Original Article
Influencing outcomes with automated time zero for sepsis through statistical validation and process improvement
Abstract
Background: Sepsis is a life threating complication of infection acquired by more than 1.5 million people in the United State annually. Each year, sepsis claims the lives of at least 250,000 people. Real-time, automated surveillance for sepsis among hospitalized patients is of critical importance, given that one in three people who die in hospitals have sepsis. The early identification and treatment of sepsis is associated with reduced mortality and costly intensive care bed days. The objective of this analysis was to improve the performance of an electronic medical record based sepsis algorithm (early identification) and improve evidence based bundle compliance (early intervention) with the addition of a real-time, automated time zero calculation.
Methods: Data from our enterprise-wide health information systems were landed in a data lake platform and was used to statistically validate existing sepsis algorithms. An additional algorithm calculating time zero was introduced and a post-hoc comparison of measures of test performance, alert timing, bundle compliance, ICU length of stay, and all-hospital mortality were performed.
Results: A total of 55,918 alerts for sepsis were generated over the one-year study period across 30 inpatient facilities. The addition of an automated time zero algorithm improved several key indicators including superior positive predictive value (37% to 52%), enhanced timing of the alert (79% occurred within six hours, 77% within the critical 180-minute SEP-1 window, 47% within an hour of time zero), a 14% increase in bundle compliance, a 10% reduction in ICU length of stay, and a decrease in mortality from sepsis.
Conclusions: The addition of a real-time, automated sepsis time zero calculation improved the performance and timeliness of a predictive sepsis alert provided through a system developed mobile application for clinicians and administrators.
Methods: Data from our enterprise-wide health information systems were landed in a data lake platform and was used to statistically validate existing sepsis algorithms. An additional algorithm calculating time zero was introduced and a post-hoc comparison of measures of test performance, alert timing, bundle compliance, ICU length of stay, and all-hospital mortality were performed.
Results: A total of 55,918 alerts for sepsis were generated over the one-year study period across 30 inpatient facilities. The addition of an automated time zero algorithm improved several key indicators including superior positive predictive value (37% to 52%), enhanced timing of the alert (79% occurred within six hours, 77% within the critical 180-minute SEP-1 window, 47% within an hour of time zero), a 14% increase in bundle compliance, a 10% reduction in ICU length of stay, and a decrease in mortality from sepsis.
Conclusions: The addition of a real-time, automated sepsis time zero calculation improved the performance and timeliness of a predictive sepsis alert provided through a system developed mobile application for clinicians and administrators.
