PharmTech AI Pulse Check: SURPASS and the Case for Simulation-Based Trials

A newly launched research initiative is testing whether digital twins and real-time analysis can reshape how clinical trials are designed, and whether regulators will accept the evidence they produce.

In PharmTech AI Pulse Check video series, host Christopher Cole discusses the question with Florin Muraru, a regulatory affairs expert specializing in EU and US strategy and AI governance; Gourav Pandey, R&D quality lead at Takeda; and Richard Jaenisch, senior director of education, outreach, and digital experience at Open Biopharm.

Can Regulators Accept Simulation-Based Clinical Trial Evidence?

The conversation centered on SURPASS (Simulation Augmented Real-time Platform Adaptive Seamless Trials),¹ a program Muraru notes (at the time of recording) was "launched one week ago, so nothing is built yet." Despite the early stage, he points to real regulatory precedent: in 2022, the European Medicines Agency qualified prognostic covariate adjustment,² a method in which a model trained on historical data predicts how patients would likely fare on control treatment, allowing randomized trials to be more precise with fewer participants. That same logic underpins SURPASS's ambition to make trials "leaner, faster, highly adaptable."

Muraru outlined the regulatory guardrails sponsors will need to meet: the adaptation rule and error control must be fixed before the first patient is enrolled, the model must be locked, and "every change should be versioned and justified." He also flagged that the program's four-year timeline may be optimistic, particularly for endpoints requiring an additional two years of follow-up. And while digital twins can reduce the number of patients needed to demonstrate efficacy, they don't change how long patients must be monitored for safety.

A chronic therapy "might hit your efficacy answer early but still need to collect safety data," Muraru says.

What Happens When Clinical Trials Generate Too Much Data?

Pandey framed the moment as part of a broader shift already underway at agencies like the FDA, predicting a future with a "very digital trail of almost everything that's happening" in a trial, generating more data that, over time, improves both efficiency and safety outcomes.

Jaenisch said digital twins are strong at prediction, but "regulators will look for the acceptance through evidence," meaning validation artifacts and benchmarks still need to mature. He also raised the question

if trials generate enormous volumes of real-time data, at what point does the FDA request it, and how will sponsors manage preparing and cleaning that data for a regulatory dossier? Muraru agreed it "will be a nightmare to clean the data," while Pandey suggested AI agents may be essential to managing that scale.

The exchange captures where simulation-driven trial design currently stands: regulatory precedent exists, ambition is high, and the open questions are no longer about whether the modeling works, but whether the data infrastructure and acceptance pathways can keep pace.

References

1. ARPA-H. SURPASS. Access Oct 19, 2026.
2. Vanderbeek A, Vidovszky A, Ross, J, et al. A Weighted Prognostic Covariate Adjustment Method for Efficient and Powerful Treatment Effect Inferences in Randomized Controlled Trials. arXiv:2309.14256v1 [stat.ME] 25 Sep 2023.