Fairness of artificial intelligence in healthcare: review and recommendations
Open accessThe DOI matches its Crossref record; 127 of the 127 references it cites were found in a registry or an index; 19 of the 73 sources its claims cite were checked for support.
Abstract
In this review, we address the issue of fairness in the clinical integration of artificial intelligence (AI) in the medical field. As the clinical adoption of deep learning algorithms, a subfield of AI, progresses, concerns have arisen regarding the impact of AI biases and discrimination on patient health. This review aims to provide a comprehensive overview of concerns associated with AI fairness; discuss strategies to mitigate AI biases; and emphasize the need for cooperation among physicians, AI researchers, AI developers, policymakers, and patients to ensure equitable AI integration. First, we define and introduce the concept of fairness in AI applications in healthcare and radiology, emphasizing the benefits and challenges of incorporating AI into clinical practice. Next, we delve into concerns regarding fairness in healthcare, addressing the various causes of biases in AI and potential concerns such as misdiagnosis, unequal access to treatment, and ethical considerations. We then outline strategies for addressing fairness, such as the importance of diverse and representative data and algorithm audits. Additionally, we discuss ethical and legal considerations such as data privacy, responsibility, accountability, transparency, and explainability in AI. Finally, we present the Fairness of Artificial Intelligence Recommendations in healthcare (FAIR) statement to offer best practices. Through these efforts, we aim to provide a foundation for discussing the responsible and equitable implementation and deployment of AI in healthcare.
Overview
CoChat Summary
TL;DR
This narrative review examines fairness and bias in clinical artificial intelligence (AI), especially radiology, and proposes the FAIR statement of 10 best-practice recommendations for equitable implementation. It synthesizes sources of bias, mitigation strategies, ethical/legal issues, and stakeholder collaboration, without presenting original empirical data.
Key findings
- Fairness in healthcare AI is multidimensional, requiring equitable distribution of resources, opportunities, and outcomes; biases arise from data, algorithms, clinician–AI interactions, and patient–AI interactions.
- Data biases include minority bias, missing data bias, informativeness bias, and training–serving skew; algorithmic biases include label bias and cohort bias. Examples include cardiovascular risk algorithms trained primarily on male patient data, a pneumonia chest X-ray model performing better on internal than external data, and a commercial healthcare-needs algorithm using cost as a proxy that underestimated the needs of Black patients.
- Mitigation strategies include diverse and representative data, independent algorithm audits and validation, and education for clinicians and patients; ethical/legal considerations include data privacy and security, liability and accountability, and transparency and explainability.
- The FAIR statement contains 10 recommendations: diverse data; independent audits; education; privacy/security; liability/accountability; transparency/explainability; collaboration between physicians, AI researchers, and developers; policymaker/regulatory involvement; patient/advocacy group participation; and professional association support.
Methods
- Narrative review; no systematic search strategy, inclusion criteria, data extraction, or quantitative synthesis is described.
- Synthesizes prior literature and examples across healthcare and radiology AI fairness; no original dataset, sample, or empirical analysis is reported.
- Main output is a conceptual framework and the FAIR recommendations.
Limitations
- The paper states that implementing each best practice is difficult, AI technology is still evolving with new challenges, explainable AI has limitations such as confirmation bias, and independent bias research is difficult because large-scale deployed algorithms are generally proprietary.
- Evident from the text: as a narrative review, it lacks a systematic search and quality appraisal and does not empirically test the FAIR recommendations; no original quantitative evidence is reported.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
8 Valid · 0 Not valid · 34 Not checked · 13 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Another bias related to interactions is the feedback loop [80]. This occurs when clinicians accept AI recommendations even if they are incorrect, leading the algorithm to relearn and perpetuate the same mistakes.”
§ Fairness of artificial intelligence in healthcare: review and recommendations > Fairness concerns in healthcare > Clinician interaction-related biases
What the sources say
- Partly supportedcites [80] Walsh CG, Chaudhry B, Dua P, Goodman KW, Kaplan B, Kavuluru R, et al. Stigma, biomarkers, and algorithmic bias: recommendations for precision behavioral health with artificial intelligence. JAMIA Open. 2020;3:9–15. https://doi.org/10.1093/jamiaopen/ooz054. sourceExperimental — accuracy not yet measured
Our copy does identify a 'harmful feedback loop' as a critical issue in algorithmic bias (CHALLENGES > Potential for algorithmic bias), but the mechanism it describes is different from the claim's: existing disparities lead to unrepresentative training data, bias seeps into predictive models, and biased predictions further exacerbate disparities — not clinicians accepting incorrect AI recommendations which the algorithm then relearns. The paper's closest related point is the recommendation that users have channels to disagree with recommendations so algorithms and designers can learn from those disagreements (RECOMMENDATIONS > Foster trust through education and transparency), plus a warning that providers 'might feel compelled to respond' to high-risk designations (CHALLENGES > Inappropriate use, interpretability/explainability, and trust); neither states the relearning/perpetuation loop the citing sentence attributes. So the existence of a feedback-loop bias is supported while the specific clinician-over-reliance mechanism is not. (LLM, conf=0.70)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“In a study training AI to diagnose pneumonia from chest X-rays, the performance on unseen data from the institution where the training data were collected was significantly higher than its performance on data collected from external hospitals [72].”
§ Fairness of artificial intelligence in healthcare: review and recommendations > Fairness concerns in healthcare > Data biases
What the sources say
- Partly supportedcites [72] Zech JR, Badgeley MA, Liu M, Costa AB, Titano JJ, Oermann EK. Variable generalization performance of a deep learning model to detect pneumonia in chest radiographs: a cross-sectional study. PLOS Med. 2018. https://doi.org/10.1371/journal.pmed.1002683. sourceExperimental — accuracy not yet measured
Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
None of the sources cited for this claim have been checked yet.
In the paper
“For example, identifying melanoma from images of patients with dark skin is more challenging than those with light skin [70, 71].”
§ Fairness of artificial intelligence in healthcare: review and recommendations > Fairness concerns in healthcare > Data biases
What the sources say
- Not checked yetcites [70] Adamson AS, Smith A. Machine learning and health care disparities in dermatology. JAMA Dermatol. 2018;154:1247–8. https://doi.org/10.1001/jamadermatol.2018.2348. source
- Could not verifycites [71] Navarrete-Dechent C, Dusza SW, Liopyris K, Marghoob AA, Halpern AC, Marchetti MA. Automated dermatological diagnosis: hype or reality? J Invest Dermatol. 2018;138:2277–9. https://doi.org/10.1016/j.jid.2018.04.040. sourceExperimental — accuracy not yet measured
The main text (page unavailable) reports differences in algorithm performance between Asian and Caucasian datasets, but does not say which performed better or compare dark and light skin. Its melanoma test used images only from Caucasian patients. A search for a dark-versus-light-skin comparison found no passage supporting the claimed direction of difficulty. (LLM, conf=0.74)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Every source the paper cites for this claim was checked and supports it.
In the paper
“For example, significant racial bias has been observed in commercially available algorithms used to predict patients' healthcare needs.”
§ Fairness of artificial intelligence in healthcare: review and recommendations > Fairness concerns in healthcare > Algorithmic biases
What the sources say
- Supportedcites [76] Obermeyer Z, Powers B, Vogeli C, Mullainathan S. Dissecting racial bias in an algorithm used to manage the health of populations. Science. 2019;366:447–53. https://doi.org/10.1126/science.aax2342. source
The paper states on p. 2 that a widely used commercial prediction algorithm for identifying patients with complex health needs exhibits significant racial bias. The citing sentence presents this as an example, rather than claiming that the paper tested every commercial algorithm. (LLM, conf=0.92)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“When healthcare professionals interact with AI, biases related to interactions can occur, potentially affecting the algorithm's performance, fairness, and adoption [11].”
§ Fairness of artificial intelligence in healthcare: review and recommendations > Fairness concerns in healthcare > Clinician interaction-related biases
What the sources say
- Partly supportedcites [11] Rajkomar A, Hardt M, Howell MD, Corrado G, Chin MH. Ensuring fairness in machine learning to advance health equity. Ann Intern Med. 2018;169:866–72. https://doi.org/10.7326/M18-1990. sourceExperimental — accuracy not yet measured
In “What Can Cause a Machine-Learning System to Be Unfair?” (page unavailable), the paper describes automation and dismissal biases when clinicians trust or ignore model predictions, and says these interactions can affect patient outcomes and resource allocation. It does not say those interactions change the algorithm’s predictive performance or address their effect on AI adoption. (LLM, conf=0.86)
plainstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
9 more claims
Create a free account to see all 55 claims and to verify the 54 references still open.
- Probably not a substantial claim. 31
- It may state two separate claims as one. 24
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“This deployment bias has been shown to lead to the dismissal of appropriate critical warnings, with potential risk of patient harm47,49.”
From Bias recognition and mitigation strategies in artificial intelligence healthcare applications (2025) · cites this paper as [49]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Studies27,29,30,42 provide empirical evidence that GenAI tools can significantly improve clinical accuracy, workflow efficiency, and decision-support capacity when implemented under structured supervision.”
From Generative Artificial Intelligence in Healthcare: Automation Bias, Deskilling, and Cognitive Implications – A Systematic Review (2026) · cites this paper as [30]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Systematic reviews of AI fairness in healthcare have demonstrated that models trained on demographically skewed datasets can produce reduced accuracy and exacerbate health disparities in underrepresented groups412.”
From Artificial intelligence-driven multi-omics approaches in Alzheimer's disease: Progress, challenges, and future directions (2025) · cites this paper as [412]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Governance frameworks that incorporate ethics boards and patient advocacy representatives throughout the development lifecycle also play a critical role in ensuring transparency, addressing algorithmic bias, and promoting equitable use of AI systems412.”
From Artificial intelligence-driven multi-omics approaches in Alzheimer's disease: Progress, challenges, and future directions (2025) · cites this paper as [412]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“However, many of these guidelines neglect the importance of addressing bias at multiple stages of model development.”
From Advancing Health Equity through Multi-Level Fairness in Health Informatics (2026) · cites this paper as [7]
Experimental — accuracy not yet measured
2 more claims about this paper
Create a free account to see every claim other papers make about this one.
Citations
Cites
Meyer, I. H. (2003) · Psychological Bulletin · doi:10.1037/0033-2909.129.5.674 · Cited by 14,736
Arrieta, A. B., Díaz-Rodríguez, N., Del Ser, J., Bennetot, A., Tabik, S., Barbado, A., García, S., Gil-López, S., Molina, D., Benjamins, R., et al. (2019) · arXiv · doi:10.48550/arxiv.1910.10045 · Cited by 11,321 · cited 1×
“Fostering transparency and account‑ ability in AI is also an important aspect of their respon‑ sibilities [5, 116].”
BackgroundRudin, C. (2019) · Nature Machine Intelligence · doi:10.1038/s42256-019-0048-x · Cited by 10,561
High-performance medicine: the convergence of human and artificial intelligence
Topol, E. J. (2019) · Nature Medicine · doi:10.1038/s41591-018-0300-7 · Cited by 10,450 · cited 1×
“Collaboration among physicians, AI researchers, and AI developers is essential to address fairness concerns in AI [108, 110].”
BackgroundBarredo Arrieta, A., Díaz-Rodríguez, N., Del Ser, J., Bennetot, A., Tabik, S., Barbado, A., Garcia, S., Gil-Lopez, S., Molina, D., Benjamins, R., et al. (2020) · Information Fusion · doi:10.1016/j.inffus.2019.12.012 · Cited by 10,423 · cited 1×
“Fostering transparency and account‑ ability in AI is also an important aspect of their respon‑ sibilities [5, 116].”
BackgroundDissecting racial bias in an algorithm used to manage the health of populations
Obermeyer, Z., Powers, B., Vogeli, C., & Mullainathan, S. (2019) · Science · doi:10.1126/science.aax2342 · Cited by 7,137 · cited 2×
“Although several biases were affected, a major contributing factor to this algorithm’s bias was its design, which used cost as a proxy for healthcare needs, leading to an underestimation of the needs of Black patients compared with those who are White with similar conditions [76].”
BackgroundThe global landscape of AI ethics guidelines
Jobin, A., Ienca, M., & Vayena, E. (2019) · Nature Machine Intelligence · doi:10.1038/s42256-019-0088-2 · Cited by 6,220 · cited 1×
“This will ensure that AI systems can better recognize, diagnose, and treat a wide range of patient conditions, reducing disparities and promoting equity in healthcare outcomes 2.”
BackgroundA Survey on Bias and Fairness in Machine Learning
Mehrabi, N., Morstatter, F., Saxena, N., Lerman, K., & Galstyan, A. (2022) · ACM Computing Surveys · doi:10.1145/3457607 · Cited by 5,812
A Survey on Bias and Fairness in Machine Learning
Mehrabi, N., Morstatter, F., Saxena, N., Lerman, K., & Galstyan, A. (2019) · arXiv · doi:10.48550/arxiv.1908.09635 · Cited by 5,812
Lipton, Z. C. (2018) · Queue · doi:10.1145/3236386.3241340 · Cited by 5,427 · cited 1×
“However, recognizing the limitations of explainability is important [104–107].”
Background
Research works printed but not linked: 4
- Binns R. Fairness in machine learning: Lessons from political philosophy. In: Friedler SA, Wilson C, editors. Proceedings of the 1st conference on fairness, accountability and transparency. PMLR; 2018. p. 149–59
- Minkler M, Wallerstein N. Community-based participatory research for health: From process to outcomes. John Wiley & Sons; 2011
- Buolamwini J, Gebru T. Gender shades: Intersectional accuracy disparities in commercial gender classification. In: Friedler SA, Wilson C, editors. Proceedings of the 1st conference on fairness, accountability and transparency. PMLR; 2018. p. 77–91
- Coulter A, Collins A. Making shared decision-making a reality: No decision about me, without me. London: The King’s Fund; 2011;621.
Cited by
Lekadir, K., Frangi, A. F., Porras, A. R., Glocker, B., Cintas, C., Langlotz, C. P., Weicken, E., Asselbergs, F. W., Prior, F., Collins, G. S., et al. (2025) · BMJ · doi:10.1136/bmj-2024-081554 · Cited by 957
AI in diagnostic imaging: Revolutionising accuracy and efficiency
Khalifa, M., & Albadawy, M. (2024) · Computer Methods and Programs in Biomedicine Update · doi:10.1016/j.cmpbup.2024.100146 · Cited by 517
Fahim, Y. A., Hasani, I. W., Kabba, S., & Ragab, W. M. (2025) · European Journal of Medical Research · doi:10.1186/s40001-025-03196-w · Cited by 329
“Furthermore, managing sensitive patient data raises privacy concerns, where breaches threaten both ethical and legal standards [160].”
Bias recognition and mitigation strategies in artificial intelligence healthcare applications
Hasanzadeh, F., Josephson, C. B., Waters, G., Adedinsewo, D., Azizi, Z., & White, J. A. (2025) · npj Digital Medicine · doi:10.1038/s41746-025-01503-7 · Cited by 320
Ethical and legal considerations in healthcare AI: innovation and policy for safe and fair use
Pham, T. (2025) · Royal Society Open Science · doi:10.1098/rsos.241873 · Cited by 261
Metadata
- Authors
- Ueda, D., Kakinuma, T., Fujita, S., Kamagata, K., Fushimi, Y., Ito, R., Matsui, Y., Nozaki, T., Nakaura, T., Fujima, N., Tatsugami, F., Yanagawa, M., Hirata, K., Yamada, A., Tsuboyama, T., Kawamura, M., Fujioka, T., & Naganawa, S.
- Affiliations
- Osaka Metropolitan University (JP)
- The University of Tokyo (JP)
- Juntendo University (JP)
- Kyoto University (JP)
- Nagoya University (JP)
- Okayama University (JP)
- Keio University (JP)
- Kumamoto University (JP)
- Hokkaido University Hospital (JP)
- Hiroshima University (JP)
- The University of Osaka (JP)
- Hokkaido University (JP)
- Shinshu University (JP)
- Tokyo Medical and Dental University (JP)
- Corresponding author
- Ueda, Daiju
- ORCID
- Venue
- Japanese Journal of Radiology
- Published
- 2023-08-04
- Volume
- 42
- Issue
- 1
- Pages
- 3-15
- ISSN
- 1867-1071 (ISSN-L)
- 1867-108X
- PMC
- PMC10764412
- OpenAlex
- W4385564466
- Semantic Scholar
- a8c6a60f0b44ec06344d6b7a9570369dc3484465
- Cited by
- 607
- References
- 120
- Citations by year
- 2026: 191 · 2025: 267 · 2024: 137 · 2023: 11
- FWCI
- 7.12
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- creativecommons.org/licenses/by/4.0 (tdm)
- creativecommons.org/licenses/by/4.0 (vor)
- Locations
- Japanese Journal of Radiology (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- Institutional Repositories DataBase (IRDB) (submittedVersion, cc-by)
- Kyoto University Research Information Repository (Kyoto University) (submittedVersion, cc-by)
- Topics
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
- Keywords
- algorithmic fairness, algorithmic bias, artificial intelligence in healthcare, deep learning, radiology, bias mitigation, algorithmic auditing, explainability, transparency, accountability, data privacy, legal and ethical considerations, misdiagnosis
- MeSH terms
- Radiologists, Algorithms, Artificial Intelligence, Delivery of Health Care, Humans, Radiology
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-05, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-05 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (qwen3.8-omni-flash and gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Fairness of artificial intelligence in healthcare: review and recommendations. https://app.cochat.ai/papers/10.1007/s11604-023-01474-3
- The paper
- Ueda, D., Kakinuma, T., Fujita, S., Kamagata, K., Fushimi, Y., Ito, R., Matsui, Y., Nozaki, T., Nakaura, T., Fujima, N., Tatsugami, F., Yanagawa, M., Hirata, K., Yamada, A., Tsuboyama, T., Kawamura, M., Fujioka, T., & Naganawa, S. (2024). Fairness of artificial intelligence in healthcare: review and recommendations. Japanese Journal of Radiology. https://doi.org/10.1007/s11604-023-01474-3
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.
