This study found few demographic or socioeconomic factors, health history, or health behaviors that were associated with stage of diagnosis of non-small cell lung cancer. Subjects who quit smoking <10 years and were college graduates had increased risks of being diagnosed with advanced stage NSCLC, compared to never smokers and non-college graduates, respectively, whereas sigmoidoscopy/colonoscopy receipt was associated with a decreased risk. These associations should not be construed as causal but may be important factors in the development and detection of NSCLC.
Cancer is detected based on symptoms, incidental findings, or active screening and differences in stage at diagnosis must operate via one or more of these mechanisms . In terms of symptoms, individuals may delay care despite the presence of symptoms and present with advanced stages. Patients and clinicians may delay symptomatic care for several reasons that include health beliefs, limited access to healthcare, and/or competing causes (e.g. attributing a cough in a smoker to a benign cause). Also, factors that alter symptoms for a particular stage could be associated with a differential stage at diagnosis (e.g. causing hemoptysis at an earlier stage). Second, lung cancer is commonly found incidentally on imaging studies . Individuals with decreased access to care may have fewer opportunities to have asymptomatic, early stage tumors discovered incidentally [14–17]. Comorbid diseases might also alter the likelihood of incidental findings (e.g. a patient with congestive heart failure is found to have an incidental early stage tumor on a chest x-ray). A third factor is active screening to identify asymptomatic cases.
Stage at diagnosis is a strong predictor of lung cancer mortality  so identifying individual factors associated with stage at diagnosis is important for several reasons. First, identification may aid the development of disease progression biomarkers through better understanding of potential confounders . Second, these factors may be in the causal pathway for previously studied factors, including race/ethnicity , census tract-level socioeconomic status , rural versus urban location , and insurance status , that have been associated with lung cancer care disparities. Finally, it is likely that some lung cancers grow too slowly to cause death [20–22]. Understanding factors associated with slow-growing tumors, those with a higher chance of being found incidentally at earlier stages, may aid research into underlying mechanisms of tumor growth and development.
It is intriguing that a screening behavior was associated with lower risk of an advanced stage at diagnosis after adjusting for many potential confounders such as access to care variables, comorbid diseases, and other health behaviors. In addition, we compared in-situ and local disease versus advanced disease; local stages are rarely symptomatic  so factors that are mediated by symptoms are unlikely to explain our findings. Smoking status has not been evaluated in terms of its association with stage of diagnosis but a recent study from Sweden did not find an association with education status . Patients who received a sigmoidoscopy/colonoscopy as of 2002, and their clinicians, may have been "early adopters" of screening as there was no solid evidence of its benefit for preventing colon cancer mortality at that time . At the time of questionnaire administration lung cancer screening was not recommended . Computed tomography (CT) may soon have a role in early detection based on a preliminary report from the National Lung Cancer Screening Trial that reports to show a 20% decrease in lung cancer mortality in the screened group . Even before the results of this study became available, some experts and advocacy groups recommend that high risk groups discuss the utility of screening with their clinicians [27–29]. Similar to the widespread adoption of prostate cancer screening prior to evidence of its benefit [30, 31], lung cancer screening was advocated by some groups without knowledge of its risks and benefits . Patients may currently undergo CT screening for lung cancer based on their own or clinician beliefs about its efficacy [5, 33, 34] and our results raise this possibility as well. Understanding screening behaviors prior to the implementation of widespread recommendations for CT screening will assist evaluation of its implementation.
Our study has several strengths. We used a large, prospective, population-based cohort study design. We controlled for comorbid disease and factors associated with access to care. Although we were unable to directly adjust for insurance status, we adjusted for age in the primary analysis and did not observe effect modification for subjects over and under age 65, the age where there is essentially universal coverage through Medicare. Finally, the SEER database is complete and accurate, so there is minimal risk of outcome misclassification.
Residual and unmeasured confounding by access to care and comorbid disease may influence our findings. In addition, there are other limitations of this study. First, the baseline survey did not discriminate between sigmoidoscopy/colonoscopy performed for colon cancer screening or symptoms. While it is likely most studies were performed for screening, there is undoubtedly exposure misclassification that likely attenuates the results. The findings of this study may not be generalizable to other populations since the VITAL cohort includes fewer smokers, is predominantly white, and is more highly educated than the general US population. However, although the response rate to the survey was only 21%, it is unlikely that selection bias could have affected our results because in a prospective design, subjects cannot participate jointly based on exposure and future (unknown) disease status.