Clinical trials become increasingly more expensive and time-consuming. High failure rates especially in late phase trials put a substantial financial burden on healthcare systems and pharmaceutical companies. Hence, a paradigm shift is urgently needed in clinical trial design.
Generative artificial intelligence has celebrated numerous success stories within the last few years. We hypothesized that apart from image and text generation, such models could also be used to generate synthetic clinical trial cohorts.
We used two distinct technologies, Normalizing Flows and CTAB-GAN+, to generate two high dimensional data sets of patients with acute leukemia including clinical, laboratory and genetic variables based on a large multicenter cohort.
We found AI-generated synthetic patients to closely mimic the behavior of real patients both biologically and clinically. All relationships between patient variables were sustained in the synthetic cohort, highlighting a faithful representation of disease biology in synthetic patients. Further, patient privacy is fully maintained.
Synthetic trial cohorts can openly be shared (In fact, our cohort of 3200+ synthetic patients is fully available).
Plus, the availability of synthetic patients opens new avenues in trial designs with possible trial simulations, augmentation or substitution of control cohorts and customization of control cohort features for precision medicine.
Check out our new paper in npj Digital Medicine and stay tuned for much more to come in synthetic trial designs!
Eckardt, JN., Hahn, W., Röllig, C. et al. Mimicking clinical trials with synthetic acute myeloid leukemia patients using generative artificial intelligence. npj Digit. Med. 7, 76 (2024). https://doi.org/10.1038/s41746-024-01076-x
