Patients with heart failure are prone to hospitalization. In fact, one in five patients that are discharged from the hospital are readmitted within 30 days. Physicians are not able to adequately predict the risk of hospitalization. Implantable devices that measure pulmonary artery pressure have proven the ability to reduce hospitalization in this patient group. However, these devices are expensive, and the implantation is risky. Hence the need for cost-effective methods that can predict and thereby prevent hospitalization.
One potential method for hospitalization prediction is using artificial intelligence(AI) on data derived from the electronic health record (EHR) or administrative data. In our recent review we summarized the current literature on hospitalization prediction using AI. What did we find? Sixteen research papers described methods for 30 day hospital admission resulting in moderate performance. Six research papers described methods to predict hospitalization within a longer than 30 day window and also reported moderate performance. Only one study prospectively evaluated a home monitoring device for 10 day hospital admission and achieved good performance.
What are the conclusions? These results indicate that hospitalization predicting in the future might be possible. However, all articles are based on retrospective data. In order to value these methods prospective and external validations are warranted. In the future, combining multiple data sources, including tele-monitoring devices might enhance the accuracy of prediction models potentially reducing hospital admission.