Alexandra Smirnova (1), Boris Shulgin (1), Yuri Kosinsky (1), Gabriel Helmlinger (2), Kirill Peskov (1,3)
(1) M&S-Decisions LLC, Russia, (2) Clinical Pharmacology & Toxicology, Obsidian Therapeutics, Cambridge, MA, USA (3) Computational Oncology group, I.M. Sechenov First Moscow State Medical University, Russia
Objectives: Immune checkpoint inhibitors (ICIs) are promising anti-cancer therapies; yet, they are associated with immune-mediated adverse events (imAE), especially in combination therapies. The objective of this study was to apply a Bayesian model-based meta-analysis (BMA) towards the quantification of dose dependence of different AEs for PD-1 (nivolumab, pembrolizumab) and CTLA-4 (ipilimumab, tremelimumab) inhibitor monotherapies and combinations. In addition, as reported previously for standard meta-analyses, use of a statistical model with normal distributions for both inter- and intra-trial variabilities (normal-normal, N-N) can lead to biased results for rare events. In such cases, a binomial-normal (B-N) model should be used. We thus compared N-N and B-N models when studying rare imAEs for ICIs therapies.
Methods: We searched PubMed and TrialTrove databases, ASCO and ESMO abstracts to identify relevant ICI safety data. We introduced a novel meta-analysis methodology based on modeled drug exposure normalized by drug potency, for an integrative, comparative study across different doses and drugs. To explore dose dependence, we split the dosing interval into lower and higher dosing subgroups and compared their AE rates. The BMA was performed for all-grade and high-grade (Grades 3-4) treatment-related (trAEs) adverse events and for imAEs associated with specific organ groups (gastrointestinal, dermatological, hepatic, endocrine, pulmonary). Software STAN and the brms package in R software were used to perform the BMA. B-N and N-N models were compared using clinical trial data and simulated datasets.
Results: A total of 163 articles were identified, representing 33,887 patients in 199 dosing cohorts treated with ICI monotherapies or combinations. No relationship between AE rates and dose of anti PD-1 monotherapy was observed. In contrast, dose dependence was observed for total grade 3/4 trAEs (AE rates for lower and higher doses subgroups – 22% vs. 34%, respectively) and hepatic imAEs (1.4% vs. 7%) for CTLA-4 inhibitor monotherapy. Dose dependence was also observed for total grade 3/4 trAEs (34% vs. 53%), and for organ imAE (gastrointestinal (6% vs. 17%), hepatic (8% vs. 15%) and skin (3.5% vs. 5%)) for {CTLA-4 + PD-1} inhibitor combination therapy. AE rates for {CTLA-4 + PD-1} inhibitor combination regimens were supra-additive vs. respective monotherapies. Use of a N-N model led to significant overestimation (up to two fold) of rare imAEs rates, e.g., endocrine or pulmonary, while use of a B-N model gave accurate results confirmed by simulation and external validation exercises.
Conclusions: Significant AE dependences on ICIs dose/exposure for anti CTLA-4 monotherapy, as well as {anti CTLA-4 + anti PD-1} and {ICI + chemotherapy} combinations were observed for multiple AE types. The B-N model showed better performance and should be used for the meta-analysis of rare imAEs. This novel model-based meta-analysis methodology provides a quantitative framework for selecting ICI doses and dosing regimens with respect to specific AE rates.
Reference: PAGE () Abstr 9483 [www.page-meeting.org/?abstract=9483]
Poster: Oral: Drug/Disease Modelling