How UCLA Health used artificial intelligence to reduce preventable hospital admissions by 27%
The Shift from Reactive to Proactive Healthcare
The healthcare industry is undergoing a fundamental transformation. Rather than waiting for patients to get sick and then treating them, innovative health systems are now using artificial intelligence to identify patients at risk and intervene before they require hospitalization.
UCLA Health has demonstrated the power of this approach through their groundbreaking implementation of AI-driven care management, which resulted in a remarkable 27% reduction in preventable hospital admissions. This success story provides a roadmap for how healthcare systems can leverage AI to improve patient outcomes while optimizing resource utilization.
How UCLA Health Built Their AI Risk Model
UCLA Health developed a sophisticated AI model to identify patients at highest risk for preventable hospital and emergency department visits. The model draws from multiple data sources:
- Electronic health records (Epic) for clinical information
- Administrative claims data from health plans
- Socioeconomic factors from the University of Wisconsin’s Area Deprivation Index
The final model incorporated 144 independent variables and was specifically calibrated to predict potentially preventable hospital admissions and emergency department visits. This comprehensive approach allowed UCLA Health to effectively stratify their approximately 400,000 primary care patients according to risk levels.
From Prediction to Action: The Care Management Workflow
Identifying at-risk patients is only the first step. The real innovation in UCLA Health’s approach was their systematic workflow that translated AI predictions into meaningful interventions. They established a graduated care management system with different levels of support based on patient risk:
For “highest-risk” patients:
- Clinical advisers (registered nurses or licensed clinical social workers) conducted thorough chart reviews
- Consulted with primary care physicians
- Engaged directly with patients to identify adherence barriers and social determinants affecting their health
- Addressed medical aspects like symptom monitoring and medication management or behavioral health concerns depending on their specialty
For “high-risk” patients:
- Administrative specialists called comprehensive care coordinators provided support
- Focused on ensuring adherence to treatment plans
- Addressed prioritized care gaps such as cancer screenings
- Facilitated access to care by coordinating transportation assistance and prescription refills
This stratified approach optimized resource utilization by matching intervention intensity to patient risk levels, creating an efficient system for preventive care delivery across a large patient population.
The Results: Significant Reductions in Hospital Admissions
The implementation of UCLA Health’s AI-driven proactive care management program yielded substantial improvements:
- 27% reduction in potentially preventable hospital admissions
- Trend toward decreased emergency department visits (7% faster deceleration)
- Nonsignificant but promising reductions in all hospital admissions (-19%) and emergency department visits (-15%)
Interestingly, the effectiveness varied across different patient risk segments:
- High-risk patients showed greater reductions in hospital admissions than highest-risk patients
- Highest-risk patients showed greater reductions in emergency department visits
- Patients in the “rising-risk” subgroup experienced significant reductions in both hospital admissions (-30%) and emergency department visits (-42%)
These findings challenge the conventional wisdom that care management resources should be concentrated exclusively on the highest-risk patients. Instead, they suggest substantial benefits may be achieved by targeting interventions across the risk spectrum, with particular attention to those patients whose risk trajectories indicate potential for deterioration.
Economic and Quality-of-Life Benefits
While UCLA Health’s study didn’t explicitly quantify the economic impact, the implications are significant. In 2017 alone, approximately 3.5 million adult hospital admissions in the United States were considered potentially preventable, costing almost $34 billion.
A 27% reduction in preventable admissions therefore represents not only improved clinical outcomes but also substantial potential cost savings. Beyond direct healthcare costs, preventable hospitalizations impose significant burdens on patients in terms of out-of-pocket spending, lost productivity, and diminished quality of life.
Implementation Framework for Other Healthcare Organizations
Healthcare organizations seeking to replicate UCLA Health’s success can benefit from this implementation framework:
- Assemble the right team
- Clinical leaders
- Data scientists
- Care management personnel
- Operational experts
- Identify and integrate relevant data sources
- Electronic health records
- Claims data
- Social determinants of health
- Develop a predictive model emphasizing both accuracy and interpretability
- Create clear risk stratification categories
- Highest-risk
- High-risk
- Rising-risk
- Establish distinct care management protocols for each risk level
- Implement communication and alert systems
- Monitor outcomes and continuously improve
- Clinical outcomes (reduced hospitalizations, improved disease control)
- Operational efficiency (intervention timeliness, resource utilization)
- Patient experience measures
The Future: Reinforcement Learning for Personalized Interventions
While UCLA Health’s model represents the current state of the art, the future holds even more promise with reinforcement learning (RL). Unlike traditional supervised learning, reinforcement learning enables AI systems to learn optimal decision-making strategies through an iterative process of action and feedback.
In the context of proactive patient intervention, reinforcement learning could:
- Determine the most effective intervention approaches for specific patient profiles
- Develop personalized and adaptive interventions that evolve based on patient responses
- Dynamically adjust intervention strategies based on observed outcomes
Researchers at Microsoft have already developed a methodology called Dead-end Discovery (DeD) that focuses on identifying high-risk treatments to avoid rather than prescribing optimal interventions, offering a complementary approach to proactive care.
Conclusion
The UCLA Health case study demonstrates that AI-driven proactive intervention is not merely theoretical but achievable within current healthcare systems, offering tangible benefits for patients, providers, and healthcare organizations alike.
As healthcare continues to grapple with rising costs, capacity constraints, and the growing burden of chronic disease, AI-driven proactive intervention represents a promising pathway toward more efficient, effective, and patient-centered care delivery.
By continuing to refine these models and implementation approaches, the healthcare community can accelerate the transition from reactive to preventative care paradigms, ultimately improving health outcomes while optimizing resource utilization across the healthcare ecosystem.
References
- Proactive Care Management of AI-Identified At-Risk Patients Decreases Preventable Admissions. (2023). The American Journal of Managed Care. https://www.ajmc.com/view/proactive-care-management-of-ai-identified-at-risk-patients-decreases-preventable-admissions
- How Proactive, Predictive, AI-Powered Patient Oversight Can Boost Health Outcomes. (2023). Healthcare IT News. https://www.healthcareitnews.com/news/how-proactive-predictive-ai-powered-patient-oversight-can-boost-health-outcomes
- Using Reinforcement Learning to Identify High-Risk States and Treatments in Healthcare. (2023). Microsoft Research Blog. https://www.microsoft.com/en-us/research/blog/using-reinforcement-learning-to-identify-high-risk-states-and-treatments-in-healthcare/
- Technology Will Improve How Health Care Services Are Distributed to At-Risk Patients. (2023). UCLA Health News. https://www.uclahealth.org/news/article/technology-will-improve-how-health-care-services-are-distributed-to-at-risk-patients
- Predictive Analytics in Healthcare. (2023). Gleecus. https://www.gleecus.com/blogs/predictive-analytics-healthcare/
- The Next Frontier of AI in Healthcare: Prediction and Proactive Care. (2023). Health Management. https://healthmanagement.org/c/it/News/the-next-frontier-of-ai-in-healthcare-prediction-and-proactive-care
- AI-Powered Healthcare: Shifting from Reactive to Proactive. (2023). HealthLeaders Media. https://www.healthleadersmedia.com/technology/ai-powered-healthcare-shifting-reactive-proactive
- Reinforcement Learning in Health Care: Why It’s Important and How It Can Help. (2023). CapeStart. https://www.capestart.com/resources/blog/reinforcement-learning-in-health-care-why-its-important-and-how-it-can-help/