Summary:
UCLA researchers in the Department of Medicine have developed an AI/machine learning model which incorporates patient data from over a hundred features (variables) to predict patients who will have hospitalization and emergency department visits over the next year. This prospective model helps in providing proactive care to prevent future avoidable hospitalization and visits to the emergency room, lowering cost and increasing the quality of healthcare.
Background:
Unnecessary hospitalization and emergency department visits drain healthcare resources while incurring significant cost. Predictive models aid healthcare professionals in identifying at-risk patients who require preventive care in order to avoid costly hospitalization and emergency department visits. Unfortunately, current methods (rules-based models) of identifying at-risk patients often have low predictive performance as they are formulated from only a few factors, are non-scalable, and are not tuned for a specific population. Therefore, an adaptive and precise model for assessing at-risk patients is greatly needed to increase the quality and decrease the cost of healthcare accessible to these populations.
Innovation:
UCLA researchers in the Department of Medicine have developed a prospective AI/ machine learning model to predict patients at risk for future hospitalization and visits to the emergency department over the next year. This model can help to lower the cost and increase the quality of healthcare by quantifying this risk through a simple 0 to 100 risk score. This score identifies individual patients who are likely to have poor outcomes without further intervention such as through care coordination, closing care gaps, and connecting patient to required care. The model allows turning care from reactive to proactive. Moreover, the model outperforms historic rule-based models by using over a hundred features from a number of data domains as needed to make the predictions precise and specific to a given population. This model is scalable and adaptable and can be tuned to the population of a specific health system, medical group, payer, or geography. The model uses clinical, administrative, and social data.
Potential Applications:
- Payers
- Managed Care entities with capitation risk for hospitalization and emergency room visits
- Medical practices – clinics and offices including primary and specialty care
- Hospitals/health systems
- At-risk entities in alternative payment models such accountable care organizations (ACO), Primary Care First, Direct Contracting, and Medicare Advantage
Advantages:
- Adaptable and scalable model
- Population-specific (customized and tuned to a specific population of patients)
- Can lead to lower healthcare utilization/cost
- Predicts utilization of hospitalization/emergency departments visit up to one year ahead in individual patients
- Quantification of risk and factors causing risk in individual patients
Development to Date: The UCLA Population Risk Model has been implemented at fifty UCLA Health primary care practices across Southern California