糖尿病发病风险预测模型的构建与验证

Construction and validation of a prediction model for risks in diabetes

目的利用金昌队列人群搭建数据库,构建糖尿病发病风险预测模型并进行验证。方法选取基线和2014—2019年3期随访匹配人群中的31463例为研究对象,按照7∶3的比例随机分为训练集和验证集,其中训练集22025例,验证集9438例。训练集中的数据通过单因素和多因素Cox比例风险模型筛选预测因子并构建列线图预测模型,使用受试者工作特征曲线下面积评价模型区分度,通过绘制校准曲线评价模型准确度,绘制决策曲线评价模型临床应用价值。同时对预测模型的区分度、准确度和临床应用价值进行内部验证。结果以性别、年龄、体重指数、饮酒、戒酒、高血压、甘油三酯、高密度脂蛋白胆固醇、谷氨酰转移酶、糖尿病家族史、胆囊炎、胆囊缺如为预测因子构建糖尿病发病风险预测模型。训练集和验证集中模型3、5、7年的受试者工作特征曲线下面积分别为0.783、0.825、0.842和0.782、0.805、0.807,表明模型区分度较好。校准曲线均接近对角线,表明模型的准确度较高。决策曲线显示净获益水平较高,表明模型临床实用性较好。结论本研究所构建的列线图预测模型具有良好的预测能力和临床实用性,为筛查未确诊糖尿病患者或高危人群提供了一种方便且实用的方法。

Objective To use the Jinchang cohort population to build a database for constructing a risk prediction model for diabetes and to validate it.Methods A total of 31463 patients from baseline and a three-phase follow-up from 2014 to 2019 were selected as the study subjects.According to the ratio of 7∶3,they were randomly divided into a training set and validation set,with 22025 in the former and 9438 were in the latter.The data in the training set screened predictors through univariate and multivariate Cox proportional hazards models,and a nomogram prediction model based on the Cox model was established.The area under the receiver operating characteristic curve was used to evaluate the model discrimination.The accuracy of the model was evaluated by plotting a calibration curve.The clinical application of the model was evaluated by decision curve analysis.At the same time,the differentiation,accuracy and clinical application value of the prediction model were internally verified.Results Gender,age,body mass index,alcohol consumption,hypertension,triglyceride,high density lipoprotein cholesterol,gamma-glutamyl transpeptidase,family history of diabetes,cholecystitis and gallbladder removal were used as predictors to construct the risk prediction model for diabetes.The area under the receiver operating characteristic curve of the model in the training set and the validation set were 0.783,0.825,0.842,and 0.782,0.805,0.807 in the 3-year,5-year and 7-year models,respectively.The results showed that the model had a good discrimination.The calibration curves were all close to the diagonal,indicating that the accuracy of the model was high.In the decision curve analysis,the model curve indicated a higher level of net benefit and the predictive model being with better clinical utility.Conclusion The nomogram prediction model constructed in this study has good predictive ability and clinical practicability.It can provide a convenient and cost-effective method for screening patients with undiagnosed diabetes or high-risk groups in China.