Data Sheet 1_Development and validation of an explainable neural network model for predict

349 patient records from a single-center retrospective cohort study at Quzhou People's Hospital in China, collected between June 2022 and June 2025. The dataset likely contains 36 baseline clinical characteristics used to identify four core predictors for diabetic kidney disease progression. Author Binfeng Xiong published the data sheet on figshare in May 2026.

Use Cases

• Training machine learning models to predict diabetic kidney disease progression based on routine clinical parameters.
• Evaluating feature importance and model interpretability using SHAP analysis on identified predictors like MASLD and AST.
• Comparing performance of different ML models (e.g., neural networks) for clinical risk prediction based on AUC and calibration metrics.
• Developing cost-effective clinical tools for early intervention in primary care settings based on the identified core predictors.

Strengths

• Model validation includes specific performance metrics: AUC of 0.742, Hosmer-Lemeshow P=0.2020, and Brier score of 0.2105.
• Feature selection was performed using LASSO regression and multivariable logistic regression from 36 baseline characteristics.
• SHAP analysis confirmed stable feature importance rankings between training and test datasets (Pearson r = 0.976).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data reflects geographic bias inherent to a single-center study in China.

Provenance

Source

Quzhou People's Hospital, China.

Collection Method

Single-center retrospective cohort study.

Time Range

June 2022 to June 2025.

Freshness

Last updated 2026-05-28 04:50:49; freshness should be verified.

Geography

China.