A risk‐score model for predicting risk of type 2 diabetes mellitus in a rural Chinese adult population: A cohort study with a 6‐year follow‐up

Background Several prediction tools have been developed to identify people with type 2 diabetes mellitus (T2DM) and to quantify the probability of developing T2DM. However, most of the risk models were constructed based on cross-sectional studies and tea-drinking was not included. Methods A total of 15 768 participants without known T2DM were followed up from 2007-2008 to 2013-2014; 12 654 were randomly assigned to the derivation dataset and 3114 to the validation dataset. We constructed a risk-score model for T2DM by using a Cox proportional-hazards model. Risk scores were calculated by multiplying β by 10 in the derivation cohort and were verified in the validation dataset. The model's accuracy was assessed by the area under the receiver operating characteristic curve (AUC). Results Predictors for T2DM risk in the derivation dataset were drinking tea frequently, body mass index ≥28.0 kg/m2, waist to height ratio ≥ 0.5, triglycerides level 1.70 to 2.25 and ≥2.26 mmol/L, and fasting plasma glucose 5.6 to 6.0 and ≥6.1 mmol/L. The corresponding scores were −2, 7, 7, 4, 6, 11, and 25, respectively. The sensitivity, specificity, and AUC (95% confidence interval) for this full model were 69.63%, 75.56%, and 0.791 (0.783-0.799), respectively. The ability of the non-invasive models to predict T2DM was not superior to that of the full model. With the validation dataset, the predictive performance was better for our full model than the Framingham risk-score model (AUC 0.731 vs 0.525, P < .001). Conclusions Our risk-score model has fair efficacy for predicting 6-year risk of T2DM in a rural adult Chinese population.