Development and validation of a simplified pre-screening model for diabetic foot ulcer identification in diabetic...

This study aimed to develop a simplified cross-sectional pre-screening tool for diabetic foot ulcer (DFU) using routinely available clinical indicators, with particular focus on the albumin-to-glycated hemoglobin ratio (Alb/HbA1c) as a composite marker of nutritional status and glycemic control.

We retrospectively analyzed 1,854 hospitalized patients with type 2 diabetes (training set, January 2012-December 2020) and 678 outpatients (validation set, January 2021-December 2025) from a geographically distinct branch of the same hospital system. Demographic characteristics, lifestyle factors, and laboratory parameters were collected. LASSO regression and multivariate logistic regression were used for predictor selection. Model performance was assessed by area under the curve (AUC), calibration curves, Hosmer-Lemeshow test, and Brier score. Risk was visualized using nomograms and heatmaps.

DFU prevalence was 20.9% in the training set and 15.8% in the validation set. Four independent predictors were identified: age (OR = 1.029/year), history of injury (OR = 7.57), alcohol consumption (OR = 0.48, see Discussion for interpretation), and Alb/HbA1c ratio (OR = 0.49). The model showed good discrimination in the training (AUC 0.807, 95% CI 0.790-0.825) and validation sets (AUC 0.817, 95% CI 0.782-0.852), with acceptable calibration (Hosmer-Lemeshow P > 0.05; Brier scores 0.127 and 0.118). Internal validation confirmed stability (optimism-corrected AUC = 0.803). Risk heatmaps revealed synergistic interactions between age and Alb/HbA1c, with injury history amplifying risk across all strata.

This simplified pre-screening model using age, injury history, alcohol consumption, and Alb/HbA1c demonstrates a high negative predictive value, making it suitable for ruling out DFU in resource-limited primary care settings.