Existing biomarker studies profile plasma during acute illness, leaving pre-illness proteomic susceptibility to critical care largely uncharacterised. We mapped the baseline proteomic architecture of critical-care outcomes in UK Biobank and tested for sex differences. We performed a proteome-wide association study of 2923 plasma proteins in 53,000 UK Biobank participants linked to Hospital Episode Statistics critical-care records, analysing three outcomes: ever critical-care exposure, prolonged critical care, and in-hospital death during a critical-care hospitalisation. Primary analyses used logistic regression; sensitivity analyses included Cox models, cross-validated LASSO, sex-interaction testing, and reverse-causation proxy analyses. Of 53,000 participants, 2986 had at least one critical-care episode. For ever critical-care exposure, 545 proteins survived Bonferroni correction, led by GDF15 (OR per SD 1.37, 95% CI 1.33–1.42, P = 2.82 × 10⁻68). In-hospital death yielded 18 Bonferroni-significant proteins, again led by GDF15. Prolonged critical care showed a markedly attenuated signal. Sex-interaction effects were predominantly stronger in women. An exploratory reverse-causation analysis yielded null estimates, indicating unresolved residual confounding. Baseline plasma proteins measured a median of more than 10 years before any critical-care episode were associated with later critical-care exposure and in-hospital death during a critical-care hospitalisation, with a biologically coherent pattern spanning stress-response, renal, epithelial, inflammatory, and tissue-remodelling pathways and a secondary female-predominant interaction signal among the leading proteins. The incremental discrimination over demographics (ΔAUC 0.040 on a 0.672 baseline) was modest but detectable over a > 10-year horizon. We interpret the signal as a cumulative-morbidity proteomic axis of critical-illness susceptibility rather than as validated isolated pre-morbid biomarkers; reverse-causation risk could not be excluded and external-cohort validation is required.
Ren et al. (Wed,) studied this question.