Economics and Equity of Large Language Models: Health Care Perspective.

Area of Science:

• Health informatics
• Artificial intelligence in healthcare
• Large language models (LLMs)

Background:

• Large language models (LLMs) show increasing promise in healthcare, but successful adoption hinges on digital readiness, infrastructure, workforce training, privacy, and regulatory frameworks.
• Healthcare ecosystems present unique challenges and variations, influencing the optimal pathway for LLM implementation and equitable access.

Purpose of the Study:

• To discuss and compare three distinct LLM implementation pathways: training from scratch (TSP), fine-tuned pathway (FTP), and out-of-the-box pathway (OBP).
• To analyze the risks, benefits, and economics of these pathways across major cloud providers (Amazon, Microsoft, Google, Oracle).
• To guide health systems in strategically adopting LLMs for equitable access and improved healthcare outcomes.

Main Methods:

• Comparative analysis of three LLM implementation pathways (TSP, FTP, OBP).
• Examination of economic factors, including on-demand and spot pricing across cloud service providers.
• Discussion of the utility of managed services and cloud enterprise tools like federated learning for healthcare data.

Main Results:

• Training from scratch (TSP) offers maximum customization and performance but is the most resource-intensive.
• Fine-tuned pathway (FTP) balances customization, cost, and performance, though pretrained models may introduce bias.
• Out-of-the-box pathway (OBP) allows rapid deployment with minimal customization and transparency, posing potential long-term availability and integration challenges.

Conclusions:

• The choice of LLM implementation pathway (TSP, FTP, OBP) depends on specific health system needs and affordability.
• Managed services and enterprise tools can mitigate challenges in LLM implementation, particularly with sensitive healthcare data.
• Strategic understanding of pathway economics and trade-offs is essential for successful LLM adoption, value demonstration, and positive impact on healthcare outcomes.