Data Sheet 2_A clinically interpretable machine learning model for early detection of diab

A retrospective cohort of 1,475 diabetic patients from three community health centers in China, used to develop and validate machine learning models for predicting diabetic retinopathy risk. The dataset includes a development set of 1,177 patients and a held-out test set of 298 patients. The model results, including performance metrics and feature importance analysis, were authored by Juncheng Tong and published on figshare in May 2026.

Use Cases

• Benchmarking machine learning models for clinical risk prediction based on the described performance metrics (AUROC, AUPRC, Brier score).
• Studying feature importance for diabetic retinopathy based on the SHAP analysis identifying urine glucose as a key predictor.
• Evaluating model calibration and decision curve analysis for clinical utility based on the reported calibration intercept, slope, and net benefit thresholds.
• Developing risk stratification tools for primary care based on the use of routinely collected health variables described in the study.

Strengths

• Model performance is quantitatively reported with specific metrics: AUROC of 0.770 (95% CI 0.671–0.856) and AUPRC of 0.452 for the GLMNET model.
• The cohort is clearly defined with 1,475 patients and a split into development (n=1,177) and test (n=298) sets.
• The study provides interpretability insights via SHAP analysis, identifying the most influential predictor.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the underlying data is unknown, which may limit suitability assessment.
• The dataset scope is limited to 46.9 KB, suggesting it contains model documentation or summary results rather than the full patient-level data.

Provenance

Source

figshare, authored by Juncheng Tong.

Collection Method

Retrospective study using routinely collected health data from three community health centers.

Time Range

null

Freshness

Last updated 2026-05 04:24:13; freshness should be verified.

Geography

China (three community health centers).