Our project investigates the feasibility, impact, and investment potential of AI-driven virtual control groups as a transformational technology in Phase III clinical trials. Traditional clinical trials are slow, expensive, and often fail due to recruitment challenges, high per-patient costs, and insufficient statistical power. We propose replacing or reducing real control-arm participants with simulated control trajectories generated from machine learning models trained on historical trial data, electronic health records, and real-world evidence
Our proposed system focuses initially on cardiology clinical trials, which are long, costly, and data-rich, ideal for model development and impact. By generating individualized control predictions using multimodal patient inputs (demographics, clinical history, genomics, imaging, and wearable data), AI models can simulate realistic disease progression under placebo or standard-of-care conditions. This approach can reduce per-patient control costs from ~$15,000 to ~$2,000, enabling a 30% overall trial cost reduction and enabling sponsors to recruit more treatment-arm patients, increasing statistical power by 10–20%.
We evaluate multiple implementation strategies, including reduced control groups calibrated through real patient outcomes and dual modeling approaches that generate both treatment and control trajectories for internal bias correction. Through expert interviews with clinicians, digital twin companies, and AI researchers, we identify limitations of current industry solutions, such as limited therapeutic focus, outdated modeling architectures, and challenges aligning with current standards of care.
We conclude by mapping regulatory considerations, go-to-market pathways, and investment opportunities. As drug discovery accelerates through AI, late-stage trials risk becoming the bottleneck. Our project demonstrates that AI-powered virtual controls offer a clear, investible opportunity to make clinical trials faster, less expensive, and more likely to succeed. Our final report can be found here.