上海市女性乳腺癌危险因素分析与风险预测模型研究

目的 乳腺癌风险预测模型可将人群分为不同的风险等级,有助于降低筛查成本,使乳腺癌筛查效益最大化.本研究分析了上海市女性乳腺癌的危险因素,初步建立了符合该人群流行病学特征的风险预测模型,为乳腺癌高危人群的筛选提供依据.方法 2008-05-23-2012-09-30,采用调查表对上海市闵行区149 577名35~74岁女性开展乳腺癌初筛,内容包括人口学、月经生育史、乳腺疾病史和家族史等信息,具备任一明确定义危险因素者为初筛阳性.将所有对象的个人信息与上海市肿瘤登记系统和生命统计系统进行记录联动,收集2015-06-30前乳腺癌确诊和全死因死亡信息.采用Cox比例风险模型,建立乳腺癌风险预测模型,计算乳腺癌5年发病风险,并采用5折交叉验证法,分别计算期望病例数与观察病例数比值(ratio of the expected to the observed number,E/O)和受试者工作特征曲线下面积(areas under the receiver operating characteristic curve,AUC),评价模型的校准度和区分力.结果 经过774 333人年(中位随访人年5.05年)随访,共发现新发乳腺癌病例973例,粗发病率(crude incidence rate,CIR)和年龄标化率(age-standardized incidence rate,ASR)分别为125.66/10万和112.55/10万,初筛阳性者的粗率和标化率分别为133.91/10万和121.83/10万,显著高于初筛阴性者的119.76/10万和106.91/10万.年龄、教育程度、乳腺癌家族史、患重度乳腺小叶增生、有乳房肿块、患乳腺导管内乳头状瘤与乳腺癌呈正向关联,哺乳和月经周期规律与乳腺癌呈负向关联.基于这些因素建立的风险预测模型估计该人群乳腺癌5年绝对发病风险高峰出现在55岁,在0.19%~1.10%之间变化.模型的E/O值为0.98(95%CI为0.92,1.04),AUC为0.596(95%CI为0.538,0.654).进一步按年龄分层,发现55岁以下组和55岁及以上组的E/O值分别为0.96(0.88,1.03)和1.01(0.91,1.16),AUC分别为0.627(0.514,0.701)和0.567(0.518,0.630).结论 本研究建立的风险评估模型主要基于自我报告的乳腺症状及体征,总体校准度较好,而总体区分力不理想,但在55岁以下女性中有所提高,可用于社区人群尤其是55岁以下人群的乳腺癌风险分级.

Risk factors of breast cancer and a risk predictive model for Chinese women in Shanghai,China

OBJECTIVE Identifying women at high risk of breast cancer using risk predictive model can improve cost-effectiveness of breast cancer screening.This study was designed to establish a risk predictive model for breast cancer based on self-reported risk factors in Chinese women in Shanghai,and to provide a utility to identify individuals at high risk.METHODS In-person interview was conducted in 149 577 women participating in a breast cancer screening program in Minhang District during the period of May 23,2008 and September 30,2012.Information on demographic and reproductive factors,as well as history of breast diseases and family history of breast cancer were collected using a structured questionnaire.Women with any one defined risk factor were considered with positive results and recommended further examinations.Incident breast cancers and all-cause deaths up to June 30,2015 were identified for all subjects by a record linkage with the Shanghai Cancer Registry and the Vital Statistics.Cox proportional hazards regression was used to establish a risk predictive model of breast cancer.Absolute 5-year risk of developing breast cancer was established based on the established model.A five-fold cross-validation was performed to obtain estimates of model predictive accuracy.The discriminatory power was evaluated using the areas under the receiver-operating characteristic curve (AUC).The ratio of the expected to the observed number (E/O) of incident breast cancer cases was calculated to assess overall calibration.RESULTS A total of 973 women were newly diagnosed with breast cancer after an average of 5.05 person-years of follow-up,with a crude incidence rate (CIR) and age-standardized incidence rate (ASR) of 125.66/100 000 and 112.55/100 000.CIR and ASR of breast cancer were 133.91/100 000 and 121.83/100 000,respectively,in women with positive results,significantly higher than 119.76/100 000 and 106.91/100 000 in negative women.A significant positive association was observed for age,education,family history,lobular hyperplasia,breast lumps,ductal papilloma with breast cancer risk,while a negative association was observed for breastfeeding and regular menstrual cycle.According to the established model based on these risk factors,the absolute 5-year risk reached peak at the age of 55 years,which ranged from 0.19% to 1.10%.The model was observed to slightly underestimate the risk (E/O,0.98;95%CI,0.92 to 1.04) and have modest discriminatory accuracy (AUC,0.596;95%CI,0.538 to 0.654).Further stratified analysis showed that E/O and AUC were 0.96(95%CI:0.88,1.03) and 0.627(95%CI:0.514,0.701),respectively,among women less than 55 years old,and were 1.01(95%CI:0.91,1.16) and 0.567(95%CI:0.518,0.630),respectively,in women aged 55 years or above.CONCLUSIONS The established risk predictive model mainly based on self-reported symptoms performs well in calibration but is not satisfied in discriminatory accuracy,particularly in women aged 55 years or above.The model may help to identify women at high risk of breast cancer,particularly in younger women.