Fig. 4. Personalized antibiogram coverage rate as a function of the antibiotic budget parameters.

a Held out Stanford test set of emergency hospital admissions treated with one of the twelve most common empiric IV/IM antibiotic selections. b Antibiotic sweep simulations showing the trade off between coverage rate and broad-spectrum antibiotic use when antibiotic selection is optimized with personalized antibiograms. For example, the plot titled “Vanc/Pip-Tazo - Pip-Tazo” illustrates the change in coverage rate as more historical prescriptions for Vancomycin + Piperacillin-Tazobactam (Vanc/Pip-Tazo) are exchanged for Pip-Tazo, prioritized by personalized antibiogram predictions. On the far left of the plot, starting with 0 change in the actual amount of antibiotics prescribed, the figure illustrates the green region where personalized antibiograms could potentially have achieved an even better antibiotic coverage rate than the actual prescriptions by clinicians (represented by the dark dashed line) with equal or fewer Vanc/Pip-Tazo.