Recent progress in Artificial Intelligence and Machine Learning provides an avenue for using historical data to create more efficient clinical trials without introducing bias. Rather than incorporating data from external sources directly into the trial, we leverage predictions from AI-based prognostic models — called Digital Twins — trained on historical control data to reduce uncertainty in estimated treatment effects. In this whitepaper, we describe how this novel approach enables optimally efficient clinical trials that require fewer subjects to achieve pre-specified power while rigorously controlling bias and type-I error rates.
