基于Logistic回归、决策树、神经网络构建住院老年患者肌少症相对风险预测模型-现代医学

基于Logistic回归、决策树、神经网络构建住院老年患者肌少症相对风险预测模型

单位：新疆医科大学第一附属医院老年病科, 新疆乌鲁木齐 830054

分类号：R685

出版年·卷·期（页码）：2023·51·第八期（1134-1143）

摘要：

目的：分析住院老年患者肌少症的影响因素，构建肌少症相对风险预测模型，并对结果进行比较分析。方法：采用便利抽样法，连续入选2020年7月至2021年9月在新疆医科大学第一附属医院住院的老年患者372例，应用多因素Logistic回归、决策树CHAID算法、神经网络分析肌少症的危险因素，构建Logistic回归预测模型、决策树预测模型、神经网络预测模型，采用受试者工作特征(ROC)曲线下面积(AUC)比较3种预测模型的效果。结果：住院老年患者肌少症检出率为18.82%，其中男性检出率为23.84%，女性为14.50%。3种模型的结果均显示体质指数、步速、性别为住院老年患者发生肌少症的主要影响因素。Logistic回归预测模型的AUC为0.882(95%CI 0.836～0.928)，风险预测正确率为87.8%；决策树预测模型的AUC为0.874(95%CI 0.832～0.916)，风险预测正确率为86.7%；神经网络预测模型的AUC为0.890(95%CI 0.848～0.931)，风险预测正确率为85.8%；3种模型的预测价值均>0.7，预测效果较好。结论：多种模型可从不同的层面挖掘肌少症的危险因素，多模型有效结合能更充分地了解不同因素之间的相互作用，构建预测模型具有较好的预测作用，可为早期筛查和干预提供参考。

Objective: The influencing factors of sarcopenia in hospitalized elderly patients were analyzed, the relative risk prediction model of sarcopenia was constructed, and the results were compared. Methods: A total of 372 elderly patients hospitalized in the First Affiliated Hospital of Xinjiang Medical University from July 2020 to September 2021 were selected by convenience sampling method. Multivariate Logistic regression, decision tree CHAID algorithm and neural network were used to analyze the risk factors of sarcopenia. Logistic regression prediction model, decision tree prediction model and neural network prediction model were constructed,the area under receiver operating characteristic(ROC) curve(AUC) was used to compare the effect of the three prediction models. Results: The incidence of sarcopenia in hospitalized elderly patients was 18.82%, 23.84% in males and 14.50% in females. The results of the three models showed that body mass index, walking speed, age and gender were the main influencing factors of sarcopenia for hospitalized elderly patients. The AUC of Logistic regression prediction model was 0.882(95%CI 0.836-0.928), and the accuracy of risk prediction was 87.8%. The AUC of decision tree prediction model was 0.874(95%CI 0.832-0.916), and the accuracy of risk prediction was 86.7%. The AUC of neural network prediction model was 0.890(95%CI 0.848-0.931), and the accuracy of risk prediction was 85.8%. The prediction value of the three models is all >0.7, and the prediction effect is good. Conclusion: Multiple models can excavate the risk factors of sarcopenia from different levels, and the effective combination of multiple models can more fully understand the interaction among different factors. The models have good prediction effects, and can provide references for early screening and intervention.

参考文献：

[1] MARQUES J,SHOKRY E,UHL O,et al.Sarcopenia:Investigation of metabolic changes and its associated mechanisms[J].Skelet Muscle,2023,13(1):2.
[2] 杨明,游利.肌少症发病机制[J].中华骨质疏松和骨矿盐疾病杂志,2018,11(4):408-414.
[3] PAPADOPOULOU S K.Sarcopenia:a contemporary health problem among older adult populations[J].Nutrients,2020,12(5):1293.
[4] KAWADA T.Mortality risk of sarcopenia in older subjects[J].J Am Med Dir Assoc,2021,22(9):1883.
[5] MELLEN R H,GIROTTO O S,MARQUES E B,et al.Insights into pathogenesis,nutritional and drug approach in sarcopenia:a systematic review[J].Biomedicines,2023,11(1):136.
[6] HAN P,KANG L,GUO Q,et al.Prevalence and factors associated with sarcopenia in suburb-dwelling older chinese using the asian working group for sarcopenia definition[J].J Gerontol A Biol Sci Med Sci,2016,71(4):529-535.
[7] 吴琳瑾,李静欣.中国社区老年人肌少症患病率的meta分析[J].现代预防医学,2019,46(22):4109-4112,4140.
[8] 武力,郑燕蓉,柴源.老年住院患者衰弱程度与肌肉减少症严重程度的相关性分析[J].现代医学,2020,48(5):647-651.
[9] 王焕如,于翰,邵晋康.肌肉减少症研究进展[J].中国骨质疏松杂志,2022,28(2):304-307.
[10] 肌少症共识[J].中华骨质疏松和骨矿盐疾病杂志,2016,9(3):215-227.
[11] 刘杰,盛逸澜,余波.老年原发性肌少症的诊断和评估新进展[J].实用老年医学,2023,37(1):1-4.
[12] 刘思琪,赵一璟,杨昱,等.GLP-1受体激动剂对2型糖尿病肥胖患者肌少症影响的研究进展[J].东南大学学报(医学版),2021,40(4):552-556.
[13] 刘娟,丁清清,周白瑜,等.中国老年人肌少症诊疗专家共识(2021)[J].中华老年医学杂志,2021,40(8):943-952.
[14] 戚艳艳,郑欣,毕丽娜,等.老年2型糖尿病患者合并肌少症的情况及影响因素分析[J].广西医学,2021,43(13):1553-1557.
[15] 姚思宏.老年住院患者肌少症和骨质疏松症患病率、危险因素及相关性分析[D].吉首:吉首大学,2019.
[16] 耿佳旭,魏雅楠,王晶桐.相位角与住院老年慢病患者肌少症的相关性分析[J].中国骨质疏松杂志,2022,28(4):499-504.
[17] 车雅洁.乌鲁木齐市社区老年人肌少症现状及相关因素研究[D].乌鲁木齐：新疆医科大学,2021.
[18] 张栌尹,莫永珍,欧阳晓俊,等.住院老年患者肌少症患病率及相关因素分析[J].老年医学与保健,2021,27(1):64-67.
[19] 程群,郑丽丽,章振林.肌少症流行病学及发病机制[J].中华骨质疏松和骨矿盐疾病杂志,2016,9(3):228-235.
[20] 岳益兵,于颖,沈磊,等.基于体检数据的肌少症患病风险预测[J].杭州师范大学学报(自然科学版),2022,21(1):14-22.
[21] 宋健.Logistic回归模型、神经网络模型和决策树模型在肺癌术后心肺并发症预测中的比较[D].合肥：安徽医科大学,2014.
[22] 陈静华.Logistic回归模型、神经网络模型和决策树模型在轻度认知功能障碍向阿尔茨海默症转归预测中的比较[D].南昌：南昌大学,2018.
[23] 刘艳平,谭明杨,徐超强,等.社区老年慢性病患者肌少症风险预测模型的构建[J].中国护理管理,2022,22(12):1814-1819.
[24] 韩婷,钱绪芬,王庆芳,等.基于Logistic回归和决策树模型的老年住院患者肌少症风险的影响因素分析[J].护理学报,2022,29(12):56-62.
[25] 李倩,刘芸宏,吴晓慧,等.基于决策树和Logistic回归预测出血性脑卒中手术后医院感染风险[J].中华医院感染学杂志,2021,31(23):3556-3561.
[26] 覃伟,吕芯芮,王子尧,等.基于logistic回归和随机森林的急性缺血性卒中3个月预后预测模型的构建[J].现代预防医学,2021,48(2):193-197.
[27] 宋祖玲,刁莎,严兰平,等.联合决策树及logistic回归建立乳腺癌相对风险预测模型[J].现代预防医学,2019,46(7):1156-1160，1175.
[28] 梁冰倩,黄志碧,赖银娟,等.随机森林模型和Logistic回归模型在高尿酸血症预测中的应用效果比较[J].广西医学,2020,42(6):729-733.

服务与反馈：

【文章下载】【发表评论】【查看评论】【加入收藏】

提示：您还未登录，请登录！点此登录