Main Session
Sep
30
PQA 09 - Hematologic Malignancies, Health Services Research, Digital Health Innovation and Informatics
3632 - Association of Differential Censoring with Primary Endpoint Superiority in Radiation Oncology Cooperative Group Trials
Presenter(s)
Eric Hsu, MD, PhD - Stanford Health Care, Stanford, CA
E. J. Hsu1, A. D. Sherry2, T. A. Lin2, and E. B. Ludmir3; 1Department of Radiation Oncology, Stanford University School of Medicine, Stanford, CA, 2The University of Texas MD Anderson Cancer Center, Houston, TX, 3Department of Gastrointestinal Radiation Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX
Purpose/Objective(s):
In oncology clinical trials, time-to-event endpoints are often used to evaluate therapeutic benefit of treatment arms. Presence of differential censoring (DC), which is censoring imbalance between treatment arms, may lead to falsely inflated treatment benefit, thus potentially creating bias and misinterpretation of therapeutic benefit. Whether DC is prevalent in clinical trials related to radiotherapy and how it may potentially bias outcomes is unclear. This study aims to identify the presence and incidence of DC in Radiation Therapy Oncology Group (RTOG) and Trans-Tasman Radiation Oncology Group (TROG) trials and whether it is associated with observed clinical benefit of the experimental arms in question.Materials/Methods:
ClinicalTrials.gov and PubMed.gov were queried to identify RTOG and TROG trials with at least two randomized experimental treatment arms with a time to event surrogate primary endpoint (PEP) (such as progression free survival) and overall survival (OS) secondary endpoint. Kaplan-Meier (KM) curves were reconstructed from each trial using standard published methods, and reverse KM plots were generated from reconstructed KM curves. DC was detected if the Cox regression or the restricted mean survival time difference (in the setting of non-proportional hazards) calculated from the reverse KM curves was statistically significant at P of 0.05. Fisher’s or Mann-Whitney U tests were used to compare characteristics between different groups.Results:
We identified and reconstructed 37 pairs of surrogate PEP/OS KM curves from 23 RTOG and 8 TROG trial publications from 2005 to 2024. For the PEP, five (14%) trials exhibited DC in the control arm, and 3 (8%) trials exhibited DC in the experimental arm. Of the 8 trials with DC, 6 (75%) showed PEP superiority, compared to 8 out of 29 trials (28%) that did not exhibit DC (Fisher’s P=.035). More specifically, 4 out of 5 (80%) trials that exhibited DC in the control arm showed PEP superiority, compared to 10 out of 32 (31%) that did not (P=.057). Only 2 out of 29 trials exhibited OS superiority, for which both trials exhibited PEP control arm DC (P=.015). Trials with at least one KM pair showing control arm DC exhibited a longer enrollment to publication time compared to those without control arm DC (median 16 vs 12 years, Mann-Whitney P=.044).Conclusion:
In RTOG and TROG clinical trials, DC, especially control arm DC, may be associated with experimental arm superiority in both surrogate PEP and secondary OS endpoints. This suggests a potential source of bias in trials showing superiority, although the present analysis is limited by small numbers and univariable analysis. Regardless, improved efforts to minimize loss of follow up and informative censoring by investigators, particularly in trials involving radiotherapy, may reduce bias in interpreting surrogate and OS outcomes.