Houston Methodist achieves 95.6% accuracy predicting dementia patient hospitalization outcomes with ML model

Geriatric patients with dementia experience longer hospital stays and higher healthcare costs than other patients. Clinicians lacked early-warning tools to identify high-risk patients and modifiable risk factors at admission, making timely intervention difficult across Houston Methodist's eight-hospital system.

Researchers developed a machine learning model trained on 10 years of records from 8,407 geriatric dementia patients. The model identifies predictive risk factors and their ranked importance for undesirable hospitalization outcomes on day one or two of admission, covering multiple dementia subtypes including Alzheimer's, Parkinson's, vascular, and Huntington's dementia.

The model achieved 95.6% accuracy, outperforming all other prevalent risk assessment methods. The team plans to deploy it as a mobile app integrated into Epic EHR for system-wide use, alerting ICU and floor staff to high-risk patients and guiding interventional steps to reduce adverse outcomes and hospitalization costs.

• Early AI-based risk stratification on day 1–2 of admission enables timely, targeted clinical interventions for a vulnerable population.
• Identifying modifiable risk factors (e.g., encephalopathy, UTIs, falls, anemia) gives clinicians actionable levers to improve outcomes.
• EHR integration (Epic) is the critical path to scaling research models into routine clinical practice.