Mayo Clinic’s AI-enabled CARE Risk Score is designed to address a lingering challenge.
By John Halamka, M.D., M.S., Dwight and Dian Diercks President, Mayo Clinic Platform and Paul Cerrato, MA, senior research analyst and communications specialist, Mayo Clinic Platform
Preventable hospital readmissions remain a challenge for many hospitals, a challenge that can have a significant impact on patient outcomes and a health system’s financial well-being. The Centers for Medicare and Medicaid Services has a Hospital Readmissions Reduction Program that will reduce reimbursements to acute care hospitals that allow excessive 30-day unplanned readmissions. The Program focuses on the following conditions:
- Heart failure
- Acute myocardial infarction
- Pneumonia
- Chronic obstructive pulmonary disease (COPD)
- Coronary artery bypass graft (CABG) surgery
- Elective primary total hip and/or knee arthroplasty
Mayo Clinic is developing the CARE Risk Score (Clinical Assessment of REadmission Risk Score) to address this problem. The score is designed to integrate with the Mayo Clinical Platform and Epic (Plummer Chart), embedding predictive insights directly into clinical workflows, updating four times daily across the patient encounter after the first day. This work aligns with Mayo Clinic’s broader digital health and AI strategy and supports the BOLD FORWARD 2030 goals by delivering explainable, actionable analytics to multidisciplinary care teams. The AI‑powered tool uses an XGBoost model to analyze structured data, assign each patient a risk percentile from 1 to 100, and classify the drivers of risk across five categories: medications and treatment plan, medical history, social and demographic drivers, physiological stability, and care complexity.
Each of these drivers includes a long list of specific parameters. Under medications and treatment plan, the model will monitor obvious things like active drugs but will also track and incorporate frequent hospitalizations. The predictive model takes a deep dive into a patient’s medical history as well. It tracks recent admissions, including ED visits, and the number of days since the patient was last discharged. It also includes any chronic conditions they may be experiencing, like kidney and liver disease, cancer, and diabetes. To measure care complexity, the model records the Braden Activity Score and Mobility Score, their fall risk score, in addition to other metrics.
Notably, the Readmission Prediction Score monitors social parameters that are often overlooked, including evidence captured through Mayo Clinic’s Social Determinants of Health Score that includes: intimate partner violence, inadequate housing, food insecurity, financial strain, education level and employment, social isolation, physical activity levels, and level of neighborhood social economic status.
To accomplish this goal, the CARE score will be evaluated using adoption and engagement metrics, including score reviews at discharge, interaction with the tool over time, and user feedback to judge usability and how effective it blends in with current workflows.
Preventable readmissions to the hospital continue to trouble clinicians and administrators. But with carefully developed AI-assisted tools, we are optimistic about reducing this burden on hospitals and patients alike.
Share this:
