A Data-Driven Approach to Predicting Septic Shock in the IntensiveCare Unit

摘要

Early diagnosis of sepsis and septic shock has been unambiguously linked to lowermortality and better patient outcomes. Despite this, there is a strong unmetneed for a reliable clinical tool that can be used for large-scale automatedscreening to identify high-risk patients. We addressed the following questions:Can a novel algorithm to identify patients at high risk of septic shock 24 hoursbefore diagnosis be discovered using available clinical data? What areperformance characteristics of this predictive algorithm? Can current metricsfor evaluation of sepsis be improved using novel algorithm? Publicly availabledata from the intensive care unit setting was used to build septic shock andcontrol patient cohorts. Using Bayesian networks, causal relationships betweendiagnosis groups, procedure groups, laboratory results, and demographic datawere inferred. Predictive model for septic shock 24 hours prior to digitaldiagnosis was built based on inferred causal networks. Sepsis risk scores wereaugmented by de novo inferred model and performance was evaluated. A novelpredictive model to identify high-risk patients 24 hours ahead of time, witharea under curve of 0.81, negative predictive value of 0.87, and a positivepredictive value as high as 0.65 was built. The specificity of quick sequentialorgan failure assessment, systemic inflammatory response syndrome, and modifiedearly warning score was improved when augmented with the novel model, whereas noimprovements were made to the sequential organ failure assessment score. We useda data-driven, expert knowledge agnostic method to build a screening algorithmfor early detection of septic shock. The model demonstrates strong performancein the data set used and provides a basis for expanding this work towardbuilding an algorithm that is used to screen patients based on electronicmedical record data in real time.