Development and assessment of an early diagnostic approach for painful diabetic peripheral neuropathy using basic clinical and laboratory parameters

Abstract

Objective

The objective of this study is to construct a predictive model for the onset of painful diabetic peripheral neuropathy (PDPN) in patients with diabetic peripheral neuropathy (DPN).

Methods

The clinical and laboratory data of 783 patients with DPN were retrospectively analyzed to form the modeling group. A Douleur Neuropathique 4 score of ≥4 was used to identify neuropathic pain (NP), and such patients were categorized into the PDPN group. Potential predictive variables were screened using least absolute shrinkage and selection operator (LASSO) regression. Logistic regression was subsequently used to construct a predictive model for PDPN. The discriminatory power and calibration of the model were assessed using receiver operator characteristic (ROC) curve analysis and calibration plots. For validation, the model was tested on an independent group of 350 patients with DPN.

Results

The final modeling and validation groups comprised of 359 and 162 patients with PDPN, respectively. The inclusion of five clinical variables resulted in an optimal predictive model: hemoglobin A1c (HbA1c) (odds ratio [OR] = 1.173, P < 0.001), triglycerides (TG) (OR = 1.813, P < 0.001), body mass index (BMI) (OR = 1.081, P = 0.002), disease duration (OR = 1.066, P < 0.001), and 24-hour urine microalbumin (UMA) (OR = 1.003, P < 0.001). The areas under the ROC curve for the modeling and validation groups were 0.812 and 0.850, respectively. The calibration plot demonstrated a close fit between the calibration curve and the ideal curve, with Hosmer–Lemeshow P values of 0.4153 for the modeling group and 0.8413 for the validation group.

Conclusion

These findings indicate that our nomogram can effectively predict the occurrence of PDPN in patients with DPN, thereby assisting clinicians in identifying patients at risk.