Abstract
Hospital acquired complications (HACs) are serious problems affecting modern day healthcare institutions. It is estimated that HACs result in an approximately 10% increase in total inpatient hospital costs across US hospitals. With US hospital spending totaling nearly $900 billion per annum, the damages caused by HACs are no small matter. Early detection and prevention of HACs could greatly reduce strains on the US healthcare system and improve patient morbidity & mortality rates. Here, we describe a machine-learning model for predicting the occurrence of HACs within five distinct categories using temporal clinical data. Using our approach, we find that at least $10 billion of excessive hospital costs could be saved in the US alone, with the institution of effective preventive measures. In addition, we also identify several keystone features that demonstrate high predictive power for HACs over different time periods following patient admission. The classifiers and features analyzed in this study show high promise of being able to be used for accurate prediction of HACs in clinical settings, and furthermore provide novel insights into the contribution of various clinical factors to the risk of developing HACs as a function of healthcare system exposure.