Predicting High-risk and High-cost Patients for Proactive Intervention.

Area of Science:

• Health Services Research
• Health Informatics
• Predictive Analytics

Background:

• A small percentage of patients account for a disproportionately large share of healthcare expenditures.
• Proactive intervention for high-risk, high-cost (HRHC) patients is a key strategy for cost reduction and improved health outcomes.
• Accurate identification of HRHC patients is crucial for effective resource allocation and care management.

Purpose of the Study:

• To develop and validate a predictive model for identifying HRHC patients with enhanced accuracy.
• To assess the predictive power of various statistical models and comorbidity datasets for cost prediction.
• To provide a tool for value-based health systems to proactively manage high-cost patient populations.

Main Methods:

• Observational study utilizing administrative data from the Veterans Health Administration for fiscal years 2018 and 2019.
• Analysis included over 5.6 million patients with continuous care across both fiscal years.
• Split-sample validation was employed to evaluate 5 statistical models and 3 sets of patient comorbidities, including expanded Clinical Classifications Software Refined (CCSR) groups.

Main Results:

• Box-Cox regression utilizing expanded CCSR groups as predictors demonstrated the highest predictive accuracy.
• The model achieved an R-squared value of 0.51 for transformed cost and 0.37 for raw scale cost.
• The developed model significantly outperformed previously reported predictive models in the literature.

Conclusions:

• The study successfully developed a highly predictive model for identifying HRHC patients.
• The algorithm leverages readily available administrative data and a public classification system (CCSR).
• This model offers a practical and implementable solution for healthcare systems aiming for value-based care and proactive patient management.