Expected versus observed effect sizes for survival endpoints in phase 3 oncology trials
Menée à partir d'une revue systématique de 111 essais de phase III en cancérologie ayant fait l'objet d'une publication en 2023 dans l'une des 10 meilleures revues médicales, cette étude met en évidence, pour les critères d'évaluation de la survie, une surestimation de la taille de l'effet attendu par rapport à la taille de l'effet observé
More than half of randomized phase 3 oncology trials fail to meet their primary endpoints, often despite favorable results from phase 1 and phase 2 trials. One potential reason for this high failure rate is effect size selection practices for powering these trials. In a systematic review of phase 3 oncology trials published in ten top medical journals in 2023, we identified a pattern of effect size overestimation for survival endpoints, where the expected hazard ratios (average HR = 0.66) were stronger than those observed in the primary analyses (average HR = 0.72; two-sided signed-rank test p = .0035). Across 111 trials, 82 of 143 observed hazard ratios for primary survival endpoints (57.3%) were weaker-than-expected; among five journals with 2023 impact factors between 9.9-56.7 (vs 58.7-98.4 for the top five journals), this ratio was 70.2% (59/84). These results suggest that phase 3 oncology trials are likely underpowered, contributing to the high failure rates in oncology research.
Journal of the National Cancer Institute , résumé, 2025