应用人工神经网络预测个体血脂异常患病危险度

目的建立个体血脂异常患病危险度的预测模型,探讨并评价预测个体血脂异常的新方法。方法选择8914例社区居民流行病学调查资料,按3∶1分为训练集(6686例)与检验集(2228例),分别用于筛选变量、建立预测模型及对模型的检测和评价。应用人工神经网络(ANN)和logistic回归分别建立血脂异常预测模型,受试者工作曲线(ROC)评价预测模型的优劣。结果 ANN预测模型的特异度(64.79%)较低,但灵敏度(94.86%)、约登指数(59.65%)、一致率(81.23%)均优于logistic回归预测模型(特异度=77.49%、灵敏度=53.51%、约登指数=31.00%、一致率=81.23%);ANN预测模型ROC曲线下面积(Az=0.824±0.009)明显大于logistic回归预测模型(Az=0.655±0.012)(P<0.05)。结论在预测个体血脂异常方面,ANN模型较logistic回归模型具有更好的预测判别效能。

Application of artificial neural network model to predict the health risk of dyslipidemia in individuals

Objective To establish models to predict the risk of dyslipidemia in individuals,and then to explore and evaluate new prediction models.Methods The epidemiological survey data of 8914 community residents was selected and divided into a trained group(6686 cases) and a test group(2228 cases).Artificial neural network(ANN) and Logistic regression analysis were used to establish prediction models,and then the results were evaluated by receiver operating characteristic(ROC) curve.Results The specificity(64.79%) of AAN model forecasting the results of test group was lower,but the sensitivity(94.86%),Youden’s index(59.65%) and consistency rate(81.23%) of AAN model was higher than the Logistic regression predicted model(specificity = 77.49%,sensitivity = 53.51%,Youden’s index = 31.00% and consistency rate = 81.23%,respectively).Moreover,the area under ROC curve of ANN prediction model(Az = 0.824 ± 0.009) was significantly bigger than the Logistic regression prediction model(Az = 0.655 ± 0.012).Conclusion The discrimination performance of ANN model is better than Logistic regression in the prediction of health risk of dyslipidemia in individuals.