Fig. 3

Results of BCAUS on Diabetes dataset. a Standardized differences for all 21 covariates in Diabetes dataset for one arm: Treatment = Insulin + Metformin + GLP-1 Agonist + SGLT2 Inhibitor + Sulfonyurea; Control = Insulin. Cohort sizes are 125 and 44,600 respectively, implying a class-imbalance ratio ~ 1:350. Green trace is raw unadjusted data and violet trace shows data adjusted by IPW. Dashed line represents threshold at 0.1. Inset shows normalized histograms of the distribution of propensity scores i.e. output of BCAUS, for control (green) and treatment (violet) groups. ZCTA = Zip-Code Tabulation Area. b Histogram of number of balanced covariates (standardized difference, Δ < 0.1 between control and treatment) in each intervention arm prior to BCAUS training and IPW adjustment. Shown here for 133 intervention arms. c Same as (b), but after BCAUS training and adjustment. For a majority of intervention arms (124 of 133) all covariates are balanced