2382 - Development of an Oncology Trainee-Focused Educational Tool to Characterize Enrollment Bias In Early Stage Breast Cancer Trials
Presenter(s)
R. Smith1, C. Rodriguez-Russo1, and S. R. Alcorn2; 1Department of Radiation Oncology, University of Minnesota Medical School, Minneapolis, MN, 2Department of Radiation Oncology, University of Minnesota School of Medicine, Minneapolis, MN
Purpose/Objective(s): One challenge in resident education is understanding how trial inclusion criteria and characteristics of enrolled populations shape the external generalizability of study results. Here, we apply a database tool originally designed for patient-specific metaanalysis to characterize enrollment patterns for patient populations in early stage breast cancer trials relative to 1) inclusion criteria for those trials and 2) reported real-world population and disease characteristics for trial-unenrolled populations.
Materials/Methods: A database of 101,954 patients enrolled in early stage breast cancer trials was assembled from 29 trials selected on the basis of recent citation patterns and inclusion in NCCN guidelines. 31 clinical features were collected from each trial and compared to population-based distributions from patients with early stage breast cancer in two large published cohorts: 2,423,875 patients in the National Cancer Database (NCDB) between 2004 and 2015 and 244,810 patients in the SEER database between 2010 and 2013. Analysis of the SEER cohort was limited to the 151,766 patients with Stage I-III disease. Demographic and disease-specific characteristics were analyzed descriptively.
Results: Comparison of reported patient characteristics between the clinical trial database and NCDB and SEER cohorts was constrained to mutually reported variables. Many variables demonstrated reasonable agreement between the trial database vs. the population-based cohorts: left-sided disease in 51.07% vs. 50.6%, LVSI in 20.8% vs. 16.4%, ER-positive disease in 88.7% vs. 84.2%, HER2-positive disease in 11.2% vs. 14.9%, grade 1 disease in 23.9% vs. 21.5%, grade 2 disease in 49.9% vs. 42.9%, and grade 3 disease in 26.2% vs. 32.4%, respectively. Trial reporting of race was infrequent but notable for 1.1% vs. 0.26% Asian race and 3% vs. 11.2% Black race in the trials vs. the population-based cohorts, respectively. Reporting of age by trials was heterogeneous but consistent with low enrollment of patients under 40 years of age: only 2 included trials reported either a median or mean age under 50. For comparison, 136,525 patients (5.6%) of the NCDB cohort were under 40.
Conclusion: We demonstrate feasibility of an oncology trainee-focused education tool that allows interactive comparison of inclusion criteria vs. actual enrolled patient populations within clinical trials as well as vs. trial-unenrolled population characteristics. The reported analysis shows a systematic bias towards lower enrollment of several populations in key clinical trials. This overrepresentation of clinical features generally associated with lower-risk disease (e.g., ER-positive and grade 1-2) leaves unanswered questions regarding the standard of care for patients with higher risk features. Our next steps are to formally evaluate didactic effectiveness of this tool in increasing trainee awareness of trial enrollment bias.