Prediction of Hospital Readmission from Longitudinal Mobile Data Streams.

Chen Qian,Afsaneh Doryab,Carissa Low,Matthew Landers,Patraporn Leelaprachakul,Anind K Dey

doi:10.3390/s21227510

Abstract

Hospital readmissions impose an extreme burden on both health systems and patients. Timely management of the postoperative complications that result in readmissions is necessary to mitigate the effects of these events. However, accurately predicting readmissions is very challenging, and current approaches demonstrated a limited ability to forecast which patients are likely to be readmitted. Our research addresses the challenge of daily readmission risk prediction after the hospital discharge via leveraging the abilities of mobile data streams collected from patients devices in a probabilistic deep learning framework. Through extensive experiments on a real-world dataset that includes smartphone and Fitbit device data from 49 patients collected for 60 days after discharge, we demonstrate our framework’s ability to closely simulate the readmission risk trajectories for cancer patients.

Highlights

Hospital readmissions are associated with high mortality rate and suffering for both patients and family members, and they cost the US health system billions of dollars
Our research addresses the challenge of daily readmission risk prediction after the hospital discharge through leveraging the abilities of mobile devices and deep learning models
We designed a deep learning framework based on LSTM to predict the daily risk of readmission in patients after hospital discharge

Summary

Introduction

Hospital readmissions are associated with high mortality rate and suffering for both patients and family members, and they cost the US health system billions of dollars. Traditional measures use static information at the time of discharge and do not take into account other factors in patients’ daily life that may contribute to the increase or decrease of their readmission risk. These factors among others include a patient’s daily behavioral activities that affect their recovery after treatment. There is no mechanism to objectively measure readmission risk nor is there a method by which readmission risk can be updated over time based on changes in patients’ daily activity and behavior. Carrol et al [15] demonstrated that there is a significant decrease of mortality for pancreatic adenocarcinoma in the first 60 days, and the risk of death does not change significantly between 60 days and 2 years, which indicates that 60 days may be an important endpoint

Methods

Results

Discussion

Conclusion