Clinical Data for Identifying Atrial Fibrillation in Diabetic Kidney Disease Patients

787 patient records from a retrospective two-center cohort study conducted between January 2021 and October 2025. The dataset was used to develop an interpretable machine-learning model for identifying comorbid atrial fibrillation in patients with diabetic kidney disease. The model was created by Xiaoran Li and is shared under a CC-BY-4.0 license.

Use Cases

• Training interpretable machine-learning models for disease identification based on routine clinical and echocardiographic variables.
• Benchmarking feature selection methods like LASSO regression based on the 14 retained clinical predictors.
• Studying the relationship between renal dysfunction, proteinuria, and atrial fibrillation risk based on the SHAP-identified major contributors.
• Developing preliminary clinical calculators for AF identification based on the translated web-based tool.

Strengths

• Model performance is documented with specific metrics, including an AUC of 0.927 and accuracy of 0.886 on the test set.
• The dataset underpinning the model includes 787 unique patient records from a defined time period (January 2021 to October 2025).
• The model's interpretability is supported by SHAP analysis, identifying key contributors like proteinuria burden and age.

Limitations

• The dataset is very small at 29.3 KB, indicating limited scope and a small number of records.
• The model has undergone internal validation only; external and prospective validation is required for clinical application.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

Xiaoran Li

Collection Method

Retrospective two-center cohort study using routinely collected clinical data.

Time Range

January 2021 to October 2025

Freshness

Last updated 2026-05-26 04:39:11; freshness should be verified.