2349 - Representativeness and Generalizability of NCI-Funded Multi-Modality Randomized Clinical Trials (RCTs): A Case Study of Locally-Advanced Non-Small Cell Lung Cancer (LA-NSCLC) Based on RTOG-0617 and the National Cancer Database (NCDB)
Presenter(s)
C. F. P. M. de Sousa1, H. Parikh2, J. D. Bradley3, E. Stuart2, and C. Hu4,5; 1Department of Radiation Oncology and Molecular Radiation Sciences, Johns Hopkins University School of Medicine, Baltimore, MD, 2Johns Hopkins University Bloomberg School of Public Health Department of Biostatistics, Baltimore, MD, 3University of Pennsylvania/Abramson Cancer Center, Philadelphia, PA, 4Division of Biostatistics and Bioinformatics, Sidney Kimmel Comprehensive Cancer Center, Johns Hopkins University School of Medicine, Baltimore, MD, 5NRG Oncology Statistics and Data Management Center, Philadelphia, PA
Purpose/Objective(s): NCI-funded trials set treatment standards but may not fully reflect real-world populations. We assessed the generalizability of RTOG 0617, which found dose escalation worsened overall survival (OS) in LA-NSCLC, relative to NCDB. We examined treatment effects, survival differences, and representativeness with and without incorporating outcome variability.
Materials/Methods: We analyzed 411 RTOG 0617 patients (pts) and 8,095 NCDB stage III NSCLC pts (TNM 6th) who received 57–74 Gy concurrent chemoradiation from 2007 to 2011. Given real-world dosing variations, NCDB pts were categorized into two dose groups aligned with RTOG 0617: standard (=57 Gy to =66 Gy) and high (>66 Gy to =74 Gy). OS was evaluated using Kaplan-Meier estimates and Cox models with inverse probability weighting. Subgroup analyses examined chemotherapy timing and facility type. Representativeness was assessed using absolute standardized mean difference (ASMD <0.2) and the generalizability index (GI =0.95). The Rashomon Set of Optimal Trees (ROOT) method identified underrepresented subgroups.
Results: NCDB pts had worse OS than RTOG 0617 pts (median OS: 18.3 vs. 25.0 months; 12-month OS: 66.6% vs. 75.4%). Dose escalation (=66 Gy) had no significant effect on NCDB overall (HR 0.96, 95% CI 0.86–1.06, p = 0.418). However, in academic centers without chemotherapy delays, dose escalation worsened OS (HR 1.28, 95% CI 1.02–1.61, p = 0.034), supporting RTOG 0617’s conclusion. ASMD (<0.2) and a GI (=0.95) confirmed baseline trial representativeness. When accounting for short-term survival, ROOT further identified "other NSCLC histologies" as underrepresented. We could not characterize underrepresented subgroups with hampered precise treatment effect estimation.
Conclusion: RTOG 0617 was representative of real-world pts at enrollment, but contemporaneous NCDB pts prior to immunotherapy era had worse OS, highlighting trial-related benefits, such as adherence, standardized care, toxicity management, and underlying health status. The absence of a dose-escalation effect in NCDB overall does not contradict RTOG 0617, as treatment variability, selection bias, and unmeasured confounders likely diluted any survival difference. Subgroup analyses in academic centers confirm the external validity of RTOG 0617 in controlled settings. ROOT results support further investigation into time-dependent generalizability assessments to better characterize subgroups with heterogeneous treatment effects who are insufficiently represented in the trial. Future research should capture real-world toxicity and treatment standardization data, improve adherence evaluations, and refine methods to address unmeasured confounding in observational studies.