Top left: Histogram of baseline (standalone antibiotic therapy) mortality rates for 500 test patient parameterizations (N = 50 instances per test patient parameterization). Top right: Histogram of mortality rates using the learned policy (combination antibiotic cytokine mediation therapy) for 500 test patient parameterizations. Bottom: Performance of the learned policy on each of the 500 test patient parameterizations (defined as the difference in mortality rate with and without the policy, normalized by the larger of the two), sorted by increasing performance. Thus, positive values can be interpreted as the fraction of patients who were healed by the policy that otherwise would have died, whereas negative values indicate the fraction of patients who died from the policy that otherwise would have healed. Since the policy did not cause an increase in mortality rates for any patient parameterization, there are no negative values.