Automated Machine Learning for Prediction of...
Leveraging data from the Patient-Centered Outcomes Prognostication (PCOP) study (Detsky et al., 2017), this secondary analysis will compare the predictive performance of various types of machine learning algorithms for predicting long-term outcomes among patients admitted to 5 intensive care units (ICUs) across 3 hospitals at the University of Pennsylvania Health System between October 2013 and May 2014. The algorithms will use clinical and demographic variables as well as the predictions made by physicians and nurses collected from the PCOP study. Examining the competitive performance of each approach separately-- and when the data are used in a cooperative framework between humans and machines -- this study will elucidate key opportunities to improve performance of prediction methods.