An Algorithm for Early Detection of Sepsis Using Traditional Statistical Regression Modeling

Abstract

Sepsis is the final common pathway for many infections, whereby the body’s immune response leads to organ failure, and eventually death. It is associated with high mortality rates and, if survived, significant morbidity. Early detection is imperative to improve outcomes. Yet, there is also a need to avoid a high false alarm rate. The aim of this study was to develop and evaluate a simple algorithm for early sepsis detection. Significant missing data were encountered in the dataset, which were forward-filled or substituted with population means. Clinically relevant variable combinations were added along with transformation features including dichotomization, z-scores, first derivative, and changes from baseline. A logistic regression model was used to identify candidate features and build the overall risk score function for prediction. The final candidate score had areas under the receiver operating characteristic curve of 0.747, 0.760, and 0.783 for the three test data sets. It had accuracies of 0.795, 0.889, 0.815, respectively, and an overall utility score for the full test set of 0.249 using a cutoff of 0.024. Evaluation indicated significant potential for further optimization, including reduction of false-positive predictions. Adding features capturing change over time is expected to provide scope for further investigation.