Editorial Commentary: Advancing the Value of Artificial Intelligence in Health Care Means Rethinking What We Are Measuring It Against.

Abstract

Despite an increasing emphasis on value-based health care, inappropriate resource utilization remains prevalent, with unnecessary referrals for imaging and surgical procedures. Recent studies have explored the use of large language models (LLMs) as copilots for streamlining data acquisition and optimizing decision-making with respect to these areas of patient care. Emerging data suggest that patient questionnaires, when parsed by LLMs, may allow for decision-making regarding the appropriate next steps in evaluation and treatment that is comparable to human providers. While encouraging, such use cases merit caution about the biases inherent in generalized LLMs whose recommendations fundamentally do not reflect informed clinical decision making but rather a byproduct of models trained on vast amounts of data from unverified sources that include medical information. Further, the question remains: should we be comparing LLM performance to the outcomes of the current health care system we function within or the outcomes that are necessary to advance patient care moving forward? When researchers choose to repurpose LLMs for these use cases, they must consider which options provide the greatest value.