Clinical Scorecard: Evolving Readmission Risk Factors in Type 2 Diabetes: A Machine Learning Approach with SHAP Interpretability Over Short, Medium, and Long-Term Periods

At a Glance

Key Highlights

• Nine machine learning models were developed to predict readmission at 30, 60, and 365 days.
• Age was the primary predictor for 30-day readmission; inflammatory markers for 60-day; chronic complications for 365-day.
• Artificial Neural Networks (ANN) showed the highest performance for short/medium-term predictions.
• Random forest was effective for long-term prediction.
• SHAP analysis provided insights into the dynamic nature of risk factors.

Guideline-Based Recommendations

Diagnosis

• No specific predictive tools recommended for different time windows.

Management

• Dynamic risk monitoring should be considered for T2DM patients.

Monitoring & Follow-up

• Regular assessment of risk factors is essential as they evolve over time.

Risks

• Readmission rates for T2DM patients are significantly higher than non-diabetic individuals.

Patient & Prescribing Data

14,048 hospitalized T2DM patients

Focus on individualized follow-up and intervention strategies based on evolving risk factors.

Clinical Best Practices

• Utilize machine learning models for predicting readmission risk.
• Incorporate SHAP analysis for understanding risk factor contributions.
• Adapt clinical interventions based on the time-specific risk factors identified.

Related Resources & Content

• IDF Diabetes Atlas 2025
• Centers for Medicare & Medicaid Services (CMS)