Longitudinal Synthetic Data Generation by Artificial Intelligence to Accelerate Clinical and Translational Research...

Area of Science:

• Oncology
• Biomedical Informatics
• Artificial Intelligence

Background:

• Real-world data (RWD) is crucial for breast cancer (BC) research but faces limitations like privacy concerns, missing information, and fragmentation.
• Advanced generative models offer a potential solution for creating robust, longitudinal datasets.

Purpose of the Study:

• To explore the use of synthetic data (SD) generated by advanced AI models to overcome RWD limitations in breast cancer research.
• To create harmonized, privacy-preserving longitudinal datasets for improved research and clinical applications.

Main Methods:

• Utilized a dataset of 1052 BC patients (HER2-positive and triple-negative) from the i2b2 platform.
• Applied generative adversarial networks (GANs), variational autoencoders (VAEs), and language models (LMs) to generate synthetic longitudinal data.
• Evaluated SD using the Synthetic Validation Framework (SAFE) for fidelity, utility, and privacy across three settings: i2b2 integration, disease modeling, and synthetic control group generation.

Main Results:

• SD demonstrated high fidelity (0.94 score) and ensured data privacy, with validated temporal patterns.
• Integration of SD with the i2b2 platform preserved privacy while mirroring RWD.
• SD improved multistate disease progression model performance (up to 10% C-index increase) and replicated clinical trial endpoints for synthetic control arm generation.

Conclusions:

• AI-generated longitudinal SD effectively address RWD challenges in breast cancer research.
• This approach enhances translational research and clinical trial design with robust privacy protection.
• Integration with platforms like i2b2 shows scalability for broader oncology applications.