Mortality study of a PFOA-exposed cohort in the mid-Ohio Valley

Abstract

Background:

The C8 Science Panel previously conducted studies of perfluorooctanoic acid (PFOA) exposure in Ohio Valley residents, finding probable links with kidney and testicular cancers, raised cholesterol, ulcerative colitis, thyroid disease, and pre-eclampsia. Findings were largely based on a cohort of 32,254 persons with high PFOA exposure followed through 2011 for disease incidence. We have now followed this cohort through 2021 for mortality.

Methods:

Deaths were ascertained via the National Death Index (NDI). We extended cumulative PFOA serum estimates from 2011 until 2021. We used internal comparisons via Cox regression models to analyze cause-specific mortality for 22 causes of death in relation to natural log transformed cumulative PFOA exposure (continuous and in quartiles) with a variety of lags, controlling for confounders. We also considered multiple cause of death.

Results:

There were 5709 deaths in the cohort through 2021. We found a positive significant trend (p<0.05) in underlying cause of death analyses between the log cumulative serum level and brain cancer [HR 1.25 (1.05, 1.50)] and a suggestive but non-statistically significant association with bladder cancer [HR 1.17 (0.98, 1.39). . In stratified analysis, pancreatic cancer showed a positive significant trend among older participants (>72 years) [HR 1.23 (1.05, 1.43)]. No other causes showed marked positive trends, except for stroke among workers (13% of total cohort).

Conclusion:

PFOA exposure was significantly associated with mortality from brain cancer, and pancreatic cancer among older subjects. Pancreatic cancer has been associated with PFOA in animal studies, with equivocal evidence in other cohort studies. There was some suggestion of an increase of bladder cancer Caution is warranted due to multiple comparisons.

Introduction

Perfluorooctanoic acid (PFOA) has been used in many commercial applications since the early 1950s and is one of a class of perfluoroalkyl substances (PFAS), which are ubiquitous in the environment. PFOA is considered a ‘legacy’ PFAS, partly because it is the one of the earliest created and used, and partly because it is part of a group of the early longer-chained PFAS which have longer half-lives (approximately 2–4 years for PFOA) (ATSDR 2021). PFOA, like other PFAS, is known as a ‘forever’ chemical, because it persists indefinitely in the environment. It is currently present at low levels in virtually everyone’s blood in the US and other industrialized countries; it is absorbed from the diet, from contaminated water systems, and in the past from some commercial non-stick products (ATSDR 2021). Concerns about exposure to PFOA have increased over time as more people have become aware of its presence in blood and potential adverse health effects.

PFOA’s impact on chronic disease remains unclear (Steenland et al. 2020, Steenland and Winquist 2021, ATSDR 2021). PFOA has been positively associated primarily with kidney and testicular cancer incidence, increased cholesterol (but not with increased heart disease), ulcerative colitis (an auto-immune disease), immunosuppression, altered liver enzymes (but not clearly with liver disease), gestational hypertension, and non-malignant thyroid disease (ATSDR 2021). However, these health effects have not been consistently observed across studies (ASTDR 2021).

The International Agency for Research on Cancer (IARC) in 2023 declared PFOA to be ‘carcinogenic to humans’ (Group 1), based on sufficient animal evidence, strong mechanistic evidence, and limited human evidence (Zahm et al. 2024). This conclusion was based on the evidence for kidney and testicular cancer. The U.S. Environmental Protection Agency (EPA) has determined that PFOA is “Likely to Be Carcinogenic to Humans” (EPA 2024). Meanwhile the EPA has recently issued regulations to sharply reduce allowable PFOA levels in drinking water, to 4 ppt, down from a previous recommendation of 70 ppt (www.epa.gov/sdwa/and-polyfluoroalkyl-substances-pfas).

Epidemiologic findings about PFOA have generally come from 1) general population studies with relatively low exposure contrasts, 2) occupational studies with high exposure contrasts but typically small sample sizes, and 3) studies in a small number of highly exposed communities, with high exposure contrasts and relatively large numbers of exposed subjects. Among the last group is the exposed community near a chemical production plant in the mid-Ohio Valley, which has been studied by the C8 Science Panel (c8sciencepanel.org). The C8 Science Panel was created in 2004 as part of a legal settlement. PFOA was measured in 69,000 participants during 2005/2006 as part of the legal settlement. Participants had mean and median PFOA serum levels of 82.9 ng/ml and 28.2 ng/ml, respectively (Frisbee et al. 2009) compared to mean levels of 4 ng/ml in the general adult US population during 2005–2006 (Jain 2018). The C8 Science Panel conducted a variety of studies for subsets of this population through 2012 and determined that six incident outcomes were probably linked to PFOA exposure: kidney and testicular cancer, ulcerative colitis, high cholesterol, thyroid disease, and pregnancy-induced hypertension.

One of the key studies of the C8 Science Panel was a follow-up of a subset of original adult participants. This 32,254-person adult cohort was followed through 2011 for disease incidence via matching with death and cancer registries and via self-reporting of chronic disease with verification through medical record review. Mortality for this cohort was not examined. Here we report on mortality in this cohort after adding 10 years of follow-up, through 2021.

Methods

Data sources and study cohort

The overall cohort consisted of adults from a community cohort and adults from a worker cohort. The detailed methods used for the combined community and worker cohort have been described previously (Winquist et al., 2013; Barry et al., 2013). The original C8 Health Project (C8HP) survey included 69,030 community participants enrolled between August 2005 and August 2006 (Frisbee et al 2009), of whom 54,457 were 20 years of age or older. Of these, approximately 74% (40,145) consented to further contact after the C8HP. Of these, 82% (32,712) then participated in one or two follow-up surveys during 2008–2011. The worker cohort (Steenland and Woskie 2012) included 5,791 workers who were employed at a manufacturing plant in Washington, West Virginia, plant for ≥ 1 day between 1 January 1948 and 31 December 2002. Of these, 4,391 workers were interviewed, but some were excluded for lack of retrospective serum estimates. Ultimately, the overall study cohort included 32,254 participants who were aged ≥ 20 years from both cohorts (note that some workers were also in the community cohort), participated in at least one survey in 2008–2011, and had exposure estimates as described below (Figure 1). Workers (n=3713) represented 13% of the cohort. During the survey, participants were interviewed on demographics, health characteristics, occupational and residential history, and medical history. All participants gave informed consent to participate, to match personal information to state cancer registries and the National Death Index, and to release medical records to study personnel. Medical records were protected in accordance with the Health Insurance Portability and Accountability Act (HIPAA) regulation. The study was approved by the Emory University Institutional Review Board.

Figure 1.

Cohort enrollment flowchart.

Retrospective serum PFOA concentration estimates

Retrospective serum PFOA estimates were originally calculated for every community participant for each year beginning from 1952 or year of birth, whichever came later, through 2011, based on a multistage modeling procedure (Shin et al. 2011a, 2011b). Briefly, estimates were based on historic amounts of PFOA emitted from the facility, the physicochemical properties of PFOA, and local geologic and meteorological data. Yearly serum PFOA concentration estimates for each individual considered each person’s residential history, drinking-water source, tap-water consumption, workplace water consumption, and a PFOA absorption, distribution, metabolism, and excretion model. Exposure estimates for worker cohort participants considered occupational exposure using a job–exposure matrix for specific jobs or departments. Estimates were based on 2,125 historical serum PFOA measurements, knowledge of process changes, and each participant’s work history (Woskie et al. 2012). After work at the plant ended, exposure estimates were decayed at a rate of 18% per year based on a presumed half-life of 3.5 years (Olsen et al. 2007) until they reached levels predicted for that individual by the community exposure model. The occupational model had a correlation of 0.80 with measured serum levels (Woskie et al. 2012). The modeled estimates for 2005/2006, incorporating both the community and worker exposure estimates had a Spearman correlation of 0.71 with 30,303 measured serum levels in 2005/2006 (Winquist et al. 2013).

Due to the decrease in plant emissions and improvement of drinking water systems (installation of filters for public water) by the end of 2007, exposure was assumed to have ceased in 2008 for community residents and workers, and serum concentration estimates for 2009 to 2011 were again based on a constant decay rate of 18% per year (Olsen et al. 2007). To estimate exposure for the years 2012 through 2021, we continued to assume a decay rate of 18% per year. When the estimated yearly concentration reached the value of minimum concentration in 2011 across the cohort (1 ng/ml), we stopped the decay and that level remained the concentration in the following years.

Death data and confirmation process

We identified deaths through the end of 2021 by matching with the National Death Index (NDI). The underlying cause of disease (UCOD) and multiple causes of death (2^nd to 5^th) were collected. We used ICD-9 [International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) (National Center for Health Statistics 2006)] and ICD-10 [International Statistical Classification of Diseases and Related Health Problems, 10th Revision (World Health Organization 1992)] to classify diseases into categories used by the NIOSH/CDC life table analysis system (Robinson et al. 2006). In total, we present results for 22 specific causes of death, including esophageal cancer, intestinal cancer, liver cancer, pancreatic cancer, lung cancer, breast cancer, prostate cancer, kidney cancer, bladder cancer, melanoma, brain cancer, Non-Hodgkin lymphoma, leukemia and aleukemia, multiple myeloma, diabetes mellitus, ischemic heart disease, cerebrovascular disease, chronic obstructive pulmonary disease (COPD), other disease of the heart, cirrhosis, chronic renal disease, and non-malignant thyroid disease (ICD codes used to define each cause of death are shown in Table S1). Diseases included in analyses/presentation were chosen among those which had at least 30 deaths as the underlying cause. We conducted analyses using both the underlying cause of death and multiple causes of death. In the analyses of multiple causes of death, if a cause of interest occurred as one of the four possible multiple causes, then that the death was counted as an occurrence for that cause. Time of death remained unchanged. In effect the use of multiple cause of death information increased the numbers of death for each cause compared to using only underlying cause– more so for causes which occur more frequently as contributory rather than underlying cause. Hence these analyses are particularly valuable for such causes [e.g., diabetes or thyroid disease (diseases which are often not fatal) vs. pancreatic cancer (almost always fatal)] (Steenland et al. 1992).

For non-malignant thyroid disease, which is generally not the underlying cause on death certificates, we only conducted multiple cause of death analyses.

Statistical Analysis

To examine the associations between time-varying cumulative PFOA serum concentrations and specific causes of death, Cox proportional hazards regression models were used with time-varying natural log transformed cumulative PFOA concentrations as our primary exposure metric, and age as the time scale. Tests for trend were calculated for both log transformed and untransformed cumulative PFOA concentrations.

Participants were followed from the age of 20 years or age in 1952 (the year after the first PFOA emissions), whichever came later, to age of death or age at end of follow-up in 2021, whichever came first. Participants who were born before 1920 were excluded (n=173). Each model was adjusted for the following potential confounders: time-varying alcohol consumption, time-varying smoking, gender, race (white/non-white), education, BMI and 10-year birth year period. Alcohol consumption, smoking information, and BMI were based on our last survey available, typically in 2011, and therefore involved some misclassification for years at risk from 2011–2021. Missing data were rare. Missing data on education (n=3) and BMI (n=25) were imputed using the most frequently reported categories, which were high school and overweight (25 ≤ BMI < 30), respectively.

Cumulative PFOA serum concentration (in nanograms per milliliter-years) was calculated as the sum of all yearly serum PFOA estimates for each year through 2021. We considered both untransformed and the natural log of cumulative PFOA concentration as a continuous variable and also used categorical variables for cumulative concentration quartiles. We also considered the p-value of the coefficient for continuous (untransformed or log) cumulative exposure as the test for trend.

We used 2-piece linear splines, and also restricted cubic splines with 3 knots as per Harrell et al. (1988). Quartile cut points were calculated based on cases for each cause-specific group of deaths and hence varied by cause-of-death. We also made quartile cut points based on all deaths and generally found similar results to those using cause-specific cut points. In addition, we considered models that lagged cumulative PFOA serum concentration by 5, 10 and 20 years in order to consider scenarios in which death could have been caused by exposure further in the past. We chose the best fitting lag for each cause of death (based on the AIC) for presentation of results. P-values for trends were considered statistically significant when they are <=0.05, and borderline significant when >=0.05 but <=0.10.

We checked the proportional hazard assumption for all outcomes, which requires that rate ratios be approximately constant over follow-up time (which is age in our case). It was not violated for any outcomes except for breast and pancreatic cancer (for breast cancer only at longer lags). For these two cancers, we ran additional analyses after dividing follow-up time before and after the median age at death for each of these cancers (age 72 for pancreas cancer, 64 for breast cancer).

While our basic analysis assumed prior deaths from other causes were uninformative and did not affect our estimation of exposure effects for specific causes, competing risks could occur if prior death from other causes could affect our observed exposure-response estimate for any specific cause. Therefore we conducted additional analyses using the Fine-Gray technique to compute (sub-distribution) rate ratios for several underlying causes, taking into account possible competing risks from other causes (Fine and Gray 1999, Lau et al. 2009).

Results

The final analytic sample included 32,081 participants, after excluding those born before 1920 (n=173). Their characteristics are summarized in Table 1. On average, participants were 63 years of age at the time of final follow up, with worker participants slightly older than community participants (68 years vs. 62 years). Most participants were white and non-Hispanic. Compared to the community group, the worker group was more likely to be male (80% vs. 42%) and have a bachelor degree or higher (36% vs. 15%). Participants who had worked at the plant had higher estimated annual PFOA serum levels compared to participants who never worked at the plant. The average years of follow up were 42 and 48 for community and worker participants, respectively.

Table1.

Demographic characteristics of cohort (n = 32,081) by community and worker groups, restricted to analytical files ^a.

Characteristics	Entire Cohort (n=32,081)	Community group (n=28,427)	Worker group (n=3,654)
Age at end of follow-up (Mean ± SD)	62.80 ± 14.44	62.14 ± 14.55	67.94 ± 12.41
Year of birth (Mean ± SD	1957 ± 15.30	1958 ± 15.37	1951 ± 13.45
Sex
Male	14,799 (46%)	11,902 (42%)	2,897 (79%)
Female	17,282 (54%)	16,525 (58%)	757 (21%)
Race/ethnicity a
White, non-Hispanic	31,267 (98%)	27,747 (98%)	3,520 (96%)
Other	814 (2%)	680 (2%)	134 (4%)
Education b
Less than high school	2,997 (9.34)	2,970 (10.45)	27 (0.74)
High school or certificate of equivalency	12,907 (40.24)	11,667 (41.04)	1,240 (33.94)
Some college	10,498 (32.73)	9,427 (33.16)	1,071 (29.32)
Bachelor or higher	5,676 (17.69)	4,361 (15.34)	1,315 (36.00)
BMI c
Mean ± SD	28.94 ± 6.53	29.05 ± 6.66	28.10 ± 5.30
Number of Deaths	5644	4908	736
Mean year of death	2015	2015	2012
Estimated annual PFOA serum level d
Median (range)	16.94 (2.60–3675.86)	13.11 (2.60–2453.69)	135.76 (4.12–3675.86)
Cumulative estimated PFOA level
2005: Median (range)	525 (81–205,460)	404 (81–42,440)	6278 (161–205,461
2020: Median (range)	667 (110–206,548)	511 (111–43,008)	6630 (180–206,548)

^aThe participants born prior to 1920 were excluded from analysis.

^bRace/ethnicity information missing for 257 participants (all from the occupational group) was imputed as white, non-Hispanic.

^cEducation information was missing for 3 participants.. The missing category was imputed as high school or certificate of equivalency.

^dBMI was missing 25 subjects. The missing category was imputed as the overweight category in later analysis.

^eAnnual serum level was calculated by dividing cumulative serum PFOA at the end of follow-up by number of follow up years for each subject. Community residents were followed for an average of 42 years, and workers were followed for an average of 48 years.

The adjusted proportional hazard model results for underlying causes and multiple causes are shown in Table 2, which shows results for each cause of death based on the best-fitting lag (0, 5, 10, or 20) based on the AIC. Overall, exposure did not increase the all cause mortality rate. The most positive overall trends were seen for bladder and brain cancers as underlying causes, with the trend for brain cancer being statistically significant at the 0.05 level (HR=1.25, 95%CI: 1.05–1.50); results for these cancers were similar when in the multiple-cause analysis.. A statistically significant inverse association was observed for mortality from chronic renal disease as the underlying cause (HR=0.70, 95%CI: 0.60–0.82), although this association became null when considering multiple causes, which had almost four times the number of deaths. It is likely that multiple cause analysis is more closely aligned with incidence here, and is more relevant for etiology.

Table 2.

Hazard ratios (HRs) and (95% CIs)^a for the effect of estimated log cumulative PFOA serum concentrations on underlying and multiple causes of deaths in the cohort (n= 32,081).

Cause of death	Underlying causes				Multiple causes
	N	HR (95% CI)	p-value b	p-value c	N	HR (95% CI)	p-value b	p-value c
All causes (no lag)	5694	0.98 (0.96, 0.99)	0.01	0.01
Esophageal cancer (20-year lag)	55	0.90 (0.77, 1.07)	0.23	0.15	59	0.90 (0.76, 1.06)	0.19	0.12
Intestinal cancer (20-year lag)	92	0.93 (0.82, 1.06)	0.30	0.28	103	0.91 (0.80, 1.03)	0.13	0.18
Liver cancer (20-year lag)	53	1.01 (0.85, 1.19)	0.94	0.49	58	0.99 (0.85, 1.17)	0.93	0.52
Pancreatic cancer (5-year lag)	114	1.07 (0.95, 1.20)	0.27	0.20	117	1.06 (0.95, 1.19)	0.30	0.21
Pancreatic cancer older ages (no lag)	58	1.23 (1.05, 1.44)	0.01	0.04	59	1.21 (1.04, 1.42)	0.02	0.05
Lung cancer (20-year lag)	445	1.06 (1.00, 1.12)	0.06	0.38	483	1.04 (0.99, 1.10)	0.12	0.52
Breast cancer (no lag)	116	0.94 (0.83, 1.07)	0.37	0.74	130	0.94 (0.83, 1.06)	0.33	0.62
Prostate cancer (20-year lag)	54	1.04 (0.89, 1.20)	0.64	0.92	76	1.00 (0.88, 1.14)	0.95	0.64
Kidney cancer (10-year lag)	31	1.08 (0.87, 1.34)	0.49	0.44	39	1.03 (0.85, 1.25)	0.79	0.31
Bladder cancer (20-year lag)	39	1.17 (0.98, 1.39)	0.09	0.70	49	1.15 (0.99, 1.34)	0.08	0.69
Melanoma (20-year lag)	22	0.92 (0.70, 1.20)	0.54	0.35	26	0.90 (0.70, 1.16)	0.42	0.45
Brain cancer (no lag)	40	1.25 (1.05, 1.50)	0.01	0.76	43	1.21 (1.02, 1.43)	0.03	0.37
Non-Hodgkin’s Lymphoma (10-year lag)	47	0.94 (0.79, 1.12)	0.50	0.14	59	0.92 (0.78, 1.08)	0.29	0.17
Leukemia and Aleukemia (10-year lag & 5-year lag)	75	0.93 (0.81, 1.07)	0.29	0.26	89	0.98 (0.86, 1.12)	0.81	0.28
Multiple Myeloma (no lag)	33	0.96 (0.77, 1.19)	0.73	0.57	44	0.96 (0.79, 1.16)	0.66	0.41
Diabetes Mellitus (20-year lag)	205	1.02 (0.94, 1.11)	0.62	0.62	641	1.01 (0.97, 1.06)	0.55	0.40
Ischemic Heart Disease (no lag)	789	0.96 (0.92, 1.00)	0.07	0.39	1285	0.98 (0.95, 1.02)	0.26	0.15
Other Diseases of the Heart (no lag)	295	0.98 (0.90, 1.05)	0.51	0.47	1455	0.95 (0.92, 0.98)	0.01	0.01
Cerebrovascular Disease (20-year lag & no lag)	267	0.95 (0.88, 1.03)	0.20	0.95	435	0.98 (0.92, 1.04)	0.53	0.77
COPD (no lag)	363	0.96 (0.90, 1.03)	0.28	0.18	719	0.97 (0.93, 1.02)	0.24	0.11
Cirrhosis (20-year lag)	80	1.08 (0.94, 1.24)	0.26	0.40	128	1.08 (0.97, 1.20)	0.19	0.61
Chronic renal disease (20-year lag & no lag)	86	0.70 (0.60, 0.82)	<0.01	0.05	310	0.94 (0.87, 1.01)	0.11	0.16
Thyroid Disease (20-year lag)	4	-	-	-	58	1.04 (0.88, 1.22)	0.65	0.88

^aHazard ratios are for log transformed cumulative PFOA serum concentrations with either no lag, lag 5, 10 or 20 years, from Cox regression models. Based on AIC, the best fitting lag was chosen for each cause for this Table (it appears after the cause; if two, the former refers to the best fit for underlying causes and the latter refers to the best fit for multiple causes). . Detailed results for different lags can be found in Table S4 and S5. All models adjusted for time-varying alcohol consumption and smoking, age, gender, education, BMI, and 10-year birth year period.

^bP-value for trend using log cumulative exposure.

^cP-value for trend using cumulative exposure

Breast and pancreatic cancer analyses were found to violate the proportional hazards assumption (for breast cancer, the violation was only for longer lags), and supplemental analyses were conducted after dividing follow-up time (age) above and below the median age at death for each cancer. Pancreatic cancer at older ages (>72) showed a significant positive trend with increasing exposure (HR 1.23 (1.05, 1.43) (Table 2); breast cancer results were not substantially different from the overall analysis (see Table S2 for further details).

The lag-specific analyses for underlying and multiple causes can be found in Table S3 and S4. HRs and 95% CIs generally were similar between models where estimated cumulative PFOA was unlagged, 5 years lagged, 10 years lagged and 20 years lagged, which was to be expected as many cancers occurred years after large decreases in exposure.

Results for underlying causes and multiple causes of death using estimated cumulative PFOA serum concentration quartiles are shown in Table 3. The positive trends noted for bladder and brain cancer were not monotonic, in both cases dropping off in the fourth quartile Brain cancer was significantly elevated in both third and fourth quartiles, for both underlying and multiple analysis. Bladder cancer was significantly elevated for both underlying and multiple cause in the third quartile. For kidney cancer, a site of a priori interest, we observed a statistically significant increased risk in the third (vs. first) quartiles in both the underlying and multiple cause analyses (HRs of 3.99 and 2.68, respectively), though the trends were not monotonic. Pancreatic cancer at older ages (>72 years) showed a monotonic positive trend. There was some elevation of mortality rates among higher categories of cirrhosis (significant elevation in 2^nd and 4^th quartile), although trend tests were not significant.

Table 3.

HRs (95% CIs)^a for the effect of estimated cumulative PFOA serum concentrations on underlying and multiple causes of deaths by quartile ^b in the cohort (n = 32,081).

Cause of death	Underlying causes			Multiple causes
	Quartile 2	Quartile 3	Quartile 4	Quartile 2	Quartile 3	Quartile 4
All casuse (no-year lag)	0.98 (0.91, 1.06)	0.97 (0.90, 1.05)	0.91 (0.85, 0.99)
Esophageal cancer (20-year lag)	1.26 (0.57, 2.79)	0.86 (0.39, 1.89)	0.81 (0.36, 1.83)	1.29 (0.60, 2.77)	0.90 (0.42, 1.93)	0.83 (0.38, 1.82)
Intestinal cancer (20-year lag)	0.67 (0.36, 1.26)	0.99 (0.53, 1.87)	0.61 (0.32, 1.16)	0.78 (0.43, 1.42)	0.88 (0.48, 1.59)	0.56 (0.31, 1.03)
Liver cancer (5-year lag)	1.02 (0.48, 2.20)	0.68 (0.31, 1.47)	1.00 (0.45, 2.22)	1.06 (0.51, 2.21)	0.71 (0.34, 1.47)	1.01 (0.47, 2.16)
Pancreatic cancer (10-year lag)	0.64 (0.38, 1.08)	0.93 (0.54, 1.57)	0.93 (0.54, 1.62)	0.58 (0.34, 0.98)	0.85 (0.50, 1.44)	0.89 (0.52, 1.53)
Pancreatic cancer older ages (no lag)	1.05 (0.45, 2.48)	1.16 (0.52, 2.60)	2.44 (1.14, 5.22)	1.10 (0.48, 2.56)	1.13 (0.50, 2.56)	2.32 (1.09, 4.92)
Lung cancer (5-year lag)	1.61 (1.24, 2.10)	1.24 (0.95, 1.62)	1.30 (0.99, 1.70)	1.66 (1.29, 2.14)	1.22 (0.95, 1.58)	1.26 (0.97, 1.64)
Breast cancer (no lag)	0.92 (0.55, 1.53)	0.75 (0.45, 1.26)	0.67 (0.40, 1.13)	0.97 (0.60, 1.58)	0.77 (0.47, 1.25)	0.71 (0.43, 1.15)
Prostate cancer (10-year lag)	1.48 (0.69, 3.18)	1.13 (0.53, 2.43)	1.75 (0.79, 3.86)	1.85 (0.97, 3.52)	1.23 (0.64, 2.37)	1.31 (0.68, 2.53)
Kidney cancer (no lag)	3.80 (1.42,10.17)	3.99 (1.47,10.79)	1.37 (0.48, 3.87)	1.76 (0.73, 4.26)	2.68 (1.10, 6.53)	1.07 (0.42, 2.70)
Bladder cancer (no lag & 10-year lag)	1.52 (0.63, 3.70)	3.08 (1.24, 7.67)	1.56 (0.61, 3.99)	1.63 (0.73, 3.62)	2.50 (1.11, 5.67)	1.97 (0.86, 4.51)
Melanoma (10-year lag & no lag)	0.44 (0.31, 0.63)	0.22 (0.14, 0.34)	0.11 (0.07, 0.18)	1.25 (0.42, 3.75)	4.26 (1.48, 12.31)	0.79 (0.26, 2.41)
Brain cancer (20-year lag)	1.64 (0.66, 4.06)	4.09 (1.64,10.23)	2.72 (1.04, 7.09)	1.93 (0.81, 4.61)	2.70 (1.13, 6.44)	2.33 (0.92, 5.90)
Non-Hodgkin’s Lymphoma (no lag)	2.93 (1.31, 6.58)	1.26 (0.56, 2.84)	0.93 (0.40, 2.16)	2.04 (0.99, 4.19)	1.30 (0.63, 2.68)	0.77 (0.36, 1.63)
Leukemia and Aleukemia (20-year lag & 5-year lag)	1.17 (0.58, 2.35)	0.74 (0.37, 1.48)	0.68 (0.33, 1.40)	0.87 (0.48, 1.59)	0.97 (0.53, 1.78)	0.96 (0.52, 1.77)
Multiple Myeloma (no lag & 10-year lag)	1.00 (0.38, 2.60)	1.52 (0.58, 4.02)	0.68 (0.26, 1.83)	1.19 (0.51, 2.78)	1.60 (0.68, 3.75)	0.85 (0.35, 2.02)
Diabetes Mellitus (20-year lag & no lag)	1.08 (0.72, 1.63)	1.49 (0.99, 2.23)	1.04 (0.68, 1.58)	0.97 (0.78, 1.21)	1.22 (0.98, 1.53)	0.95 (0.76, 1.19)
Ischemic Heart Disease (20-year lag)	1.10 (0.89, 1.35)	1.25 (1.01, 1.54)	0.90 (0.73, 1.12)	1.03 (0.87, 1.21)	1.19 (1.01, 1.39)	0.98 (0.83, 1.16)
Other Diseases of the Heart (10-year lag & no lag)	0.88 (0.63, 1.22)	0.98 (0.71, 1.37)	0.90 (0.64, 1.27)	0.94 (0.82, 1.09)	0.82 (0.70, 0.95)	1.00 (0.86, 1.16)
Cerebrovascular Disease (20-year lag & 10-year lag)	0.95 (0.66, 1.35)	0.92 (0.64, 1.31)	0.75 (0.52, 1.09)	1.32 (1.01, 1.73)	1.24 (0.95, 1.63)	1.06 (0.80, 1.40)
COPD (5-year lag & no lag)	1.29 (0.96, 1.72)	1.09 (0.81, 1.46)	0.93 (0.68, 1.25)	1.03 (0.63, 1.69)	0.78 (0.48, 1.28)	0.97 (0.58, 1.62)
Cirrhosis (20-year lag)	2.17 (1.15, 4.10)	1.15 (0.61, 2.17)	2.18 (1.12, 4.25)	1.24 (1.00, 1.53)	1.08 (0.87, 1.33)	1.00 (0.80, 1.24)
Chronic renal disease (20-year lag & 10-year lag)	0.60 (0.31, 1.17)	0.57 (0.30, 1.10)	0.28 (0.14, 0.54)	0.79 (0.57, 1.09)	0.73 (0.52, 1.00)	0.71 (0.51, 1.00)
Thyroid Disease (20-year lag)	-	-	-	2.23 (1.06, 4.67)	1.77 (0.84, 3.69)	1.38 (0.65, 2.94)

^aQuartiles of cumulative PFOA serum concentrations. . Based on AIC, the best fitting lag was chosen for each cause of death for presentation of results, and is listed after the cause. When two best lags are noted, the first is for the underlying cause and the second for multiple cause. All models adjusted for time-varying alcohol consumption and smoking, age, gender, education, BMI, and 10-year birth year period. Relevant tests for trend are shown in Table 2, for both log cumulative exposure, and untransformed continuous variable. Bolded 95% CIs indicate that the lower bound is equal or greater than 1.00 or the upper bounds is equal or less than 1.0,.

^bQuartiles defined by using data from each cause-specific group of deaths, and hence varied by cause-of-death.

While we found little evidence of an increased trend for ischemic heart disease in our data, with continuous exposure having a rate ratio less than 1.0. There was evidence of a significantly increased risk in the third quartile of quartile analysis for both underlying and multiple cause (HRs of 1.25 and 1.19, respectively), although trends were not monotonic). Other causes of heart disease death also showed no positive association with exposure, and indeed were significantly decreased with higher exposure in multiple causes analyses.

Two-piece linear splines and restricted cubic splines for kidney and brain cancer (underlying cause) are shown in Figures 2 and 3, both using log transformed cumulative PFOA serum concentration, as it fit better. The 2-piece linear spline fit slightly better than the cubic spline for kidney cancer, while the opposite was true for brain cancer.

Figure 2.

2-piece linear spline of cumulative PFOA(ng/m^l -years) for underlying cause kidney cancer (Left) and brain cancer (Right), with the best knot being at the 84th percentile of cumulative PFOA for kidney cancer and the 75th percentile for brain cancer. For a better fit, cumulative exposure was used at a lag of 10 years for kidney cancer and at no lag for brain cancer and models used natural log transformed cumulative PFOA concentrations. AIC: kidney cancer: 443.856; brain cancer: 531.873. Hazard ratios on the log scale were linear, but are here transformed to non-logged scale.

Figure 3.

Hazard ratio for underlying cause kidney cancer (Left) and brain cancer (Right) using natural log transformed cumulative PFOA serum exposure (ng/ml-years) via a cubic spline. For a better fit, cumulative exposure was used with a 10 year lag for kidney cancer and no lag for brain cancer. AIC: kidney cancer: 444.295; brain cancer: 529.312.

Results by community vs. workers subjects can be found in Table S5 with results for log cumulative exposure as a continuous variable. Lung cancer showed a significant positive trend among community residents with a 20 year lag [HR per unit of log cumulative exposure of 1.08 (1.01–1.16)], and workers also showed a positive trend [1.18 (0.92–1.51)]. The HR for brain cancer was significantly elevated in the community group with a 20 year lag [1.28 (1.01–1.62)]. The pancreatic cancer HR was primarily elevated among workers [HR 10 year lag 1.53 (0.96–2.47)]. Kidney and bladder cancer HR elevations were mostly among community participants, although they did not reach statistical significance at the 0.05 level. Stroke showed a significant positive trend among workers [1.86 (1.13–3.05)] with no lag, and similar marked trends at 10 and 20 years lags.

Table S6 shows results by sex. Lung, kidney, and pancreatic excesses at higher PFOA levels were predominantly among women, while the brain cancer excess was seen in both genders.

After adjusting for competing risks, results for underlying cause for continuous exposure were similar to our original rate ratios. The rate ratios (using the same optimal lags from Table 2) for brain cancer, kidney cancer, bladder cancer, pancreatic cancer (at older ages, 0 lag), and ischemic heart disease were 1.27 (1.07, 1.51), 1.10 (0.90, 1.34), 1.21 (1.06, 1.38), 1.26 (1.09, 1.46), and 0.96 (0.92, 1.01) respectively, after competing risk adjustment. While not strictly comparable to our original rate ratios for these causes, these results are reassuring in that they are quite similar to our original findings, suggesting that competing risks are not playing a big role in our results. Bladder cancer became slightly more positive, and the stronger association was statistically significant.

Discussion

Results of the analyses of mortality showed only one positive trend that was statistically significantly (at the 0.05 level) associated with cumulative PFOA serum levels, which was for brain cancer. Brain cancer has not previously been associated with PFOA in the literature, although Barry et al. (2013) in their study of cancer incidence in this cohort did find a non-significant positive trend for log transformed serum PFOA based on 17 cases, with no lag (HR per 1 unit increase in natural log-transformed cumulative exposure:1.13 [95% CI 0.84–1.51, p=0.43]. Many studies have not presented results for brain cancer, as it is a relatively rare cancer. PFOA can cross the blood-brain barrier (Wang et al. 2018). A companion study of cancer incidence in this cohort, conducted at the same time as this one, also found a positive association with aggressive brain cancer (Hofmann et al. 2025).

The positive trend with kidney cancer mortality was not marked, and the p value for trend was 0.49, but the size of the coefficient for log transformed cumulative exposure (1.08) was similar to that found for kidney cancer incidence in this cohort, in our earlier analysis (1.10 (0.98–1.24), based on 105 cases) (Barry et al. 2013). Kidney cancer excesses have also been found in some prior studies, including a prospective nested case-control study in the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial (Shearer et al. 2021), among women in an incidence study among the Cancer Prevention Study-II (CPS-II) Lifelink Cohort (Winquist et al. 2023), among non-Hispanic white participants in the Multiethnic Cohort (Rhee et al, 2023), and in recent years in the more contaminated areas around Veneto (Biggeri et al. 2024) Other studies in the mid-Ohio Valley (with some overlap with the population in the current investigation) have also noted associations with kidney cancer, including a prior mortality study of workers at this manufacturing plant workers (Steenland and Woskie 2012) and a case-control study of mid-Ohio Valley residents (Vieira et al. 2013). The companion study of cancer incidence in this cohort also found a significant positive trend with kidney cancer among younger cohort participants (Hofmann et al. 2025). However, we note that there are a relatively small number of kidney cancer deaths (n=31 underlying cause) in our study, weakening the precision of our risk estimates.

The positive but non-statistically significant association between PFOA exposure and bladder cancer mortality found here has not been observed in most other studies. For example, in the previous incidence study of this cohort, Barry et al. (2013) found no trend for cumulative serum PFOA in relation to bladder cancer incidence in this cohort. Other negative studies include Vieira et al. (2013), Eriksen et al. (2009), and Steenland and Woskie (2012). However, there has been one positive occupational study (Raleigh et al. 2014), with a positive trend using a worker referent group. Furthermore, the companion study of cancer incidence in this same cohort also found a borderline positive association with aggressive bladder cancer (Hofmann et al. 2025).

We found a statistically significant positive trend between PFOA and pancreatic cancer among older subjects (person time when age >72, which was the median age at death). This finding conforms to the known relationship between PFOA and pancreatic tumors in animals (EPA 2024). Pancreatic cancer is rare, so that most epidemiologic studies have had only small numbers of pancreatic cancer cases. One exception is the large population-based cancer study by Eriksen et al. (2009), which did find a positive (although not statistically significan) trend of higher PFOA with pancreatic cancer (RRs by quartile of PFOA, 1.00, 0.88 (0.49–1.57), 1.33 (0.74–2.38) and 1.55 (0.85–2.80)). Another exception is a study of pancreatic cancer in two other population-based cohorts; in one cohort the authors found positive association between PFOA and pancreatic cancer (Zhang et al. 2024). Our companion cancer incidence study also found an excess of pancreatic cancer, again in older cohort members (>72 years) (Hofmann et al. 2025).

The three cancers, including kidney, bladder, and brain cancer, showing stronger positive trends in this cohort all had non-monotonic trends in quartile analyses, with HRs trailing off in the last quartile. While this non-monotonicity can be considered an argument against causality, several authors have noted that exposure-response trends may plateau at higher exposure, for reasons including greater measurement error at higher exposures, biological saturation of pathways, and depletion of susceptible individuals (Stayner et al. 2003, Lanphear et al. 2017). The tailing off at high exposures is reflected in our data in the better fit of log transformed cumulative serum PFOA compared to untransformed PFOA. For incident kidney cancer, a pooled analysis of two key studies has also shown a plateau effect (Steenland et al. 2022).

We were unable to examine mortality from several other cancer outcomes of interest, in particular testicular and thyroid cancer, both previously associated with PFOA in incidence data in this cohort (Barry et al. 2013), due to the small number of fatal cases.

Our largely negative findings for heart disease are consistent with our earlier findings of no increased heart disease incidence in this cohort, despite having an observed positive association between cholesterol and PFOA in this cohort (Winquist et al. 2014). The over literature is mixed with regard to PFOA and heart disease, with one recent paper finding excess cardiovascular mortality in the highly contaminated Veneto area in Italy (Biggeri et al. 2024).

The availability of historical serum estimates, enabling the use of cumulative serum levels as our exposure metric, is a strength of this cohort. The majority of studies which have serum estimates of exposure have them at only one point in time, which may not be accurate historically. Measurement error is always a concern in model-based estimates such as ours. However, here there are several aspects which partially allay this concern. First, as noted above, the modeled serum estimates correlated well (rho=0.71) with measured serum estimates for 30,303 residents of the mid-Ohio Valley in 2005/2006. Second, given the large exposure contrast in our study (which is another important strength), mismeasurement of quantitative exposure is unlikely to change the relative ranking of individuals’ exposures in the cohort, as has been demonstrated quantitatively by Avanasi et al. (2016). However, it is likely that measurement error in this cohort is non-differential and may have resulted in some bias to the null in our findings.

The lack of associations with other non-cancer outcomes in this cohort is not entirely surprising, as there is no strong a priori evidence for a number of these outcomes. An exception was the positive finding for stroke among workers in our cohort. Stroke (like heart disease) is associated with high cholesterol, which has consistently been associated with PFOA (Steenland et al. 2020). However, we found no evidence in this analysis of a PFOA-heart disease association, and a prior incidence study in this cohort (Winquist and Steenland 2014) also found no evidence of associations of PFOA with heart disease. On the other hand, a prior study of disease incidence among the workers in our cohort did find elevated rate ratios for stroke in higher quartiles in categorical analyses (Steenland et al. 2015).

There were significant excesses of cirrhosis mortality in 2^nd and 4^th quartiles, but trend tests were not significant and another study in this cohort of incident liver disease was also negative (Darrow et al. 2016). Our lack of a positive finding for non-malignant kidney disease is consistent with a prior negative finding for incident non-malignant kidney disease (Dhingra et al. 2016). A number of other health effects of concern with PFOA are for diseases which are generally non-fatal, and hence not observable in our study, such as preeclampsia, thyroid disease, ulcerative colitis, and high cholesterol. We were able to study thyroid disease as a contributing cause of death in this cohort and found no positive trend; however, as noted above, mortality is not the most sensitive way to study thyroid disease.

A potential weakness in this study is the fact that our cohort is largely a survivor cohort, with most subjects known to be alive in 2005/2006, while follow-up began in most cases much earlier. However, Barry et al. (2015) showed that in this cohort results are not likely to be biased from missing early cases/deaths unless survival after disease incidence differs between those with high vs. low exposure, which seems unlikely.

In conclusion, in this mortality study of a large cohort with high PFOA exposure, we found significant (at p<=0.05) associations with higher serum PFOA for brain cancer overall and for pancreatic cancer among older participants, as well as a suggestive but non-statistically significant association with bladder cancer. There was also some suggestion of an increase in kidney cancer mortality with higher PFOA, which has been observed in prior studies. However, we recognize that there are many outcomes presented here, and there is always the danger of chance findings due to multiple comparisons. Hence we believe caution should be exercised in interpretation, especially for outcomes with little or no a priori evidence.

Data availability.

De-identified data will be made available upon request.