人工神经网络在个体患原发性高血压预测中的应用

Application of Artificial Neural Networks to Predict Individual Health Risk of Essential Hypertension

目的在流行病学调查资料的基础上,探讨并评价预测个体患原发性高血压的新方法。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回归分别建立高血压患病预测模型,用受试者工作曲线(ROC)评价预测模型的优劣。结果 ANN预测模型的灵敏度(95.94%)、特异度(85.04%)、约登指数(80.98%)、一致率(88.78%)优于logistic回归预测模型(灵敏度=51.31%、特异度=95.56%、约登指数=46.87%、一致率=80.39%)。通过ROC曲线下面积比较模型的预测能力:ANN预测模型曲线下面积(Az=0.904±0.007)明显大于logistic回归预测模型(Az=0.734±0.012)。结论利用ANN模型进行疾病分类预测,较logistic回归模型能获得更好的预测效果。

Objective：To explore the potential application of artificial neural network（ANN） on epidemiological classification of disease,and then to evaluate the predicted models.Methods：The epidemiological survey data of 8 914 community residents was selected and divided into trained group（6 686 cases） and inspected group（2 228 cases）.The prediction models were established using ANN and logistic regression analysis,and then evaluated by receiver operating characteristic（ROC） curve.Results：The prediction model of ANN has better sensitivity（95.94%）,specificity（85.04%）,Youden＇s index（80.98%） and consistency rate（88.78%） than logistic regression prediction model（sensitivity=51.31%,specificity=95.56%,Youden＇s index=46.87% and consistency rate=80.39%,respectively）.Moreover,the area under ROC curve of ANN prediction model（0.904±0.007） was significant higher than logistic regression prediction model（0.734 ± 0.012）.Conclusion：The ANN model was better than logistic regression in the prediction of individual health risk of essential hypertension.