BACKGROUND: The prevention of common chronic diseases is hampered by the lack of tools to identify at-risk individuals. The plasma proteome could enable a systems-level assessment of health, but its clinical translation is limited by platform-specific biases and a focus on single diseases. METHODS: We aimed to develop and validate a clinically deployable proteomic classifier for multi-system risk prediction. Using data from 11,900 disease-free adults, we defined five ProteoHealth Archetypes (PHAs) via unsupervised learning. Critically, we trained a classifier using paired protein ratios to ensure robustness across measurement platforms. RESULTS: The ratio-based classifier successfully transferred PHA signatures in an internal validation cohort (n = 3,570) and, importantly, in three external cohorts profiled by different technologies (SomaScan and mass spectrometry). Each PHA exhibited distinct, highly reproducible risks for developing cardiometabolic, inflammatory/immune, neurovascular, and psychiatric diseases. Individuals in high-risk archetypes experienced significantly steeper declines in survival. Genome-wide association analyses identified loci associated with PHA liability, and two-sample Mendelian randomisation produced effect directions consistent with the observational disease association. The differential proteomes and enriched pathways aligned with the specific disease profiles of each archetype. CONCLUSIONS: PHAs provide an externally transferable, mechanistically interpretable map of baseline proteomic health that forecasts multisystem disease and survival, offering a scalable substrate for prevention, risk communication, and biomarker-guided stratification.
Zhang et al. (Tue,) studied this question.