基于社区居民大肠癌筛查的高危人群发病风险分级模型构建  

Construction of risk classification model for high-risk population in community-based colorectal cancer screening

作　　者：闫蓓[1,2] 林涛 陈建平[3] 王莹莹 林上群 杨琛 YAN Bei;LIN Tao;CHEN Jianping;WANG Yingying;LIN Shangqun;YANG Chen(Center for Disease Control and Prevention,Pudong New Area,Fudan University Pudong Institute of Preventive Medicine,Shanghai 200136,China;Xi’an International Medical Center Hospital,Xi’an 710100,Shannxi Province,China;Shanghai First Maternity and Infant Hospital,School of Medicine,Tongji University,Shanghai 201204,China)

机构地区：[1]上海市浦东新区疾病预防控制中心,复旦大学浦东新区预防医学研究院,上海200136 [2]西安国际医学中心医院,陕西西安710100 [3]同济大学附属第一妇婴保健院,上海201204

出　　处：《肿瘤》2022年第8期570-579,共10页Tumor

基　　金：上海市浦东新区卫计委卫生计生科研项目资助(PW2017A-7)

摘　　要：目的:基于上海市社区大肠癌筛查模式,针对筛查人群建立危险因素评分系统和大肠癌的风险预测模型,旨在从筛查对象中识别出大肠癌高风险人群,以期对其进行分级管理。方法:对上海市浦东新区2013—2017年大肠癌筛查及随访数据进行分析,以筛查后2年内的大肠癌发生为结局变量拟合logistic回归模型,建立大肠癌危险因素评分系统;对筛查对象进行危险因素评分并根据总得分建立大肠癌的预测模型,用受试者操作特征(receiver operator characteristic,ROC)曲线评估其效能,并用2018年的筛查数据进行外部验证。结果:研究表明,男性、黏液便和血便史、恶性肿瘤史及一级亲属大肠癌史与2年内大肠癌的发生风险有关;随着年龄的增长,大肠癌的发生风险逐渐升高;粪便隐血检测阳性与大肠癌发病风险上升有关,粪便隐血检测2次阳性的对象大肠癌的发生风险升高25倍。通过拟合多因素回归模型,建立大肠癌危险因素的评分系统,对每个筛查对象进行危险因素评分并建立大肠癌风险预测模型:Y=-7.4743+0.101×总得分。用ROC曲线评价模型发现,当危险因素总得分为20分时,模型预测效果最佳,敏感度为63.6%,特异度为86.6%,其ROC曲线下面积为0.819(95%置信区间:0.808~0.831);在5%和10%假阳性率时的检出率分别为44.6%和57.8%。外部数据验证结果显示,模型的校准斜率为0.9997,校准截距为-0.2352,ROC曲线下面积为0.815,可见模型的预测效果较好。在此基础上,尝试提出对筛查人群进行大肠癌风险分级,根据危险因素总得分将筛查人群分为一般风险、中度风险Ⅰ级、中度风险Ⅱ级、高度风险和极高风险5个等级并分别实施不同的随访管理方式,以期在有限的医疗资源下提高大肠癌筛查的成效。结论:本研究基于上海市大肠癌的筛查模式,建立了大肠癌危险评分系统和风险预测模型,尝试提出对大肠癌筛查人群进行�Objective:To establish a risk factor scoring system and a prediction model for colorectal cancer based on the colorectal cancer population screening implementation in Shanghai so as to identify high-risk groups of colorectal cancer from the participants for hierarchical management.Methods:Data of colorectal cancer screening and related follow-up from 2013 to 2017 in Pudong New Area of Shanghai were analyzed.The occurrence of colorectal cancer within 2 years after screening was used as the outcome,and logistic regression was fitted to establish a scoring system for colorectal cancer risk factors.The subjects were scored according to the risk factor scoring system,and a predictive model for colorectal cancer was established based on the subjects’total risk scores.Its prediction efficacy was evaluated using the ROC(receiver-operating characteristic)curve and externally validated using the screening data in 2018.Results:This study showed that male,history of mucus or blood in stool,history of cancer and history of colorectal cancer in first-degree relatives were associated with the occurrence of colorectal cancer.The risk of colorectal cancer gradually increased with advancing age.Positive fecal occult blood test was associated with increased risk of colorectal cancer.Notably,a 25-fold increase in the risk of colorectal cancer was observed for subjects who were tested positive twice.A scoring system for colorectal cancer risk factors was established using the multivariate regression model.A predictive model(Y=-7.4743+0.101×total risk score)was established based on the total risk scores of all subjects screened.The ROC curve showed that the performance of the model was optimum with a sensitivity and specificity of 63.6%and 86.6%respectively when the cut-off value was 20,and the area under the ROC curve was 0.819(95%confidence interval:0.808-0.831).The detection rate was 44.6%and 57.8%respectively at the false positive rate of 5%and 10%.External validation showed that the calibration slope of the model was 0.9997,th

关 键 词：大肠癌 筛查 风险预测模型 分级管理 

分 类 号：R735.34[医药卫生—肿瘤]