University of Kansas Health System

The University of Kansas Health System had plateaued in its readmission reduction efforts and was failing to meet internal targets. Clinicians lacked timely insight into which patients were at highest risk of readmission before discharge, and there was high variability across inpatient units due to a lack of standard discharge workflows and limited interdisciplinary transparency. The two most common root causes identified were inadequate discharge plans and chronic disease progression.

The health system deployed the Health Catalyst Analytics Platform—including its Late-Binding Data Warehouse—to automate aggregation of clinical, financial, and administrative data and apply machine learning across 30+ published readmission models. A custom multi-variate risk algorithm was developed using variables such as MS-DRG, tobacco use, zip code, medication count, age, and comorbidities. Alongside the predictive model, a Continuum of Care Advisory Team was chartered and a hospital-to-home follow-up program was launched for high-risk patients, including post-discharge calls and home visits.

The combined machine learning, predictive analytics, and lean care redesign initiative produced a 39% relative reduction in all-cause 30-day readmissions and a 52% relative reduction for patients with a principal diagnosis of heart failure. Patients enrolled in the hospital-to-home program achieved a 42% relative reduction in all-cause readmissions and a 49% reduction for heart failure patients specifically.

• Counterintuitive risk factors emerged from the ML model—patients on zero medications and younger patients were at highest risk, overturning clinical assumptions.
• Combining a predictive model with standardized care protocols and a multidisciplinary oversight team was essential; analytics alone was insufficient without process redesign.
• Expanding the hospital-to-home follow-up program beyond heart failure to all high-readmission patients amplified results across the broader population.