Unsupervised dietary-behavioral profiling identifies PFAS (per- and polyfluoroalkyl substances) and metal exposure...

Beyond "what and how much you eat," modern nutritional epidemiology is paying attention to how dietary and behavioral patterns, which combine food selection and the context of cooking, packaging, and ingestion, organize health and environmental exposure. This cross-sectional secondary analysis used adult data from the 4th Korea National Environmental Health Survey (KoNEHS Cycle 4, 2018-2020; n = 4239). Multi-question variables related to diet, cooking, packaging, and ingestion behaviors were mapped into a latent space through unsupervised learning, clusters were derived, and the exposure levels of PFAS, bisphenols, urinary cadmium, and blood lead biomarkers were compared by cluster. As a result, serum PFAS concentrations were higher in the "basic diet/seafood propensity" cluster, with weighted geometric means ranging from 10.998 to 21.767 µg/L for PFOS and from 5.082 to 8.323 µg/L for PFOA across clusters. In the "packaging/contact propensity" cluster, higher PFOA was observed in the adjusted analysis, while blood lead showed an elevated direction but was not robust after FDR correction. On the other hand, the differences between clusters for bisphenols and urinary cadmium were not robust when considering covariates and multiple comparisons, suggesting the possibility that the variability of single urine measurements, uncertainty in exposure assessment, and complexity of indicator characteristics may have affected the results. This study provides an empirical basis for identifying population-level dietary-behavioral exposure profiles and informing targeted risk communication by showing how lifestyle-based dietary and behavioral structures are associated with environmental exposure patterns before directly linking dietary patterns with diseases.