Derivation and validation of a time-dependent risk prediction model for in-hospital mortality

Abstract

Accurate risk prediction models for in-hospital mortality are important for unbiased comparisons of hospital performance (by producing risk-adjusted mortality rates) and improved patient outcomes (by identifying high-risk patients in need of special medical attention). No previous risk prediction models have properly used post-admission information to predict risk of death in-hospital. In this study, we used administrative and laboratory data to derive and internally validate a Cox regression model (the "Escobar +" model) that predicts the risk of in-hospital death at any point during the admission. The model had excellent discrimination (c-statistic 0.895,95% confidence interval [CI] 0.889-0.902) and calibration. The Escobar+ model is a powerful risk-adjustment methodology that can be used in studies where the start of observation occurs post-admission. The model could also improve the quality and timeliness of patient care by providing health care workers with highly specific and accurate estimates of in-hospital death risk during the patient's stay.