First, Do No Harm: Predictive Analytics to Reduce In-Hospital Adverse Events

Abstract

Inadequate patient safety is a serious issue in current medical practice. Medical errors cause adverse events (AEs) for patients and lead to premature deaths, unintended complications, prolonged hospital stays, and higher medical costs. Although the importance of AE prediction and prevention is well recognized in the information systems literature, there is a dearth of research on modeling and predicting AEs caused by medical errors. Following the design science research paradigm, this study describes the search, design, and evaluation of a novel in-hospital AE prediction model, called StoChastic AutoRegressions for LatEnt Trajectories (SCARLET). The proposed model integrates generalized linear mixed model with multitask learning and stochastic time-series processes. Results from our large-scale empirical evaluation show that SCARLET outperforms prior state-of-the-art techniques in predicting harms from medical errors during patients’ hospital stays. Through a simulated experiment, we further demonstrate significant cost savings potential when hospitals implement and integrate SCARLET in their inpatient clinical workflow.