基于中国健康与养老纵向研究数据库的糖尿病老年人失能预测模型构建

Construction of a disability prediction model for the elderly with diabetes based on CHARLS database

目的基于CHARLS数据库的资料,建立糖尿病老年人失能的预测模型,并分析其效用。方法选取2018年发布的2015年中国健康与养老纵向研究(China health and retirement longitudinal study,CHARLS)调查中的4797例患有糖尿病老年人的相关资料,包括基本特征、体格测量、生活习惯、伴随症状和疾症、血液和血生化指标、功能检测等。根据ADL将其分为失能与无失能2组,在比较2组各指标的差异后,通过Logistic回归分析筛选失能有关因素,建立失能预测模型,并采用受试者工作特征(ROC)评价模型的效用。结果通过Logistic回归,共筛选到腰围、舒张压、饮酒、疼痛、下肢功能评分、上肢功能评分、血脂异常、卒中、情绪障碍、共病数量、居住状况、工作为相关因素,据此建立了失能预测模型,即模型公式为:ln=-4.880+0.012×腰围(cm)-0.010×舒张压(mmHg)-0.250×饮酒+0.854×疼痛+0.235×下肢功能评分+0.431×上肢功能评分-0.278×血脂异常+0.809×卒中+1.169×情绪障碍+0.165×共病数量-0.542×居住状况-0.083×工作(其中P为失能状态)。ROC分析结果显示,AUC为0.89(P<0.005)、敏感度0.82、特异度0.74,截断值0.26。结论依据CHARLS以Logistic回归遴选因素所构建的中国糖尿病老人失能状态预测模型具有较好的预测能力。

Objective To establish a disability prediction model for the elderly with diabetes based on data from the China Health and Retirement Longitudinal Study(CHARLS)database and analyze its utility.Methods The data of 4797 elderly diabetic patients from the 2015 CHARLS survey released in 2018 were selected,including their basic characteristics,physical measurements,lifestyle habits,accompanying symptoms and diseases,blood and biochemical indicators,functional tests,etc.According to activities of daily living(ADL),they were divided into two groups:disabled group and non-disabled group.After comparing the difference of each index between the two groups,the disability-related factors were screened by logistic regression analysis,and a disability prediction model was established.The utility of the model was evaluated by receiver operating characteristic(ROC).Results Through logistic regression,waist circumference,diastolic blood pressure,alcohol drinking,pain,lower limb function score,upper limb function score,dyslipidemia,stroke,emotional disorder,number of comorbidities,living status,and work were identified as relevant factors.Based on these factors,a disability prediction model was established as follows:ln=-4.880+0.012×waist circumference(cm)-0.010×DBP(mmHg)-0.250×alcohol drinking+0.854×pain+0.235×lower limb cumulative score+0.431×upper limb cumulative score-0.278×dyslipidemia+0.809×stroke+1.169×emotional disorder+0.165×number of comorbidities-0.542×living status-0.083×work(P was the disabled state).The results of ROC analysis showed that the AUC was 0.89(P<0.005),the sensitivity was 0.82 the specificity was 0.74 and cutoff value was 0.26.Conclusion The disability prediction model established based on CHARLS database and the factors selected by Logistic regression has good predictive ability.