C-LACE2: computational risk assessment tool for 30-day post hospital discharge mortality

Abstract

This paper describes a computational approach to assessing 30-day post-hospital discharge mortality risk. The idea of the constructed model is partially based on the popular LACE (Length of stay, Acuity, Comorbidities and Emergency visits) model, and combines elements of APACHE II (Acute Physiology and Chronic Health Evaluation) and IPEC (Inpatient Evaluation Center of Veterans Health Administration) mortality models. The resulting C-LACE2 model includes length of stay, acuity, comorbidities, selected lab values and medications. The process of construction of the model and its validation are presented in details. The constructed final model consists of 101 attributes, and its minimum version of 20 attributes. C-LACE2 has been constructed by applying machine learning methods to MIMIC III inpatient EHR data. The model achieved accuracy (AUC) of 0.779. Detailed analysis of the C-LACE2 model has been performed to check its sensitivity to inputs.