In silico trials platform for exploration of trial design space in atopic dermatitis
Simon Arsène, Natacha Go, Diane Lefaudeux, Yishu Wang, Loic Etheve, Claudio Monteiro, Caterina Sansone, Christian Pasquali, Lorenz Lehr, Alexander Kulesza
Novadiscovery, OM Pharma
Objectives:
Atopic dermatitis (AD) clinical trials are stymied by a complex pathophysiology characterized by a wide spectrum of endotypes and phenotypes, and many possibilities for trial design driven by complicated clinical management such as rules for rescue treatment or choices of washout periods [1]. Trial design in AD is thus not spared from the issue of the high number of dimensions and parameters involved. This is why, in this context, mechanistic in silico approaches seem well suited to guide clinical trial optimization by allowing in-depth exploration of trial design space.
Methods:
We developed a knowledge-based computational model of AD representing the interplay between the skin barrier and the immune system in a pediatric population. We focused on developed pathological states, neglecting the natural history of the disease. This allows to integrate a wide spectrum of AD endotypes and phenotypes for which the model successfully reproduced responses to standards of care (emollient and topical corticosteroids) as well as responses to the oral immunomodulating bacterial lysate OM-85 [2].
We used this trial simulation setup to interrogate a large ensemble of potential trial designs, informing the n-dimensional space of design choices with predicted trial power and effect size. Intersecting, these two constructed n-dimensional surfaces allowed us to identify optimal design choices in an otherwise conceptually intractable space.
Results:
We simulated a large ensemble of 6075 trial designs in order to identify optimal trial design choices. To construct this trial design space, 18 trial parameters were varied according to various levels for each parameter: duration of treatment (2, 6, or 9 months), number of days of administration at the beginning of each treatment month (1, 10 or 30 days), topical corticosteroid (TCS) induction phase length (2, 8 or 32 days), frequency of TCS administration during maintenance (every 3, 7 or 14 days), included patients’ disease severity (mild, moderate or severe) and the level of 12 skin biomarkers (one-at-a-time, low or high). For each point in the trial design space, simulations of 1000 randomized placebo-controlled trials with 100 patients per arm were conducted, sampled from a pre-run database of more than 100 000 simulated virtual patients.
The distribution of absolute benefit (mean severity difference between placebo and treated group) for the 6075 trial designs displays a heavy right tail showing that a fraction of trial designs are able to capture a larger clinical benefit. A threshold at 5 SCORAD-equivalent unit improvement is defined as the minimal absolute benefit which can be considered clinically relevant. This value is chosen arbitrarily as a proof-of-context and can be adapted for the specific context of interest. The distribution of empirical power is the second important measure of trial success and is centered around low power values with a right tail showing that a large proportion of trial designs are not sufficiently powered with the chosen realistic sample size. A threshold of minimal power is set to 0.9 for this analysis. A candidate optimal trial design should satisfy both thresholds on absolute benefit and on power. Intersection of the two selected ensembles results in the selection of 364 trial designs over the 6075 initial designs.
Conclusions:
Here, we show how the combination of mechanistic disease and treatment modeling with high computing capacity can be used to exhaustively explore the high dimensional space of trial parameters. Contrary to purely statistical models, here this search is not limited to trial parameters which do not impact treatment efficacy. For example, dosing regimen or patient characteristics considerations can be included. As a result, this represents a powerful tool for model-assisted drug development especially in the field of complex immune diseases such as AD.
References:
[1] Silverberg, J. I., Simpson, E. L., Armstrong, A. W., de Bruin-Weller, M. S., Irvine, A. D., & Reich, K. (2022). Expert Perspectives on Key Parameters that Impact Interpretation of Randomized Clinical Trials in Moderate-to-Severe Atopic Dermatitis. American journal of clinical dermatology, 23(1), 1–11. https://doi.org/10.1007/s40257-021-00639-y
[2] Bodemer, C., Guillet, G., Cambazard, F., Boralevi, F., Ballarini, S., Milliet, C., Bertuccio, P., La Vecchia, C., Bach, J. F., & de Prost, Y. (2017). Adjuvant treatment with the bacterial lysate (OM-85) improves management of atopic dermatitis: A randomized study. PloS one, 12(3), e0161555. https://doi.org/10.1371/journal.pone.0161555