Risk adjustment using automated ambulatory pharmacy data: the RxRisk model

Area of Science:

• Health Services Research
• Medical Informatics
• Pharmacoeconomics

Background:

• Accurate prediction of future healthcare costs is crucial for resource allocation and financial planning in managed care organizations.
• Existing risk adjustment models often rely on diagnosis codes (ICD-9-CM) or demographic data, which may not fully capture patient complexity.
• Automated pharmacy data offers a potentially valuable, underutilized source for identifying chronic conditions and assessing health risk.

Purpose of the Study:

• To develop and evaluate the RxRisk model, a novel risk assessment instrument utilizing automated ambulatory pharmacy data.
• To compare the predictive performance of RxRisk against demographic-only models, Ambulatory Clinical Groups (ACGs), and Hierarchical Coexisting Conditions (HCCs) for future healthcare costs.
• To assess the utility of RxRisk in forecasting healthcare resource utilization.

Main Methods:

• A retrospective cohort study was conducted using health services utilization and cost data from approximately 1.5 million individuals across five US Health Maintenance Organizations (HMOs).
• The RxRisk algorithm was developed to classify prescription drug fills into chronic disease categories for both adult and pediatric populations.
• Model performance was evaluated by comparing the variance in prospective healthcare costs explained (R-squared) by RxRisk, ACGs, and HCCs.

Main Results:

• Hierarchical Coexisting Conditions (HCCs) demonstrated the highest accuracy in forecasting total healthcare costs, explaining 15.4% of the variance.
• The RxRisk model explained 8.7% of the prospective cost variance, performing similarly to the Ambulatory Clinical Groups (ACGs) model, which explained 10.2%.
• Despite differences in overall accuracy, RxRisk, ACGs, and HCCs generated comparable cost predictions for the middle 60% of the healthcare cost distribution.

Conclusions:

• While HCCs provide more accurate total cost forecasts, the pharmacy-based RxRisk model serves as a viable alternative to diagnosis-based risk assessment instruments.
• RxRisk's reliance on readily available pharmacy data makes it a potentially more appropriate and accessible option for certain medical risk analysis applications.
• The choice of risk assessment model should consider the specific application and the trade-offs between data sources and predictive accuracy.