摘要
目的构建健康素养预测模型,为快速筛检低健康素养水平人群提供参考。方法2020年,采用多阶段随机整群抽样方法,抽取苏北X市15~69岁居民为调查对象,采用国家健康素养调查问卷开展调查,建立logistic回归模型和决策树模型,以实际调查结果为金标准,比较两种模型预测效果,并通过增加预测变量方法构建新的logistic回归健康素养预测模型。结果共纳入1933名调查对象,不具备健康素养1580人,占81.74%。logistic回归分析结果显示,年龄、文化程度、职业和吸烟情况是居民健康素养水平的影响因素(P值均<0.05)。建立的logistic回归模型AUC为0.79(95%CI:0.77~0.81),构建的决策树模型AUC为0.78(95%CI:0.75~0.80)。以健康素养是否具备分别与56道问题知晓情况进行统计学比较,将问题按照各题χ^(2)值由大到小排列,纳入问卷中前6道题目(C09、C05、D01、C08、C12、C10)构建新的logistic回归预测模型:Logit(P)=ln[P/(1-P)]=-3.104+0.325[年龄(25~34)]-0.149[年龄(35~44)]-0.857[年龄(45~54)]+0.367[年龄(55~64)]-2.748[年龄(65~69)]+2.531[文化程度(小学及以下)]+1.597[文化程度(初中)]+0.303[文化程度(高中/职高/中专)]+1.444[题目C05(错误)]+1.744[题目C08(错误)]+1.621[题目C09(错误)]+2.025[题目C10(错误)]+1.806[题目C12(错误)]+1.958[题目D01(错误)],该预测模型灵敏度为85.90%,特异度为90.04%,AUC为0.95(95%CI:0.94~0.96)。在约登指数取最大值为0.76时,logit(P)对应P值为0.831,即预测概率超过0.831,可判定为不具备健康素养。结论利用改进的健康素养预测模型中的切点值,可在现场调查时快速鉴别出不具备健康素养人群。
Objective To develop predictive models for health literacy,so as to provide insights into rapid screening of individuals with low health literacy.Methods Residents at ages of 15 to 69 years were sampled from X city in southern Jiangsu Province using a multi-stage random cluster sampling method in 2020,and subjects′health literacy was surveyed using National Health Literacy Questionnaires.Logistic regression model and decision tree model were created,and the predictive values of these two models were compared with the actual survey results as a gold standard.In addition,new logistic regression models for prediction of health literacy were created through addition of predictive variables.Results A total of 1933 subjects were enrolled,and there were 1580 subjects without health literacy(81.74%).Logistic regression analysis identified age,educational level,occupation and smoking status as factors affecting residents′health literacy levels(all P values<0.05).The areas under the receiving operating curve were 0.79[95%confidential interval(CI):(0.77,0.81)]for the logistic regression model and 0.78[95%CI:(0.75,0.80)]for the decision tree model,respectively.The associations of presence of health literacy with the awareness of 56 questions were statistically analyzed,and the questions were arranged according to chi-square values in a descending order.The first six questions in the questionnaire(C09,C05,D01,C08,C12 and C10)were included to create a new logistic regression model for prediction of health literacy:Logit(P)=ln[P/(1-P)]=-3.104+0.325[age(25 to 34 years)]-0.149[age(35 to 44)years]-0.857[age(45 to 54 years)]+0.367[age(55 to 64)years]-2.748[age(65 to 69 years)]+2.531[educational level(primary school and below)]+1.597[educational level(junior high school)]+0.303[educational level(senior high school/vocational high school/secondary specialized school)]+1.444[question C05(mistake)]+1.744[question C08(mistake)]+1.621[question C09(mistake)]+2.025[question C10(mistake)]+1.806[question C12(mistake)]+1.958[question D01(mistake)].The sensitivity,specificity and AUC of this predictive model were 85.90%,90.04%and 0.95[95%CI:(0.94,0.96)]for prediction of health literacy,and if the maximal Youden index was 0.76,the corresponding P value of logit(P)was 0.831,indicating that no health literacy was defined if a predictive probability was over 0.831.Conclusions Residents without health literacy may be rapidly identified in field surveys using the cut-off in improved predictive models for health literacy.
作者
蔡婷
卓琳
苗春霞
汪秀英
卓朗
CAI Ting;ZHUO Lin;MIAO Chun-xia;WANG Xiu-ying;ZHUO Lang(Xuzhou Municipal Center for Disease Control and Prevention,Xuzhou Jiangsu 221006,China;不详)
出处
《江苏预防医学》
CAS
2024年第3期283-287,共5页
Jiangsu Journal of Preventive Medicine
基金
国家社会科学基金(19BGL251)。
