Fig. 2: PWR complements disease-specific prognostic markers.

A Univariate and multivariate mortality hazard ratios were calculated for PWR and disease-specific prognostic markers in the MGH cohort: D-dimer for COVID-19, NT-proBNP for heart failure, and HS-TnT for myocardial infarction. Disease-specific biomarkers for heart failure and myocardial infarction generated higher univariate hazard ratios than PWR. For all three cohorts, PWR provided statistically significant prognostic information in multivariate modeling suggesting it provided complementary information. B Percentage of patients for whom each disease-specific marker was available within 48 h of admission. Error bars (A) reflect 95% confidence intervals on the hazard ratios, with each cohort n given in Table 1. For HS-TnT data availability was calculated only after Apr-2018, when MGH switched to use of a high-sensitivity troponin assay. PWR hazard ratios may differ slightly from those in Fig. 1, as they are calculated only in the subset of patients who have the alternative marker available. In (A) univariate refers to hazard ratios that are corrected for age, but not for the other marker, and multivariate refers to jointly fitting both markers along with age. D-dimer, NT-proBNP and HS-TnT have long-tailed distributions and were log-scaled before analysis. Source data are provided as a Source Data file. Marker abbreviations are defined in Table 1.