10044 Background: Accurate and reproducible response assessment is critical for evaluating efficacy in clinical trials. Central response review is frequently used to reduce inter-site variability but is resource intensive and may delay data availability. Whether centralized review meaningfully alters response endpoints compared with remote source verification of site assessment remains unclear. Methods: We evaluated response assessments from eight NANT phase 1–2 trials conducted between February 2013 and December 2022 in relapsed/refractory high-risk neuroblastoma. Initial investigator-assessed response, remote source-verified response (remote review of radiology/bone marrow BM reports), and centrally-reviewed responses (review of radiographic images and BM biopsy slides) were assessed. Each patient’s best overall response (BOR) and individual component responses (BM, MIBG/bone, and soft tissue ST) at time of BOR as determined by these three review methods were assessed. Only patients with MIBG-avid disease were included in bone response. Results: Among 244 evaluable patients, concordance between remote source-verified response and central review response exceeded the concordance between initial investigator-assessed response and central review response (BOR: 89% vs 79%, BM: 91% vs 74%, MIBG/bone: 92% vs 83%, ST: 82% vs 74%). Concordance between initial investigator-assessed response and remote source-verified response was 86% for BOR and similar for component response (BM: 83%, MIBG/bone: 91%, ST: 86%). Remote source-verified assessment compared to central review demonstrated higher level of concordance than initial investigator assessment for complete response (CR) (90.0% vs 81.0%), partial response (PR) (85.7% vs 72.7%), minor response (MR) (80.6% vs 71.0%), and stable disease (SD) (92.1% vs 86.7%), indicating improved response classification consistency across all major BOR categories. Overall, the initial investigator-assessed response for CR, PR, and MR agreed with central review 74.0% of the time, while remote source-verified agreed 84.7%. Most discordance reflects relabeling within adjacent categories (e.g., CR vs PR or PR vs MR). A change in response from CR/PR/MR to SD/PD when compared to central review was uncommon: 9.4% for initial investigator-assessed response, and decreased with remote source verification down to 7.1%. A change from CR/PR to MR/SD/PD was more frequent but also improved with remote source verification (from 20% to 11.3%). Conclusions: Remote source verification substantially improves agreement with central review and reduces responder misclassification. These findings demonstrate that structured remote source verification improves response assessment reliability and harmonization with central review, reinforcing its role as an important quality-control mechanism in multicenter trials.
Migotsky et al. (Wed,) studied this question.