基于随机森林算法的海洛因成瘾者美沙酮维持治疗剂量分类  

Classification of Methadone Maintenance Treatment Dose for Heroin Addicts Based on Random Forest Algorithm

作　　者：李鹏 曾庆[1] LI Peng;ZENG Qing(Department of Health Statistics,School of Public Health and Management,Chongqing Medical University,Chongqing,400016 China)

机构地区：[1]重庆医科大学公共卫生与管理学院卫生统计教研室,重庆400016

出　　处：《系统医学》2020年第12期111-114,共4页Systems Medicine

摘　　要：目的美沙酮具有治疗海洛因依赖脱毒和替代维持治疗的药效作用,但其用药剂量直接影响其疗效和是否产生不良反应,目前国内临床上MMT剂量方案尚无规范化的标准。该文选取贵州省贵阳市某个美沙酮维持治疗试点的2015年1月—2017年3月的110例美沙酮维持治疗患者,建立新入组患者的服药剂量分类模型,为患者的服药剂量提供参考。方法采用SOM-K组合聚类模型对美沙酮患者的剂量进行聚类,将聚类结果作为分类数据的类标号建立基于随机森林算法的分类模型。结果聚类结果为2类,低剂量组和高剂量组,低剂量组平均剂量均值(48.7±10.5)mg/d,高剂量组平均剂量均值(74.5±11.2)mg/d,经wilcoxon符号秩和检验两组比较,差异有统计学意义(W=51.5,P<0.05)。基于随机森林算法的分类模型准确率0.894,Kappa系数0.789,AUC值0.990。结论随机森林算法在海洛因成瘾者美沙酮维持治疗剂量分类问题中表现优异,较之其他基准分类器分类精度更高,分类效果更好。Objective Methadone has the effect of treating heroin dependence detoxification and replacement maintenance therapy, but its dosage directly affects its efficacy and whether it produces adverse reactions. At present,there is no standardized standard for MMT dosage regimen in China. This paper selects 110 methadone maintenance treatment patients from January 2015 to March 2017 in a methadone maintenance treatment pilot in Guiyang, Guizhou Province, and establishes a classification model of the dose of newly enrolled patients to provide reference for the patient’s dose. Methods The SOM-K combined clustering model was used to cluster the doses of methadone patients,and the clustering results were used as the class labels of the classification data to establish a classification model based on the random forest algorithm. Results The clustering results were in two categories: low-dose group and highdose group. The average dose of the low-dose group was(48.7±10.5)mg/d, and the average dose of the high-dose group was(74.5 ±11.2)mg/d. By Wilcoxon symbolic rank, compared with the two test groups, the difference is statistically significant(W=51.5,P<0.05). The classification model based on random forest algorithm has an accuracy rate of 0.894,a Kappa coefficient of 0.789, and an AUC value of 0.990. Conclusion The random forest algorithm performs well in the methadone maintenance treatment dose classification problem for heroin addicts. Compared with other benchmark classifiers, the classification accuracy is higher and the classification effect is better.

关 键 词：美沙酮 剂量 随机森林 聚类 分类 

分 类 号：R749.64[医药卫生—神经病学与精神病学]