Dynamic Surveillance of the Cardiovascular Risk and Identification of Treatment Responders in Type 2 Diabetes Using A...

Area of Science:

• Cardiology
• Endocrinology
• Artificial Intelligence in Medicine

Background:

• Type 2 diabetes significantly increases cardiovascular disease (CVD) risk.
• Accurate prediction of CVD events and treatment response is crucial for managing patients with type 2 diabetes.
• Existing risk scores may not fully capture dynamic risk changes or individual treatment effects.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting cardiovascular risk in type 2 diabetes patients.
• To identify patients who exhibit a positive response to specific interventions, such as canagliflozin.
• To enable dynamic risk surveillance and personalized therapeutic guidance.

Main Methods:

• Utilized data from 11,677 patients across four clinical trials (ACCORD, CANVAS, CANVAS-R, CREDENCE).
• Developed an ML-CVD Primary model using baseline and 1-year changes in interim factors.
• Validated the model using holdout and external testing datasets, assessing predictive performance with Harrell's C-index.

Main Results:

• The ML-CVD Primary model showed strong predictive performance (C-index 0.66-0.71), outperforming traditional scores in primary prevention.
• Intensive interventions, including canagliflozin, significantly reduced ML-CVD Primary risk score progression.
• Patients identified as 'Responders' (decreased score) on canagliflozin had a 45% lower risk of cardiovascular outcomes compared to 'Non-Responders'.

Conclusions:

• The ML-CVD Primary model provides dynamic cardiovascular risk prediction for type 2 diabetes patients.
• It effectively identifies individual responses to therapies, aiding in personalized treatment strategies.
• This approach supports ongoing risk monitoring and the optimization of cardiovascular protective therapies.