LAUSR

BACKGROUND AND AIM Trial designs using multiple primary endpoints (MPEs) are increasing in phase III cancer trials. Our objectives were to describe the incidence of MPEs in recently published phase III trials testing systemic treatments in patients with advanced cancer; the main characteristics of trials adopting MPEs; the presence of mature results for all endpoints in the primary publication; consistency between results of each endpoint and authors' conclusions. METHODS Articles of randomised phase III trials conducted in patients with advanced cancer, published between 2017 and 2020, were retrieved from PubMed. The main outcome was the proportion of trials with MPEs. In principle, according to regulatory agencies, we considered two distinct cases: (i) MPEs correspond to 'multiple chances' for the success of experimental treatment, needing adjustment for multiplicity, and (ii) a positive result depends on the success in all MPEs ('co-primary' endpoints). RESULTS Out of 235 eligible trials, 27 trials (12%) adopted MPE, mostly overall survival (OS) and progression-free survival (PFS). The proportion of trials with MPEs increased over time, from 6% in 2017 to 20% in 2020 (p = 0.025). MPEs were adopted in 16% of for-profit trials versus 4% of non-profit trials (p = 0.006). The proportion of trials adopting MPEs was particularly high with immunotherapy (53%, p < 0.00001). Out of 27 trials with MPEs, 10 (37%) adopted an explicit definition of 'co-primary' endpoints, but only 1/10 declared the positivity of both endpoints critical for interpretation. Most trials (23, 85%) planned correction for multiplicity. Of 21 publications with positive conclusions, only 12 had a statistically significant positive result in both primary endpoints. In four cases (15%), positive conclusions were based on PFS results alone. CONCLUSIONS Adoption of MPEs in randomised trials in oncology is quite common. Only a minority of trials respect recommendations by regulatory agencies about the adoption of MPEs, definition of 'co-primary' endpoints and correction for multiplicity.