Accurate and complete ascertainment of incident cancer outcomes is critical for precisely estimating exposure-risk associations in observational cohort studies. In the United States, owing to the lack of a national cancer registry, methods frequently used for cancer incidence ascertainment in observational cohort studies include self-report with or without validation from medical records and linkage with population-based cancer registries. Each of these ascertainment methods has strengths and limitations. When comparing cancer incidence ascertainment using self-report/death certificates with that from cancer registry linkage, we [1] and others [2], [3], [4], [5], [6] have found that the latter is generally more complete and accurate, while the former is more likely to be affected by misreporting. The misclassification or misreporting of cancer site, behavior, or histologic or molecular subtype can lead to systematic biases in the estimation of exposure-cancer associations [7]. However, it is unclear to what extent choice of cancer ascertainment method affects the accuracy of exposure-risk associations.

The U.S. Radiologic Technologists (USRT) Study cohort presents a unique opportunity to compare the performance of different cancer ascertainment methods in estimating associations of risk factors with cancer incidence. Initially, incident cancers were identified from self-administered mailed questionnaires, and cancer-related deaths were ascertained with death certificates [8], [9], [10]. Questionnaire data collected demographic, lifestyle, medical and reproductive history, and work history characteristics during four questionnaire surveys conducted every 8–10 years from 1983 to 2014. More recently, the cohort has been linked to 43 state/regional U.S. population-based cancer registries. Our previous comparison of cancer registry linkage to both self-report alone or self-report supplemented with death certificates revealed that these latter methods of cancer incidence follow-up captured only 46.5 % and 63.0 % of total cancer diagnoses identified by the registries, respectively [1]. In the USRT, most cancer incidence studies to-date relied on self-reported outcomes [11], [12], [13], [14], while studies focused on more fatal cancers used death certificate data [15], [16], [17], [18].

The objectives of this paper are to: (1) compare the performance of several methods of cancer ascertainment for estimating associations of known thyroid and lung cancer risk factors in USRT, and (2) use simulation studies to investigate the impact of imperfect outcome ascertainment on cancer risk models. We evaluated two cancer outcomes in USRT: thyroid cancer, which has a very high survival rate (thus likely to be identified through self-report every 8–10 years), and lung cancer, which has relatively poor survival (thus unlikely to be captured completely through periodic self-report). To compare cancer ascertainment methods, we estimated associations for thyroid and lung cancer incidence in relation to the following well-established risk factors [19], [20]: age, sex, body mass index (BMI), smoking and alcohol consumption. Some associations (e.g., age and BMI with thyroid cancer, age and smoking with lung cancer) are relatively strong in magnitude, thus maximizing the ability to detect modest differences by cancer ascertainment method.