The DOI matches its Crossref record; 41 of the 41 references it cites were found in a registry or an index; 19 of the 76 sources its claims cite were checked for support.
Considerations for addressing bias in artificial intelligence for health equity
Open accessAbstract
Health equity is a primary goal of healthcare stakeholders: patients and their advocacy groups, clinicians, other providers and their professional societies, bioethicists, payors and value based care organizations, regulatory agencies, legislators, and creators of artificial intelligence/machine learning (AI/ML)-enabled medical devices. Lack of equitable access to diagnosis and treatment may be improved through new digital health technologies, especially AI/ML, but these may also exacerbate disparities, depending on how bias is addressed. We propose an expanded Total Product Lifecycle (TPLC) framework for healthcare AI/ML, describing the sources and impacts of undesirable bias in AI/ML systems in each phase, how these can be analyzed using appropriate metrics, and how they can be potentially mitigated. The goal of these “Considerations” is to educate stakeholders on how potential AI/ML bias may impact healthcare outcomes and how to identify and mitigate inequities; to initiate a discussion between stakeholders on these issues, in order to ensure health equity along the expanded AI/ML TPLC framework, and ultimately, better health outcomes for all.
Overview
CoChat Summary
TL;DR
The paper proposes expanding the Total Product Lifecycle (TPLC) framework to address bias and health equity across all phases of AI/ML-enabled medical devices. It argues that bias can arise or persist at every phase and should be quantified with equity metrics such as subgroup accuracy disaggregation and population-achieved sensitivity/specificity.
Key findings
- Bias in AI/ML healthcare can exacerbate existing inequities or create new disparities; the paper cites an example where an algorithm intended to identify patients needing extra help actually predicted healthcare costs, resulting in sicker Black patients receiving similar care to healthier White patients despite needing higher acuity care.
- Equity impact can occur at every TPLC phase and is independent across phases: even when bias is mitigated in earlier phases, a later phase can still introduce undesirable equity effects.
- Proposed equity metrics include sensitivity disaggregated by demographic subgroup, differential clinical outcome across subgroups, and population-achieved sensitivity/specificity to measure the impact of access and so-called “invisible populations.”
- Mitigation opportunities include diverse creator teams, racially invariant priors, representative training and validation datasets, diverse clinical sites, and monitoring real-world access, cost, and treatment follow-through.
Methods
- Conceptual/perspective paper with no empirical study, dataset, or statistical analysis; it synthesizes existing literature and regulatory frameworks, including Char et al., FDA’s TPLC approach, IMDRF, and Good Machine Learning Practice.
- The authors map TPLC phases—Conception, Design, Development, Validation, Access and marketing, and Monitoring—to sources of bias, equity metrics, and potential mitigation strategies.
- Analysis is narrative and illustrative, using Fig. 1 examples rather than original data collection or quantitative evaluation.
Limitations
- The paper states its framework is not intended to be comprehensive for all bias risks and mitigations and is meant to initiate discussion among stakeholders.
- The paper states it is beyond its scope to exhaustively list metrics for ethics and instead gives examples to illustrate the principle.
- Evident from the text: the proposed expanded TPLC framework is not empirically validated or quantitatively tested in this paper, and it relies on cited examples and conceptual reasoning.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures
Verification
Claims the paper makes
11 Valid · 0 Not valid · 30 Not checked · 28 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Thus, the inherent bias in the algorithm appeared to contribute to worse outcomes for Black patients by influencing the likelihood they would receive the appropriate level of care25.”
§ Considerations for addressing bias in artificial intelligence for health equity > Introduction
What the sources say
- Partly supportedcites [25] Obermeyer, Z., Powers, B., Vogeli, C. & Mullainathan, S. Dissecting racial bias in an algorithm used to manage the health of populations. Science 366, 447–453 (2019). sourceExperimental — accuracy not yet measured
The paper’s opening summary (p. 2) reports that the algorithm’s racial bias affected which Black patients were identified to receive additional help; correcting it would increase the share of Black patients receiving that help from 17.7% to 46.5%. This supports the claim about likelihood of care, but the cited passages do not establish that the algorithm caused worse clinical outcomes for those patients. (LLM, conf=0.85)
hedgedstates 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
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“As an example, where there is concern about access to a diagnostic test, population-achieved sensitivity and specificity, which measures the impact of both access and sensitivity, has been proposed as a way to understand the impact of bias on population health, when including so-called invisible populations32.”
§ Considerations for addressing bias in artificial intelligence for health equity > Introduction > Measuring bias in AI systems
What the sources say
- Partly supportedcites [32] Abramoff, M. D. et al. Foundational considerations for artificial intelligence using ophthalmic images. Ophthalmology 129, e14–e32 (2022). sourceExperimental — accuracy not yet measured
The Equity section of our copy does state that "population-achieved sensitivity, or access-corrected sensitivity, is used to analyze such effects on equity," explicitly combining sensitivity (sc) with compliance/access (c) to give the effective sensitivity at the population level, and frames this around inappropriate bias and health disparities. However, two elements of the claim are absent: our copy only ever defines population-achieved SENSITIVITY (PAS) — there is no "population-achieved specificity" anywhere (searched: population-achieved specificity, access-corrected specificity) — and it never refers to "invisible populations"; the populations it corrects for are the "nonadherent"/differentially-served subgroups defined by race, ethnicity, age, sex and income. So the access-plus-sensitivity framing and the bias/equity-at-population-level purpose are supported, while the specificity extension and the 'invisible populations' qualifier are not. (LLM, conf=0.78)
plain- 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
Every source the paper cites for this claim was checked and supports it.
In the paper
“The impact of such differential access can be measured for example through population-achieved sensitivity32 compared to overall sensitivity, i.e., the fraction of correctly identified disease cases in a sample, without regard to representativeness of that sample (or lack thereof).”
§ Considerations for addressing bias in artificial intelligence for health equity > Introduction > Conception phase
What the sources say
- Supportedcites [32] Abramoff, M. D. et al. Foundational considerations for artificial intelligence using ophthalmic images. Ophthalmology 129, e14–e32 (2022). source
The Equity section (page not provided) says population-achieved sensitivity is used to analyze the effects of access on equity. It contrasts an AI system’s sensitivity among people who receive the examination with the lower effective sensitivity at the population level when access is limited. (LLM, conf=0.93)
plain- 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
“As an example, if clinicians are used as the reference standard, their potential bias in their diagnosis may lead to bias in the training data, ultimately persisting as bias in the AI system37.”
§ Considerations for addressing bias in artificial intelligence for health equity > Introduction > Development phase
What the sources say
- Partly supportedcites [37] Sun, M., Oliwa, T., Peek, M. E. & Tung, E. L. Negative patient descriptors: documenting racial bias in the electronic health record. Health Aff. 41, 203–211 (2022). sourceExperimental — accuracy not yet measured
The Policy Implications section (page unavailable) says clinicians may introduce biased language into records, and the Limitations section says training-data trends may influence the machine-learning results. The paper does not address clinicians’ diagnoses being used as a reference standard or show diagnostic bias persisting in an AI system. (LLM, conf=0.76)
hedgedstates 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
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Thus, such bias reduces the bioethical principle of “Justice,” as Char et al.28 and Abramoff et al.32 described.”
§ Considerations for addressing bias in artificial intelligence for health equity > Introduction > Measuring bias in AI systems
What the sources say
- Partly supportedcites [28] Char, D. S., Abràmoff, M. D. & Feudtner, C. Identifying ethical considerations for machine learning healthcare applications. Am. J. Bioeth. 20, 7–17 (2020). sourceExperimental — accuracy not yet measured
Our copy of Char et al. (2020) extensively treats bias in ML-HCAs as an ethical problem — 'Development: Perpetuation of Bias within Training Data, Risk of Harm due to Group Membership' describes latent biases in training data being 'perpetuated or even amplified' with 'racially discriminatory outcomes' — and the 'Calibration' section names 'equity (that the advantages of the application are shared fairly by all)' as a cardinal design feature, so the substance of the claim (bias as an ethical harm falling along group/fairness lines) is present. However, no passage invokes the bioethical principle of 'Justice' (searched: justice, principlism, standard moral principles, fairness/equity), so the specific attribution that bias 'reduces' that named principle is not stated; indeed the 'Allure of Highly Restricted Focus' section criticises reducing ML-HCA ethics to the single consideration of fairness, and the paper hedges bias effects with 'may'/'possibility', which the citing sentence states as a settled reduction. (LLM, conf=0.70)
- Supportedcites [32] Abramoff, M. D. et al. Foundational considerations for artificial intelligence using ophthalmic images. Ophthalmology 129, e14–e32 (2022). source
The Equity section (Bioethical Foundation, same section) states that the authors use 'equity' rather than the traditional bioethical term 'justice' for the same concept, defined as freedom from bias or favoritism, so substituting 'Justice' is faithful to this source. The same section asserts that 'Inappropriate bias, increase in health disparities, and thus decreased equity can manifest across the entire AI pipeline,' which is exactly the paraphrased point attributed here (the pipeline-wide account is credited to Char et al, but the bias-to-decreased-equity link is stated by this paper itself). (LLM, conf=0.85)
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
18 more claims
Create a free account to see all 69 claims and to verify the 57 references still open.
- It may state two separate claims as one. 20
- Probably not a substantial claim. 35
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
“While these reviews have made valuable contributions to our understanding of bias in medical AI, they differ in scope, depth, and focus, leaving a gap in understanding how bias permeates the development of AI for clinical decisions.”
From Bias in medical AI: Implications for clinical decision-making (2024) · cites this paper as [20]
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, the 100% care-gap closure rate by autonomous AI in this racially and ethnically diverse pediatric population could have the potential to reduce health disparities for vision loss from diabetes, furthering health equity, and deserves further study42,43.”
From Autonomous artificial intelligence increases screening and follow-up for diabetic retinopathy in youth: the ACCESS randomized control trial (2024) · cites this paper as [43]
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, the same technology can reproduce or worsen inequity if devices are unavailable, internet connectivity is weak, staff are inadequately trained, costs are shifted to patients, or algorithms perform differently across populations and camera systems (10,(31)(32)(33).”
From From validation to public health impact: implementing equitable AI-enabled diabetic retinopathy screening (2026) · cites this paper as [31]
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
“As AI solutions have emerged, a wide variety of governance best practices in the form of frameworks, governance structures, checklists, and guidelines have been proposed7–23.”
From Advancing healthcare AI governance through a comprehensive maturity model based on systematic review (2026) · cites this paper as [7]
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
“A recurring theme across multiple articles was the necessity for an overarching governance body7–10,17. This entity would not only assist in selecting AI tools and software for implementation but also guide the validation, implementation, and monitoring processes.”
From Advancing healthcare AI governance through a comprehensive maturity model based on systematic review (2026) · cites this paper as [7]
Experimental — accuracy not yet measured
8 more claims about this paper
Create a free account to see every claim other papers make about this one.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Considerations for addressing bias in artificial intelligence for health equity > Introduction > The AI total product lifecycle
“the principles outlined in the International Medical Device Regulators Forum’s (IMDRF)47 Software as a Medical Device Clinical Evaluation”
§ Considerations for addressing bias in artificial intelligence for health equity > References
“47. International Medical Device Regulators Forum - Software as a Medical Device (SaMD) Working Group. “Software as a Medical Device”: Possible Framework for Risk Categorization and Corresponding Considerations.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Considerations for addressing bias in artificial intelligence for health equity > Introduction > Development phase
“For considerations around which reference standard to use, see Abramoff et al. 202132”
§ Considerations for addressing bias in artificial intelligence for health equity > References
“32. Abramoff MD Foundational considerations for artificial intelligence using ophthalmic images Ophthalmology 2022 129 e14 e32”
Citations
Cites
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 3×
“Multiple studies have shown examples of the use of AI in healthcare (not evaluated by regulatory agencies) exacerbating, rather than mitigating, health disparities 24,25 .”
BackgroundSelf-citationInternational Organization for Standardization (ISO)
Nahler, G. (2009) · Dictionary of Pharmaceutical Medicine · doi:10.1007/978-3-211-89836-9_723 · Cited by 3,677
Implementing Machine Learning in Health Care — Addressing Ethical Challenges
Char, D. S., Shah, N. H., & Magnus, D. (2018) · New England Journal of Medicine · doi:10.1056/nejmp1714229 · Cited by 1,857 · cited 1×
“These ethical and other concerns with AI in healthcare have been shown by a number of research studies 27 – 29 .”
BackgroundSelf-citationAbràmoff, M. D., Lavin, P. T., Birch, M., Shah, N., & Folk, J. C. (2018) · npj Digital Medicine · doi:10.1038/s41746-018-0040-6 · Cited by 1,638
Self-citationPotential Biases in Machine Learning Algorithms Using Electronic Health Record Data
Gianfrancesco, M. A., Tamang, S., Yazdany, J., & Schmajuk, G. (2018) · JAMA Internal Medicine · doi:10.1001/jamainternmed.2018.3763 · Cited by 1,395 · cited 1×
“Gianfrancesco et al. similarly limited their analysis to bias derived from characteristics of retrospective training sets derived from existing Electronic Health Record (EHR) data 44 .”
BackgroundEnsuring fairness in machine learning to advance health equity.
Rajkomar, A., Hardt, M., Howell, M. D., Corrado, G., & Chin, M. H. (2018) · Annals of Internal Medicine · doi:10.7326/m18-1990 · Cited by 1,341 · cited 1×
“Thus, AI systems are increasingly in a position to help improve patient and population health outcomes and drive down cost, increase physician job satisfaction, and address health disparities 20,21,3,22,23 .”
BackgroundWhat are Health Disparities and Health Equity? We Need to Be Clear
Braveman, P. (2014) · Public Health Reports® · doi:10.1177/00333549141291s203 · Cited by 860
Inequalities in health: definitions, concepts, and theories
Arcaya, M. C., Arcaya, A. L., & Subramanian, S. V. (2015) · Global Health Action · doi:10.3402/gha.v8.27106 · Cited by 696
Fairness in Machine Learning: A Survey
Caton, S., & Haas, C. (2020) · arXiv · doi:10.48550/arxiv.2010.04053 · Cited by 646 · cited 3×
“More recently, the Foundational Considerations on Algorithmic Interpretation (FPOAI) workgroup of the Collaborative Community on Ophthalmic Imaging published their “ Foundational Considerations ” 32 on AI as a start to developing metrics for ethics , including metrics for “ Equity ” 33 , in order to be able to evaluat…”
BackgroundFairness in machine learning: A survey
Caton, S., & Haas, C. (2024) · ACM Computing Surveys · doi:10.1145/3616865 · Cited by 646 · cited 3×
“More recently, the Foundational Considerations on Algorithmic Interpretation (FPOAI) workgroup of the Collaborative Community on Ophthalmic Imaging published their “ Foundational Considerations ” 32 on AI as a start to developing metrics for ethics , including metrics for “ Equity ” 33 , in order to be able to evaluat…”
Background
Research works printed but not linked: 2
- American Medical Association. Board of trustees policy summary. Augmented Intelligence In Healthcare. https://www.ama-assn.org/system/files/2019-08/ai-2018-board-policy-summary.pdf (2019).
- Chen, I., Johansson, F. D. & Sontag, D. Why is my classifier discriminatory? arXiv 1805, 12002 (2018).
Cited by
Bias in medical AI: Implications for clinical decision-making
Cross, J. L., Choma, M. A., & Onofrey, J. A. (2024) · PLOS Digital Health · doi:10.1371/journal.pdig.0000651 · Cited by 607
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
“Mitigating Bias in Healthcare AI: A Model Life-cycle Approach Establishing standardized and repeatable approaches for mitigating bias is an expanding societal responsibility for AI-healthcare developers and providers 23 .”
InfluentialEthical 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
A Scoping Review of Reporting Gaps in FDA-Approved AI Medical Devices
Muralidharan, V., Adewale, B. A., Huang, C. J., Nta, M. T., Ademiju, P. O., Pathmarajah, P., Hang, M. K., Adesanya, O., Abdullateef, R. O., Babatunde, A. O., et al. (2024) · npj Digital Medicine · doi:10.1038/s41746-024-01270-x · Cited by 201
Artificial intelligence for personalized nanomedicine; from material selection to patient outcomes
Mazumdar, H., Khondakar, K. R., Das, S., Halder, A., & Kaushik, A. (2025) · Expert Opinion on Drug Delivery · doi:10.1080/17425247.2024.2440618 · Cited by 159
“Furthermore, Abràmoff et al. [204] offered an extended Total Product Lifecycle (TPLC) paradigm for AI/ML in healthcare, outlining the causes and effects of undesired bias in AI/ML systems at each stage, how to assess them with suitable metrics, and how to possibly reduce them.”
Metadata
- Authors
- Abràmoff, M. D., Tarver, M. E., Loyo-Berrios, N., Trujillo, S., Char, D., Obermeyer, Z., Eydelman, M. B., Foundational Principles of Ophthalmic Imaging and Algorithmic Interpretation Working Group of the Collaborative Community for Ophthalmic Imaging Foundation, W., & Maisel, W. H.
- Affiliations
- University of Iowa (US)
- United States Food and Drug Administration (US)
- Ochin (US)
- Stanford Medicine (US)
- University of California, Berkeley (US)
- Corresponding author
- Abràmoff, Michael D.
- Venue
- npj Digital Medicine
- Published
- 2023-09-12
- Volume
- 6
- Issue
- 1
- Pages
- 170-170
- ISSN
- 2398-6352 (ISSN-L)
- PMC
- PMC10497548
- OpenAlex
- W4386637335
- Cited by
- 349
- References
- 41
- Citations by year
- 2026: 106 · 2025: 141 · 2024: 98 · 2023: 4
- FWCI
- 13.47
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- cc-by
- creativecommons.org/licenses/by/4.0 (tdm)
- creativecommons.org/licenses/by/4.0 (vor)
- Locations
- npj Digital Medicine (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- eScholarship (California Digital Library) (submittedVersion, cc-by)
- 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
- Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
- Keywords
- algorithmic bias, artificial intelligence in healthcare, health equity, bias mitigation, health disparities, machine learning, digital health technologies, bias metrics
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 (gpt-6-sol and 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: Considerations for addressing bias in artificial intelligence for health equity. https://app.cochat.ai/papers/10.1038/s41746-023-00913-9
- The paper
- Abràmoff, M. D., Tarver, M. E., Loyo-Berrios, N., Trujillo, S., Char, D., Obermeyer, Z., Eydelman, M. B., Foundational Principles of Ophthalmic Imaging and Algorithmic Interpretation Working Group of the Collaborative Community for Ophthalmic Imaging Foundation, W., & Maisel, W. H. (2023). Considerations for addressing bias in artificial intelligence for health equity. npj Digital Medicine. https://doi.org/10.1038/s41746-023-00913-9
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.
