Jacob Leander (1), David Ramsay (2*), Eduard Molins (3), Mark Pilling (4), José Sánchez (5), Sofia Tapani (2), Ulrika Wählby Hamrén (1)
(1) Clinical Pharmacology and Quantitative Pharmacology, Clinical Pharmacology and Safety Sciences, R&D, AstraZeneca, Gothenburg, Sweden (2) R&I Biometrics & Statistical Innovation, Late R&I, BioPharmaceuticals R&D, AstraZeneca, Gothenburg, Sweden (3) R&I Biometrics & Statistical Innovation, Late R&I, BioPharmaceuticals R&D, AstraZeneca, Barcelona, Spain (4) R&I Biometrics & Statistical Innovation, Late R&I, BioPharmaceuticals R&D, AstraZeneca, Cambridge, UK (5) CVRM Biometrics, Late CVRM, BioPharmaceuticals R&D, AstraZeneca, Gothenburg, Sweden (*) Statistical contractor
Introduction: The primary analysis of a Ph2b dose-finding study is typically based on pairwise statistical tests of the study arms, using for example a mixed model for repeated measures (MMRM) approach. Dose-response analysis provides the potential to increase the precision in estimates, which would result in better informed Ph3 dose selection, and allows for optimization of other design aspects like sample size, doses to be explored and allocation ratios.
A promising approach for analysing Ph2b dose-finding trials is dose-response MMRM (DR-MMRM), introduced by Wellhagen et al. [1]. The DR-MMRM method makes limited assumptions about the shape of the longitudinal response while still borrowing information across doses.
Objectives:
- To compare the performance of DR-MMRM and MMRM for a Ph2b study design in rheumatoid arthritis (RA) with DAS28-CRP as endpoint using simulations
Methods: A longitudinal simulation model for DAS28-CRP was constructed. This model consisted of 4 components: placebo model, drug effect model, between-subject variability and within-subject variability. The placebo model was constructed from published data together with placebo data from four clinical trials in RA from the TransCelerate database [2]. The drug effect was described by an Emax model, where a drug effect of -1.2 units was assumed to be achieved at Week 12 at a dose of 300 mg [3]. To reflect an unknown true dose-response relationship, ED50 was varied from 4 to 128 mg. The total variability in change from baseline DAS28-CRP was separated into between and within-subject. The between-subject variability was modelled using an additive normal distributed random effect, and the within-subject variability structure was modelled using an AR(1) process parametrized by correlation parameter ρ and standard deviation σ.
Four clinical study designs were simulated, with varying dose-range and number of doses. Together with 6 different ED50s, this yielded a total of 24 simulation scenarios. The same sampling time points (Week 1, 2, 4, 6, 8, 12) were used across designs, and the same sample size of 26 subjects per arm was used (based on standard sample size calculations).
For each simulation scenario, 2000 virtual studies were simulated. The performances of the different methods (MMRM and DR-MMRM) were evaluated using absolute bias, standard deviation of estimate and root-mean square error (RMSE). In addition, the percentage of successful estimations (i.e., model convergence) for both methods were evaluated. The MMRM and DR-MMRM model were estimated using the nlme package in R [4].
Results: The MMRM and DR-MMRM method yielded similar mean estimates across simulation scenarios. As expected, MMRM had a 100% success rate. DR-MMRM had a success rate ranging between 84-100%, with numerical problems emerging for the least informative study design.
The bias introduced by DR-MMRM was in general low and dependent on ED50 and study design. DR-MMRM yielded higher precision of the effect size estimate, with a corresponding improvement in RMSE compared to the MMRM approach across all simulation scenarios. For the 300 mg dose at Week 12, the statistical power increased from 83% using MMRM to 94-97% using the DR-MMRM approach.
Conclusions: This analysis further supports the value of using DR-MMRM for the analysis of Ph2b dose-finding studies, here exemplified with DAS28-CRP as clinical endpoint. By leveraging information across doses, the precision of effect size estimates is increased which enables better informed dose selection for Phase 3.
References:
[1] Wellhagen, G.J., Hamrén, B., Kjellsson, M.C. et al. (2020). Dose-Response Mixed Models for Repeated Measures – a New Method for Assessment of Dose-Response. Pharm Res 37, 157. https://doi.org/10.1007/s11095-020-02882-0
[2] Buckley, C. D., Simón-Campos, J. A., Zhdan, V., Becker, B., Davy, K., Fisheleva, E., Gupta, A., Hawkes, C., Inman, D., Layton, M., Mitchell, N., Patel, J., Saurigny, D., Williamson, R., & Tak, P. P. (2020). Efficacy, patient-reported outcomes, and safety of the anti-granulocyte macrophage colony-stimulating factor antibody otilimab (GSK3196165) in patients with rheumatoid arthritis: a randomised, phase 2b, dose-ranging study. The Lancet. Rheumatology, 2(11), e677–e688. https://doi.org/10.1016/S2665-9913(20)30229-0
[3] England BR, Tiong BK, Bergman MJ, Curtis JR, Kazi S, Mikuls TR, O’Dell JR, Ranganath VK, Limanni A, Suter LG, Michaud K. (2019). Update of the American College of Rheumatology Recommended Rheumatoid Arthritis Disease Activity Measures. Arthritis Care Res (Hoboken). 2019 Dec;71(12):1540-1555. https://doi.org/10.1002/acr.24042
[4] Pinheiro J, Bates D, R Core Team (2023). nlme: Linear and Nonlinear Mixed Effects Models. R package version 3.1-164, https://CRAN.R-project.org/package=nlme
This publication relies on data from TransCelerate’s Historical Trial Data Sharing Initiative, to which AbbVie, Allergan, Amgen, AstraZeneca, Astellas, Boehringer Ingelheim, Bristol Myers-Squibb, Eli Lilly, EMD Serono, GSK, Johnson & Johnson, Novo Nordisk, Pfizer, Roche, Sanofi, Shionogi, and UCB Pharma (“Data Providers”) have contributed data. None of those companies nor TransCelerate have contributed to, approved, or are in any way responsible for the author’s research results.
Reference: PAGE 32 (2024) Abstr 10803 [www.page-meeting.org/?abstract=10803]
Poster: Methodology - New Modelling Approaches