Data Sheet 1_Machine learning-based risk predictive models for depression in patients with

14 studies comprising 64 distinct machine learning models for predicting depression risk in diabetes patients were systematically reviewed. The pooled analysis of the best-performing models reported a pooled AUC of 0.822, indicating relatively good overall predictive performance. The review, authored by Xingxin Cai and last updated in April 2026, was conducted using a search of databases from their inception to January 2026.

Use Cases

• Benchmarking machine learning model performance (AUC) for depression prediction in diabetic cohorts based on the reported pooled metrics.
• Identifying common predictors like age, sex, and BMI for feature engineering in similar clinical models based on the description.
• Comparing validation strategies (e.g., 5-fold vs. 10-fold cross-validation) based on the reported AUC outcomes for model generalizability assessment.
• Analyzing trends in model type usage (logistic regression, general ML, deep learning) and their reported performance based on the subgroup analysis.

Strengths

• Includes a quantitative meta-analysis of 64 distinct ML models from 14 studies.
• Reports specific pooled performance metrics, including an overall AUC of 0.822 with a 95% CI.
• Compares subgroup performance for traditional regression (AUC 0.765), general ML (AUC 0.789), and deep learning (AUC 0.802) models.
• Analyzes validation strategies, reporting AUCs for 5-fold (0.913), 10-fold (0.819), and random split (0.747) validation.

Limitations

• The primary document is a 2.3 MB DOCX file; column-level data and sample rows are unavailable for direct analysis.
• All included studies were assessed as having a high risk of bias, and the meta-analysis reported substantial heterogeneity (I² = 97.4%).
• The description notes most existing studies had relatively small sample sizes and lacked external validation.

Provenance

Source

figshare, author Xingxin Cai.

Collection Method

Systematic review and meta-analysis of clinical studies from PubMed, Embase, Cochrane Library, and Web of Science.

Time Range

Literature search from database inception to January 2026.

Freshness

Last updated 2026-04-15 04:36:36.

Geography

null