| 孔蕴馨,董 栋,马 悦,等.江苏省徐州市结直肠癌筛查风险评分预测模型建立与验证[J].中国肿瘤,2023,32(12):925-934. |
| 江苏省徐州市结直肠癌筛查风险评分预测模型建立与验证 |
| Development and Validation of Colorectal Cancer Screening Risk Score Prediction Model Based on Local Data in Xuzhou City， Jiangsu Province |
| 投稿时间：2023-04-21 |
| DOI：10.11735/j.issn.1004-0242.2023.12.A006 |
|
 |
| 中文关键词: 结直肠癌 筛查 风险评估 危险因素 预测模型 |
| 英文关键词:colorectal cancer screening risk assessment risk factors prediction model |
| 基金项目:中国医学科学院医学与健康科技创新工程项目（2017-I2M-1-006）；国家重大公共卫生服务项目（城市癌症早诊早治项目） |
|
| 摘要点击次数: 276 |
| 全文下载次数: 129 |
| 中文摘要: |
| 摘 要：［目的］ 使用危险因素构建本地结直肠癌(colorectal cancer，CRC)筛查风险评分预测模型，并在CRC筛查数据中对模型效果进行验证。 ［方法］ 在徐州市2014—2015年参与城市癌症早诊早治项目流行病学调查的50~69岁居民中选择符合研究条件的受试者作为推导人群研究对象，利用徐州市肿瘤信息化管理平台进行6.5年连续被动随访。以CRC为预测指标，参考亚太结直肠癌筛查评分(Asia-Pacific Colorectal Screening score，APCS)模型的构建方法建立本地模型。使用χ2检验对推导人群进行单因素分析，P<0.2或既往模型中存在的危险因素纳入多因素Logistic模型，使用危险因素的比值比（odds ratio，OR）构建CRC风险评分模型，个体风险得分为其所有危险因素得分的总和。模型得分范围为0～8分，得分0～3分的受试对象定义为一般风险，得分4分及以上的受试对象定义为高风险。使用2015—2022年徐州市完成CRC筛查的50~69岁居民数据对新建模型进行外部验证，并与APCS模型进行预测效果比较。［结果］ 推导人群中共纳入符合条件的受试对象36 270名，经6.5年随访，CRC累积发病率为0.43%(157/36 270)。多因素分析结果显示，性别和年龄是CRC发病的主要危险因素。男性患CRC的概率为女性的1.54倍［95%置信区间（confidence interval，CI）：1.06～2.25，P=0.025］，60～64岁和65～69岁人群患CRC的概率为50～54岁人群的2.31倍(95%CI：1.48～3.62，P<0.001)和1.92倍(95%CI：1.15～3.21，P=0.013)。使用性别、年龄、吸烟、身体质量指数（body mass index，BMI）、一级亲属CRC家族史、食用红肉构建CRC风险评分模型。验证人群中共纳入符合条件的受试对象3 053名，进展期肿瘤检出率为4.00%(122/3 053)。使用新建风险评分模型和APCS模型对验证人群进行风险评分，高风险率分别为55.19%(1 685/3 053)和49.92%(1 524/ 3 053)，进展期肿瘤检出率分别为5.52%(93/1 865)和5.45%(83/1 524)，风险比（risk ratio，RR）分别为2.60 (95%CI：1.73～3.93，P<0.001)和2.14 (95%CI：1.47～3.10，P<0.001)，受试者工作特征曲线的曲线下面积（area under curve，AUC）分别为0.61 (95%CI：0.57～0.65)和0.59 (95%CI：0.55～0.64)，灵敏度分别为76.23% (95%CI：67.68%～83.47%)和68.03% (95%CI：58.98%～76.18%)，特异度分别为46.67% (95%CI：44.85%～48.50%)和52.17% (95%CI：50.34%～53.99%)。新模型在RR、AUC、灵敏度方面高于APCS模型，在特异度上低于APCS模型。新模型和APCS模型结肠镜资源负载均为18，远低于未使用风险评分的结肠镜资源负载25。 ［结论］ 徐州市CRC筛查风险评分预测模型可以帮助确定无症状人群患CRC的风险，诊断效果略优于APCS模型，相比未使用风险评分模型能减少结肠镜资源负载，可作为开展大规模人群CRC筛查的本地化工具。 |
| 英文摘要: |
| Abstract：［Purpose］ To develop a colorectal cancer(CRC) screening risk score prediction model and to validate its prediction effectiveness in CRC screening. ［Methods］ Subjects eligible for the study were selected as the derivation cohort from the residents aged 50～69 years old who participated in the questionnaire survey of Cancer Screening Program in Urban China(CanSPUC) in Xuzhou from 2014 to 2015. The continuous passive follow-up was conducted for 6.5 years using the Xuzhou Cancer Management System. Based on CRC as a predictive index and referring to the construction method of the Asia-Pacific Colorectal Screening score(APCS) model， a localized model was constructed. χ2 test was used for univariate analysis of the derivation cohort. Risk factors with P value <0.2 or present in previous models were incorporated into the multivariate Logistic model for multivariate analysis. The odds ratio(OR) values of risk factors were rounded as risk factor assignment to build a risk score model for CRC screening. Each individual’s risk score was the sum of their risk factor scores. The score range of the model was 0～8 points， subjects with scores of 0～3 were defined as average risk， and those with scores of 4 or above were defined as high risk. The data of Xuzhou residents aged 50 to 69 who completed CRC screening from 2015 to 2022 were used for external validation cohort of the local model， and the prediction effectiveness was compared with the APCS model. ［Results］ A total of 36 270 eligible subjects were included in the derivation cohort. After 6.5 years of follow-up， the cumulative incidence of CRC was 0.43%(157/36 270). The results of multivariate analysis showed that gender and age were the risk factors for CRC. The incidence of CRC in male was 1.54 times higher than that in female(95%CI：1.06～2.25， P=0.025)； and the incidence of CRC in people aged 60~64 and 65~69 years old was 2.31 times and 1.92 times higher than that in people aged 50～54 years old(95%CI：1.48～3.62， P<0.001 and 95%CI：1.15~3.21， P=0.013)， respectively. The CRC risk scoring model was constructed using gender， age， smoking， body mass index(BMI)， family history of CRC in first-degree relatives， and red meat intake. A total of 3 053 eligible subjects were included in the validation cohort， and the detection rate of advanced neoplasm was 4.00%(122/3 053). The constructed risk scoring model and APCS model were used to score the risk of the validation cohort， and the high risk rates were 55.19%(1 685/3 053) and 49.92%(1 524/3 053)， respectively. The detection rates of advanced neoplasm were 5.52%(93/1 865) and 5.45%(83/1 524)， the risk ratio(RR) was 2.60(95%CI：1.73～3.93， P<0.001) and 2.14(95%CI：1.47～3.10， P<0.001)， and the area under curve(AUC) was 0.61(95%CI：0.57～0.65) and 0.59(95%CI：0.55～0.64)， the sensitivity was 76.23%(95%CI：67.68%～83.47%) and 68.03%(95%CI：58.98%～76.18%)， the specificity was 46.67%(95%CI：44.85%～48.50%) and 52.17%(95%CI：50.34%～53.99%)， respectively. The constructed local model had higher RR， AUC and sensitivity than APCS model， but lower specificity than APCS model. The colonoscopy resource load of both models was 18， much lower than that without using the risk score(25). ［Conclusion］ The CRC screening risk prediction model constructed in the study can help determine the risk of CRC in asymptomatic population， and its diagnostic effect is slightly better than that of the APCS model. Compared with non-use of risk score as primary screening， it can reduce the load of colonoscopy resources. It can be used as a localized tool to conduct CRC screening in large populations. |
|
在线阅读
查看全文 查看/发表评论 下载PDF阅读器 |