C5.0决策树模型在严重精神障碍患者服药依从性预测中的探讨

目的 研究C5.0 决策树模型在严重精神障碍患者服药依从性影响因素中的应用，并探讨其预测效果。方法 抽样采取多阶段分层随机抽样方法，患者基本信息及相关因素使用自编调查问卷采集，运用C5.0决策树模型对患者是否依存服药进行预测，模型性能采用ROC曲线下面积、预测正确率、约登指数、灵敏度和特异度进行评价。结果 C5.0决策树模型和logistic回归模型预测性能比较显示，模型正确预测率分别为80.9%和71.4%，C5.0决策树模型ROC曲线下面积均数±标准误为0.808±0.011（95%CI:0.790~0.826），logistic回归模型为0.713±0.012（95%CI:0.692~0.733），两模型比较差异有统计学意义（Z=8.766，P<0.001），C5.0决策树模型约登指数、灵敏度和特异度均高于logistic回归模型，C5.0决策树模型显示，病种、治疗费用支出、收入水平、监护人是否与患者共同生活对严重精神障碍患者服药依从性影响较大。结论 C5.0决策树模型在严重精神障碍患者服药依从性预测中性能良好，经济水平和监护人是否与患者共同生活对患者服药依从性起重要作用。

The influence factors of drug compliance among serious mental patients based on the model of C5.0 decision tree

To study the influence factors of drug compliance among serious mental patients based on the model of C5.0 decision tree, and toexploreit’s prediction effect. Methods To apply stratified random sampling method. The basic information and the related factors were collected by self-compiling questionnaires. The model of C5.0 decision tree was used to predict compliance of taking medicine. The area under ROC curve (AUC), correct forecast rate, Youden index, sensitive and specificity were used to assessment the model performance. Results The predictive accuracy of C5.0 decision tree model and logistic regression model were 80.9% and 71.4% respectively, the AUC mean±standard error of the mean(SEM) of two models were 0.808±0.011(95%CI:0.790-0.826) and 0.713±0.012(95%CI:0.692-0.733), and it has statistical significance. The Youden index, sensitive and specificity of C5.0 decision tree model were all higher than logistic regression model. The result showed that disease category, treatment cost, income level, etal has great influence on drug compliance among serious mental patients from C5.0 decision tree model. Conclusion C5.0 decision tree model has a good performance to predict the drug compliance among serious mental patients,the economic level and whether guardian living together with patients play important roles in drug compliance.