Establishment and Validation of a Clinical Prediction Model for Diabetic Lower Extremity Deep Venous Thrombosis
WANG Zan,XIE Xiaoyun,DING Anle,GUO Peng,ZHOU Mengdie,LIU Yang,YIN Shiwu
DOI: https://doi.org/10.16833/j.cnki.jbmc.2022.12.004
2022-01-01
Abstract:Objective:To construct a clinical prediction model for the indication of DVT risk in patients with diabetes, and to provide early warning for clinical practice.Methods:A total of 281 inpatients were selected from November 2018 to November 2021 in Shanghai Tongji Hospital Affiliated to Tongji University, among which 111 patients were diagnosed of diabetes mellitus complicated with lower extremity and deep vein thrombosis and set as the experimental group, and the other 170 patients during the same period diagnosed of diabetes without deep vein thrombosis were selected as the control group. The model was evaluated and optimized using the least absolute contraction and selection operator regression model for the feature selection of the diabetic lower extremity deep vein thrombosis risk model. Univariate and multivariate logistic regression analysis combined with the least absolute shrinkage and features selected in the selection operator regression model were applied to build a predictive model. C-index, calibration curve plots, and decision curve analysis were used to assess the discriminability, calibration, and clinical utility of predictive models.Result:Single factor analysis results showed significant differences（P＜0.05）among factors such as age, 2 hours after meal, glycosylated hemoglobin, total cholesterol, high-density lipoprotein cholesterol, fibrinogen, prothrombin time, D2 polymers, hemoglobin, c-reactive protein, total protein, neutrophil count, lymphocyte count, alkaline phosphatase, and aspertate aminotransferase. Multivariate analysis showed that age, two-hour postprandial blood glucose, glycosylated hemoglobin, D2 polymers, C-reactive protein, alkaline phosphatase, lymphocyte count, and alanine aminotransferase were independent factors associated with lower extremity deep vein thrombosis in patients with diabetes mellitus. The model is discriminative with a C-index of 0.928, which is with well calibrated. The subject curve showed good predictive power.Conclusion:The nomogram plotted with age, two-hour postprandial blood glucose, glycosylated hemoglobin, D2-mer, hemoglobin, C-reactive protein, alkaline phosphatase, and aspartate aminotransferase, can be conveniently used for risk prediction of diabetes complicated with deep vein thrombosis.