论文部分内容阅读
目的:探讨人工神经网络(ANN)模型在个体2型糖尿病患病风险预测中的应用。方法:通过横断面调查对河南某农村社区8 640名居民进行流行病学调查,按3∶1的比例随机分为训练集(6 480人)与检验集(2 160人),分别用于筛选变量、建立预测模型及对模型的检测和评价。分别应用ANN和logistic回归建立2型糖尿病预测模型,应用受试者工作特征曲线(ROC)评价预测模型的检验效能。结果:ANN预测模型的灵敏度(95%CI)=86.93(81.41~91.29)%、特异度(95%CI)=79.14(77.18~81.02)%、阳性预测值(95%CI)=31.86(28.60~35.03)%、阴性预测值(95%CI)=98.18(97.37~98.81)%优于logistic回归预测模型[灵敏度(95%CI)=62.81(55.73~69.47)%、特异度(95%CI)=71.70(69.52~73.79)%、阳性预测值(95%CI)=19.94(17.00~22.99)%、阴性预测值(95%CI)=94.50(93.32~95.57)%];ANN预测模型AUC(95%CI)=0.891(0.877~0.905)明显大于logistic回归预测模型[AUC(95%CI)=0.742(0.722~0.763)]。结论:在预测个体患2型糖尿病方面,ANN模型较logistic回归模型具有更好的预测效能。
Objective: To explore the application of artificial neural network (ANN) model in predicting the risk of type 2 diabetes mellitus in individuals. Methods: The epidemiological survey of 8 640 residents in a rural community of Henan province was conducted by cross-sectional survey, and randomly divided into training set (6 480) and test set (2 160) according to the ratio of 3: 1, Variables, build predictive models and test and evaluate models. The prediction models of type 2 diabetes were established by using ANN and logistic regression respectively. The test performance of the prediction model was evaluated by receiver operating characteristic curve (ROC). Results: The sensitivity (95% CI) of ANN prediction model was 86.93 (81.41-91.29)%, the specificity (95% CI) = 79.14 (77.18-81.02)%, the positive predictive value (95% CI) = 31.86 (95% CI) = 98.18 (97.37 ~ 98.81)% better than the logistic regression model [sensitivity (95% CI) = 62.81 (55.73-69.47)%, specificity (95% CI) = 19.94 (17.00-22.99)%, negative predictive value (95% CI) = 94.50 (93.32-95.57)%; ANN predictive model AUC (95% CI) = 0.891 (0.877 ~ 0.905) was significantly higher than that of logistic regression model [AUC (95% CI) = 0.742 (0.722-0.763)]. CONCLUSIONS: The ANN model has better predictive power than the logistic regression model in predicting individuals with type 2 diabetes.