Fig. 2: D-CLEF to predict mortality of patients with COVID-19 from University of California (UC) Health COVID Research Data Set (CORDS) data.

a The UC CORDS data (dataset X) with n = 15,279 patients and m = 100 covariates, including patient and COVID-19 information (dataset X1) and drugs prescribed (dataset X2). The patients were from five UC Health medical centers (datasets X3–X7). b Overview of the setup on dataset X to estimate the performance of D-CLEF. c Main results of the horizontal scenario on dataset X, comparing D-CLEF with siloed and centralized LR models. d Main results of D-CLEF in the vertical scenario on dataset X, compared with siloed/centralized LR models. e Main overall comparison on dataset X, using two-sided, two-sample Wilcoxon signed rank-test. All box plots are derived from 30 trials, with median as center line, upper and lower quartiles as box limits, 1.5x interquartile range as whiskers, and outliers as points.