Anonymized but Useful Synthetic Tabular Health Data for AI based Fall Risk Assessment

Area of Science:

• Health Informatics
• Artificial Intelligence in Healthcare
• Data Science

Background:

• Artificial Intelligence (AI) holds promise for enhancing healthcare, contingent upon accessible, realistic, and valuable data.
• Developing AI models in healthcare necessitates robust datasets that respect patient privacy.

Purpose of the Study:

• To introduce SynTabFall, a novel synthetic tabular dataset designed for fall risk assessment.
• To enable the training of AI models for fall risk prediction using synthetic data, bypassing the need for original patient records.
• To present a responsible data-sharing framework combining anonymization and generative AI.

Main Methods:

• Generation of a synthetic tabular dataset (SynTabFall) with 745,380 samples and 44 attributes related to fall risk factors.
• Utilizing generative AI (genAI) methods for synthesizing tabular data.
• Developing and evaluating a data-sharing process in collaboration with healthcare professionals and technical experts.

Main Results:

• The synthetic SynTabFall dataset facilitates the training of fall risk prediction models.
• Models trained on SynTabFall demonstrate predictive performance on par with models trained on real patient data.
• The developed data-sharing approach allows for responsible healthcare data sharing without compromising utility.

Conclusions:

• SynTabFall provides a valuable resource for AI-driven fall risk assessment.
• The proposed methodology enables the responsible and effective sharing of sensitive healthcare data.
• This work supports the advancement of AI applications in healthcare through open-access synthetic data.