AI firms test virtual drug trials, but human studies remain essential
Reuters examined early predictions by BioinvestGPT and the wider use of simulated patients; a small set of correct calls does not validate virtual trials as replacements for clinical studies.
Testing predictions against later results
AI companies are using simulated patients to estimate how experimental medicines might perform before costly human trials, Reuters reported on October 7. BioinvestGPT shared detailed predictions for several prominent drug studies with Reuters in July, before their results were known. The news agency says five of six subsequent calls were correct, including a negative prediction for Novartis’ del-desiran trial. One call was wrong: the system expected a positive result for another Novartis drug, pelacarsen, which later failed to reduce major cardiovascular events in a late-stage study.
What virtual trials can and cannot establish
BioinvestGPT says its model uses DNA sequencing to simulate a body matching a trial’s eligibility criteria. Other companies, including QuantHealth, use different data and methods to model patient responses. The five-of-six record is a small, selected set described in Reuters’ reporting, not proof of general predictive accuracy across drug development. Reuters also reported BioinvestGPT’s forecasts for upcoming Biogen and Takeda studies; those are unverified predictions, not trial outcomes. A Takeda research executive cautioned that AI tools cannot yet make definitive predictions. Researchers may use simulations to decide which programs merit further investment, but human studies remain necessary to establish safety and clinical benefit.