A Model Based Meta-Analysis for Bridging Treatment Doses of Rheumatoid Arthritis with Axial Spondyloarthritis
Monica Simeoni, Jaap Mandema, Stefano Zamuner, Anubha Gupta
GlaxoSmithKline, Certara
Objectives: Anti-TNF, anti-IL17 and JAK inhibitors are the drug classes which have shown efficacy in both rheumatoid arthritis (RA) and axial spondyloarthritis (axSpA). AxSpA includes radiographic and non-radiographic axSpA. The r-axSpA is also called as ankylosing spondylarthritis (AS). In general, wider dose range is studied for in RA compared to axSpA. A model based meta-analysis (MBMA) has been conducted on summary level data with the purpose of bridging doses for the treatment of RA with axSpA. In particular, the objective was to compare the dose–response relationship for the efficacy endpoints ACR20 and ASAS20 for RA and AS respectively, with the ultimate aim of supporting an expedite drug development of first in class compounds in a new indication (i.e. axSpA).
Methods: TCertara’s clinical trial outcome database for RA and AS was utilized [1]. All trials which included anti-TNF drugs (i.e. adalimumab, etanercept, golimumab, certolizumab and infliximab), anti-IL-17 drugs (i.e. ixekizumab and secukinumab) and JAK inhibitors drugs (i.e. tofacitinib and filgotinib), either as primary drug, or as a control arm with placebo, and reported ACR20 and ASAS20 results, were part of the analysis dataset. In total 79 and 25 trials were included for RA and AS respectively.
ACR20 and ASAS20 data were analyzed using a nonlinear regression for binary endpoints, implemented in the generalized nonlinear least squares (gnls) and nonlinear mixed-effects (nlme) routines in R (version 3.6.1) [2].
The number of patients achieving response y (ACR20 or ASAS20) in the jth treatment arm of the ith study (NACR20/ASAS20ij) was assumed to follow a binomial distribution according to the probability of the event P(ACR20 or ASAS20)ij and the sample size Nij.
In a first model P(ACR20 or ASAS20)ij was described as a function of a study specific placebo response (E0,i) and a dose–response relationship for the treatment effect:
P(ACR20 or ASAS20)ij=logit-1(E0,i+Emax,k·Doseij/(exp(ED50,h)+Doseij)
E0,i represents placebo response for each study and accounts for trial-to-trial variability in overall response. A similar maximal drug effect, Emax, was assumed for drugs with the same mechanism of action k [2], but a separate Emax was estimated for each endpoint (ACR20 and ASAS20). Dose is the total daily/weekly/monthly dose normalized to the standard regimen for each drug, and ED50 estimated (in log domain) for each drug, h, and each endpoint is the dose required to achieve 50% of Emax.
In a second model P(ASAS20)ij was modified such that the ED50 for ASAS20 was derived by estimating an ED50 ratio between the two endpoints:
P(ASAS20)ij=logit-1(E0,i+Emax,k,ASAS20·Doseij/(exp(ERh+ED50,ACR20,h)+Doseij)
where ERh is the log of the ED50 ratio for the drug class h, by which ED50 of each drug varies between the RA and AS indications. An exp(ER) of 1 will indicate same dose-response for a drug in RA and AS.
Models were selected based on the model fit to the observed data, and likelihood ratio test.
Results: Both models were applied first to the anti-TNF compounds, which encompass the majority of approved doses. Both described the observed data well and the model-predicted profiles were similar for AS and RA indications, except for certolizumab. This could be due to limited data for certolizumab in AS and comparatively higher placebo response in that trial. When excluding the certolizumab data the two models were not significantly different (p = 0.2131). An analysis also including anti-IL17 and JAK inhibitors was then conducted with the second model. The confidence interval for the estimate of ED50 ratio, includes 1 for the three mechanisms indicating that there is no evidence of ED50 for a specific drug being different for RA and AS. The parameter estimates of ER have high uncertainty, but this is likely due to limited data available in AS compared to RA.
Conclusions: TWhen agents with the same mechanism of action are efficacious in related diseases, there is an implicit assumption that dose-requirements are also similar. This is based on pharmacological principles that similar levels of target inhibition are required to elicit a clinical response due to similar pathophysiology. This MBMA supports this principle, and the evaluation of the same dose regimen(s) in axSpA, without conducting a dose ranging study, provided that the risk of the drug class not being efficacious in the new indication is appropriately discharged.
Disclosure: MS, SZ and AG are GSK employees and have stocks in the company, JM is employed by Certera and provided consulting service for the work. This study was sponsored by GSK
References:
[1] https://www.certara.com/data-and-informatics/codex-clinical-trial-outcomes-databases/
[2] Mandema JW, Gibbs M, Boyd RA, Wada DR, Pfister M (2011). Model-based meta-analysis for comparative efficacy and safety: application in drug development and beyond. Clin Pharmacol Ther. 90(6):766-9.
