before diagnosis, may not accurately represent the exposure levels and windows most relevant to disease causation. EPA also specifically noted that declining background serum PFOS levels in the U.S. population since the early 2000s (following the 3M production phase-out) complicate the interpretation of studies using only recent NHANES cycles as representative of historical peak exposure, and the Agency’s own dose-response modeling therefore relied on exposure windows more likely to capture historical U.S. peak PFOS exposure.

Co-exposure to other chemicals and confounding by demographic and lifestyle factors. Beyond co-occurring PFAS specifically, some PFOS study populations have documented co-exposure to other contaminants or health conditions that complicate interpretation: for example, several of the serum lipid and blood pressure studies discussed in Section 3 were conducted in populations with high rates of diabetes, obesity, or participation in weight-loss or lifestyle-intervention clinical trials, and study authors themselves noted uncertainty about whether these underlying health conditions explained some of the heterogeneity in findings across studies. As with PFOA, incomplete adjustment for socioeconomic status and lifestyle factors in some individual studies remains a general source of residual uncertainty, though the pattern of replication across occupational, community, and general-population studies with different confounding structures provides some reassurance that residual confounding alone is unlikely to fully explain the more consistent findings (hepatic ALT, serum lipids, birth weight, and vaccine antibody response).

5. Independent Expert Panel and Agency Weight-of-Evidence Conclusions

The U.S. EPA’s 2024 final toxicity assessment, developed under the IRIS program and subjected to external peer review by the EPA Science Advisory Board, concluded that PFOS is “Likely to Be Carcinogenic to Humans” under EPA’s Guidelines for Carcinogen Risk Assessment, based on a weight-of-evidence evaluation that explicitly characterized the epidemiological cancer evidence as showing “mixed results across tumor types” that provide “supporting,” rather than primary, evidence, with the central support instead coming from a single chronic rat bioassay demonstrating statistically significant, dose-related, multi-site (liver and pancreatic islet cell) and multi-sex tumor findings. This is a meaningfully different evidentiary structure than EPA’s parallel PFOA cancer classification, where epidemiological findings for kidney and testicular cancer played the primary role; for PFOS, EPA’s cancer slope factor was derived entirely from the animal bioassay data because the epidemiological data were not considered suitable for quantitative cancer dose-response derivation. Separately, EPA concluded that the human epidemiological evidence “indicates” — the Agency’s second-highest confidence tier — that PFOS causes hepatic injury, immune suppression, adverse cardiovascular/lipid effects, and developmental toxicity, mirroring its PFOA conclusions for these same four noncancer outcome categories. EPA additionally identified endocrine (thyroid) and nervous system effects as having evidence “suggestive” of an association bordering on “indicates,” a somewhat different pair of borderline outcomes than the reproductive/endocrine outcomes flagged in EPA’s parallel PFOA assessment. Unlike PFOA, RoseyFields is not aware of an independent expert panel comparable to the C8 Science Panel that has issued a formal “probable link” determination specifically for PFOS and a named disease outcome; the C8 Science Panel’s original “probable link” determinations were focused on PFOA given that PFOA, not PFOS, was the dominant contaminant in the Ohio/West Virginia community water supply, though the C8 Health Project cohort’s biomonitoring data on PFOS has since been used extensively in the broader epidemiological literature discussed above.

6. Use of Epidemiological Evidence to Support Causation Opinions

Epidemiological evidence of the kind summarized above forms one of the two principal evidentiary pillars — alongside toxicological and mechanistic evidence from animal and in vitro systems — that support both general and specific causation opinions in PFOS litigation, and functions similarly to how it is used for PFOA for the four shared noncancer outcome categories (hepatic injury, immune suppression, cardiovascular/lipid effects, and developmental toxicity). For these outcomes, the consistency of findings across occupational, community, and general-population studies conducted by independent research groups, using different study designs and exposure assessment approaches, supports core Bradford Hill considerations for causal inference — consistency, dose-response gradient, temporality (particularly from the prospective birth-cohort literature), and biological plausibility, independently reinforced by coherent mechanistic and animal findings such as PPAR-mediated hepatic effects and altered lipid homeostasis in rodent and non-human primate models.

For cancer specifically, however, the PFOS epidemiological evidence supports a general causation analysis that is necessarily more qualified than for PFOA. Because the epidemiological findings vary meaningfully by tumor type — more consistent for bladder cancer in high-exposure occupational settings and kidney cancer before adjustment for co-occurring PFAS, more mixed for liver and prostate cancer, and highly heterogeneous (with subgroup-specific and even directionally inconsistent findings) for breast cancer — and because EPA itself relied on animal rather than epidemiological data for its quantitative cancer dose-response assessment, general causation opinions addressing PFOS and a specific cancer type should be calibrated to the tumor type at issue rather than treated uniformly, and should account explicitly for the confounding-by-co-occurring-PFAS issue that measurably attenuated the kidney cancer association in the largest available study. For the four consistent noncancer outcomes, by contrast, the epidemiological record is robust enough to support general causation opinions with a similar evidentiary foundation to those available for PFOA.

For specific causation in an individual case, this epidemiological literature again provides the quantitative building blocks — dose-response relationships, effect-size estimates, and characterized exposure ranges — needed to evaluate whether a plaintiff’s individual exposure history falls within ranges associated with increased risk in the epidemiological studies, and to characterize the magnitude of any resulting contribution to a diagnosed condition, in conjunction with individual medical and exposure history and differential diagnosis. Because PFOS and PFOA frequently co-occur in the same contaminated water supplies, consumer products, and human serum samples, specific causation analysis in cases involving both chemicals should account for their shared and independent contributions where the science permits that distinction (as it does, to varying degrees, for the outcomes discussed in Sections 3 through 5), and should be appropriately cautious where — as with kidney cancer — the available evidence indicates that a substantial portion of an apparent single-chemical association may in fact reflect co-exposure to correlated PFAS.

Bibliography