甘肃省2019-2020年出生儿童预防接种行为评分卡模型构建  

Construction of a behavior scorecard model for vaccination of children born in 2019-2020 in Gansu province

作　　者：梁雪枫[1] 安婧[1] 金娜[1] 焦永卓[1] 邓琳[1] 张晓曙[1] 唐宇[1] Liang Xuefeng;An Jing;Jin Na;Jiao Yongzhuo;Deng Lin;Zhang Xiaoshu;Tang Yu(Gansu Provincial Center for Disease Control and Prevention,Lanzhou 730000,Gansu,China)

机构地区：[1]甘肃省疾病预防控制中心,甘肃兰州730000

出　　处：《中国疫苗和免疫》2022年第4期465-469,共5页Chinese Journal of Vaccines and Immunization

基　　金：中华预防医学会预防接种信息化与服务能力提升项目(CPMA-JY2021-005)。

摘　　要：目的建立儿童预防接种行为评分卡模型以识别流失接种对象。方法将甘肃省免疫规划信息系统中2019-2020年出生儿童数据集按7∶3比例随机分为训练集和测试集;通过疫苗接种的时间-频率-花费(Recency frequency monetary,RFM)模型识别流失儿童,采用相关性检验和信息值筛选模型变量,采用决策树方法对变量分组;对证据权重(Weight of evidence,WOE)转换后的训练集使用Logistic回归进行拟合,然后将模型转换为行为评分卡;绘制两个数据集的柯尔莫哥洛夫-斯米尔诺夫(Kolmogorov-Smirnov,KS)、受试者工作特征(Receiver operating characteristic,ROC)和群体稳定性指标(Population stability index,PSI)曲线,评估模型拟合效果。结果甘肃省2019-2020年出生儿童共626653名,RFM模型识别出43465名(6.94%)流失儿童。构建的行为评分卡模型包括13个变量,模型基础分为479分;训练集和测试集的模型KS值分别为0.6064、0.6063,均>0.60;ROC曲线下面积(Area under ROC curve,AUC)分别为0.8718、0.8714,均>0.87;模型PSI值为0.0001。结论本研究建立的预防接种行为评分卡模型识别流失接种对象的效果和稳定性好;可在免疫规划信息管理系统中集成应用。Objective To establish a childhood vaccination behavior scorecard model to identify children lost to follow-up.Methods We randomly divided a dataset of 2019-2020 birth cohorts from the Gansu Immunization Program Information System into a training set and a test set in a ratio of 7∶3.We used the recency-frequency-monetary(RFM)model to identify children lost to follow-up.Model variables were screened for applicability by correlation testing and by information values(IVs),and were grouped into a decision tree.The training set was transformed into weights of evidence(WOEs)which were fitted by logistic regression.The fitted model was transformed into a behavior scorecard.We plotted Kolmogorov-Smirnov(KS),receiver operating characteristic(ROC),and population stability index(PSI)curves for the two datasets to evaluate model fit.Results A total of 626653 children were born in Gansu in 2019-2020,and the RFM model identified 43465(6.94%)children lost to follow-up.The constructed behavior scorecard model included 13 variables with a 479 basic scores.KS values for the training and test sets were 0.6064 and 0.6063(both greater than 0.60),respectively,with area under the ROC curve(AUC)values of 0.8718 and 0.8714(both greater than 0.87).The PSI value of the model was 0.0001.Conclusions The vaccination behavior scorecard model established in this study has good effectiveness and stability in identifying children lost to follow-up.The model can be integrated into the Immunization Program Information System.

关 键 词：预防接种 信息管理系统 行为评分卡模型 流失儿童 

分 类 号：R186[医药卫生—流行病学]