Alert Timing in Sepsis Prediction Models—An Opportunity to Tailor Interventions

Abstract

bundle compliance).If all alerting patients cannot actually be evaluated, the clinical benefits of these tools diminish greatly.4 Thus, clinicians overseeing the use of these tools must not only focus on selecting the optimal threshold or the best-performing tool.They must also determine whether any tools are sufficiently high performing to be useful at all.Recently, there are sepsis prediction models whose implementation has led to improvements in clinical outcomes, 5,6 including at least 1 study involving the SPM.However, performance of the SPM can vary substantially across hospitals (at least in version 1), 7 and we do not know the minimal performance required within a clinical setting before the use of a sepsis tool will translate to clinical benefits.Better clinical outcomes require both effective models and effective interventions, and both may benefit from tailoring to a local context.Through their analysis, Schertz et al 1 have identified a critical aspect of that tailoring: timing of alerts relative to sepsis onset.By estimating when alerts are expected to fire relative to sepsis onset, we can select interventions most likely to be useful at the time of the alert, hopefully aiding the translation of these tools into better clinical outcomes.