Abstract: Acute Kidney Injury (AKI) is associated with increased mortality, morbidity, length of stay, and hospital cost. Since AKI is sometimes preventable, there is great interest in prediction, with recent focus on EHR-based models. In this study, we mine counts of 926 different medications occurring in 124,518 hospitalizations from 34,505 patients. We use iterated, semi-nested, grouped cross validation with machine learning methods lasso and gradient boosting (GB) to reveal novel predictors for AKI in this cohort.
Learning Objective 1: Take a data- rather than knowledge-driven approach to EHR-based predictive modeling.
samuel weisenthal (Presenter)
university of rochester
Martin Zand, university of rochester