基于随机森林的男性急性心肌梗死诊断模型建立及验证

目的利用随机森林建立及验证男性急性心肌梗死诊断模型。 方法检测2016年1至6月于武汉大学人民医院心内科住院的205例心绞痛或急性心肌梗死男性患者的血清生化及生物标志物指标，其中151例患者作为训练集，54例患者作为验证集。用随机森林对指标预测急性心肌梗死的重要性进行排序。根据袋外数据误差，赤池信息量准则和贝叶斯信息量准则对排序指标进行筛选并构建诊断模型；多维标度法(MDS)观察模型对急性心肌梗死和心绞痛的区分能力；用验证集数据验证模型对心绞痛和急性心肌梗死的鉴别能力。 结果19个指标根据平均准确度下降程度和平均基尼(Gini)指数下降程度进行重要性排序。用袋外数据误差，赤池信息量准则和贝叶斯信息量准则筛选出C-反应蛋白、中性粒细胞绝对值和血糖3个变量，并纳入模型。通过MDS法观察到多半样本得到良好的区分，但部分样本仍难以区分开。在外部验证中，31例急性心肌梗死患者有26例(83.87%)被正确识别；在23例心绞痛患者中有19例(82.61%)被正确识别。 结论基于随机森林的诊断模型建立能较好区分急性心肌梗死与心绞痛。

Establishment and validation of diagnosis model for acute myocardial infarction based on random forest classification in men

ObjectiveTo establish and validate the model of forecasting acute myocardial infarction in men. MethodsFrom January to June 2016, 205 male patients admitted to the department of cardiology of the People′s Hospital of Wuhan University with angina pectoris or acute myocardial infarction were included in our study. Among them, 151 patients served as training set and 54 patients served as validation set. Random forest was used to rank the importance of predicting acute myocardial infarction. According to the OOB error, AIC and BIC criterion, the sorting indexs were screened and the prediction model was constructed. Multidimensional scaling (MDS) was used to observe the ability of the model to differentiate acute myocardial infarction from angina pectoris, and validation set data was used to investigate whether the random forest could distinguish between acute myocardial infarction and angina pectoris. ResultsThe 19 indicators were ranked according to mean Decrease Accuracy and mean Decrease Gini index. C-reactive protein, neutrophil absolute value and blood sugar inclusion model were screened by OOB error, AIC criterion and BIC criterion. In external validation, 26 of 31(83.87%) patients with acute myocardial infarction were correctly identified, and 19 of 23(82.61%) patients with angina pectoris were correctly identified. ConclusionRandom forest-based predictive model can well distinguish between acute myocardial infarction and angina pectoris.