Fairness in multimodal machine learning applications in clinical decision support: a systematic review
- 1School of Public Health, Faculty of Medicine and Health, University of Sydney, Sydney, Australia
- 2School of Computer Science, Faculty of Engineering, University of Sydney, Sydney, Australia
Abstract
Multimodal AI models are increasingly used in clinical decision support systems, but fairness appears to be rarely measured. In this systematic review, our aim was to assess studies that examine fairness in multimodal AI-based clinical decision support (CDS) systems. CDS systems that are rule-based or knowledge-based are considered out of scope. Two searches were used to identify literature focused on evaluating fairness in multimodal CDS systems and to determine the rate at which fairness is included in studies for two focus areas—chest x-rays and sepsis. The review was registered with PROSPERO (CRD42024579923). From 3059 search results, 160 articles included fairness evaluations, 11% (18/160) used multimodal data, and 29 different fairness measurement techniques were used. From 1422 search results related to the two focus areas, 88 studies using multimodal data for CDS related to chest x-ray and sepsis were included. Fairness evaluations were found in 8% (7 of 88) of these, despite 83% (74 of 88) including data on sensitive attributes that could have enabled fairness evaluation. The results show that fairness evaluations are uncommon for studies in multimodal CDS systems. AI-based multimodal CDS systems research would benefit from guidance and tools that support fairness evaluations.
Related
- Choriocapillaris Dysfunction in Retinal Diseases: Multimodal Imaging Biomarkers, Pathophysiological Insights, and Emerging Clinical Applications
- COMBINING CHEST X-RAY AND CLINICAL COVARIATES FOR OSTEOPOROSIS SCREENING: A MULTIMODAL DEEP LEARNING APPROACH
- Advances in Multimodal Artificial Intelligence in Radiology: Data Integration, Foundation Models, and Clinical Applications-A Narrative Review.
- Multimodal ensemble machine learning pipeline for oocyte fertilization prediction using first polar body morphology and clinical factors
- Machine Learning and Deep Learning for Early Disease Detection Across Modalities: A Comprehensive Review of Methods, Applications, and Clinical Translation
BibTeX
@article{saha2026fairness,
title = {{Fairness in multimodal machine learning applications in clinical decision support: a systematic review}},
author = {Amit Saha and Lexuan Shao and Jinman Kim and Adam G. Dunn},
journal = {NPJ Digital Medicine},
year = {2026},
month = jul,
doi = {10.1038/s41746-026-03000-x},
url = {https://doi.org/10.1038/s41746-026-03000-x},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - Fairness in multimodal machine learning applications in clinical decision support: a systematic review
AU - Amit Saha
AU - Lexuan Shao
AU - Jinman Kim
AU - Adam G. Dunn
PY - 2026
DA - 2026/07/14
JO - NPJ Digital Medicine
DO - 10.1038/s41746-026-03000-x
UR - https://doi.org/10.1038/s41746-026-03000-x
AB - Multimodal AI models are increasingly used in clinical decision support systems, but fairness appears to be rarely measured. In this systematic review, our aim was to assess studies that examine fairness in multimodal AI-based clinical decision support (CDS) systems. CDS systems that are rule-based or knowledge-based are considered out of scope. Two searches were used to identify literature focused on evaluating fairness in multimodal CDS systems and to determine the rate at which fairness is included in studies for two focus areas—chest x-rays and sepsis. The review was registered with PROSPERO (CRD42024579923). From 3059 search results, 160 articles included fairness evaluations, 11% (18/160) used multimodal data, and 29 different fairness measurement techniques were used. From 1422 search results related to the two focus areas, 88 studies using multimodal data for CDS related to chest x-ray and sepsis were included. Fairness evaluations were found in 8% (7 of 88) of these, despite 83% (74 of 88) including data on sensitive attributes that could have enabled fairness evaluation. The results show that fairness evaluations are uncommon for studies in multimodal CDS systems. AI-based multimodal CDS systems research would benefit from guidance and tools that support fairness evaluations.
SN - 2398-6352
ER -