AI Predicts Risk of Readmission for Heart Failure Patients
By HospiMedica International staff writers
Posted on 19 Dec 2017
Image: New AI technology can very accurately predict the risk of hospital readmissions within 30 days among patients with heart failure (Photo courtesy of iStock).
Hitachi, Ltd. (Tokyo, Japan) and Partners Connected Health (Somerville, MA, USA) have collaborated to develop artificial intelligence (AI) technology, which can very accurately predict the risk of hospital readmissions within 30 days among patients with heart failure. The AI technology helps choose appropriate patients to participate in a readmission prevention program after hospital discharge and provides an explanation as to why the patients have been identified as high risk ones.
Partners Connected Health, at Partners HealthCare, is leveraging information technology – mobile phones, tablets, wearables, sensors and remote health monitoring tools – to deliver quality patient care outside of traditional medical settings. The Connected Health team creates and deploys mobile technologies in a number of patient populations and care settings, and is conducting innovative research studies to test the effectiveness of mobile health technologies in various clinical applications, including medication adherence, care coordination, chronic disease management, prevention and wellness.
Hitachi's new AI technology uses deep learning to construct the risk prediction model. The company’s technology for risk prediction analyzes the results presented by deep learning and extracting the several dozens of actionable factors for each patient from the vast amount of data collected from heart failure patients. Through a standard statistical approach based on this risk prediction model, the extracted factors are used to calculate the risk of hospital readmission, and the relevance of the factors is calculated.
As part of a study, the Partners Connected Health Innovation team simulated the readmission prediction program among heart failure patients participating in the Partners Connected Cardiac Care Program (CCCP), a remote monitoring and education program designed to improve the management of heart failure patients at risk for hospitalization. These results were compared to data from approximately 12,000 heart failure patients hospitalized and discharged from the Partners HealthCare hospital network in 2014 and 2015. The analysis showed the prediction algorithm achieved a high accuracy of approximately AUC 0.71, and can significantly reduce the number of patient readmissions. (AUC, area under the curve, is a measure of prediction model performance with an ideal value range from 0 to 1.) As a result, an additional amount of approximately USD 7,000 savings per patient per year can be expected among the cohort of CCCP patients.
The technology is an example of explainable AI, a new term currently defined as enabling machines to explain their decisions and actions to human users, and enabling them to understand, appropriately trust and effectively manage AI tools, while maintaining a high level of prediction accuracy. Hitachi and the Partners Connected Health Innovation team plan to jointly conduct a prospective study to evaluate the prediction program by clinicians, and study how to integrate this within clinical workflows. By using this new AI technology, Hitachi will provide solutions for the medical field, including solutions for insurance and pharmaceutical companies, emergency services, and other healthcare services where prediction-based on medical data can be utilized.
"Traditional machine learning can help us predict events, but as end-users, we can't tell why the machine is predicting something a certain way," said Kamal Jethwani, MD, MPH, Senior Director, Partners Connected Health Innovation. "With this innovation, doctors and nurses using the algorithm will be able to tell exactly why a certain patient is at high risk for hospital admission, and what they can do about it. We want to enable our providers to act on this information, which is a step beyond the state-of-the-art today, in terms of machine learning algorithms."