Scoping Review Will Map How Intersectionality Is Used to Shape Health Care AI
Artificial intelligence and machine learning are moving deeper into routine health care, but the same tools that promise faster triage, diagnosis, risk prediction, and treatment planning can also reproduce the social hierarchies embedded in the data they learn from. This protocol sets out to examine how intersectionality is being used to confront that problem in health care AI and ML research, with a specific interest in whether investigators move beyond simple identity labels and actually build equity into design, implementation, and interpretation.
The review is not a technical audit of model performance alone. It is an examination of whether the people designing these systems are asking the harder questions: whose data are included, whose experiences are flattened, and whose priorities are missing when an algorithm is built to make clinical decisions.
Why Identity Variables Alone Cannot Capture Equity in Algorithm Design
Health systems commonly treat race, gender, class, disability, and other characteristics as separate variables in predictive models. In practice, that approach can strip away the overlap between social positions that shape access to care, exposure to harm, and the likelihood of being misclassified by an algorithm. Intersectionality offers a different lens. It asks how overlapping structures of power influence health and how those structures should shape the very architecture of AI and ML tools.
This matters because bias in health care algorithms rarely appears as a single, isolated error. It often emerges when training data reflect existing inequities in access, diagnosis, referral, and treatment, then translate those inequities into automated recommendations. A model can look statistically strong and still distribute risk unfairly if it was trained on populations whose lives are systematically underrepresented or misread.
The protocol frames this issue directly: equity-oriented AI requires more than adding demographic fields to a dataset. It demands a contextual understanding of how social identities and systems of power intersect in real clinical settings.
The Review Will Examine Both Conceptual Use and Practical Application
Investigators will search MEDLINE, Embase, Emcare Nursing, APA PsycInfo, the Cochrane Database of Systematic Reviews, and Cochrane Central Register of Controlled Trials for English-language primary studies. Eligible articles must apply intersectionality to guide the design and implementation of AI or ML in health care. The review will follow the Joanna Briggs Institute framework for scoping reviews, with two independent reviewers screening titles, abstracts, and full texts against prespecified criteria.
Preliminary searches identified 4191 records across six databases. That is a large evidence base for a field still defining its vocabulary. The volume alone suggests that AI and equity are now being discussed in the same sentence far more often than before, but the protocol is designed to find out whether that language is substantive or merely decorative.
The team will extract data from included studies and summarize the findings narratively, supported by tables and charts. The key output is not a pooled effect size or a single performance metric. It is a map of how researchers define intersectionality, where they apply it, and which methods or pedagogical tools they use to operationalize it in actual AI and ML projects.
Participatory Practice Sits at the Center of the Protocol
The protocol gives unusual weight to coproduction. Patient and family partners will be engaged throughout the review process to refine the scope, interpret findings, and support knowledge translation. In a field where marginalized communities are often treated as data sources rather than decision-making partners, that design choice is consequential.
Coproduction changes what gets asked and what gets counted. Community-based collaborators can surface blind spots that academic teams may not recognize, including whether a study’s language is genuinely equity-focused or simply using equity as a framing device while leaving power relationships untouched. Their involvement also matters at the level of implementation. An AI system built without input from affected communities may optimize for institutional convenience while failing the people most likely to bear its errors.
The protocol links participatory practice to the quality of the review itself. By involving patient and family partners, the team is trying to ensure that the synthesis does not remain abstract or institution-centered. Instead, it aims to produce recommendations that are responsive to community priorities and grounded in lived experience.
A Field Moving Faster Than Its Methods
AI in health care has already entered diagnostic workflows, triage systems, and treatment planning pipelines. The pace of deployment has outstripped the pace at which many groups have defined ethical guardrails for fairness, representation, and accountability. Intersectionality offers one of the few frameworks capable of examining multiple forms of disadvantage at once, rather than forcing researchers to reduce people to one demographic axis at a time.
That is especially relevant when algorithms are trained on clinical records that already reflect unequal access to care. If one group is less likely to receive specialty referral, later-stage diagnosis, or consistent follow-up, a model may learn those patterns as if they were neutral facts. The result is a tool that can amplify preexisting disparities while appearing objective.
The review is expected to clarify where the literature is conceptually thin and methodologically uneven. It will also show whether researchers are using intersectionality as a genuine analytic framework or as a broad synonym for diversity and inclusion. Those are not the same thing, and the difference has direct consequences for model development, evaluation, and deployment.
October 2026 Marks the Expected Completion Point
The scoping review is slated for completion by October 2026. Until then, the main contribution is the structure of the inquiry itself: a systematic attempt to determine how intersectionality is being defined, enacted, and shared across AI and ML work in health care.
If the literature proves sparse or inconsistent, that finding will be clinically and operationally important. It would mean that much of the current equity conversation in health care AI remains aspirational, with limited evidence of robust methods or meaningful community participation. If the literature is more mature than expected, the review may identify usable frameworks that can be taken into future model design, teaching, and governance.
Either way, the practical stakes are clear. Algorithms that shape health decisions should not be built on flattened identities or disconnected assumptions about fairness. The unresolved challenge is whether current AI and ML research in health care has moved beyond acknowledging inequity to actually designing against it.