OBJECTIVES: Cerner has developed the St John’s Sepsis Alert, an evidence-based real-time algorithm that alerts clinicians to the presence of the Systemic Immune Response Syndrome or sepsis. Using simulation, we estimated the Alert’s performance and determined the optimal cut-offs for ~17 included quantitative parameters (e.g., blood pressure).  METHODS: We estimated the operating characteristics of the alert by applying its logic to 3 years of real-world data on adults from Cerner Health Facts, a time-stamped database extracted from electronic medical records.  We evaluated the base-case and performed a 10,000-iteration uncertainty analysis.  Each run used a different set of thresholds, each drawn randomly from the range of reasonable values using a Latin hypercube sampling design under the assumption of an independently distributed joint beta distribution. Each run provided a point on the Receiver Operator Characteristic curve.  We constructed an extended dominance curve from the resulting point cloud and determined the optimal values as those associated with that curve. RESULTS: Data from ~69,000 hospitalizations with a 5% incidence of sepsis were available.  Using baseline values for the Alert, we estimated a Sensitivity of 56%, Specificity of 90%, Positive Predictive Value of 22%, and Negative Predictive Value of 98%.  The uncertainty analysis found that 10 sets of cut-offs dominated all the others.  The C-statistics for these ranged from 67% to 75%, Sensitivity from 39% to 73%, Specificity from 96% to 76%, PPV 33% to 13%, NPV 95% to 98%, and the posterior probability of sepsis increased from 2.4- to 7-fold. The most accurate thresholds were not necessarily optimal for implementation; certain sets with only slightly lower C-statistics greatly decreased false positive rates. CONCLUSIONS:  Simulation can usefully inform in the design of clinical algorithms prior to implementation.  Examining a range of thresholds can identify statistical optimality and characterize trade-offs to deviating from it.