Objective To explore the prediction efficiency of multilayer perception (MLP) model in prediction of diabetes mellitus (DM) complicated with coronary heart disease (CHD) by TCM personality and constitutions; To provide a new method for objective prediction. Methods This research utilized single factor logistic regression to filter out variables, which were significant factors of TCM personality and constitutions as analytic variables for MLP and multivariate logistic regression to establish TCM prediction model of personality and constitutions for DM complicated with CHD. The prediction efficiency of the above models were tested by receiver operating characteristic curve (ROC curve). Results The sensitivity, specificity and AUC of MLP were 0.915 (0.862, 0.968), 0.846 (0.793, 0.912) and 0.913 (0.806, 0.987) respectively, which was better than the logistic regression, while these indexes of logistic regression were 0.834 (0.695, 0.953), 0.762 (0.623, 0.901), and 0.869 (0.730, 0.941) respectively. Conclusion The MLP model is better than logistic regression model in prediction of DM complicated with CHD by TCM personality and constitutions.%目的 探讨多层感知器(MLP)模型在中医人格、体质预测糖尿病性冠心病患病风险中的预测效能,为该病客观化预测提供新方法.方法 采用单因素Logistic回归进行变量筛选,筛选有统计学意义的中医人格、体质类型指标,作为MLP及多因素Logistic回归的分析变量,以构建中医人格、体质对糖尿病性冠心病患病风险的预测模型.采用受试者工作特征曲线(ROC曲线)比较2种模型的预测效能.结果 MLP预测模型的灵敏度OR95％CI=0.915(0.862,0.968)、特异度OR95％CI=0.846(0.793,0.912)、AUC OR95％CI=0.913(0.806,0.987),均优于Logistic回归预测模型[灵敏度OR95％CI=0.834(0.695,0.953)、特异度OR95％CI=0.762(0.623,0.901)、AUC OR95％CI=0.869(0.730,0.941)].结论 在中医人格、体质对糖尿病性冠心病患病风险的客观化预测中,MLP模型较Logistic回归模型具有更好的预测效能.
展开▼
