Generative Artificial Intelligence Augmentation of a Clinical Trial of a Dermo-cosmetic Intervention for Acne Remission Maintenance
November 2026 | Volume 25 | Issue 11 | 1014 | Copyright © November 2026
Published online October 29, 2026
Marco Ardigò PhDa,b, Luigi Gargiulo MDa,b, Raphael Fontaine MSc,c, Olivier Thuillart MSc,c, Jean-Louis Fraysse PhDc, Adeline Bacquey MSc,d, Pauline Lacoste MSc,d, Stéphanie Allassonnière PhDe,f, Aline Stennevin MSc,g, Gautier Doat MDg
aDermatology Unit, IRCCS Humanitas Research Hospital, Rozzano (MI), Italy
bDepartment of Biomedical Sciences, Humanitas University, Pieve Emanuele (MI), Italy
cBOTdesign, Toulouse, France
dClinical Skin Research and Development Center, Pierre Fabre Dermo-Cosmétique & Personal Care, Toulouse, France
eHeKA INSERM, INRIA Paris, Centre de Recherche des Cordeliers Paris, Université Paris Cité, Paris, France
fUniversité Paris Cité, UFR Médecine, Paris, France
gEau Thermale Avène, Pierre Fabre Dermo-Cosmetics & Personal Care, Lavaur, France
Abstract
Background: A practical application of artificial intelligence (AI) in healthcare is to generate artificial datasets that capture the characteristics of a real source dataset without exposing protected health information. This approach is particularly relevant to research settings where recruitment and long-term follow-up are barriers to generating robust clinical evidence.
Objective: To describe the protocol of a study incorporating augmented patient datasets to assess the efficacy of a dermo-cosmetic product (Cleanance Comedomed+®) in maintaining acne remission after discontinuation of oral isotretinoin.
Methods: In a randomized, placebo-controlled, blinded clinical trial, patients will receive the dermo-cosmetic product or placebo for up to 12 months. Study outcomes are the number of observed relapses and time to first relapse. To overcome recruitment constraints and reach the 400 patients required to ensure sufficient precision for a time-dependent event criterion, a cohort augmentation strategy using artificial data is implemented. The technology relies on generative models of the Riemannian Hamiltonian Variational Autoencoder type, which learn a compact and geometrically structured latent representation of real clinical data. Generation of artificial patients will occur iteratively, and the initial dataset will be progressively enriched as new real inclusions become available. Reliability of generated data will be assessed and validated through human oversight using quantitative and qualitative metrics covering 4 dimensions: (1) statistical fidelity; (2) diversity and utility; (3) confidentiality; and (4) clinical plausibility.
Discussion and Conclusion: Structured integration of artificial data into this clinical trial illustrates the potential of generative AI as a methodological tool complementing conventional approaches in clinical research.
INTRODUCTION
Generative artificial intelligence (AI) is increasingly being adopted in healthcare settings to accelerate innovation.1 Realistic artificial datasets can be generated to capture the characteristics of a real source dataset (distributions, non-linear relationships, noise) without exposing protected health information.2,3 Artificial data can overcome patient recruitment difficulties and mitigate the impact of prolonged follow-up and low-frequency or stochastic events already observed in the real dataset by stabilizing their statistical representation, while providing statistical power equivalent to that obtained with exclusively real data. Probabilistic and generative models have been shown to faithfully reproduce not only marginal distributions, but also multivariate dependencies and certain clinical trajectories observed in limited datasets.3-7
The medicalization of dermo-cosmetic products requires the production of robust, comparative, and reproducible clinical evidence. In the setting of recalcitrant moderate to severe acne, systemic treatment with oral isotretinoin remains the therapeutic gold standard.8,9 However, a substantial proportion of patients relapse after treatment discontinuation,10,11 underscoring the need to develop effective strategies for the post-isotretinoin maintenance phase. As reasons for relapse are myriad and individual,11,12 demonstrating the efficacy of dermo-cosmetic
