Abstract: 【Abstract】 Objective To analyze risk factors for diabetes mellitus in patients with vitiligo, and to construct and validate a prediction model. Methods A total of 110 vitiligo patients with diabetes mellitus （comorbidity group） and 4 505 vitiligo patients without diabetes mellitus （control group） were collected from the medical record database in Xijing Hospital, the Fourth Military Medical University from January 2010 to October 2021, and matched for gender and age at a ratio of 1∶4 by using a propensity score method. After matching, the matched pairs were randomly divided into a training set and a test set at a ratio of 4∶1. Univariate and multivariate logistic regression analyses were used to assess demographic and clinical characteristics of patients in the training set, screen differential factors, and construct a prediction model. A five-fold cross-validation method was used for internal validation after construction of the prediction model. The discrimination （area under the curve ［AUC］）, calibration （Hosmer-Lemeshow test） and accuracy （sensitivity, specificity, positive predictive value, and negative predictive value） of the prediction model were evaluated in the test set. Results A total of 107 cases in the comorbidity group and 428 cases in the control group were successfully matched. The training set included 430 cases, and the test set included 105 cases. Based on multivariate logistic regression results, a total of 6 factors were included in the prediction model, including course of vitiligo （odds ratio ［OR］ = 1.04, 95% confidence interval ［CI］: 1.02 - 1.07, P＜0.001），high-sugar/high-fat/high-salt diet （OR = 3.19, 95% CI: 1.38 - 7.38, P = 0.007）, family history of diabetes （OR = 23.23, 95% CI: 9.72 - 55.50, P＜0.001）, metabolic comorbidities （OR = 12.53, 95% CI: 5.60 - 28.07, P＜0.001）, autoimmune comorbidities （OR = 5.89, 95% CI: 2.52 - 13.76, P＜0.001）, and acral vitiligo （OR = 3.84, 95% CI: 1.45 - 10.19, P = 0.007）. Five-fold cross-validation results showed a good predictive performance of the prediction model, with the AUC being 0.902 （95% CI: 0.864 - 0.940）in the training set and 0.895 （95% CI: 0.815 - 0.974）in the test set. The prediction model also showed favourable discrimination（AUC =0.814, 95% CI：0.715 - 0.913）, calibration （Hosmer-Lemeshow test, P = 0.068）, and accuracy （sensitivity = 0.810, 95% CI: 0.574 - 0.937; specificity = 0.786, 95% CI: 0.680 - 0.865; positive predictive value = 0.486, 95% CI: 0.317 - 0.657; negative predictive value = 0.943, 95% CI: 0.853 - 0.982） in the test set. Conclusion A risk prediction model was constructed for diabetes mellitus in patients with vitiligo based on 6 factors （course of vitiligo, high-sugar/high-fat/high-salt diet, family history of diabetes, metabolic comorbidities, autoimmune comorbidities, and acral vitiligo）, which showed favourable discrimination, calibration and accuracy, and might provide a reference for screening the high-risk diabetic population in vitiligo patients.

Key words: Vitiligo, Diabetes mellitus, Comorbidity, Risk factors, Prediction model