Mimicking clinical trials with synthetic acute myeloid leukemia patients using generative artificial intelligence
Problem
Clinical research is hindered by the high cost of collecting large patient datasets and privacy/regulatory barriers to accessing existing data.
Why it matters
Restricted data access slows hematological cancer research and limits synthetic control cohort options.
What the AI did
Two generative AI methods, CTAB-GAN+ and normalizing flows, synthesized patient cohorts from 1606 acute myeloid leukemia patients treated in four multicenter trials, capturing demographics, labs, molecular/cytogenetic variables and outcomes.
Results
Both models yielded high fidelity and usability scores; synthetic cohorts (n=1606 each) closely matched original variable distributions and survival curves
Impact
Demonstrated synthetic AML cohorts resembling real trial data; broader adoption not shown.
Record
| Venue | npj Digital Medicine |
|---|---|
| Published | 2024 |
| Work type | article |
| Authors | Jan‐Niklas Eckardt, Waldemar Hahn, Christoph Röllig, Sebastian Stasik, Uwe Platzbecker, Carsten Müller‐Tidow, Hubert L. Serve, Claudia Dorothea Baldus et al. |
| Citations | 64 (OpenAlex, at import time) |
| References | 62 |
| Model or method | CTAB-GAN+ and normalizing flows |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41746-024-01076-x |
| Open access | gold |
| Keywords | generative AI, synthetic data, acute myeloid leukemia, CTAB-GAN+, normalizing flows |
Importance rating
real but narrower advance
Privacy and cost barriers hinder clinical data access for hematological cancer research; synthetic cohorts from 1606 AML patients shown as proof-of-concept, a real but narrower methodological advance.