基于随机森林算法构建精神分裂症患者复发的预测模型  

Random forest algorithm-based prediction model for recurrence of schizophrenia patients

作　　者：魏芳 杨小燕 熊佳 熊芳 魏欢 WEI Fang;YANG Xiaoyan;XIONG Jia;XIONG Fang;WEI Huan(Jiangxi Provincial Veterans Hospital,Jiangxi 330103,China)

机构地区：[1]江西省荣军优抚医院,江西省南昌市330103

出　　处：《护理实践与研究》2024年第8期1121-1127,共7页Nursing Practice and Research

摘　　要：目的探讨精神分裂症患者复发的影响因素,基于随机森林算法构建复发的预测模型。方法回顾性选取2020年3月—2022年6月在医院治疗的精神分裂症患者192例为研究对象,根据1年内患者有无复发精神分裂症分为复发组(n=58)和无复发组(n=134),通过单因素分析得出对应的影响因素,应用R语言软件,基于随机森林算法构建精神分裂症患者1年内复发的预测模型,并采用受试者工作特征(Receiver Operating Characteristic,ROC)曲线评价随机森林算法构建的模型对复发的预测价值。结果单因素分析显示,饮酒史、病程、就业情况、社会支持评定量表(Social Support Rating Scale,SSRS)评分、7项广泛性焦虑障碍量表(7-tiem Generalized Anxiety Disorder Scale,GAD-7)评分、匹兹堡睡眠质量指数量表(Pittsburgh Sleep Quality Index,PSQI)评分、服药情况、精神病家族史均与精神分裂症复发有关,差异有统计学意义(P<0.05)。采用随机森林算法构建的预测模型显示,前6位影响因素的重要性排序为GAD-7评分(GiNi=16.64)>病程(GiNi=15.09)>SSRS评分(GiNi=14.96)>PSQI评分(GiNi=14.00)>就业情况(GiNi=3.94)>饮酒史(GiNi=3.72)。ROC结果显示,随机森林模型的ROC曲线其曲线下面积(area under curve,AUC)为0.840,最佳截断值0.255对应的敏感度为82.81%,特异度为73.18%。结论精神分裂症患者复发受GAD-7评分、病程、SSRS评分、就业情况、PSQI评分、饮酒史因素影响,基于随机森林构建的精神分裂症患者复发风险预测模型具有较好的预测效能,可进一步推广应用以验证预测模型的效能。Objective To explore the influencing factors of recurrence in schizophrenia patients and build a prediction model of recurrence based on random forest algorithm.Methods 192 patients with schizophrenia admitted in hospital from March 2020 to June 2022 were retrospectively selected as the study subjects.According to whether the patients had relapsed schizophrenia within 1 year,they were divided into a relapse group(n=58)and a relapse-free group(n=134).The corresponding influencing factors were obtained through univariate factor analysis.The R language software was used to construct a predictive model for the relapse of schizophrenia within 1 year based on the random forest algorithm,and the receiver operating characteristic(ROC)curve was used to evaluate the predictive value of the model constructed by the random forest algorithm for the relapse.Results Univariate factor analysis showed that history of drinking,duration of illness,employment status,Social Support Rating Scale(SSRS)score,7-item Generalized Anxiety Disorder Scale(GAD-7)score,Pittsburgh Sleep Quality Index(PSQI)score,medication use,and family history of psychosis were associated with relapse of schizophrenia.Index(Pittsburgh Sleep Quality Index,PSQI)score,medication taking,and family history of psychosis were all associated with schizophrenia relapse,with statistically significant differences(P<0.05).The prediction model constructed using the random forest algorithm showed that the order of importance of the top 6 influencing factors was GAD-7 score(GiNi=16.64)>duration of illness(GiNi=15.09)>SSRS score(GiNi=14.96)>PSQI score(GiNi=14.00)>employment status(GiNi=3.94)>history of alcohol consumption(GiNi=3.72).The ROC results showed that the ROC curve of the random forest model had an Area Under Curve(AUC)of 0.840,and the optimal cut offvalue of 0.255 corresponded to a sensitivity of 82.81%and a specificity of 73.18%.Conclusion Relapse in schizophrenia patients is affected by GAD-7 score,disease duration,SSRS score,employment status,PSQI score,and drinki

关 键 词：精神分裂症 复发 预测模型 随机森林 服药情况 社会支持 睡眠质量 焦虑状态 

分 类 号：R473.5[医药卫生—护理学]