A new generation of patient-reported outcome measures with large language models

Abstract

Background: Patient-reported outcome measures (PROMs) are cornerstones of patient-centered clinical medicine and reflect patients’ abilities, difficulties, perceptions and behaviors. The highly structured questionnaire format of PROMs currently limits their real-world validity and acceptability to patients, which becomes increasingly relevant with the high clinical interest in PROM data. In this short commentary, we aim to demonstrate the potential use of large language models (LLMs) in the context of PROM data collection and interpretation. Main body: The popularization of LLMs enables the development of a new generation of PROMs generated and administered through digital technology that interact with patients and score their responses in real time based on artificial intelligence. LLM-PROMs will need to be developed with multi-stakeholder input and careful validation against established PROMs. LLM-PROMs could complement traditional PROMs particularly in real-world clinical applications. Conclusion: LLM-PROMs could allow quantifying patient-relevant dimensions based on less structured contents and foster the use of patient-reported data in digital, clinical applications of PROMs.