Predictive models of medication non-adherence risks of patients with T2D based on multiple machine learning...

Area of Science:

• Diabetes Mellitus Research
• Machine Learning in Healthcare
• Predictive Analytics

Background:

• Medication adherence is critical for managing type 2 diabetes (T2D).
• Predicting non-adherence is vital for personalized care, especially in resource-limited settings.
• Existing predictive models for T2D non-adherence require further development.

Purpose of the Study:

• To evaluate multiple machine learning algorithms for predicting T2D medication non-adherence.
• To identify the most effective model for identifying high-risk patients.
• To provide a tool for individualized diabetes management.

Main Methods:

• A real-world registration study involving 401 T2D patients.
• Data collected via face-to-face questionnaires covering demographics, disease, treatment, lifestyle, and adherence.
• Fourteen machine learning algorithms were applied and evaluated using Area Under the Receiver Operating Characteristic Curve (AUC).

Main Results:

• 21.20% of patients exhibited poor medication adherence.
• 16 variables were selected for modeling, resulting in 300 models.
• The best-performing machine learning model achieved an AUC of 0.866±0.082.

Conclusions:

• An accurate and sensitive adherence prediction model was developed using real-world data.
• Data imputation, balanced sampling, and larger sample sizes enhance predictive accuracy.
• This model offers a potential technical solution for personalized diabetes care.