Anubha Gupta1, Olivier Collignon2, Sofia Fernandes3, Monica Simeoni 1, Jaap Mandema4
1 Clinical pharmacology modelling and simulation, GSK UK; 2Clinical Stats, GSK, UK; 3Clinical Science GSK, UK; 4Certara USA, Inc., Princeton NJ
Objectives: Axial spondyloarthritis (axSpA) is a chronic inflammatory disease characterised by inflammation of the sacroiliac joints and spine. Clinical response is assessed using binary endpoints like ASAS40 (common primary endpoint in registration trials) and continuous endpoints like BASDAI and back pain [1]. The objectives of the work presented were 1) to quantify the relationship between the magnitude of early (e.g. week 4 or 8) treatment effect and at later time points (week 16), as compared to placebo across different mechanisms of action and 2) Use the modelled trajectory of the treatment effect over time to inform the design of a randomised controlled trial (RCT) in axSpA and benchmark the operating characteristics of an early futility analysis.
Methods: Certara’s clinical trial outcome database was used as data sources in the analysis (supported by GlaxoSmithKline) [2]. The dataset consisted of the time course of summary level ASAS40, BASDAI and back pain change from baseline from all active and placebo-controlled trials that evaluated TNF, IL-6, IL-17, IL-12/23, and JAK inhibitors. The data was analysed using random effects meta-regression models. Different models for the onset of treatment effect relative to placebo were tested. This included no time course, exponential, sigmoidal onset or a nonparametric model. The covariates assessed were drug class, percent radiographic-axSpA patients and TNF experience in total axSpA population. Change in BASDAI score was analysed jointly with ASAS40. A scaling factor between the magnitude of treatment effect for the two endpoints was estimated. Effect of time and drug class on the scaling factor allowed different speed of onset between the endpoints and variation with drug class. A similar analysis strategy was used for the joint analysis of back pain and ASAS40. A simulation framework was developed to estimate the impact of different assumptions and design features on the timing and operating characteristics of an early futility analysis based on the time-course of the treatment effect on ASAS40.
Results:
A total of 42 RCTs that evaluated TNF (adalimumab, certolizumab, etanercept, golimumab, infliximab), IL-6 (sarilumab, tocilizumab), IL-17 (secukinumab, ixekizumab, bimekizumab), IL-12/23 (ustekinumab), and JAK inhibitors (tofacitinib, upadacitinib, filgotinib) were included.
A non-parametric model was used to estimate the onset of treatment effect (ASAS40 difference from placebo). The model estimated the fractional change in treatment effect for separate time windows after start of treatment; week 1, 2, 3-4, 6-8, 10-12, 18-20, and >22 anchored to the effect at week 14-16. There was a statistically significant (p<0.001) impact of time on the ASAS40 treatment effect. The ASAS40 response was on average significantly lower at week 1 (-59% [-95 to -53]), week 2 (-45% [-51 to -39]), week 4 (-36% [-42 to -30]), and week 8 (-8.4% [-16 to -0.5]). There was a statistically significant (p<0.001) mechanism by time course interaction with TNF and JAK inhibitors showing a slower onset compared to IL-17 inhibitors. The percent of r-axSpA patients and TNF experienced patients did not significantly affect treatment effect.
There was a close correlation between ASAS40, BASDAI and back pain. There was no significant impact of time and drug class on the scaling factor between endpoints (p=0.065).
Based on the results above, an efficacious treatment for which a true treatment difference greater than 20% on top of placebo is expected at Week 16, should achieve a true treatment difference of at least 12% on top of placebo by week 4. Conducting a futility analysis enabling to stop the study if there is high chance (e.g. 90%) not to achieve this milestone would therefore allow to significantly derisk the program early using Week 4 data.
Trial simulations showed such an early futility analysis can offer strong operating characteristics, which were also robust across various assumptions around the speed of onset treatment effect. Increasing the sample size could improve the operating characteristics of the futility analyses but would trigger delay in decision-making. There was no benefit of the use of BASDAI or back pain compared to the ASAS40 endpoint.
Conclusions: Model based meta-analysis based on public domain data enable quantitative understanding of important factors like time course, patient characteristics and endpoints correlations to support design of RCTs.
References:
- Dougados, K-G Hermann, R Landewé, W Maksymowych and D van der Heijde J Sieper, M Rudwaleit, X Baraliakos, J Brandt, J Braun, R Burgos-Vargas, M; The Assessment of SpondyloArthritis international Society (ASAS) handbook: a guide to assess spondyloarthritis. Ann Rheum Dis 2009;68;ii1-ii44
- https://www.certara.com/data-and-informatics/codex-clinical-trial-outcomes-databases/. Accessed 06/04/2022
Reference: PAGE 30 (2022) Abstr 10172 [www.page-meeting.org/?abstract=10172]
Poster: Methodology - Study Design