Zrinka Duvnjak 1,2,3, Matt Zierhut, Ahmed A. Suleiman, Zhigang Wang 1
1 Sanofi-Aventis Deutschland GmbH (, Germany), 2 Dept. of Clinical Pharmacy and Biochemistry, Inst. of Pharmacy, Freie Universität Berlin (, Germany), 3 Graduate Research Training program PharMetrX (, Germany), 4 Certara (, USA)
Objectives: Asthma patients exhibit significant heterogeneity in disease presentation driven by involvement of type I (allergic) and type II (eosinophilic) inflammation. Currently, six biologics are approved for severe asthma, with indications often stratified by patient’s eosinophil counts. Model-based meta-analysis (MBMA) can provide valuable insights for efficacy benchmarking in different subpopulations, informing go/no-go decisions, and optimizing probability of success for emerging therapies. The aim of this work was to conduct the MBMA of biologics’ effects on longitudinal forced expiratory volume in 1 second (FEV1) change in severe asthma, with special emphasis on the impact of eosinophil levels on treatment efficacy.
Methods: Certara’s Asthma CODEx database (August 2025 update) was utilized for the analysis, containing publicly available summary-level randomized clinical trial results in severe asthma [1]. Initial analysis prioritized using dense longitudinal summary results with available population covariates; preferring stratified results but comprising mainly non-stratified data. An unstructured placebo approach within the drug-effect model was used to estimate placebo response by study and visit [2]. In addition, for simulation purposes, the longitudinal placebo effect was characterised parametrically in a separate mixed-effects placebo model. Multiple effect-onset and dose-response models were evaluated for each drug separately. Both fixed-effects and random effects assumptions were considered with different correlation-within-arm structures. Based on initial covariate search and heterogeneity evaluation results, if needed, the covariate analysis was reconducted on dataset with stratified study results, to broaden the covariate range per drug. Model selection was based on statistical significance, goodness of fit and plausibility of parameter estimates. Model-based arm-level simulations were performed to predict probability of success in virtual non-inferiority clinical trials [3]. Analysis was performed in R 4.4.1 using nlme package (version 3.1-168).
Results: The analysis encompassed 19 biologics from 44 studies (72 active arms, 10,904 patients), with variable study demographics, ranging in mean exacerbations: 0.30-5.27/year, eosinophil counts: 45.04-763.11 cells/μL, FEV1: 52.29-75.76% predicted. Emax effect onset was identified for tralokinumab and benralizumab (after accounting for placebo), reaching 90% maximum effect at week 17. The model included dose-response Emax model for dupilumab, linear for astegolimab, enokizumab. Two separate drug effects were estimated for two tralokinumab doses. Initial univariate covariate analysis showed statistically significant impact of baseline eosinophil on relative drug effect, however heterogeneity could not be identified. To obtain more reliable eosinophil effect estimates, study results stratified by eosinophil count were extracted from the database and covariate analysis was repeated. For patients with eosinophils ≥300 cells/μL, dupilumab showed the highest FEV1 improvements (0.24 L), followed by tezepelumab (0.18 L) and others ranging from 0.02 to 0.14 L. However, efficacy for dupilumab, tezepelumab, and benralizumab dropped by 70–76% in patients below the 300 cells/μL threshold. After inclusion of covariate effect, additional heterogeneity was not identified. Separate placebo model was a random-effect Emax model (Emax: 0.13 L, Et50: 1 week, sd: 0.10 L) also with a positive covariate effect of baseline eosinophils. Simulated non-inferiority trials comparing dupilumab versus tezepelumab (550 patients/arm) showed eosinophil-dependent outcomes. In patients with baseline eosinophils ≥300 cells/μL, dupilumab had a 60% probability of superiority and 40% non-inferiority, compared to only 25% superior and 15% non-inferior in patients with baseline eosinophils <300 cells/μL. Conclusion: This MBMA demonstrates the power of model-based approaches in integrating heterogeneous clinical trial data to quantify and compare treatment effects on FEV1 across diverse patient populations. By leveraging longitudinal efficacy data and systematically evaluating patient-level covariates, like eosinophils, MBMA enabled robust efficacy benchmarking across multiple biologics and probabilistic predictions of clinical trial outcomes. These findings illustrate how MBMA can support go/no-go decisions and optimize clinical trial design for emerging therapies in complex disease areas such as severe asthma References: [1] https://codex.certara.com/codex/ [2] Mawdsley et al. (2016), doi: 10.1002/psp4.12091 [3] FDA: Non-Inferiority Clinical Trials to Establish Effectiveness, Guidance for Industry
Reference: PAGE 34 (2026) Abstr 12301 [www.page-meeting.org/?abstract=12301]
Poster: Drug/Disease Modelling - Other Topics