Innovative AI Model Aims to Enhance Patient Care Management

In a significant advancement for healthcare, a new artificial intelligence (AI) model developed by UC Davis Health is poised to improve the identification of patients requiring care management services, ultimately aiming to reduce hospitalizations.

The model, known as BE-FAIR (Bias-reduction and Equity Framework for Assessing, Implementing, and Redesigning), is a product of a collaborative effort by a multidisciplinary team. This predictive model is designed to forecast which patients are likely to need immediate assistance, allowing healthcare providers to address health issues proactively before they escalate to emergency department visits or hospital admissions.

Details of the BE-FAIR model were published in the Journal of General Internal Medicine, where the authors emphasize its potential to promote health equity. The paper also offers guidance for other health systems on developing customized AI predictive models that enhance patient care.

As noted by a key expert involved in the project, population health initiatives often rely on predictive models to allocate limited resources effectively. Unfortunately, many existing generic AI models fail to adequately consider diverse patient populations, which can worsen health disparities. The creation of BE-FAIR sought to address these gaps by developing a tailored AI system capable of continual evaluation and improvement.

The development process for the BE-FAIR model spanned two years and involved a structured nine-step framework. This framework enables care managers to predict each patient's likelihood of future hospitalizations or emergency visits based on specific risk factors. Patients identified as being at high risk are then contacted by healthcare staff for needs assessments and possible enrollment in targeted care management programs.

After a year of implementation, the team conducted a thorough evaluation of the model's performance. They discovered that it initially underestimated hospitalizations and emergency visits among African American and Hispanic patients. Through careful analysis and adjustments, the team refined the model to improve its predictive accuracy for these populations.

Healthcare providers have a responsibility to ensure their practices are effective and equitable. By scrutinizing their AI model and enhancing data collection, they managed to implement more impactful population health strategies. Studies highlight the necessity for ongoing evaluations of AI models to ascertain their effectiveness for the diverse patient groups they serve.

Furthermore, experts advocate that AI should not only streamline resource allocation but also promote fairness within healthcare delivery. The BE-FAIR framework integrates equity at every step to prevent the reinforcement of existing health disparities.

AI systems have been increasingly adopted across the United States healthcare landscape, with approximately 65% of hospitals utilizing predictive models developed by electronic health record vendors or third-party developers. It is crucial to recognize that the effectiveness of AI models is directly correlated with the quality of the data they utilize. If a model is not specifically tailored for a particular patient demographic, there is a significant risk that some individuals may be overlooked.

To assist healthcare organizations lacking the resources to create bespoke population health AI models, the UC Davis team has provided a framework that leaders can follow to develop their own systems.