Intelligent type 2 diabetes risk prediction from administrative claim data

Area of Science:

• * Computational epidemiology
• * Health informatics
• * Machine learning in healthcare

Background:

• * Type 2 diabetes is a significant global health burden.
• * Early detection and prevention are crucial for managing type 2 diabetes.
• * Administrative claims data offer a potential resource for predictive modeling.

Purpose of the Study:

• * To develop and evaluate supervised machine learning models for predicting type 2 diabetes risk.
• * To identify key predictors of type 2 diabetes from administrative claims data.
• * To assess the performance of different machine learning algorithms in diabetes prediction.

Main Methods:

• * Utilized administrative claim data and 31 variables based on the Elixhauser Comorbidity Index.
• * Applied five supervised machine learning algorithms: Random Forest (RF), k-nearest neighbor, and others.
• * Employed Principal Component Analysis (PCA) for variable importance ranking.

Main Results:

• * Random Forest (RF) demonstrated the highest prediction accuracy at 85.06%.
• * k-nearest neighbor achieved 84.48% accuracy, closely following RF.
• * Patient age and solid tumor without metastasis were identified as the most significant predictors.

Conclusions:

• * Supervised machine learning, particularly RF, is effective for predicting type 2 diabetes risk from administrative data.
• * The developed models can support automated surveillance systems for at-risk populations.
• * Findings provide valuable insights for healthcare regulators and insurers for proactive diabetes management.