Discordance in pleural mesothelioma response classification and modelling of impact on clinical trials

ABSTRACT Introduction Agreement between radiologists regarding treatment response in Pleural Mesothelioma (PM) is acknowledged to be poor, but downstream effects in clinical trials have not been quantified. Methods We performed a mixed methods study, composed of a multicentre, retrospective cohort study and in silico modelling. CT images and data were retrieved from 4 UK centres regarding chemotherapy-treated patients. Expert radiologists classified response using modified Response Evaluation Criteria In Solid Tumours criteria (mRECIST) v1.1, generating discordance rate (%) and agreement. In silico modelling simulated two-arm trials of an active therapy with intended 80% power and confidence intervals for four endpoints (objective response rate (ORR), disease control rate (DCR), progression-free survival (PFS), overall survival (OS)) covering 95% of the true effect. Actual power and endpoint coverage were modelled against mRECIST misclassification rate (a single reporter equivalent of discordance rate). Consecutive simulations varied misclassification rate from 0-100% in 1% increments, each repeated 10,000 times. Results 172 cases were included. Discordance rate was 35% (60/172), kappa=0.456. In silico modelling demonstrated reduced power and endpoint precision with increasing misclassification. At 17% misclassification, corresponding to the observed 35% discordance, power dropped from 80% to 55% for ORR, 53% for DCR, 65% for PFS and 66% for OS, with endpoint coverage reduced to 88%, 89%, 92% and 92%, respectively. 50/60 (83%) discordances reflected interpretation or measurement differences intrinsic to mRECIST. Discordance was not associated with tumour volume. Conclusions Inconsistent response classification is common in PM and substantially reduces statistical power and endpoint precision in clinical trials.