糖尿病心血管并发症预测模型的构建

Establishment of predictive model for cardiovascular complication of diabetes mellitus

目的 筛选糖尿病心血管并发症的危险因素并构建预测模型,为早期预防和减缓进展提供依据。方法∶基于京津冀社区自然人群慢性病队列,选取基线(2017—2019年)糖尿病患者作为研究对象,根据自报患病时间先后顺序判定是否发生心血管并发症作为结局指标。按7Lasso结合Logistic回归模型方法筛选危险因素,将其纳入多因素Logistic回归模型构建糖尿病心血管并发症发生风险的预测模型,并绘制列线图进行可视化。绘制受试者工作特征曲线并计算曲线下面积,Hosmer-Lemeshow拟合优度检验和绘制校准曲线评估校准度,对预测模型进行评价和验证。结果 共纳入813例2型糖尿病患者,平均年龄为(62.6±10.4)岁,其中训练集569例,测试集244例,两组患者除超敏C反应蛋白水平具有差异(P=0.028)外,其他基本特征差异均无统计学意义。多因素Logistic回归分析显示,糖尿病心血管并发症的危险因素包括年龄[OR=1.040,95%CI(1.010,1.073),P=0.010]、高血压[OR=2.211,95%CI(1.263,3.975),P=0.006]、糖尿病病程[OR=1.063,95%CI(1.028,1.099),P <0.001]、空腹血糖水平[OR=1.186,95%CI(1.075,1.309),P=0.001]、血脂异常[OR=2.051,95%CI(1.167,3.583),P=0.012]、心血管病家族史[OR=2.794,95%CI(1.650,4.774),P <0.001]和吸烟[OR=1.975,95%CI(1.133,3.462),P=0.017];而保护因素为血清胆红素[OR=0.940,95%CI(0.889,0.991),P=0.027]。列线图显示2型糖尿病患者可根据模型中8个预测因素的动态变化计算出发生糖尿病心血管并发症的概率。训练集和测试集ROC曲线下面积分别为0.803和0.820,HosmerLemeshow检验P值分别为0.776和0.554,校准曲线与理想曲线的平均绝对误差为0.013,表明预测模型区分度和校准度均较好。结论 本研究基于社区自然人群构建的糖尿病心血管并发症风险预测模型效果较好,为糖尿病患者心血管并发症的早期预测预警提供了便利可行的工具。

Objective To screen the risk factors of cardiovascular complications of diabetes mellitus,and build a predictive model to provide a basis for early prevention and slow down the progression of the disease.Methods Based on the chronic disease cohort of the natural population in the Beijing-Tianjin-Hebei community,the baseline diabetic patients from 2017 to 2019 were selected as the study objects,and the occurrence of cardiovascular complication was determined as the outcome index based on the chronological order of the self-reported illness.The model was randomly divided into a training set and a test set at 7∶3 to build and verify the model,respectively.Lasso-Logistic regression model was used to screen risk factors,which were incorporated into the multi-factor Logistic regression model to build a prediction model for the risk of cardiovascular complication of diabetes,and a nomogram was drawn for visualization.The receiver operating characteristic curve(ROC)was drawn and the area under the curve(AUC)was calculated.The Hosmer-Lemeshow tests and plotting calibration curve to assess the calibration degree,and the prediction model was evaluated and verified.Results A total of 813 patients with type 2 diabetes were included,with an average age of(62.6±10.4)years,including 569 patients in the training set and 244 patients in the test set.There were differences in the level of hypersensitive C-reactive protein between the two groups(P=0.028),and there were no statistically significant differences in other basic characteristics.Multivariate Logistic regression analysis showed that the risk factors for cardiovascular complications of diabetes mellitus included age[OR=1.040,95%CI(1.010,1.073),P=0.010],hypertension[OR=2.211,95%CI(1.263,3.975),P=0.006],duration of diabetes[OR=1.063,95%CI(1.028,1.099),P<0.001],level of fasting blood glucose[OR=1.186,95%CI(1.075,1.309),P=0.001],dyslipidemia[OR=2.051,95%CI(1.167,3.583),P=0.012],family history of cardiovascular disease[OR=2.794,95%CI(1.650,4.774),P<0.001]and smoking[OR=1.975,95%CI(1.133,3.462),P=0.017];the protective factor was level of serum bilirubin[OR=0.940,95%CI(0.889,0.991),P=0.027].The nomogram shows that patients with type 2 diabetes can calculate the probability of developing cardiovascular complications of diabetes based on the dynamic changes in the eight predictors in the model.The AUC of the training set and the test set were 0.803 and 0.820,the P-values of the Hosmer-Lemeshow test were 0.776 and 0.554,respectively,and the average absolute error between the calibration curve and the ideal curve was 0.013,which proved that the differentiation and calibration degree of the prediction model were good.Conclusions In this study,the risk prediction model of diabetic cardiovascular complication built on the basis of community natural population has good effect,and provides a convenient and feasible tool for the early prediction and early warning of cardiovascular complication in diabetic.