Simulate before exposing patients

A clinical trial involves more than administering a medicine. Researchers must choose patients, dose, duration, success criteria and a comparison group. A design mistake can make a result unusable even when the treatment has a genuine effect.

Simulation platforms combine data from earlier trials, health records, publications and biological models. They create digital cohorts to test different scenarios: which dose looks most appropriate? Which patient profiles respond better? How many participants are needed to detect an effect?

Reuters reports that some companies have begun publishing predictions of real trial outcomes. These examples are interesting, but still too limited to establish general reliability. A model that performs well in one disease or development phase may fail in another.

What the FDA is actually examining

In 2026, the FDA launched work on real-time trials and AI-based technologies. The aim is to improve monitoring, speed and decision quality during early studies. The agency also operates a model-informed drug development programme that includes trial simulation.

This does not mean automatically approving a medicine because an algorithm predicts success. Regulators expect documented methods, representative data, independent validation and clear reporting of uncertainty. The model’s role must match the risk of the decision.

The danger of digital patients that are too perfect

A simulation learns from the past. If historical trials underrepresented women, older people, certain ethnic groups or patients with multiple conditions, a digital twin may reproduce those gaps. A seemingly precise result may then be wrong for people missing from the data.

Clinical data also vary across hospitals, countries and practices. Missing measurements, changing definitions and concomitant treatments can create biases that are hard to detect. Transparency about sources and limitations is therefore as important as average performance.

Decision support, not a substitute for evidence

The most credible near-term use is to compare protocols before a trial opens, identify inconsistencies and adjust monitoring. A simulation can also provide an additional reference scenario when historical data are robust.

For participants, this could mean fewer poorly designed trials and better-targeted visits. For pharmaceutical companies, an earlier decision to stop could save years. But a mistaken model could also discard a useful treatment or steer a study in the wrong direction.

Progress will therefore be measured not by how many virtual patients are generated, but by whether the tools improve real trials, with reproducible results overseen by regulators.