Opportunity 

Nearly one in five patients discharged from the hospital are readmitted within 30 days. If health systems could predict which patients are at high risk of readmission, providers would potentially be able to offer targeted interventions before discharge.  

However, a critical limitation of existing hospital readmission prediction models is that they use data collected until discharge but do not leverage information on patient behaviors at home. 

Intervention 

We designed a clinical trial called PREDICT that compared ways to use remotely monitored patient activity levels after hospital discharge to enhance readmission prediction models. We enrolled 500 patients who were randomly assigned to track activity using either a smartphone or a wearable device for six months. 

To create the enhanced prediction model, we incorporated remote monitoring data for physical activity and sleep into a standard model that uses patient characteristics such as demographics, comorbidities, hospital length of stay, and vitals before discharge. We also tested prediction using traditional regression versus nonparametric machine learning regression.

Way to Health was used to administer and manage the study. 

Impact 

The best predictions of 30-day readmission were made using the enhanced prediction model plus machine learning approaches. Data from wearables yielded slightly better predictions than data from smartphones.

In terms of usage, patients assigned to the smartphone-tracking arm transmitted data for a greater duration and proportion of time, with a 32 percent relative increase in patients completing the 180-day period compared with those using wearables. Because smartphones are ubiquitous, these findings indicate that these devices could be a scalable approach for remotely monitoring patient health behaviors.