Ahead of the Curve: Predictive Analytics Strategies for Medical Management
The recent report from major payers such as Humana and CVS highlight an escalating trend in healthcare utilization among seniors, driven in part by specialty expenditures. This increase is multifaceted, stemming from an aging population with complex medical needs, the socioeconomic determinants of health, and the potential deferral of elective procedures during the COVID-19 pandemic. Traditional population health and medical management approaches that focus on historically high utilizers have shown limited effectiveness in curbing the upward trajectory of costs in such dynamic environments. Yet, the advent of artificial intelligence (AI) and machine learning (ML) models provides a transformative approach, sharpening the focus on accurately predicting future high utilizers and enabling timely, personalized interventions.
Analyzing Healthcare Trends
Historical models have typically identified a high utilizer patient based on numerous emergency department (ED) or inpatient visits in the preceding year, or healthcare expenditures surpassing a specific threshold. In the subsequent year, a portion of these patients may regress to the mean with reduced risk levels, while others may unfortunately pass away. Consequently, medical management teams enroll only a fraction of these high-risk patients into misaligned intervention programs, which leads to modest impacts on population health outcomes.
Predictive analytics empowers healthcare organizations to analyze historical data to uncover patterns, trends, and anomalies. By harnessing advanced data modeling techniques, predictive analytics not only forecasts healthcare utilization patterns with increased precision but also enables organizations to identify sub-groups that are likely to experience cost spikes in modifiable clinical areas. This targeting of resources towards clinically effective interventions increases program engagement rates by focusing on patients' current and future needs rather than past utilization. A tailored managed care approach significantly affects clinical outcomes, utilization patterns, and patient experiences for both patients and risk-bearing entities.
Core Components of Predictive Analytics in Medical Management
The rise of AI and ML technologies is critical in overcoming the constraints of conventional models. By analyzing vast datasets, including historical and current healthcare claims, electronic health records (EHRs), and social determinants of health (SDOH), AI and ML models can pinpoint individuals at risk of becoming future high utilizers of healthcare services.
The integration of cutting-edge AI and ML tools into medical management involves several essential steps:
Initial hypothesis: Engage with stakeholders to confirm a genuine issue that can be incrementally addressed by AI tool. Ensure the solution is viable and the tool’s output is actionable.
Data Access: Integration of pertinent healthcare claims, clinical data, and other timely demographic or SDOH information to create an extensive dataset for analysis.
Feature Generation: In collaboration with data scientist and subject matter experts, extract relevant features from raw data that can be utilized in model inputs.
For example, one may use raw claims data to develop the feature “number of non-urgent outpatient specialty visits in the last 12 months” in a model that predicts future utilization.
Model Development: Application of sophisticated AI and ML algorithms to construct unbiased predictive models customized for the target population.
For example, a classifier model would describe which of a class a given sample is most likely to belong to, such as the likelihood of cancer vs non-cancer.
Model Refinement and Testing: Rigorous testing and refinement of these models to ensure accuracy and reliability across multiple settings.
Compliance and Certification: Attaining Health Information Trust Alliance (HITRUST) or other certifications and ensuring adherence to relevant compliance standards.
Demonstration Projects: Silent runs of the model can prospectively validate prediction strategies on cohorts without interventions. Feasibility pilots demonstrate technical and operational aspects of utilizing new prediction values existing workflows.
Intervention Strategies: Leveraging learnings from demonstration projects, deployment of personalized intervention strategies informed by the insights derived from predictive models.
Continuous Learning: Commitment to the ongoing development and improvement of AI and ML models, informed by new data and outcomes.
Conclusion
The integration of AI and ML into the fabric of managed care strategies is not merely an opportunity for innovation but a necessity for sustainability amidst escalating healthcare costs. Moreover, these technologies can change existing practice patterns and facilitate the generation of new hypotheses that were previously inconceivable.
Advanced AI models could elucidate the granular sub-etiologies of recent healthcare trends and prioritize modifiable populations. For example, one could assess the proportional impact of COVID-related procedure postponements versus the effects of undiagnosed long-COVID on unexpected healthcare utilization. Post-COVID cost ballooners could be prospectively identified and referred to their symptom specific solutions.
Written By: Reza Alavi, MD, MHS, MBA and Shreya Jain, MBS at Quintuple Aim.
