Predicting Potentially Preventable Hospital Admissions for an ACO's Patient Pool

Potentially preventable hospital admissions are one of the largest sources of avoidable cost and patient harm that accountable care organizations (ACOs) manage. In 2017 alone, roughly 3.5 million potentially preventable adult hospital stays cost the U.S. health system nearly $34 billion — and Medicare bore 67% of that, about $20.1 billion (AHRQ HCUP Statistical Brief). Identifying which patients are most at risk — early enough to intervene — is a long-standing challenge in value-based care.

In this pilot study, a collaboration between MIT, Harvard Medical School, and Skovos, we developed and evaluated a model to predict the probability of potentially preventable hospital admissions across an ACO's patient pool, using historical clinical, demographic, and utilization data.

The approach

The model scores patients by their likelihood of a potentially preventable admission, so that care teams can prioritize proactive outreach — for example, chronic care management and in-person follow-up — toward the patients most likely to benefit, before an avoidable admission occurs.

Why it matters

Risk stratification of this kind points toward a more proactive, data-driven model of population health management: instead of reacting to admissions after they happen, ACOs and the health systems they work with can direct limited care-management capacity to where it has the greatest preventive impact. This pilot is part of how Skovos grounds its enterprise platform in healthcare-native research alongside academic collaborators.

Authors: Vikas Bommineni, Will Haver, and Arthur Belkin — a collaboration across MIT, Harvard Medical School, and Skovos. Presented as a poster at the MIT × Mass General Brigham AI Cures conference.

To learn how Skovos brings governed, healthcare-native AI to ACOs and health systems, contact our team.