Partners AI System Gives Clinicians Better Information

Posted on January 25, 2018 I Written By

Anne Zieger is veteran healthcare consultant and analyst with 20 years of industry experience. Zieger formerly served as editor-in-chief of FierceHealthcare.com and her commentaries have appeared in dozens of international business publications, including Forbes, Business Week and Information Week. She has also contributed content to hundreds of healthcare and health IT organizations, including several Fortune 500 companies. Contact her at @ziegerhealth on Twitter or visit her site at Zieger Healthcare.

While HIT professionals typically understand AI technology, clinicians may not. After all, using AI usually isn’t part of their job, so they can be forgiven for ignoring all of the noise and hype around it.

Aware of this problem, Partners Connected Health and partner Hitachi have come together to create an AI-driven process which isolates data physicians can use. The new approach, dubbed ‘explainable AI,’ is designed to list the key factors the system has relied upon in making projections, making it easier for physicians to make relevant care decisions.

Explainable AI, a newer term used by the two organizations, refers not only to the work being done to develop the Partners system, but also a broader universe in which machines can explain their decisions and actions to human users. Ultimately, explainable AI should help users trust and use AI tools effectively, according to a Hitachi statement.

Initially, Partners will use the AI system to predict the risk of 30-day readmissions for patients with heart failure. Preventing such readmissions can potentially save $7,000 per patient per year.

The problem is, how can organizations like Partners make AI results useful to physicians? Most AI-driven results are something of a black box for clinicians, as they don’t know what data contributed to the score. After all, the algorithm analyses about 3,000 variables that might be a factor in readmissions, drawing from both structured and unstructured data. Without help, there’s little chance physicians can isolate ways to improve their own performance.

But in this case, the AI system offers much better information. Having calculated the predictive score, it isolates factors that physicians can address directly as part of the course of care. It also identifies which patients would be the best candidates for a post-discharge program focused on preventing readmissions.

All of this is well and good, but will it actually deliver the results that Partners hoped for? As it turns out, the initial results of a pilot program are promising.

To conduct the pilot, the Partners Connected Health Innovation team drew on real-life data from heart failure patients under its care. The patients were part of the Partners Connected Cardiac Care Program, a remote monitoring education program focused on managing their care effectively in reducing the risk of hospitalization.

The test compared the results calculated by the AI system with real-life results drawn from about 12,000 heart failure patients hospitalized and discharged from the Partners HealthCare network in 2014 in 2015. As it turned out, there was a high correlation between actual patient readmissions and the level predicted by the system. Next, Partners will share a list of variables that played the biggest role in the AI’s projects. It’s definitely a move in the right direction.