Health Gym

Health Gym generates realistic synthetic medical datasets to support development, evaluation, and comparison of machine learning methods, with emphasis on reinforcement learning for clinical conditions such as hypotension, sepsis, and HIV antiretroviral therapy.

Key Features:

• Realistic Synthetic Datasets: Synthetic datasets mimic variable distributions, correlations between variables, and temporal trends observed in clinical data.
• Clinical Focus: Includes cohorts for acute conditions such as hypotension and sepsis in intensive care units, and for individuals undergoing antiretroviral therapy for human immunodeficiency virus (HIV).
• Generative Adversarial Network (GAN) Methodology: Uses a novel Generative Adversarial Network (GAN) approach to generate synthetic clinical data that captures complex patterns and relationships.
• Low Disclosure Risk: Estimates a very low risk of sensitive information disclosure when datasets are publicly distributed.
• Algorithm Prototyping and Evaluation: Provides datasets intended for prototyping, evaluation, and comparison of machine learning algorithms, with a specific emphasis on reinforcement learning.

Scientific Applications:

• Reinforcement Learning: Enables exploration of decision-making and policy learning in dynamic clinical environments such as intensive care units.
• Algorithm Development and Comparison: Supports prototyping and benchmarking of new and existing machine learning algorithms across clinical scenarios.
• Reproducibility and Generalizability: Provides standardized synthetic datasets to improve reproducibility and assess model generalizability to real-world healthcare settings.

Methodology:

Synthetic datasets are generated using a novel Generative Adversarial Network (GAN) approach, and the resource includes an estimate of very low sensitive information disclosure risk.