AI Case Study

Partners HealthCare is developing a new AI tool that predicts the risk of hospital readmissions within 30 days for patients with heart conditions

Partners Connected Health and Hitachi, Ltd., have developed a deep learning model to predict the risk of hospital readmissions within 30 days for cardiac care patients. The algorithm achieved AUC of .71 when applied to 12,000 patient records. Additionally, Hitachi enhanced the algorithm to achieve compliance with regulations by being able to explain the reasoning behind actions.

Industry

Healthcare

Healthcare Providers And Services

Project Overview

"Partners Connected Health and Hitachi, Ltd., today announced a collaboration to develop a new artificial intelligence (AI) tool that can predict with high accuracy, the risk of hospital readmissions within 30 days for patients with heart failure. The tool will help identify appropriate patients to participate in a readmission prevention program study following hospital discharge, and can explain the reason why patients were identified as being at high risk.

To understand why the AI predicted the particular outcome, Tokyo, Japan-based Hitachi developed a technology for risk prediction with analyzing 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. These are elements familiar to clinicians and can support medical decision-making in clinical practice. Through a standard statistical approach based on this risk prediction model, the extracted factors were used to calculate the risk of hospital readmission, and the relevance of the factors was calculated. Thus, this explainable AI technology can enhance prediction accuracy and the quality of medical decision-making.

The custom platform uses Deep Learning to predict the risk of readmission using data points collected as part of Partner HealthCare's remote monitoring study and offer a number of insights to doctors and nurses. Another differentiating factor is that the model is expected to provide a deeper understanding of reasoning behind the prediction."

Reported Results

The prediction algorithm achieved a high accuracy of approximately AUC 0.71, and can be used to significantly reduce the number of patient readmissions

Technology

"With conventional deep learning models, it is difficult for users to understand why the AI predicted a particular outcome.

This presents a challenge for its adoption in healthcare. To address this problem, Tokyo, Japan-based Hitachi developed a technology for risk prediction with analyzing 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. These are elements familiar to clinicians and can support medical decision-making in clinical practice. Through a standard statistical approach based on this risk prediction model, the extracted factors were used to calculate the risk of hospital readmission, and the relevance of the factors was calculated. Thus, this explainable AI technology can enhance prediction accuracy and the quality of medical decision-making."

Function

Operations

General Operations

Background

"The 30-day readmission rate is regarded as one of the important indicators in hospital management, and can carry significant penalties for hospitals via the US Centers for Medicare and Medicaid (CMS) as part of the Affordable Care Act (ACA)."

Benefits

Data

"As part of the 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 patient 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."