CHAPTER 06 / WOMEN’S HEALTH AI, CLINICAL GOVERNANCE, AND REDRESS
Beyond Bias
“A sprint that ends in ‘do not build’ is not a failed project; it is a working one.”
Johnna D. Wesley, Ph.D.
California, U.S.
Heather Stegner
Idaho, U.S.
Heather Ward, Ph.D.
Washington, U.S.
What We Want Readers to Notice
A model can perform well and still fail the patient most likely to be harmed.
The chapter begins with a patient we call Elena. She is bleeding after childbirth in a busy labor-and-delivery unit. Her vital signs remain within the range the system expects. By the time the severity of the postpartum hemorrhage becomes unmistakable, she has crossed into crisis.
AI appears to offer an obvious answer: predict the risk earlier, mobilize staff and blood products sooner, and prevent escalation. But the promise of prediction can conceal a more difficult question. What happens when the records used to build the model carry the same omissions, racialized blind spots, reproductive erasures, and histories of dismissal that already shape women’s care?
An algorithm can be technically precise and still be clinically unready. It can perform well overall while failing particular groups. It can generate a risk score while ignoring staffing conditions, reported pain, reproductive history, access barriers, or the bedside judgment of nurses and patients. And once deployed, its apparent authority can make that missing context even harder to contest.
We wrote this chapter from three intersecting perspectives: translational science, data-enabled clinical research, and institutional operations. Across those settings, we saw the same mistake repeated. Readiness was treated as a technical threshold—enough data, satisfactory validation, functioning infrastructure, and an implementation plan.
But a system is not ready merely because it can be built.
Feminist AI Readiness asks whether the women whose bodies, symptoms, histories, and risks are being modeled have real authority within the decision. Can they help define the problem? Can clinicians and advocates challenge the model? Who can pause its use? What happens when a subgroup is harmed? Can consent be withdrawn? Is there a repair path with deadlines and consequences?
We wrote this chapter to move women from the position of data source to the position of governor—and to give clinical, product, and policy teams a practical way to decide when a women’s health AI system should proceed, pause, be redesigned, or never touch a patient’s life.
By the Authors
Johnna D. Wesley, Ph.D. Heather Stegner | Heather H. Ward, Ph.D.
Johnna is a biotechnology executive and immunologist with more than 18 years of experience advancing therapies from discovery through early clinical development. Her work brings together translational science, portfolio strategy, and equitable health innovation.
Heather is an executive leader with extensive experience in strategy, operations, communications, and stakeholder alignment. She brings a systems-oriented perspective to implementation, decision authority, and the institutional conditions required to make change hold.
Heather works across multi-omics data platforms and clinical translational science programs, connecting biological science, data compliance, cross-functional coordination, and operational execution. Her experience spans cardiometabolic disease, digital pathology, and machine-learning-enabled data analysis.
FEATURED RESOURCE
The 30-Day Feminist AI Readiness Sprint
A pre-deployment governance process for women’s health AI
The 30-Day Feminist AI Readiness Sprint helps clinical, product, policy, and governance teams determine whether an AI system is ready to proceed—not simply whether it is technically capable of deployment.
The Sprint combines participatory problem framing, feminist data review, subgroup-specific equity thresholds, decision rights, contestation pathways, and binding conditions for pause, redesign, refusal, and redress.
Its purpose is not to make every project deployable. Its purpose is to produce an evidence-based readiness decision.
FOR
Clinical and health-system leaders, women’s health teams, nurses, health-data and informatics professionals, AI-product teams, patient-safety and quality groups, policymakers, advocates, and institutional governance bodies
TIME
A structured 30-day institutional review; selected worksheets can also support a 60–90-minute preliminary assessment
FORMAT
Downloadable Sprint guide, Use Case Charter, Feminist Data Inventory, Equity Impact Checklist, Governance Specification, and 30/90-Day Log
Designed to be tested and adapted for your setting—not followed as a fixed prescription. This toolkit supports institutional governance and readiness review. It does not replace clinical validation, regulatory assessment, professional judgment, patient-safety procedures, or legal obligations.
Why This Matters
NOTICE
Recognize when technical success is concealing clinical and institutional unreadiness.
Notice when aggregate performance hides subgroup harm, women’s symptoms or histories are missing or misclassified, bedside judgment carries less authority than a score, consent is difficult to withdraw, or a model is expanding into uses its evidence cannot support.
DECIDE
Determine whether the system is safe enough to touch a patient’s life.
Set equity thresholds, define prohibited uses, identify who can contest or pause the tool, and decide whether it should proceed, proceed with safeguards, pause, be redesigned, be refused, or never be built.
SUSTAIN
Ensure that readiness remains governable after the initial review.
Track subgroup effects, overrides, contestation, consent withdrawal, repair timelines, and new uses. Preserve named pause authority and connect unresolved harm to redesign, withdrawal, non-renewal, or redress.
This volume treats the gap between model accuracy and how decisions actually get made at the bedside as the central governance problem—and offers concrete tools for building validation, workflow integration, and the authority to pause or refuse into the systems themselves.”
MAX TPOAZ, PH.D., RN, MA, FAAN, FIAHSI
Associate Professor, Columbia University School of Nursing
VNS Health
“
ABOUT THIS CHAPTER
Most critiques of AI in women’s health begin with bias: women have historically been underrepresented in clinical research, sex-disaggregated data remain incomplete, and models inherit the blind spots of the records on which they are trained.
Beyond Bias argues that this diagnosis stops too early.
The deeper problem is power.
Women may supply the bodies, symptoms, histories, and risks from which an AI system learns while having little authority over the problem it is designed to solve, the data categories it recognizes, the thresholds it uses, the purposes for which it may be deployed, or the response when harm occurs.
The chapter names several mechanisms through which this failure becomes normalized:
The Technical Readiness Default, which treats data, infrastructure, model performance, and deployment capability as sufficient.
The Accuracy Trap, in which strong aggregate metrics conceal poor performance for the women most likely to be harmed.
Data Deserts, where relevant reproductive, demographic, clinical, or contextual evidence is missing, underrepresented, misclassified, or unusable.
Context Gaps, where the model can calculate what is in the record but cannot recognize what safe care requires.
Weak or absent pathways for contestation, pause, consent withdrawal, redress, and refusal.
Feminist AI Readiness provides a different standard. A system is ready only when affected women, patients, clinicians, nurses, and advocates have meaningful authority to define, contest, pause, repair, redesign, or refuse it.
In this framework, prediction is not the same as safety, and validation is not the same as legitimacy.
SUGGESTED USES
Women’s health AI development, readiness, and pre-deployment reviews.
Clinical validation, patient-safety, and quality-improvement processes.
Health-system procurement, vendor evaluation, and contract renewal.
Policy, regulatory, and institutional governance discussions.
Product-team, clinician, patient-advocate, and community workshops.
Courses and professional learning in health policy, public health, health informatics, clinical leadership, data ethics, and feminist technology studies.