Association of PD-L1 Expression and Other Variables With Benefit From Immune Checkpoint Inhibition in Advanced Gastroesophageal Cancer: Systematic Review and Meta-analysis of 17 Phase 3 Randomized Clinical Trials.

Importance: Approval by the US Food and Drug Administration of immune checkpoint inhibition (ICI) for advanced gastroesophageal cancer (aGEC) irrespective of PD-L1 status has generated controversy. Exploratory analyses from individual trials indicate a lack of meaningful benefit from ICI in patients with absent or low PD-L1 expression; however, analysis of a single variable while ignoring others may not consider the instability inherent in exploratory analyses. Objective: To systematically examine the predictive value of tissue-based PD-L1 status compared with that of other variables for ICI benefit in aGEC to assess its stability. Data Sources: MEDLINE, Embase, Scopus, Web of Science, Cochrane Central Register (2000-2022). Study Selection, Data Extraction, and Synthesis: Randomized clinical trials (RCTs) were included of adults with aGEC (adenocarcinoma [AC] or squamous cell carcinoma [SCC]) randomized to anti-PD-1 or PD-L1-containing treatment vs standard of care (SOC). Study screening, data abstraction, and bias assessment were completed independently by 2 reviewers. Of 5752 records screened, 26 were assessed for eligibility; 17 trials were included in the analysis. Main Outcomes and Measures: The prespecified primary end point was overall survival. The mean hazard ratio (HR) for ICI vs SOC was calculated (random-effects model). Predictive values were quantified by calculating the ratio of mean HRs between 2 levels of each variable. Results: In all, 17 RCTs (9 first line, 8 after first line) at low risk of bias and 14 predictive variables were included, totaling 11 166 participants (5067 with SCC, 6099 with ACC; 77.6% were male and 22.4% were female; 59.5% of patients were younger than 65 years, 40.5% were 65 years or older). Among patients with SCCs, PD-L1 tumor proportion score (TPS) was the strongest predictor of ICI benefit (HR, 0.60 [95% CI, 0.53-0.68] for high TPS; and HR, 0.84 [95% CI, 0.75-0.95] for low TPS), yielding a predictive value of 41.0% favoring high TPS (vs ≤16.0% for other variables). Among patients with AC, PD-L1 combined positive score (CPS) was the strongest predictor (after microsatellite instability high status) of ICI benefit (HR, 0.73 [95% CI, 0.66-0.81] for high CPS; and HR, 0.95 [95% CI, 0.84-1.07] for low CPS), yielding a predictive value of 29.4% favoring CPS-high (vs ≤12.9% for other variables). Head-to-head analyses of trials containing both levels of a variable and/or having similar design generally yielded consistent results. Conclusions and Relevance: Tissue-based PD-L1 expression, more than any variable other than microsatellite instability-high, identified varying degrees of benefit from ICI-containing therapy vs SOC among patients with aGEC in 17 RCTs.

Successes and lessons learned in database development for national multi-site cancer care delivery research trials: the Alliance for Clinical Trials in Oncology experience.

INTRODUCTION: Alliance for Clinical Trials in Oncology (Alliance) coordinated trials utilize Medidata Rave® (Rave) as the primary clinical data capture system. A growing number of innovative and complex cancer care delivery research (CCDR) trials are being conducted within the Alliance with the aims of studying and improving cancer-related care. Because these trials encompass patients, providers, practices, and their interactions, a defining characteristic of CCDR trials is multilevel data collection in pragmatic settings. Consequently, CCDR trials necessitated innovative strategies for database development, centralized data management, and data monitoring in the presence of these real-world multilevel relationships. Having real trial experience in working with community and academic centers, and having recently implemented five CCDR trials in Rave, we are committed to sharing our strategies and lessons learned in implementing such pragmatic trials in oncology. METHODS: Five Alliance CCDR trials are used to describe our approach to analyzing the database development needs and the novel strategies applied to overcome the unanticipated challenges we encountered. The strategies applied are organized into 3 categories: multilevel (clinic, clinic stakeholder, patient) enrollment, multilevel quantitative and qualitative data capture, including nontraditional data capture mechanisms being applied, and multilevel data monitoring. RESULTS: A notable lesson learned in each category was (1) to seek long-term solutions when developing the functionality to push patient and non-patient enrollments to their respective Rave study database that affords flexibility if new participant types are later added; (2) to be open to different data collection modalities, particularly if such modalities remove barriers to participation, recognizing that additional resources are needed to develop the infrastructure to exchange data between that modality and Rave; and (3) to facilitate multilevel data monitoring, orient site coordinators to the their trial's multiple study databases, each corresponding to a level in the hierarchy, and remind them to establish the link between patient and non-patient participants in the site-facing NCI web-based enrollment system. CONCLUSION: Although the challenges due to multilevel data collection in pragmatic settings were surmountable, our shared experience can inform and foster collaborations to collectively build on our past successes and improve on our past failures to address the gaps.