The DOI matches its Crossref record; 93 of the 93 references it cites were found in a registry or an index; 53 of the 114 sources its claims cite were checked for support.
Bias recognition and mitigation strategies in artificial intelligence healthcare applications
Open accessAbstract
Artificial intelligence (AI) is delivering value across all aspects of clinical practice. However, bias may exacerbate healthcare disparities. This review examines the origins of bias in healthcare AI, strategies for mitigation, and responsibilities of relevant stakeholders towards achieving fair and equitable use. We highlight the importance of systematically identifying bias and engaging relevant mitigation activities throughout the AI model lifecycle, from model conception through to deployment and longitudinal surveillance.
Overview
CoChat Summary
TL;DR
This critical/narrative review examines the origins of bias in healthcare AI and mitigation strategies across the AI model lifecycle, from conception through deployment and surveillance. It argues that systematically identifying and mitigating bias is necessary to uphold fairness, equity, and equality in clinical AI.
Key findings
- As of May 13, 2024, the FDA listed 191 new AI-enabled medical devices, reaching 882 total, predominantly in radiology (76%), cardiology (10%), and neurology (4%). The review’s literature search yielded 233 potentially relevant articles, narrowed to 152 after title/abstract screening and 94 after full-text review.
- Cited evidence shows substantial bias burden: Kumar et al. found 50% of 48 sampled healthcare AI studies had high risk of bias and only 20% low risk; Chen et al. found only 86 of 555 neuroimaging-based psychiatric AI models (15.5%) included external validation, 97.5% included only high-income-region subjects, and 83% were rated high risk of bias.
- Case examples include Obermeyer et al.’s algorithm using healthcare costs as a proxy: among 43,539 White and 6,079 Black patients, Black patients had 26.3% more chronic illnesses at the same risk score (4.8 vs. 3.8 conditions); recalibration using direct health indicators nearly tripled high-risk Black patient enrollment from 17.7% to 46.5%. In cardiac MRI segmentation (nnU-Net, 5,903 UK Biobank subjects, 80% White), Dice similarity was 93.5% for White subjects but as low as 84.5% for Black and Mixed-race subjects; protected group models reached 92.15% for Black and 93.17% for Mixed-race subjects.
- Deployment biases are documented: incorrect AI BI-RADS suggestions dropped mammography reporting accuracy from 79.7% to 19.8% for inexperienced radiologists, 81.3% to 24.8% for moderately experienced, and 82.3% to 45.5% for very experienced radiologists; in a COVID-19 decision-support study, only 35% given risk score alone and 52% given risk score plus explanation critically evaluated incorrect recommendations; an ICU alarm study found 1,154,201 unique alarms, 88.8% for arrhythmias, with most false positives.
Methods
- The authors followed a critical review methodology, searching English articles published from 1993 to 2024 in PubMed, Google Scholar, and publisher platforms such as Elsevier, using Boolean operators and keywords including “Medical AI,” “AI bias,” “AI ethics,” “Fairness in medical AI,” and “Bias mitigation strategies in AI.”
- The search yielded 233 potentially relevant articles; after title/abstract screening, 152 were considered directly relevant, and 94 were selected for final review. A thematic-based approach was used to identify and categorize recurring themes, patterns, and insights related to bias types and mitigation strategies.
- The review is structured around bias origins and types—human, data, algorithmic, and model deployment—and mitigation recommendations across six AI lifecycle phases: conception, data collection, pre-processing, in-processing, post-processing, and post-deployment surveillance.
Limitations
- The paper states that achieving unbiased, broadly representative healthcare datasets remains formidable: data sparsity for underrepresented populations may not be feasible, prospective data quality improvement is time-consuming, sampling and participation biases are difficult to eliminate, standardizing data collection to reduce measurement bias is resource-intensive, vulnerable groups may remain underrepresented due to abundant WEIRD-population resources, and some socio-demographic features such as gender orientation remain poorly captured.
- The paper states that specific mitigation techniques have limitations: disaggregation can create noisy overly granular datasets; multiple imputation may introduce inaccuracies; data augmentation can generate synthetic data that fails to reflect true diversity; SMOTE can produce unrealistic data points; random under-sampling risks information loss; cost-sensitive learning can overfit; fairness metrics can create a fairness-accuracy trade-off; federated learning limits data pre-processing/quality assurance access; adversarial training can be computationally intensive and reduce accuracy; transfer learning may transfer unrecognized biases; explainability tools can be inconsistent or misleading; external validation data access is often challenging; and shadow deployment can delay implementation.
- The paper does not state limitations of the review itself; evident from the text, it is a critical/narrative review using thematic synthesis and does not report a formal risk-of-bias assessment or meta-analysis of the 94 included articles.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures




Verification
Claims the paper makes
46 Valid · 4 Not valid · 55 Not checked · 23 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“Overall, 83% of studies were rated at high ROB18.”
§ Bias recognition and mitigation strategies in artificial intelligence healthcare applications > Methods > Bias
What the sources say
- Supportedcites [18] Chen, Z. et al. Evaluation of Risk of Bias in Neuroimaging-Based Artificial Intelligence Models for Psychiatric Diagnosis: A Systematic Review. JAMA Netw. Open 6, e231671 (2023). source
The Abstract and the 'Overall ROB for AI Models' section of our copy state that 461 of 555 units (83.1%; 95% CI, 80.0%-86.2%) were rated as having a high overall risk of bias by PROBAST, and the Discussion repeats '83.1% of existing AI models exhibited a high ROB for psychiatric diagnosis'. The citing sentence's 83% figure therefore matches the source's headline estimate exactly; the source's unit of appraisal is the neuroimaging-based AI models derived from the included studies (searched for a separate study-level count and found none reported), so the paraphrase carries the same magnitude, population and rating instrument. (LLM, conf=0.85)
plain- 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 sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“During model development, developers may consciously or subconsciously select, interpret, or give more weight to data that confirms their beliefs, overemphasizing certain patterns while ignoring other patterns that don’t fit expectations. This form of human-mediated bias is called confirmation bias30,31.”
§ Bias recognition and mitigation strategies in artificial intelligence healthcare applications > Methods > Human Biases
What the sources say
- Not checked yetcites [30] Elston, D. M. Confirmation bias in medical decision-making. J. Am. Acad. Dermatol. 82, 572 (2020). source
- Supportedcites [31] Althubaiti, A. Information bias in health research: definition, pitfalls, and adjustment methods. J. Multidiscip. Healthc. 9, 211–217 (2016). source
The 'Confirmation bias' section of Althubaiti (2016) states that placing emphasis on one hypothesis because it does not contradict investigator beliefs is confirmation bias, that it is a psychological bias in which decisions are made according to the subject's preconceptions, beliefs or preferences, and that it emerges from overconfidence resulting in contradictory evidence being ignored or overlooked — matching the claim's 'give more weight to data that confirms their beliefs … ignoring other patterns'. The Introduction's statement that bias is often introduced unintentionally but can also be introduced intentionally covers the 'consciously or subconsciously' element, and the section notes confirmation bias affects any aspect of investigation requiring human judgment, including inclusion/exclusion criteria (selection). The source frames this around medical researchers/investigators rather than machine-learning model developers; the developers/model-development setting is the citing authors' own application, while the definitional content the citation is offered for ('this form of human-mediated bias is called confirmation bias') is stated directly in our copy. (LLM, conf=0.82)
hedgedstates a possible 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.
Every source the paper cites for this claim was checked and supports it.
In the paper
“examined 555 published neuroimaging-based AI models for psychiatric diagnosis”
§ Bias recognition and mitigation strategies in artificial intelligence healthcare applications > Methods > Bias
What the sources say
- Supportedcites [18] Chen, Z. et al. Evaluation of Risk of Bias in Neuroimaging-Based Artificial Intelligence Models for Psychiatric Diagnosis: A Systematic Review. JAMA Netw. Open 6, e231671 (2023). source
The Abstract (FINDINGS, p. 3 of our copy) states that 517 studies presenting 555 AI models were included and evaluated, matching the claimed number exactly. The Data Extraction and Analysis section repeats the figure ('Of the 555 artificial intelligence (AI) models, 469 were tested at the internal sample and 86 at the independent external sample'), and the Retrieval Strategy/Introduction confirm the models are neuroimaging-based and aimed at psychiatric diagnosis. (LLM, conf=0.94)
plain- 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
Every source the paper cites for this claim was checked and supports it.
In the paper
“A similar temporally sensitive bias, known as “Concept shift,” can occur due to changes in the perceived meanings of data. How clinicians perceive and code diseases or outcomes can change dynamically over time, introducing a unique, human-level bias surrounding what data means and is being used for.”
§ Bias recognition and mitigation strategies in artificial intelligence healthcare applications > Methods > Human Biases
What the sources say
- Supportedcites [33] Chen, R. J. et al. Algorithmic fairness in artificial intelligence for medicine and healthcare. Nat. Biomed. Eng. 7, 719–742 (2023). source
The 'Evolving dataset shifts over time' subsection (p. under 'Healthcare disparities arising from dataset shift') defines concept shift/concept drift as arising from 'changes in diagnostic criteria and in labelling paradigms across populations', i.e. a changed relationship between X and Y when 'the criteria for diagnosing disease Y are revised over time', illustrated by the ICD-8→ICD-9→ICD-10→ICD-11 coding migrations and by repeated revisions of classification systems (Banff, WHO/ISUP, glioma taxonomy). This matches the claim that concept shift stems from changes in the perceived/coded meaning of disease or outcome data over time, and the passage's placement under temporal/evolving dataset shifts supports the citing sentence's characterisation of it as a temporally sensitive bias. (LLM, conf=0.88)
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.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“This complexity is compounded by the inadequacy of methods for routinely detecting or mitigating biases across various stages of an algorithm’s life cycle, emphasizing a need for comprehensive and holistic bias detection frameworks14,15.”
§ Bias recognition and mitigation strategies in artificial intelligence healthcare applications > Methods > Bias
What the sources say
- Supportedcites [14] Kumar, A. et al. Artificial intelligence bias in medical system designs: a systematic review. Multimed. Tools Appl 83, 18005–18057 (2024). source
Our copy treats the shortfall directly: §4.3.4 (Analytical strategy to evaluate AI-bias) reports that most reviewed studies fall in the high-bias zone because 'the studies have not considered the imperative evaluation parameters for bias analysis', 'the bias qualitative analysis is also not performed... to identify the RoB', and 'the bias mitigation steps such as feature selection and feature augmentation are rarely performed in AI-studies' — i.e. routine detection and mitigation are inadequate. The Introduction reinforces this with 'To the best of our knowledge, no work has qualitatively and quantitatively analyzed the bias in AI-based prediction models in the healthcare domain.' The life-cycle framing the claim invokes is the paper's own structure (§3 Bias life cycle, stages I–III; §5 mitigation at pre-, in- and post-processing), and §7 Conclusion recommends robust analysis plus clinical validation before deployment. The only softening is that the paper locates the inadequacy in the practice of the surveyed studies rather than in the methods as such. (LLM, conf=0.82)
- Partly supportedcites [15] Chin, M. H. et al. Guiding Principles to Address the Impact of Algorithm Bias on Racial and Ethnic Disparities in Health and Health Care. JAMA Netw. Open 6, e2345050 (2023). sourceExperimental — accuracy not yet measured
Our copy conveys the surrounding idea but not the claim as stated: it never asserts that methods for routinely detecting or mitigating bias across the algorithm life cycle are 'inadequate.' What it does say is (a) in Guiding Principle 4 that 'technical definitions and metrics of fairness often do not translate clearly or intuitively to ethical, legal, social, and economic conceptions of fairness' and that 'there is no simple cookie-cutter technical solution,' and (b) under Guiding Principle 2 that existing reporting guidelines 'do not include concrete ways to report on fairness, and they rarely make explicit mention of equity.' Its life-cycle statements are normative rather than a finding of inadequacy — e.g., 'Equity and fairness checks should be built into each phase of the algorithm life cycle' (Guiding Principle 5) and that outcomes 'should be assessed routinely' (Algorithm Life Cycle) — so the part of the claim that routine detection/mitigation methods ARE inadequate is not stated by this source, only implied by its recommendations. (LLM, conf=0.72)
plain- 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
65 more claims · 35 with their cited sources checked
Create a free account to see all 128 claims and to verify the 61 references still open.
- Probably not a substantial claim. 18
- It may state two separate claims as one. 46
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
“Moreover, frameworks developed in studies4,11,35 enhance methodological theory by incorporating AI ethics, transparency, and bias mitigation into evidence appraisal models like PRISMA-AI and CLASMOD-AI.”
From Generative Artificial Intelligence in Healthcare: Automation Bias, Deskilling, and Cognitive Implications – A Systematic Review (2026) · cites this paper as [4]
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
“In the same way, ethical and psychological aspects4,33 show that clinician trust and identity need to be actively built through open governance and clear AI design.”
From Generative Artificial Intelligence in Healthcare: Automation Bias, Deskilling, and Cognitive Implications – A Systematic Review (2026) · cites this paper as [4]
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
“Ethical governance frameworks, like the ones suggested in,4,49 stress the importance of clear rules, accountability systems, and algorithmic transparency in order to protect clinician independence and patient trust.”
From Generative Artificial Intelligence in Healthcare: Automation Bias, Deskilling, and Cognitive Implications – A Systematic Review (2026) · cites this paper as [4]
Experimental — accuracy not yet measuredIn the citing paper
“Nevertheless, we draw a tentative conclusion that a minimal level of ML and data literacy should be included, consistent with the predominant expert view and patient-safety imperatives [31-34].”
From Role-Differentiated AI Competencies and Curriculum Implications for Health Professions Education: Qualitative Study (2026) · cites this paper as [33]
Citations
Cites
The PRISMA 2020 statement: An updated guideline for reporting systematic reviews
Page, M. J., McKenzie, J. E., Bossuyt, P. M., Boutron, I., Hoffmann, T. C., Mulrow, C. D., Shamseer, L., Tetzlaff, J. M., Akl, E. A., Brennan, S. E., et al. (2021) · BMJ · doi:10.1136/bmj.n71 · Cited by 108,283
SMOTE synthetic minority over-sampling technique
Chawla, N. V., Bowyer, K. W., Hall, L. O., & Kegelmeyer, W. P. (2002) · Journal of Artificial Intelligence Research · doi:10.1613/jair.953 · Cited by 32,865 · cited 2×
“To address class imbalance, strategies such as resampling to balance class distributions 60 , synthetic data generation (such as Synthetic Minority Over-sampling Technique (SMOTE)) 61 , and the application of cost-sensitive learning to emphasize minority class errors should be considered 62 .”
A typology of reviews: an analysis of 14 review types and associated methodologies
Grant, M. J., & Booth, A. (2009) · Health Information & Libraries Journal · doi:10.1111/j.1471-1842.2009.00848.x · Cited by 10,817
Dissecting 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×
“However, ongoing surveillance is necessary, as reliance on historical data and evolving healthcare dynamics could allow biases to re-emerge 46 .”
A 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 · cited 1×
“An example of this is the “ healthy volunteer ” selection bias observed in the UK Biobank, where participants are generally healthy and therefore do not represent patients typically encountered by healthcare systems 39 .”
The PRISMA 2020 statement: an updated guideline for reporting systematic reviews
Page, M. J., McKenzie, J., Bossuyt, P., Boutron, I., Hoffmann, T., mulrow, C., Shamseer, L., Tetzlaff, J., Akl, E., Brennan, S. E., et al. (2020) · doi:10.31222/osf.io/v7gm2 · Cited by 5,177
Information bias in health research: definition, pitfalls, and adjustment methods
Althubaiti, A. (2016) · Journal of Multidisciplinary Healthcare · doi:10.2147/jmdh.s104807 · Cited by 3,018 · cited 1×
“For example, systemic bias may manifest as inadequate medical resource funding for un-insured individuals, underserved 30,31 .”
PROBAST: A Tool to Assess the Risk of Bias and Applicability of Prediction Model Studies
Wolff, R. F., Moons, K. G. M., Riley, R. D., Whiting, P. F., Westwood, M., Collins, G. S., Reitsma, J. B., Kleijnen, J., Mallett, S., & for the PROBAST Group† (2019) · Annals of Internal Medicine · doi:10.7326/m18-1376 · Cited by 2,984 · cited 1×
“A similar study, performed by Chen, et al., using the PROBAST (Prediction model Risk Of Bias ASsessment Tool) framework 17 (For further details, refer to Supplementary Table 1), examined 555 published neuroimaging-based AI models for psychiatric diagnosis, identifying only 86 studies (15.5”
InfluentialExplainable AI: a review of machine learning interpretability methods
Linardatos, P., Papastefanopoulos, V., & Kotsiantis, S. (2020) · Entropy · doi:10.3390/e23010018 · Cited by 2,922 · cited 3×
“For example, while decision tree models provide clear feature usage insights, deep learning models offer limited insights 66 .”
InfluentialImplicit bias in healthcare professionals: a systematic review
FitzGerald, C., & Hurst, S. (2017) · BMC Medical Ethics · doi:10.1186/s12910-017-0179-8 · Cited by 2,556 · cited 1×
“This can present surreptitiousbutpowerfulin fl uencesonmedicalAIsystemstrainedfrom these decisions, particularly when features contributing to this bias are not routinely captured by training data 26 .”
Research works printed but not linked: 2
- Jackson, M. C. Artificial Intelligence & Algorithmic Bias: The Issues with Technology Reflecting History & Humans Notes & Comments. J. Bus. Tech. L. 16, 299–316 (2021).
- Wachter, S., Mittelstadt, B. & Russell, C. Counterfactual explanations without opening the black box: Automated decisions and the GDPR. Harv. JL & Tech. 31, 841 (2017).
Cited by
Multimodal Large Language Models in Medical Imaging: Current State and Future Directions
Nam, Y., Kim, D., Kyung, S., Seo, J., Song, J. M., Kwon, J., Kim, J., Jo, W., Park, H. J., Sung, J., et al. (2025) · Korean Journal of Radiology · doi:10.3348/kjr.2025.0599 · Cited by 152
“Effective risk mitigation hinges on balanced, expertly curated datasets, continuous post-deployment auditing, and cross-institutional benchmarking [125,126].”
Nourazarain, A., & Vaziri, Y. (2025) · Genes & Nutrition · doi:10.1186/s12263-025-00790-9 · Cited by 90
“…Data Protection Regulation (GDPR) and Health Insurance Por-tability and Accountability Act (HIPAA), alongside the adoption of privacy-preserving techniques like federated learning, which enables model training across decentralized datasets without centralizing raw patient information [126, 151].”
Peykani, P., Ramezanlou, F., Tănăsescu, C. R., & Ghanidel, S. (2025) · Applied Sciences · doi:10.3390/app15148103 · Cited by 82
Khan, S. N., Danishuddin, Khan, M. W. A., Guarnera, L., & Akhtar, S. M. F. (2026) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2025.1743921 · Cited by 76
Racacho, K. J., Shiau, Y., Villa, R., Mahri, S., Tang, M., Lin, T., & Li, Y. (2025) · Frontiers in Immunology · doi:10.3389/fimmu.2025.1621812 · Cited by 70
Metadata
- Authors
- Hasanzadeh, F., Josephson, C. B., Waters, G., Adedinsewo, D., Azizi, Z., & White, J. A.
- Affiliations
- University of Calgary (CA)
- Morgan State University (US)
- Mayo Clinic (US)
- Stanford Medicine (US)
- Corresponding author
- White, James A.
- ORCID
- Venue
- npj Digital Medicine
- Published
- 2025-03-11
- Volume
- 8
- Issue
- 1
- Pages
- 154-154
- ISSN
- 2398-6352 (ISSN-L)
- PMC
- PMC11897215
- OpenAlex
- W4408314709
- Semantic Scholar
- d25dafe6aee6f79c641ef84b9d0982f2870a8731
- Cited by
- 320
- References
- 93
- Citations by year
- 2026: 221 · 2025: 97
- FWCI
- 49.28
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- cc-by-nc-nd
- creativecommons.org/licenses/by-nc-nd/4.0 (tdm)
- creativecommons.org/licenses/by-nc-nd/4.0 (vor)
- Locations
- npj Digital Medicine (publishedVersion, cc-by-nc-nd)
- Europe PMC (PubMed Central) (publishedVersion, cc-by-nc-nd)
- PubMed Central (publishedVersion, cc-by-nc-nd)
- PubMed (publishedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
- Keywords
- algorithmic bias, artificial intelligence in healthcare, bias mitigation, health care disparities, algorithmic fairness, model life cycle, bias detection, stakeholder responsibility, long-term monitoring
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).
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: Bias recognition and mitigation strategies in artificial intelligence healthcare applications. https://app.cochat.ai/papers/10.1038/s41746-025-01503-7
- The paper
- Hasanzadeh, F., Josephson, C. B., Waters, G., Adedinsewo, D., Azizi, Z., & White, J. A. (2025). Bias recognition and mitigation strategies in artificial intelligence healthcare applications. npj Digital Medicine. https://doi.org/10.1038/s41746-025-01503-7
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.