听力残疾老年人失能风险预测模型的构建与验证

Construction and validation of the disability risk prediction model for older adults with hearing disabilities

目的探讨听力残疾老年人发生失能的预测因素,并构建其失能风险预测模型。方法本研究采用二手数据研究方法,基于中国健康与养老追踪调查数据库,选择2015年≥60岁的374例听力残疾老年人为研究对象,按照7∶3比例分为训练集(n=261)和验证集(n=113),根据其2018年是否发生失能为依据进行分组,分为随访期间未发生失能组(n=164)和随访期间发生失能组(n=97)。采用最小绝对收缩和选择运算符(LASSO)多元逻辑回归筛选可能影响失能发生的变量。使用多因素logistic回归构建预测模型及绘制列线图,并进行外部验证。结果训练集失能率37.2%。高血压、肝脏疾病、关节炎/风湿病、文化水平、牙齿掉光、晚上睡眠时长、认知功能、吸烟、午间睡眠时长是听力残疾老年人发生失能的预测因素。训练集模型ROC曲线下面积为0.703,95%CI为0.636~0.769,敏感度为69.1%,特异度为67.1%;校准曲线与理想曲线相近,Brier得分为0.207;测试集的ROC曲线下面积为0.713,95%CI为0.606~0.821,敏感度为76.3%,特异度为68.0%。结论高血压、肝脏疾病、关节炎/风湿病、文化水平、牙齿掉光、晚上睡眠时长、认知功能、吸烟、午间睡眠时长是听力残疾老年人发生失能的独立预测因素,列线图预测模型良好,对听力残疾老年人失能风险有一定的预测价值。

Objective To explore the predictive factors of disability in elderly people with hearing disabilities,and to construct a disability risk prediction model for them.Methods This study,based on the China health and retirement longitudinal survey(CHARLS)database,used secondary data research methods select 374 elderly people who reached the age of 60 and above in 2015 as the study subjects,which were randomly divided into training set(n=261)and test set(n=113)according to the ratio of 7∶3.They were categorized into"no disability occurred during follow-up"group(n=164)and"disability occurred during follow-up"group(n=97)based on whether or not disability occurred in 2018.Least absolute shrinkage and selection operator(LASSO)multivariate logistic regression was used to screen the variables that might affect the occurrence of incapacitation.Multifactorial logistic regression was used to construct the predictive models,plot the nomogram,and externally validated.Results The training set had a 37.2%incidence of incapacitation.Hypertension,liver disease,arthritis or rheumatism,literacy level,tooth loss,length of sleep at night,cognitive function,smoking,and length of midday sleep were predictors of the occurrence of incapacitation in older adults with hearing disabilities.The area under the ROC curve for the training set model was 0.703,with 95%CI was 0.636-0.769,a sensitivity of 69.1%and a specificity of 67.1%;The calibration curve was similar to the ideal curve,with a Brier score of 0.207;And the area under the ROC curve for the test set was 0.713,with 95%CI was 0.606-0.821,a sensitivity of 76.3%and a specificity of 68.0%.Conclusion Hypertension,liver disease,arthritis or rheumatism,literacy level,tooth loss,length of sleep at night,cognitive function,smoking,and length of midday sleep are independent predictors of the occurrence of disability in older adults with hearing disabilities,and the nomogram prediction model is good,and it has a certain predictive value of the risk of disability in older adults with hearing disabilities.