Uncovering Exposure Patterns of Metals, PFAS, Phthalates, and PAHs and Their Combined Effect on Liver Injury Markers

People are exposed to mixtures of metals, per- and polyfluoroalkyl substances (PFAS), phthalates, and polycyclic aromatic hydrocarbons (PAH) rather than single chemicals, yet mixture inference is hampered by high dimensionality, correlation, missingness, and left-censoring below limits of detection (LOD). We analyzed 2013–2014 National Health and Nutrition Examination Survey (NHANES) biomarkers (n = 4367) to (i) recover latent, interpretable co-exposure structures and (ii) quantify how these mixtures relate to liver health. To denoise and handle censoring, we applied Principal Component Pursuit with LOD adjustment (PCP-LOD), decomposing the exposure matrix into a non-negative low-rank component (population co-exposure profiles) and a sparse component (individual spikes), and then used Bayesian Kernel Machine Regression (BKMR) to estimate nonlinear and interactive associations with AST, ALT, GGT, ALP, total bilirubin, and the Fatty Liver Index (FLI), retaining analytes with ≥50% detection. PCP-LOD revealed coherent clusters (e.g., long-chain PFAS grouping; shared metal loadings), while the sparse layer highlighted episodic phthalate elevations. BKMR indicated outcome-specific mixture effects: PAHs and selected phthalates showed consistently positive associations with ALP, GGT, and FLI; PFAS (PFOS, PFNA, PFOA) exhibited modest associations with ALP and bilirubin; metals displayed mixed directions. A joint increase in the overall mixture from the 25th to 75th percentile corresponded to an upward shift in FLI and a smaller rise in ALT. This censoring-aware low-rank-plus-sparse framework coupled with flexible mixture modeling recovers actionable exposure architecture and reveals clinically relevant links to liver injury and steatosis, motivating longitudinal and mechanistic studies to strengthen causal interpretation.