摘要
目的 基于机器学习算法构建高血压老年牧民轻度认知障碍(Mild cognitive impairment, MCI)风险筛查模型,为该人群MCI的筛查和预防提供参考。方法 通过横断面调查抽取新疆南山牧区高血压老年牧民2 117例,按照8∶2随机分为训练集(n=1 694)与验证集(n=423)。训练集进行单因素及多因素Logistic回归筛选变量,并构建随机森林模型与支持向量机模型。通过验证集进行模型验证,根据灵敏度、特异度、曲线下面积(Area under curve,AUC)、准确率、精确率、F1值评价模型性能并选出最优模型。结果 2 117例高血压老年牧民中,发生MCI 823例(38.88%),未发生MCI 1 294例(61.12%)。单因素及多因素Logistic回归分析显示,性别、年龄、高血压分级、握力、BMI、食用奶制品、食用鱼类、合并糖尿病、睡眠障碍、服药依从性与高血压老年牧民发生MCI有关,其中性别、年龄、高血压分级、握力、BMI、合并糖尿病、睡眠障碍为发生MCI的独立危险因素,食用奶制品、食用鱼类、服药依从性则为独立保护因素。基于训练集多因素Logistic回归分析确定的10个独立影响因素分别构建随机森林模型、支持向量机模型,经比较发现,随机森林模型性能较好,其灵敏度、特异度、准确率、精确率、F1值分别为94.51%、93.55%、94.09%、94.92%、0.95。在随机森林模型中,训练集与验证集的AUC分别为0.819(95%CI:0.799~0.839)、0.942(95%CI:0.916~0.968),提示模型区分度较好;临床决策曲线均显示净收益较高,提示模型具有临床有效性。结论 随机森林模型能有效识别高血压老年牧民MCI的发生风险,为MCI的筛查和防治提供参考。
Objective To develop a machine learning-based risk screening model for early detection of mild cognitive impairment(MCI)in elderly pastoralists with hypertension.Methods A total of 2117 cases of elderly hypertensive pastoralists in Nanshan pastoral areas of Xinjiang were sampled by cross-sectional survey and randomly divided into a training set(n=1694)and a validation set(n=423)according to 8∶2.Univariate and multifactor Logistic regression was performed to screen the variables in the training set,and the Random Forest model and Support Vector Machine model were constructed.The validation set was used for model validation,and the model performance was evaluated according to the sensitivity,specificity,AUC value,accuracy,precision and F1 value,and the optimal model was selected.Results In a training set of 2117 elderly hypertensive pastoralists,823 cases(38.88%)developed MCI,while 1294 cases(61.12%)did not.Both multifactor and univariate Logistic regression analysis showed that the differences in data regarding gender,age,hypertension grading,grip strength,BMI,consumption of dairy products,consumption of fish,presence of diabetes,sleep disorders and medication adherence were statistically significant.Gender,age,hypertension grading,grip strength,BMI,presence of diabetes and sleep disorders were identified as independent risk factors for the occurrence of MCI in elderly hypertensive pastoralists,while the consumption of dairy products,consumption of fish,and medication adherence were independent protective factors.Based on 10 independent influencing factors identified by multifactor Logistic regression analysis in the training set,the Random Forest model and Support Vector Machine model were constructed respectively,and it was found upon comparison that the Random Forest model performed better,with sensitivity,specificity,accuracy,precisionand F1 value of 94.51%,93.55%,94.09%,94.92%and 0.95,respectively.The AUC values of the training set and validation set were 0.819(95%CI:0.799-0.839)and 0.942(95%CI:0.916-0.968),respectively,suggesting that the model differentiation was better;the clinical decision curves of the training set and validation set both showed high net gain,suggesting that the model had clinical validity.Conclusion The Random Forest model effectively screens for MCI risk in elderly pastoralists with hypertension,offering a reference for MCI screening and prevention.
作者
孟娜
刘琴
吴培
艾非热·阿贝宝
王璇
张晋豪
由淑萍
MENG Na;LIU Qin;WU Pei;Aifeire Abeibao;WANG Xuan;ZHANG Jinhao;YOU Shuping(Department of Nursing,Xinjiang Medical University,Urumqi 830000,China;Provincial-Ministry Joint Establishment of State Key Laboratory of Causes and Prevention of High Morbidity in Central Asia,Urumqi 830000,China;Xinjiang Regional Population Disease and Health Care Research Centre,Urumqi 830000,China)
出处
《新疆医科大学学报》
2025年第5期701-709,共9页
Journal of Xinjiang Medical University
基金
省部共建中亚高发病成因与防治国家重点实验室项目(SKL-HIDCA-2023-HL10)
2024年国家级大学生创新训练项目(202410760031,202410760033)。
关键词
轻度认知障碍
机器学习
高血压
牧民
筛查模型
mild cognitive impairment(MCI)
machine learning
hypertension
pastoralists
screening model
作者简介
孟娜(1989-),女,在读硕士,主管护师,研究方向:社区护理。;通信作者:由淑萍,女,博士,教授,硕士生导师,研究方向:社区慢性病管理。
