The AI model that is changing clinical trial design

Clinical trials are expensive, slow and often limited by outdated design constraints. Placebo arms, in particular, create ethical and logistical hurdles, especially in areas like rare disease and oncology.

Digital twins offer a way forward.

They offer patient-specific outcome predictions, generated using machine learning models trained on real historical clinical data. These digital twins are created for each trial participant using their baseline data – regardless of whether they are assigned to the placebo or treatment arm – and simulate how that individual would have responded under control conditions. Rather than replacing control patients, these forecasts of control outcomes are integrated into the analysis to reduce sample size requirements or increase statistical power.

Unlearn is one of the few companies applying this approach at scale. Chief Executive Officer Steve Herne has spent more than 25 years in clinical research, with senior roles at WCG, Bioclinica and Covance. At Unlearn, he is focused on bringing digital twin technology into active use across both early and late-phase trials.

“Today, we are focused on scaling AI-driven clinical solutions that help sponsors optimise their clinical programmes,” Herne says. “Accelerating timelines and de-risking development decisions across the pipeline.”

What are digital twins, really

At Unlearn, digital twins are not hypothetical avatars or speculative models. They are patient-specific outcome predictions, generated using disease-specific machine-learning models trained on large, longitudinal clinical datasets. Each digital twin represents a data-driven forecast of how an individual participant would have progressed under placebo, based solely on their baseline characteristics.

“We use the term ‘digital twin’ to mean a comprehensive forecast of an individual trial participant’s longitudinal clinical outcomes,” Herne explains. “These are generated using disease-specific machine-learning models called Digital Twin Generators, or DTGs.”

By incorporating digital twins into the statistical analysis of a trial, sponsors can reduce the number of participants required in the control arm without compromising statistical power. This makes trials faster and more efficient, with clear implications for patient burden, recruitment timelines and ethical design.

“By incorporating digital twins into the analysis, sponsors can maximise the available power of their study,” Herne says. “That power can then be used in different ways: to increase confidence in the trial’s results or to reduce the number of participants needed – especially in the control arm – and speed up the study.”

This approach does not replace traditional controls but complements them, offering a validated method for extracting greater value from baseline data.

Regulatory backing

The pharmaceutical industry is often open to innovation in principle, but the demands of regulatory compliance can slow the adoption of emerging technologies – particularly those that challenge conventional trial design.

Unlearn’s early collaboration with regulators has helped it navigate this space effectively. The company worked closely with agencies including the US Food and Drug Administration (FDA) and European Medicines Agency (EMA) from the outset, aligning its methodology with evolving guidance. In 2022, the EMA formally qualified Unlearn’s PROCOVA method – a specific covariate adjustment approach – for use in Phase I and III trials with continuous outcomes. The FDA has also confirmed that Unlearn’s covariate adjustment strategy is consistent with current guidance.

“We take transparency seriously, documenting every aspect of our models – from how they are trained to how they are validated in their specific context of use,” Herne says.

That emphasis on transparency and scientific validation has supported wider adoption. Unlearn’s digital twins are now in use in both early and late-stage clinical trials, with adoption continuing to grow.

Solving old problems in new ways

Herne is well acquainted with the operational inefficiencies that persist across clinical development. One of the most persistent challenges remains the control group, which can slow recruitment, prolong timelines and present ethical concerns.

“In randomised controlled trials, digital twins reduce the number of control participants required while maintaining power,” Herne says. “Alternatively, digital twins can boost trial power without needing to increase sample sizes.”

In oncology and rare disease studies, where traditional control arms are often impractical, digital twins can serve as simulated control groups in exploratory or planning contexts. This approach enables scientifically valid comparisons without placing additional patients on placebo or withholding standard-of-care treatment, though it is not yet accepted for regulatory decision-making.

“By generating scientifically valid comparisons without the need for external or placebo controls, digital twins expand what is possible in trial design.”

This is not simply an operational improvement. It has the potential to reshape how control data is generated and used in future studies.

The AI under the hood

Unlearn’s DTGs are built on proprietary neural network architectures developed specifically for clinical prediction. Unlike more generalised models, these have been designed to reflect the complexity and variability of real-world patient data.

“These models are powered by a proprietary neural network architecture purpose-built for clinical prediction,” Herne says. “Their accuracy and granularity reflect deep investment in designing models that are optimised for complex, real-world clinical data.”

The models are designed for compatibility with existing trial infrastructure and do not require changes to endpoints, treatment arms or randomisation schemes.

“Integration is straightforward,” Herne notes.

For sponsors, this creates a clear value proposition: meaningful innovation without disruption to current workflows or regulatory expectations.

AI with a human purpose

Herne speaks about AI in clinical research with pragmatic focus. Rather than making sweeping claims about transforming healthcare, he stresses the need for practical tools that improve clinical decision-making and integrate into current systems.

“Our goal is to make adoption feel familiar – leveraging proven methods in a smarter, more personalised way that brings added confidence to clinical decision-making.”

This vision aligns with the core concept of the digital twin: not to replace traditional approaches, but to reflect and refine them with greater precision.

In a landscape where AI is often overstated, Unlearn’s work represents a case study in how measured application of machine learning can improve the efficiency, ethics and outcomes of clinical trials.

Key takeaways

• What: Unlearn builds digital twins – AI-generated predictions of how a patient would progress in a clinical trial – used to optimise trial design and execution
• Why it matters: Fewer patients in placebo arms, faster timelines and more confident data without loss of statistical power
• Regulatory backing: EMA qualification in 2022 and FDA alignment confirmed
• Use cases: Randomised trials with reduced control groups and single-arm studies in rare diseases and oncology
• Technology: Disease-specific neural networks trained on extensive clinical data
• Future direction: Integration with existing trial designs and workflows to enable smarter, more patient-centred studies.

Many in the industry remain cautious about adopting new technologies. Unlearn is taking a practical, evidence-based approach to AI using digital twins to improve trial efficiency, support ethical study designs and produce reliable clinical outcomes.