Zuckerberg San Francisco General Hospital cuts heart failure readmissions 14.3% and retains $7.2M with Epic predictive model

ZSFG, an urban safety net hospital, ranked among the worst in California for 30-day readmission rates in 2016 and faced a potential loss of substantial state and federal pay-for-performance funding annually. Over 40% of readmissions were heart failure patients, and Black patients with heart failure were readmitted at a higher rate than other groups. Existing interventions—a paper checklist and an interdisciplinary Heart Team—were reactive and could not scale to proactively identify high-risk patients.

ZSFG implemented Epic's Risk of Unplanned Readmission (Version 2) predictive model, later augmented with a custom model built on Epic's Cognitive Computing Developer Platform targeting congestive heart failure patients. The model was embedded in a standardized heart failure workspace that guided clinicians through evidence-based care paths and triggered high-priority cardiac clinic referrals for the highest-risk patients. The interdisciplinary Heart Team was given a customized dashboard surfacing model predictions to facilitate monthly case review and proactive care management planning.

From 2018 to 2023, ZSFG reduced 30-day readmissions by 14.3%, achieving one of the lowest readmission rates for any hospital in California. Patient mortality for heart failure cases decreased by 6%, and readmissions among Black patients with heart failure dropped substantially—reversing the disparity identified at project start. ZSFG retained $7.2 million in HRRP-linked funding over six years, delivering a roughly 7-to-1 return on a modest project investment.

• Embedding a predictive model directly into Epic clinical workflows drove 56–75% adoption of the heart failure workspace, demonstrating that point-of-care integration is critical for behavior change at scale.
• In safety net settings, combining medical and social care coordination (addiction, palliative, transportation) alongside AI risk scoring addresses the social determinants that are stronger predictors of readmission than clinical factors alone.
• A measured investment in predictive analytics can generate outsized financial returns by meeting pay-for-performance targets while simultaneously improving equity outcomes.