Predicting severe diabetes complications using administrative claims data in Maryland

Area of Science:

• Health Informatics
• Predictive Analytics
• Diabetes Management

Background:

• Severe type 2 diabetes complications pose a significant burden on healthcare systems.
• Accurate prediction of these complications is crucial for timely intervention and resource allocation.
• Existing risk stratification tools may not fully capture the complexity of severe diabetes complications.

Purpose of the Study:

• To develop and evaluate a large-scale predictive model for severe type 2 diabetes complications (DC) among Medicare beneficiaries in Maryland.
• To assess the operational performance and predictive accuracy of the developed model.
• To compare the model's performance against Hierarchical Condition Category (HCC) scores.

Main Methods:

• Retrospective longitudinal analysis of Medicare fee-for-service (FFS) claims data from March 2021 to July 2024.
• Development of a multivariable discrete-time survival model incorporating 219 candidate risk factors.
• Stepwise variable selection to identify 95 statistically significant risk factors for DC prediction.
• Creation and validation of DC risk scores, comparing their predictive performance against HCC scores.

Main Results:

• The developed DC risk model identified 95 significant risk factors, primarily utilization- and condition-based.
• DC risk scores demonstrated strong predictive power, with the top 10% of scores accounting for 56.9% of severe DC events.
• The DC risk model significantly outperformed HCC scores, which accounted for 37.5% of events.

Conclusions:

• A predictive model utilizing administrative claims data can effectively forecast severe diabetes complications in the Medicare FFS population in Maryland.
• The developed model offers a more accurate risk stratification tool compared to traditional HCC scores.
• This approach has the potential to improve patient outcomes and optimize healthcare resource utilization for diabetes management.