Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Evaluating healthcare quality and inequities using generative AI : a simulation study of potentially inappropriate medication among older adults analyzed via the framework analysis of individual heterogeneity and discriminatory accuracy (AIHDA)

Öberg, Johan LU orcid ; Perez-Vicente, Raquel LU orcid ; Lindström, Martin LU ; Midlöv, Patrik LU orcid and Merlo, Juan LU orcid (2025) In Discover Artificial Intelligence 5(1).
Abstract

Background: Regular monitoring of healthcare quality and equity is crucial for informing decision-makers and clinicians. This study explores the application of generative AI, more specifically large language models (LLMs), to facilitate standardized monitoring of healthcare quality using the established framework Analysis of Individual Heterogeneity and Discriminatory Accuracy (AIHDA). The study investigates whether a customized GPT can effectively apply the AIHDA-framework to assess healthcare quality in a simulated dataset. Population and methods: Using simulated data modelled on real-world healthcare information, we evaluated the quality indicator of potentially inappropriate medication (PIM). A customized GPT built on ChatGPT 4o was... (More)

Background: Regular monitoring of healthcare quality and equity is crucial for informing decision-makers and clinicians. This study explores the application of generative AI, more specifically large language models (LLMs), to facilitate standardized monitoring of healthcare quality using the established framework Analysis of Individual Heterogeneity and Discriminatory Accuracy (AIHDA). The study investigates whether a customized GPT can effectively apply the AIHDA-framework to assess healthcare quality in a simulated dataset. Population and methods: Using simulated data modelled on real-world healthcare information, we evaluated the quality indicator of potentially inappropriate medication (PIM). A customized GPT built on ChatGPT 4o was prompted via the principle TREF (Task, Requirement, Expectation, Format) to perform the analysis. Results were compared to a traditional analysis performed with Stata to evaluate accuracy and reliability. Results: The GPT successfully conducted the AIHDA analysis, producing results equal to those of the Stata analysis. The GPT provides useful visualizations and structured reports as well as interactive dialog with the end-user in real-time. However, occasional variations in the results occurred in some iterations of the analysis, highlighting potential issues with reliability. The analysis requires close supervision, as the GPT presents both errors and correct results with confidence. Conclusions: Generative AI and LLMs show promise in supporting standardized monitoring of healthcare quality and equity using the AIHDA-framework. It enables accessible analysis but requires oversight to address limitations such as occasional inaccuracies. Future and more reliable models of LLMs and local deployment on secure servers may further enhance the utility for routine healthcare monitoring.

(Less)
Please use this url to cite or link to this publication:
author
; ; ; and
organization
publishing date
type
Contribution to journal
publication status
published
subject
keywords
Epidemiological methods, Health care quality assessment, Health services evaluation, Social epidemiology
in
Discover Artificial Intelligence
volume
5
issue
1
article number
175
publisher
Springer Nature
external identifiers
  • scopus:105011402264
DOI
10.1007/s44163-025-00444-0
language
English
LU publication?
yes
additional info
Publisher Copyright: © The Author(s) 2025.
id
0986f3a0-6483-42b5-beba-ad7b290b8b39
date added to LUP
2025-08-03 13:45:09
date last changed
2025-08-04 10:35:51
@article{0986f3a0-6483-42b5-beba-ad7b290b8b39,
  abstract     = {{<p>Background: Regular monitoring of healthcare quality and equity is crucial for informing decision-makers and clinicians. This study explores the application of generative AI, more specifically large language models (LLMs), to facilitate standardized monitoring of healthcare quality using the established framework Analysis of Individual Heterogeneity and Discriminatory Accuracy (AIHDA). The study investigates whether a customized GPT can effectively apply the AIHDA-framework to assess healthcare quality in a simulated dataset. Population and methods: Using simulated data modelled on real-world healthcare information, we evaluated the quality indicator of potentially inappropriate medication (PIM). A customized GPT built on ChatGPT 4o was prompted via the principle TREF (Task, Requirement, Expectation, Format) to perform the analysis. Results were compared to a traditional analysis performed with Stata to evaluate accuracy and reliability. Results: The GPT successfully conducted the AIHDA analysis, producing results equal to those of the Stata analysis. The GPT provides useful visualizations and structured reports as well as interactive dialog with the end-user in real-time. However, occasional variations in the results occurred in some iterations of the analysis, highlighting potential issues with reliability. The analysis requires close supervision, as the GPT presents both errors and correct results with confidence. Conclusions: Generative AI and LLMs show promise in supporting standardized monitoring of healthcare quality and equity using the AIHDA-framework. It enables accessible analysis but requires oversight to address limitations such as occasional inaccuracies. Future and more reliable models of LLMs and local deployment on secure servers may further enhance the utility for routine healthcare monitoring.</p>}},
  author       = {{Öberg, Johan and Perez-Vicente, Raquel and Lindström, Martin and Midlöv, Patrik and Merlo, Juan}},
  keywords     = {{Epidemiological methods; Health care quality assessment; Health services evaluation; Social epidemiology}},
  language     = {{eng}},
  number       = {{1}},
  publisher    = {{Springer Nature}},
  series       = {{Discover Artificial Intelligence}},
  title        = {{Evaluating healthcare quality and inequities using generative AI : a simulation study of potentially inappropriate medication among older adults analyzed via the framework analysis of individual heterogeneity and discriminatory accuracy (AIHDA)}},
  url          = {{http://dx.doi.org/10.1007/s44163-025-00444-0}},
  doi          = {{10.1007/s44163-025-00444-0}},
  volume       = {{5}},
  year         = {{2025}},
}