Predicting All-condition, In-hospital Mortality of Elective Patients at Time of Scheduling
Robert JC Steele, Trevor Hillsgrove · 2019
In general, in-hospital mortality of elective patients is low, as such admissions do not correspond to an emergency or urgent admission, but rather are doctor scheduled admissions and so the death of a patient would not in general be expected. Nevertheless, there are still some cases of in-hospital death for elective admissions. In this work we have developed all-health condition, machine learning-based models to predict death for the case of elective admissions, applicable at the time of scheduling the admission, drawing upon a large cross-provider, state-wide dataset of discharge records. The best performing of the developed models demonstrated an AUC of 0.924, showing high discriminative performance. Based on a review of the literature there appears to be no equivalent previously published all-condition predictive model to predict in-hospital mortality of elective admission patients. This makes the work an important novel contribution to the field, and such models can potentially contribute to life-saving decisions for elective patients.