The Potential and Perils of Generative Artificial Intelligence for Quality Improvement and Patient Safety

TL;DR

This paper argues that synthetic data and simulators offer a scalable path to advance AI in healthcare while addressing privacy and data-sharing constraints. It classifies synthetic-data generation into physical, statistical, and hybrid paradigms and discusses sim2real transfer techniques to bridge gaps between simulated and real-world data. The authors review healthcare applications across EHRs, NLP, physiological signals, and medical imaging, highlighting potential benefits for privacy, equity, safety, and continual learning, alongside risks such as bias, unknowns, and regulatory gaps. They stress the need for standards, rigorous evaluation, and human-in-the-loop collaboration to ensure safe, effective deployment of synthetic-data solutions in clinical settings.

Abstract

Generative artificial intelligence (GenAI) has the potential to improve healthcare through automation that enhances the quality and safety of patient care. Powered by foundation models that have been pretrained and can generate complex content, GenAI represents a paradigm shift away from the more traditional focus on task-specific classifiers that have dominated the AI landscape thus far. We posit that the imminent application of GenAI in healthcare will be through well-defined, low risk, high value, and narrow applications that automate healthcare workflows at the point of care using smaller foundation models. These models will be finetuned for different capabilities and application specific scenarios and will have the ability to provide medical explanations, reference evidence within a retrieval augmented framework and utilizing external tools. We contrast this with a general, all-purpose AI model for end-to-end clinical decision making that improves clinician performance, including safety-critical diagnostic tasks, which will require greater research prior to implementation. We consider areas where 'human in the loop' Generative AI can improve healthcare quality and safety by automating mundane tasks. Using the principles of implementation science will be critical for integrating 'end to end' GenAI systems that will be accepted by healthcare teams.

Figures (3)

• Figure 1: Simulations can be A) deterministic physical simulations, B) statistical generative simulations, or C) a hybrid of two in which physical and statistical simulations are combined. Machine learned models can be trained using D) real, E) synthetic and F) a mixture of real and synthetic data from these simulations. Synthetic data can be used for testing/performance evaluation, to contrast G) testing on real data, H) systematically generated synthetic data and I) a combination of the two. Images of atrial flutter voltage distributions adapted from Dossel et al. dossel2021computer and echocardiograms adapted from Madani et al. madani2018deep.
• Figure 2: Methods of achieving sim2real include transfer learning approaches, such as (A) domain randomization or B) domain adaptation. Because it is sometimes difficult to transfer models from simulations to real data, another strategy is to improve the realism of simulators in a hybrid physics/data manner, through methods like C) differentiable simulation.
• Figure 3: A) Real and synthetic patient videos for remote monitoring estepp2014recoveringwang2022syntheticmcduff2022scamps. B) Real and synthetic electrocardiogram thambawita2021deepfake C) Real and synthesized MRI images of brain tumors shin2018medical. D) Real and synthetic hematoxylin and eosin stains from TA-PARS nonradiative absorption contrast images boktor2022virtual.