Main factors associated with variability in the assessment of the objective response rate. A re-analysis of the PHEREXA phase 3 clinical trial

Aim: Local investigators (LIs) overestimate the objective response rate (ORR) in comparison to Blinded Independent Central Reviewers (BICR) in oncology. In this study, we re-analysed data obtained in the PHEREXA trial (NCT01026142) with the following aims: i) to confirm at the single-patient level the discrepancy observed by analysing aggregated data; ii) to investigate the causes underlying such discrepancy. Methods: Individual data, available on the VIVLI platform, were analysed. For each patient, the best response was established for LIs and BICR. "Eligible" patients were those with at least one target lesion at screening. The degree of agreement was assessed through the Cohen's Kappa coefficient. Multivariate logistic regression analysis was performed to understand the reasons underlying the discrepancy. Results: Eligible subjects were 364/452 according to LIs and 371/450 per BICR. We confirmed that LIs overestimate the ORR. We found moderate agreement between the two evaluations (Cohen's kappa= 0.48, p < 0.001). The probability of discrepancy increased by 20% with the increasing number of target lesions evaluated. Selection of 8-10 lesions was associated with an increased risk of discrepancy (OR, 5.27; CI 95% = 1.21-24.92; p < 0.03). The evaluation of breast, lung and lymph nodes (only) significantly increased the probability of discrepancy in comparison to the analysis of lymph nodes + organ. Conclusion: The overestimation of the tumour response by LIs increases with the number of target lesions analysed as well as with specific evaluation sites. Thus, the adoption of the two evaluations appears necessary in uncontrolled trials to provide a better estimate of tumour response.