Abstract
Circulating plasma proteins play key roles in measuring and reflecting states of disease and health. Developments in protein metrology allow for over 2,900 proteins to be quantified in a single sample. In major epidemiological studies, this allows for profound insights into protein expression in liquid biopsies, mortality, and morbidity. Here, we have investigated the relationship between peripheral blood protein profiles and non-accident all-cause mortality within 5- and 10-year timeframes, using data on 38,150 participants from the UK Biobank. Adjusting for lifestyle and health covariates, we identified 392 proteins associated with an increase in risk for death within 5-years, and 377 associated with an increase in risk within 10-years. Proteomic signatures of cause-specific mortality (cardiovascular, cancer, all-other causes) were also identified, with 19 proteins found to overlap across those. Using logistic regression modelling, we constructed a parsimonious predictive protein panel for each respective all-cause mortality timeframe, including markers such as adrenomedullin, SERPINA1 and PLAUR. When compared to models inclusive of standalone traditional risk criteria, such as demographic and lifestyle factors, models utilising the protein panels modestly improve prediction for 5 and 10-year mortality (from AUC 0.49-0.57 to AUC of 0.62-0.68). Our results demonstrate the potential of the plasma proteome in risk stratification for all-cause mortality.