Machine learning prediction of hospital patient need for post-acute care using an admission mobility measure is robust across patient diagnoses

Abstract

ObjectiveOne-fifth of patient discharges from the acute hospital are delayed due to non-medical reasons. Prior research on small specific samples shows that patient mobility is important for predicting post-acute care (PAC) need. Our purpose was to create a disposition prediction model for PAC need in a large, clinically diverse. MethodsA random forest (RF) was constructed to analyze patient admissions at 2 hospitals. The primary outcome was discharge disposition (home or PAC). Predictors included the lowest AM-PAC ‘6-clicks’ mobility score within 48-hours of admission (primary predictor) and demographic and clinical characteristics. A global summary tree was constructed to summarize the RF. ResultsAmong 34,432 patient admissions, the most important variables for predicting PAC placement were AM-PAC, BMI, and age. The AUC was 0.80 (95% confidence interval: 0.79, 0.81). Using a predicted probability for PAC of 0.25 or higher, the sensitivity, specificity and overall accuracy was 76%, 70% and 72%, respectively. Patients 66 years or older with AM-PAC of <31 had the highest probability (0.76) for discharge to PAC. Patients with AM-PAC of >43 had the highest probability for discharge to home. ConclusionsSystematic assessment of inpatients admission mobility should be implemented and used for discharge planning. Electronic medical record systems should be designed to collect and facilitate availability of mobility data on all patients to providers who play key roles in discharge planning. Public Interest SummaryPatient's mobility status during hospitalization has been used to predict their next level of care at discharge, but this work has been done with more limited methods and focused on select patient groups. Using a machine learning technique on thousands of patients with very different medical problems, this study shows that mobility status very early in hospitalization predicts post-acute care (PAC) needs. Based on this study we recommend that early assessment of patient mobility in the hospital should occur for all patients as it can facilitate more effective discharge planning.