Predicting Diabetic Neuropathy: ML Model for Early Detection in T2DM
Summary: This retrospective study analyzes data from 1,001 T2DM patients in Xinjiang, China, to build and validate machine learning models for diagnosing diabetic peripheral neuropathy (DPN), a key precursor to foot ulcers and amputations. Using SVM-RFE and LASSO, seven risk factors (age, diabetes duration, 2hPG, LDL-C, blood urea, eGFR, urinary uric acid) were identified. The Random Forest model excelled with an external validation AUC of 0.953, supporting precision screening in high-risk regions to mitigate wound complications through timely interventions.
Key Highlights:
- Random Forest model achieved top performance (training AUC 1.000, external AUC 0.953), outperforming GBM, GLM, and SVM for DPN prediction.
- Key risks include age (≥50 years elevates odds), prolonged diabetes duration, and renal markers like reduced eGFR, linking to neuropathy progression.
- External validation on 123 patients confirmed accuracy (86.3%), sensitivity (93.0%), and no overfitting via bootstrapping and calibration.
- Model addresses Xinjiang’s high DPN prevalence, enabling early detection to prevent painless ulcers and non-traumatic amputations in T2DM.
- Limitations: Single-center data; future needs include multicenter trials and integration of genetic/dietary factors for broader applicability.
Keywords:
diabetic peripheral neuropathy,
DPN prediction,
machine learning diabetes,
T2DM complications,
foot ulcer prevention