Fig 1.
Effect sizes on all-cause mortality.
Statistically significant (p < 1.71 × 10 − 5) proteins with the largest effect sizes on all-cause mortality within 5 (n = 769) or 10 years (n = 2,100). Associations were tested using Cox proportional hazards models, with the minimally adjusted model being adjusted for age, sex and BMI only, and the fully adjusted model adjusted for age, sex, BMI, chronic condition count, alcohol, smoking, physical activity, education and Townsend deprivation index. a, 20 proteins most indicative of risk (HR > 1) for all-cause mortality within 5 years in the minimally adjusted model. b, 20 proteins most indicative of risk for mortality within 10 years in the minimally adjusted model. c, 20 proteins most indicative of risk for mortality within 5 years in the fully adjusted model. d, 20 proteins most indicative of risk for mortality within 10 years in the fully adjusted model.
Table 1.
Summary of phenotypic data used in this analysis from the 38,150 individuals who met inclusion criteria, participants who had missing entries or chose “not to answer” were excluded.
Table 2.
Top 20 categorised causes of death in 2,100 individuals with proteomic data that have died within 10 years of blood sample collection.
Fig 2.
Effect sizes on disease-specific mortality.
Statistically significant (p < 1.71 × 10 − 5) proteins with the largest effect sizes on disease-specific mortality within 10 years using the fully adjusted model. a, 20 proteins most indicative of risk (HR > 1) for cardiovascular-specific mortality within 10 years (n = 441). b, 20 proteins most indicative of risk for cancer-specific mortality within 10 years (n = 1003). c, 20 proteins most indicative of risk for all other mortality within 10 years (n = 656).
Fig 3.
Cumulative area under the curve (AUC) of 40 proteins with the largest hazard ratios from fully adjusted cox proportional hazards models.
A logistic regression model was run for each protein significantly associated with an increase in risk (hazard ratio > 1) for all-cause mortality within a, 5-year onset (n = 392) and b, 10-year onset (n = 377). A red, dashed line illustrates the protein selection cut-off for the panel.
Fig 4.
Comparison of model performance for predicting all-cause mortality for 5 and 10 years.
“Proteins” refers to the predictive protein panel for the respective timeframe. “Lifestyle” covariates include the following: smoking status, alcohol intake frequency, chronic condition count, physical activity, Townsend deprivation index and education. “Biomarkers” consists of the UK Biobank measured biochemistry biomarkers which were found to be significantly associated with the respective timeframe of death. a, ROC curves for models predicting all-cause mortality within 5 years. b, ROC curves for models predicting all-cause mortality within 10 years. c, table comparing performance metrics of each model and timeframe.