To evaluate the performance and clinical applicability of ML-based depression risk prediction models for patients with diabetes mellitus (DM).
Key Findings:
14 studies with 64 distinct ML models were included.
Pooled AUC of the best-performing models was 0.822 (95% CI, 0.789-0.858).
Substantial heterogeneity among studies (I² = 97.4%; P < 0.001).
Logistic regression was the most frequently used ML method.
Common predictors included age, sex, and BMI.
Interpretation:
The findings indicate that ML models can effectively predict depression risk in diabetic patients, although there is significant variability in model performance.
Limitations:
All included studies were assessed as high risk of bias.
High heterogeneity among studies may affect the reliability of pooled results.
Conclusion:
ML-based models show promise for predicting depression risk in diabetic patients, but further research is needed to improve model reliability and applicability.
