Vo2max based ML to predict risk of near term death post hospital discharge

<b>Background: </b>Low levels of cardiorespiratory fitness (CRF), as measured by Vo₂max, are associated with higher risks of cardiovascular disease and all-cause mortality. Despite this, CRF markers have not been incorporated into common hospital clinical morbidity and mortality risk assessment tools such the Hospital Score and LACE Index. Black patients have lower CRF levels in comparison to their white counterparts, along with higher rates of cardiovascular mortality, sudden cardiac death and excess mortality. <b>Objective:</b> This study’s aim was to develop a scalable machine learning based model using Vo₂max to identify patients in this high-risk black patient population at elevated risk for near term mortality post discharge. <b>Methods:</b> We performed a retrospective analysis of electronic health record data from the MIMIC-IV dataset to evaluate model parameters of Vo₂max, oxygen saturation, pulse pressure and length of stay. <b>Results:</b> Classification modeling was performed via PyCaret. AdaBoost (accuracy 0.7634/ROC curve AUC 0.86) demonstrated the highest performance in the black patient subset, revealing Vo₂max to be of greatest importance in this population. This study supports measuring Vo₂max prior to hospital discharge, particularly in the black patient population, and demonstrates the potential of this model as a clinical tool to assess near term mortality risk post discharge.