Data Mined Models for Predicting In-hospital Mortality of Emergency Admissions at Time of Hospital Admission Robert Steele

Emergency admissions involve unplanned admissions for which admission must occur at the earliest possible point of time. Being able to predict in-hospital mortality at the point of admission has significant clinical importance. In this work we have drawn from a large state-wide dataset to train predictive models for in-hospital mortality of emergency admissions, that can be applicable at time of admission. A number of models demonstrate high predictive performance, with AUC scores of up to 0.861. The approach taken represents an alternative to models based upon rich clinical data and may offer a number of comparative advantages in terms of very early point-in-time applicability and broad generalizability.