Bach Peter B, Kattan Michael W, Thornquist Mark D, Kris Mark G, Tate Ramsey C, Barnett Matt J, Hsieh Lillian J, Begg Colin B
The Health Outcomes Research Group, Department of Epidemiology and Biostatistics, Memorial Sloan-Kettering Cancer Center, New York, NY 10021, USA.
J Natl Cancer Inst. 2003 Mar 19;95(6):470-8. doi: 10.1093/jnci/95.6.470.
Although there is no proven benefit associated with screening for lung cancer, screening programs are attracting many individuals who perceive themselves to be at high risk due to smoking. We sought to determine whether the risk of lung cancer varies predictably among smokers.
We used data on 18 172 subjects enrolled in the Carotene and Retinol Efficacy Trial (CARET)-a large, randomized trial of lung cancer prevention-to derive a lung cancer risk prediction model. Model inputs included the subject's age, sex, asbestos exposure history, and smoking history. We assessed the model's calibration by comparing predicted and observed rates of lung cancer across risk deciles and validated it by assessing the extent to which a model estimated on data from five CARET study sites could predict events in the sixth study site. We then applied the model to evaluate the risk of lung cancer among smokers enrolled in a study of lung cancer screening with computed tomography (CT).
The model was internally valid and well calibrated. Ten-year lung cancer risk varied greatly among participants in the CT study, from 15% for a 68-year-old man who has smoked two packs per day for 50 years and continues to smoke, to 0.8% for a 51-year-old woman who smoked one pack per day for 28 years before quitting 9 years earlier. Even among the subset of CT study participants who would be eligible for a clinical trial of cancer prevention, risk varied greatly.
The risk of lung cancer varies widely among smokers. Accurate risk prediction may help individuals who are contemplating voluntary screening to balance the potential benefits and risks. Risk prediction may also be useful for researchers designing clinical trials of lung cancer prevention.
尽管尚无证据表明肺癌筛查有实际益处，但筛查项目仍吸引了许多因吸烟而自认为肺癌高危人群。我们试图确定吸烟者患肺癌的风险是否存在可预测的差异。
我们利用参与胡萝卜素与视黄醇功效试验（CARET）的18172名受试者的数据——这是一项大型肺癌预防随机试验——来推导肺癌风险预测模型。模型输入因素包括受试者的年龄、性别、石棉接触史和吸烟史。我们通过比较风险十分位数区间内肺癌的预测发生率和观察发生率来评估模型的校准情况，并通过评估基于五个CARET研究地点的数据所构建的模型对第六个研究地点事件的预测能力来验证该模型。然后我们应用该模型评估参与计算机断层扫描（CT）肺癌筛查研究的吸烟者患肺癌的风险。
该模型在内部是有效的且校准良好。CT研究参与者的十年肺癌风险差异很大，从一名68岁男性（每天吸烟两包，已吸50年且仍在吸烟）的15%到一名51岁女性（在9年前戒烟前每天吸一包，共吸28年）的0.8%不等。即使在CT研究参与者中符合癌症预防临床试验条件的子集中，风险差异也很大。
吸烟者患肺癌的风险差异很大。准确的风险预测可能有助于那些考虑自愿筛查的人权衡潜在的益处和风险。风险预测对于设计肺癌预防临床试验的研究人员也可能有用。
