Ruben Faelens(1), Martine Neyens(1), Huybrecht T’jollyn(1), Wilbert Van Duijnhoven(1), Tristan Baguet(1), Juan Jose Perez Ruixo(1)
(1) Janssen R&D, Belgium
Introduction Influenza pre-exposure prophylaxis (PrEP) drugs protect by ensuring efficacious drug concentrations at time of viral entry to inhibit virus replication. Current standard of care has a short half-life requiring daily dosing. Long-acting formulations are attractive as they may allow sustained exposure after one administration. However, depending on the half-life and the potency of the drug under investigation, doses should be well selected to offer protective concentrations throughout a 6-month influenza season.
In community-acquired infection (CAI) trials, variation in time of seasonal disease peak (early or late) makes it not only difficult to interpret the actual protective effect of the drug in the current study trial, but also to project the average effect over multiple seasons in a phase 3 study. Investigating protective effect of a compound during a season with early peak equates to testing the compound at high concentrations immediately post-administration, yielding no information on the effect at lower concentrations. Similarly, testing the compound in a season with a late peak does not inform about the drug effect at higher concentrations.
Through simulation, this work demonstrates a new methodology to interpret CAI trial results, optimizing the trial design and facilitating decision-making.
Methods DrugX is a long-acting injectable PrEP against influenza, with slow absorption yielding a C_max and C_6months of 15 to 2 ng/mL, respectively, for a single dose of 5 mg. PK was assumed known from a Phase 1 study. The probability of influenza infection was simulated through a survival model, with hazard Haz(t) = BASE(t) * (1 -C^γ/(C^γ + EC50^γ )), with γ=3 and base hazard 2% per year.
The disease incidence in virtual Phase 3 trials (N=6000 1:1-allocated intervention vs placebo) was simulated for EC50 ranging from 1 ng/mL to 60 ng/mL. For the candidate doses of 5 mg and 20 mg, the EC50 yielding 80% probability of study success (PoSS) in a proportions test evaluating superiority of treatment to placebo at p<0.05 was identified. This is the maximal EC50 allowed for that dose to ensure a successful Phase 3 trial and overall drug development success.
For single-season Phase 2 trials, hazard was scaled according to weekly incidence of influenza-like illness (ILI) reported in CDC FluView[1]. Clinical trials of size N=3000 and 1:1:1 (20 mg, 5 mg, placebo) allocation were simulated exploring (a) a range of EC50 from 1 ng/mL to 60 ng/mL, and (b) early or late seasonal influenza peak. Trial results were analyzed using two methods:
1. Using a proportions test, disease incidence per dose arm was compared to placebo arm for superiority at p<0.05. The lowest dose was selected.
2. EC50 was estimated[2] informed by individual time of disease, individual drug concentrations and known baseline hazard per week. Log-likelihood profiling was used to identify 95% CI. A dose was selected if the upper bound was below the maximum EC50 for development success of that dose. The program was stopped if the lower bound was above the maximal EC50 for development success.
Results Based on simulations of Phase 3 trials, the maximum EC50 for development success of 5 mg and 20 mg doses was EC50=7ng/mL and 28ng/mL , respectively. Beyond 28ng/mL, neither dose could ensure 80% PoSS.
Executing a proportions test on simulated Ph2 trial results, a wide variability in clinical trial results is apparent. Results are sensitive to the timing of influenza season peak. The error rate is high, e.g. stopping development in 62% of late season trials at EC50=20 ng/mL, even though a 20 mg dose would succeed in Ph3.
Using the individual PK profiles, time of individual disease occurrence, and known real-world disease incidence in each trial, EC50 could be re-estimated for each trial. All trials showed a significant drug effect. The strong sensitivity to disease season was mitigated for potent and medium drugs. For a low potency drug at EC50=60ng/mL, the method allowed to stop for futility with confidence in 79% of trials (early peak) and 57.2% of trials (late peak).
Conclusion Using simulations to predict Ph2 outcomes allows us to investigate the sensitivity of results to influenza seasons. EC50 can be estimated from a limited CAI trial by integrating real-world data, individual PK data and time of disease. This allows us to kill a short acting drug with confidence. The number of indecisive trials remains high.
References:
[1]: Charbonneau, Deborah H., and LaTeesa N. James. “FluView and FluNet: tools for influenza activity and surveillance.” Medical reference services quarterly 38.4 (2019): 358-368.
[2]: Crowther, M. J., Abrams, K. R., & Lambert, P. C. (2013). Joint Modeling of Longitudinal and Survival Data. The Stata Journal, 13(1), 165-184. https://doi.org/10.1177/1536867X1301300112
Reference: PAGE 32 (2024) Abstr 11254 [www.page-meeting.org/?abstract=11254]
Poster: Methodology - Study Design