Joint Associations of PFAS and Metal Mixtures with All-Cause and Cardiovascular Mortality

Per- and polyfluoroalkyl substances (PFASs) and toxic metals co-occur in the same people and have each been linked to cardiovascular and all-cause mortality, yet their joint contribution is poorly characterized. We examined PFOA, PFOS, lead, cadmium, and mercury jointly in NHANES 2011–2018 linked to 2019 mortality files, 3982 adults, 19,090 person-years, 236 all-cause, and 60 heart-disease deaths, using Cox models and Bayesian kernel machine regression. Heart disease was the primary cause-specific endpoint, the only category defined consistently across all four cycles. For a one-interquartile-range increase in all five exposures together, the joint hazard ratio was 0.89 (0.75, 1.06) for all-cause mortality, 1.00 (0.73, 1.37) for heart disease, and 0.94 (0.70, 1.28) for cardiovascular mortality including stroke; full socioeconomic adjustment gave 0.86 (0.69, 1.06). Individually, cadmium was the one component consistently associated with higher all-cause mortality, 1.12 (1.05, 1.20), retained across every specification examined. Mercury ran in the opposite direction, 0.91 (0.82, 1.00), consistent with blood mercury marking fish consumption and the dietary patterns accompanying it rather than toxicity; estimates below one are therefore not read as evidence of protection. PFOA, PFOS, and lead were null. The mixture model agreed with the regressions, the contrast between the joint 75th and 25th percentile profiles being −0.028 (−0.132, +0.047). How time at risk was handled mattered more than any modeling choice: with covariates and exposure contrast held fixed, a binary mortality indicator gave an odds ratio of 1.10 (0.93, 1.31), whereas a Poisson model with a log person-time offset gave 0.90 (0.75, 1.07), Cox regression 0.89 (0.75, 1.06), and restriction to a common 60-month window 0.74 (0.58, 0.95). Longer-observed participants were also more exposed, with the binary estimate potentially confounded by time at risk. Cadmium is therefore the component warranting priority in mixture surveillance, and pooled-cycle NHANES analyses should model time at risk directly.