A novel use of AI for prediction of clinical deterioration in a post-acute hospital.

There is a growing literature on the prediction of risk of deterioration in hospital settings, including by leveraging artificial intelligence (AI) models. However, this literature has focused on acute-care hospitals, rather than post-acute facilities, where the risk of deterioration remains high. Post-acute facilities tend to have lower digital maturity and poorer data foundations, as well as less rich physiologic data, making the implementation of AI tools for deterioration challenging. In this study, we demonstrate a novel use of AI for the prediction of clinical deterioration in a post-acute hospital. Despite clinical and technical barriers, we developed a meaningful predictive model (sensitivity = 82%). Further, the patterns of contribution to risk of deterioration from the primary contributing variables were remarkably physiologic. This demonstrates a novel model for the prediction of deterioration in an understudied context and provides novel evidence for physiologic predictors of deterioration consistent with previous acute-care scoring systems in the post-acute environment.