Real/artificial hybrid protocol: making the impossible possible in dermocosmetic trials
hybride

When recruiting 400 real patients is financially and time-wise out of reach, should the study be abandoned ?

Pierre Fabre and BOTdesign chose a third approach: co-designing, from the outset, a hybrid protocol integrating artificial patients, in order to obtain statistically richer results than the original study — within 12 months and at a controlled cost.

 

The structural challenge of recruitment in dermocosmetics

Patient recruitment is currently one of the main bottlenecks in clinical research, particularly in dermocosmetics. According to industry estimates, more than 80% of clinical studies experience delays due to insufficient recruitment, and the average cost of enrolling a patient in a randomized controlled trial is well above several thousand euros.

These constraints are particularly significant in indications such as post-isotretinoin acne: strict eligibility criteria, patient reluctance regarding the placebo arm, and long-term follow-up leading to attrition. In this context, many study projects are abandoned or scaled down before they have even started — not because of a lack of scientific relevance, but because it is practically impossible to assemble the required cohort within an acceptable timeframe and budget.

 

It is in response to this reality that generative AI offers a new approach: the generation of statistically consistent artificial patients, making it possible to design trials that match clinical ambitions without being constrained solely by recruitment limitations.

 

Origins and design of the Comedomed protocol

Pierre Fabre Dermo-Cosmétique (Avène brand) wanted to evaluate the efficacy of Comedomed — an anti-acne cosmetic product — used following oral isotretinoin treatment to prevent acne relapses. The post-isotretinoin relapse rate is estimated at between 15% and 20% of treated patients, fully justifying a comparative prevention trial.

Preliminary statistical calculations established that approximately 400 patients were needed to provide the study with the required statistical power. However, from the outset, Pierre Fabre’s teams identified that recruiting 400 real patients within a reasonable timeframe — due to the strict eligibility criteria, the placebo-controlled design, and the one-year follow-up period — would represent a financial and logistical challenge incompatible with the project’s constraints.

Rather than reducing the study’s ambitions or abandoning it, Pierre Fabre chose to explore a different approach and called on BOTdesign to co-design a viable hybrid protocol.

The objective: to jointly define a realistic, financially sustainable, and scientifically rigorous design within a 12-month timeframe.

The co-designed protocol is based on:

  • 100 real patients recruited across 5 investigator sites in 3 countries (France, Poland, and Italy), with 50 patients per arm (Comedomed and placebo)
  • 500 artificial patients generated by BOTdesign’s ORIGA platform, based on data from the 100 real patients
  • A 12-month follow-up period with 6 visits to the dermatologist
  • A primary endpoint focusing on the number of relapses and the time to relapse between groups

The use of 500 artificial patients — well beyond the theoretical minimum threshold — was a deliberate choice: to enrich the analytical dataset in order to achieve greater statistical power and a higher level of information granularity than a fully real-world study could have produced, even under ideal recruitment conditions.

The impossible became not only possible, but also more informative.

 

Artificial patient generation: principles and quality metrics

The ORIGA platform is based on an RHVAE (Riemannian Hamiltonian Variational AutoEncoder) architecture.

Using the profiles of the 100 real patients, ORIGA generates 500 new profiles whose multivariate distributions — demographic covariates, biological variables, and efficacy data — remain statistically faithful to the source dataset, without ever reproducing an identifiable real individual.

Two metrics ensure quality control:

  • Wasserstein distance < 5%: statistical fidelity ensuring that the data can be used in the primary analysis.
  • Zero-CAP at 99.4%: near-total protection against re-identification, in compliance with GDPR requirements and ANSM recommendations.

 

Scientific and regulatory validation

The protocol underwent rigorous multi-level validation:

  • Prior technical validation using an independent test database, before any application to the study data.

  • Validation by Pierre Fabre experts (biostatistician and AI expert) and by the clinical investigators, ensuring the statistical and clinical consistency of the generated profiles.

  • Approval by 3 European Ethics Committees, without reservations — demonstrating institutional recognition of the methodology.

  • GDPR compliance and AI governance formally validated by Pierre Fabre, with the implementation of a reusable framework for future projects.

From a regulatory perspective, the ICH E9 (R1) guidelines provide a relevant conceptual framework for integrating augmented data into primary and sensitivity analyses. The EMA is progressively recognizing the legitimacy of such approaches, subject to methodological transparency — an area that BOTdesign actively monitors. The presentation at the ICD Congress in Rome by Professor Ardigo, Principal Investigator, in June 2025 confirmed the scientific acceptability of the approach within the international dermatology community.

 

Results and perspectives

 

The hybrid protocol co-designed with BOTdesign enabled Pierre Fabre to conduct a study that would have been impossible to carry out using a fully real-patient format : 100 real patients (50 per arm) combined with 500 artificial patients, resulting in a total cohort of 600 analyzable subjects — exceeding the initial target of 400 — within a 12-month timeframe and at a controlled cost.

The measured benefits are multidimensional:

  • Feasibility: a study that was structurally impossible became achievable, without compromising statistical rigor.
  • Increased analytical power: with 600 subjects instead of 400, the granularity of the analyses and the robustness of the conclusions are greater than those of the study originally envisaged.
  • Budget optimization: the cost of an artificial patient generated by ORIGA makes it possible to reduce the cost per patient by a factor of 3 to 5, depending on the patient profile. The hybrid protocol generates substantial savings compared with recruiting the full cohort of 400 real patients, while producing a larger cohort.
  • Reduced placebo exposure: limiting the number of real patients in the placebo arm to 50 reduces unnecessary exposure to a treatment with no expected benefit, thereby addressing a major ethical requirement.
  • Reusable governance framework: the methodological, regulatory, and ethical framework implemented has been documented and can be deployed for Pierre Fabre’s future projects.

Beyond this specific case, the real/artificial hybrid protocol opens up a transformative perspective for clinical research in dermocosmetics: designing ambitious studies in areas where recruitment had previously been a limiting factor, and obtaining a statistical approximation of reality with a level of accuracy that had until now been unattainable.

 

Quote from Pierre Fabre

 

“Collaboration with BOTdesign made it possible to carry out a study that was not feasible within the imposed time and recruitment constraints. The rigor of the validation process and the responsiveness of the teams facilitated the integration of this innovative approach into our usual regulatory and ethical framework.”

— Pierre Fabre Dermo-Cosmétique / Avène

 

To learn more

BOTdesign supports clinical research teams in human and animal health in the co-design and implementation of real/artificial hybrid protocols. To discuss the feasibility of your project and receive a personalized analysis: direction@botdesign.net