Development and validation of TCM prediction model of diabetic peripheral neuropathy among type 2 diabetes mellitus population in Tianjin, China

Abstract

Abstract Aims: To determine the clinical predictors of symptoms of TCM and tongue features in type 2 diabetes mellitus (T2DM) with diabetic peripheral neuropathy (DPN), in further to verify whether these parameters of TCM can be used to develop a clinical model for predicting onset of DPN among T2DM. Methods: We collect information from a standardized questionnaire. The questionnaire survey was performed on 3590 T2DM, participants were randomly divided the training group (n = 3297) and the validation group (n = 1246). Symptoms of TCM and tongue features had used to evaluate the risk to develop DPN in T2DM. The least absolute shrinkage and selection operator (LASSO) regression analysis method and logistic regression analysis had used to optimize variable selection by running 5-fold cross-validation in the training group. Multi-factor logistic regression analysis was used to establish a predictive model. The nomogram had been developed based on the relevant risk factors. A receiver operating characteristic curve (ROC), calibration plot and decision curve analysis (DCA) were used to assess the model in training group and validation group. Results: A total of eight predictors were found to be independently associated with the DNP in multivariate logistic regression analyses, namely such as advanced age of grading (OR 1.575, 95% CI 1.236–2.006, p = 0.000), smoke (OR 2.815, 95% CI 2.079–3.811, p = 0.000), insomnia (OR 0.557, 95% CI 0.408–0.761, p = 0.000), sweating (OR 0.535, 95% CI 0.362–0.791, p = 0.002), loose teeth (OR1.713, 95% CI 1.258–2.331, p = 0.001), dry skin (OR1.831, 95% CI 1.303–2.574, p = 0.000), purple tongue (OR 2.278, 95% CI 1.514–3.428, p = 0.000) and dark red tongue (OR 0.139, 95% CI 0.044–0.441, p = 0.001). The model constructed with using these eight predictors exhibited medium discriminative capabilities, with an area under the ROC of 0.727 in the training group and 0.744 in the validation group. The calibration plot is shown that the model possesses satisfactory in goodness-of-fit. Conclusions: Introducing age of grading, purple tongue and symptoms of TCM into the risk model increased its usefulness for predicting DPN risk in patients with T2DM.