Chronic Wound Infection Prediction Model with Six Variables, Two-Center Study

Six consensus predictors—smoking, diabetes duration, wound depth, elevated C-reactive protein, elevated procalcitonin, and hypoalbuminemia—were used to build a random forest model predicting infection risk in chronic wounds. The model was developed using a primary cohort of 500 patients from two tertiary hospitals in China and validated on an external cohort of 300 patients. Yang Jiang published this research on figshare in May 2026, deploying the model as a free web calculator for clinical use.

Use Cases

• Predict infection risk in chronic wounds based on six clinical and biomarker variables.
• Benchmark machine learning algorithms for clinical prediction tasks using AUROC and calibration metrics.
• Interpret model predictions using SHAP values to understand feature importance like hypoalbuminemia.
• Validate a predictive model on an external cohort to assess generalizability.
• Deploy a clinical prediction tool as a web calculator based on routine electronic medical record data.

Strengths

• Model was validated on an external cohort of 300 patients, demonstrating generalizability.
• Performance metrics are reported with confidence intervals: AUROC 0.884 (95% CI 0.841–0.928) in testing.
• Feature selection used a rigorous overlap of Boruta and LASSO methods.
• Model interpretability was examined with SHAP, providing transparent explanations.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect geographic bias inherent to the two tertiary hospitals in China.

Provenance

Source

Two tertiary hospitals in China.

Collection Method

Two-center retrospective study using routine electronic medical record data.

Freshness

Last updated 2026-05-19 05:44:08; freshness should be verified.

Geography

China