Health Equity Strategies in AI

🌟 New Blueprint for Responsible AI in Healthcare! 🌟 Explore insights from Mass General Brigham's AI Governance Committee on implementing ethical AI in healthcare. This comprehensive study offers a detailed framework for integrating AI tools, ensuring fairness, safety, and effectiveness in patient care. Key Takeaways: 🔍 Core Principles for AI: The framework emphasizes nine key pillars—fairness, equity, privacy, safety, transparency, explainability, robustness, accountability, and patient benefit. 🤝 Multidisciplinary Collaboration: A team of experts from diverse fields established and refined these guidelines through literature review and hands-on case studies. 💡 Case Study: Ambient Documentation: Generative AI tools were piloted to streamline clinical note-taking, enhancing efficiency while addressing privacy and usability challenges. 📊 Continuous Monitoring: Dynamic evaluation metrics ensure tools adapt effectively to changing clinical practices and patient demographics. 🌍 Equity in Focus: The framework addresses bias by leveraging diverse training datasets and focusing on equitable outcomes for all patient demographics. This framework is a vital resource for healthcare institutions striving to responsibly adopt AI while prioritizing patient safety and ethical standards. #AIInHealthcare #ResponsibleAI #DigitalMedicine #GenerativeAI #EthicalAI #PatientSafety #HealthcareInnovation #AIEquity #HealthTech #FutureOfMedicine https://lnkd.in/gJqRVGc2

AI in healthcare isn’t as neutral as you think.  AI could harm the very patients it’s meant to help. Without addressing the bias, we will never be able to benefit from the good. Here’s how we can fix it.  1. 𝗜𝗺𝗽𝗿𝗼𝘃𝗲 𝗗𝗮𝘁𝗮 𝗤𝘂𝗮𝗹𝗶𝘁𝘆 AI models are only as good as the data they are trained on. Unfortunately, many datasets lack diversity, often overrepresenting patients from certain regions or demographics. Ensuring datasets are inclusive of all populations is key to reducing bias. 2. 𝗥𝗶𝗴𝗼𝗿𝗼𝘂𝘀 𝗩𝗮𝗹𝗶𝗱𝗮𝘁𝗶𝗼𝗻 AI tools must be tested across diverse populations before deployment. Studies have highlighted how biased algorithms can worsen health disparities at every stage of development. Rigorous validation ensures that these tools perform equitably for all patients. 3. 𝗧𝗿𝗮𝗻𝘀𝗽𝗮𝗿𝗲𝗻𝗰𝘆 𝗮𝗻𝗱 𝗘𝘅𝗽𝗹𝗮𝗶𝗻𝗮𝗯𝗶𝗹𝗶𝘁𝘆 Healthcare professionals need to understand how AI models make decisions. Lack of transparency can lead to mistrust and misuse. Explainable AI not only builds trust but also helps identify and correct biases in the system. 4. 𝗠𝘂𝗹𝘁𝗶-𝗦𝘁𝗮𝗸𝗲𝗵𝗼𝗹𝗱𝗲𝗿 𝗔𝗽𝗽𝗿𝗼𝗮𝗰𝗵 Bias mitigation requires collaboration between AI developers, clinicians, policy makers, and patient advocates. Diverse perspectives help identify blind spots and create solutions that work for everyone. 5. 𝗢𝗻𝗴𝗼𝗶𝗻𝗴 𝗠𝗼𝗻𝗶𝘁𝗼𝗿𝗶𝗻𝗴 Bias doesn’t stop at deployment. Continuous monitoring is needed to ensure AI tools adapt to new data and evolving healthcare needs. For instance, algorithms trained on outdated or incomplete data may maintain errors over time. Only by addressing these areas, can we see the benefits of AI in healthcare, such as reducing errors, aiding diagnoses, and personalizing treatments for all. What steps are your organization taking to ensure fairness in AI healthcare tools?

While an obvious point to many, it's not universally known - Equity 🚫 Fairness. These terms are often used interchangeably, but they may be optimizing for different values and outcomes. When it comes to health AI - fairness of an algorithm "aspires to have equal performance across all populations, with no regard for these populations’ differential needs and processes." Equity on the other hand - "requires considering that individuals with 'larger barriers to improving their health require more and/or different, rather than equal, effort to experience this fair opportunity.'" What does this look like in practice? Ziad Obermeyer's now famous Science paper gives a great illustration of an unfair algorithm. The purpose of the algorithm is to identify patients that would benefit from complex case management. However, white pts with the same health status as black pts were disproportionately offered additional support. This was due to conflating health care cost with health status. While many argued for fairness in correcting this algorithm (pts with the same health status should be equally likely to receive additional support, regardless of their race) - an equity driven algorithm might actually result in a persistently unfair algorithm but restructure it to prioritize black pts over white pts, given historical barriers to accessing care. What we prioritize is not a matter of math or computer science; but rather a deeply human, social and political question. A JAMA Health Forum perspective from earlier this year(Kevin Johnson, Eric Horvitz, Ivor Horn, MD, MPH) dives deeper into this topic, while providing specific / actionable guidance on how to include fairness + equity considerations throughout the AI lifecycle. https://lnkd.in/g5iY5kS9

⚕️ As medical AI rapidly evolves, it's critical we develop tools and resources that can be used to identify and mitigate biases that could negatively impact health outcomes. 📃 Our research paper, “A Toolbox for Surfacing Health Equity Harms and Biases in Large Language Models," is a step in this direction. 🔎 This paper provides a framework for how to assess if medical LLMs may perpetuate historical biases as well as a collection of seven adversarial testing datasets called “EquityMedQA” as a guidepost. 🤝 We used these tools to evaluate our own large language models, and now they’re available to the research community and beyond. Full paper → https://lnkd.in/e2jR7ru6 #AI #GoogleDeepMind #GoogleResearch #Research #Health #HealthEquity

This paper reviews how bias affects AI in healthcare and outlines strategies to detect and reduce such bias across the AI model lifecycle. 1️⃣ Bias in healthcare AI often originates from human, data, algorithmic, or deployment-related factors, each introducing unique risks that can worsen health disparities. 2️⃣ Implicit, systemic, and confirmation biases are introduced during data collection and model design due to unconscious attitudes or structural inequalities. 3️⃣ Data biases like representation, sampling, and measurement issues stem from underrepresented populations or inconsistent data acquisition practices. 4️⃣ Algorithmic biases, including aggregation and feature selection bias, often arise from decisions made during model development and preprocessing. 5️⃣ Deployment-related biases like automation, feedback loop, and dismissal biases emerge from how clinicians interact with AI tools in practice. 6️⃣ Mitigating bias requires a lifecycle approach—spanning from conception, data collection, preprocessing, algorithm development, deployment, to post-deployment surveillance. 7️⃣ Effective mitigation involves team diversity, use of diverse and representative data, careful feature selection, subgroup testing, and fairness metrics like equalized odds and demographic parity. 8️⃣ International bodies like WHO and regulators such as the FDA and Health Canada have issued frameworks emphasizing fairness, explainability, and ethical use in healthcare AI. 9️⃣ Future directions include embedding DEI principles in AI development, expanding bias training, and integrating AI ethics into clinical education. ✍🏻 Fereshteh Hasanzadeh Alagoz, Colin B. Josephson, Gabriella Waters, Demilade Adedinsewo, Zahra Azizi, MD, MSc, James White. Bias recognition and mitigation strategies in artificial intelligence healthcare applications. npj Digital Medicine. 2025. DOI: 10.1038/s41746-025-01503-7

‼️ 🌟 Artificial Intelligence & Cancer Health Equity 🌟 ⚖️ My co-authors (Kingsley I. Ndoh, Darlington Akogo, Hermano Alexandre Lima Rocha, Sergio Juacaba) and I are pleased to share our latest publication in *Current Oncology Reports: Artificial Intelligence and Cancer Health Equity. This paper explores the potential of AI technologies in cancer diagnosis, treatment, and patient care—while critically examining the risks of perpetuating disparities if we don’t center equity in its design, development and deployment. The promise of AI and advanced technologies in oncology is undeniable. The stark reality is that biases in training data, unequal access to technology and digital health tools, and systemic barriers may widen cancer disparities rather than close them. That’s also why it’s crucial to #spotlight organizations committed to building inclusive technologies, digital health, consumer health tools and AI solutions to address cancer care ethically and equitably. Our paper highlights leaders in this space, including: CancerIQ – Empowering healthcare providers with AI-driven risk assessment tools for early cancer detection and prevention Rede ICC Saúde / Ceará Cancer Institute – Integrating AI into oncology care to improve access in Brazil COTA, Inc. – Using real-world data to uncover and address cancer care disparities Flatiron Health – Harnessing real-world data and AI to drive precision oncology and ensure that insights from cancer research benefit all populations Freenome – Advancing early cancer detection through AI-powered multi-omics Hologic, Inc. – Advancing women's health with AI-powered breast cancer screening solutions designed for equitable access Hurone AI – Bringing AI-driven oncology solutions to underserved communities globally minoHealth AI Labs – Leveraging AI to improve cancer diagnostics and clinical decision support in Africa Patient Discovery – Using AI and patient-reported data to personalize care and reduce barriers for diverse cancer patients Vectorgram Health – Enhancing cancer diagnostics and care in sub-Saharan Africa These are just a few of the #trailblazers ensuring that advanced technologies in cancer care does not leave anyone behind. THANK YOU! Our call to action? Technology and digital health tools must be built for everyone. That means inclusive and diverse datasets, ethical frameworks, and policies that prioritize equity at every stage of AI development. Read the full paper here: #AI #CancerCare #DigitalHealth #HealthEquity #ArtificialIntelligence #PrecisionMedicine #Oncology #MachineLearning #DiversityInAI

AI "hallucinations" are predictable outputs of systems trained on biased, incomplete, and commercially curated data. We keep calling them bugs, but they're actually behaving exactly as designed. Recently, a colleague was commenting how AI governance frameworks proliferate globally, from the EU AI Act to NIST's risk management framework to OECD principles. I find myself returning to a basic question about whether we're governing the right layer, especially if trying to apply them to health. Most governance attention focuses on outputs: transparency dashboards, model cards, explainability requirements, fact-checking mechanisms. These matter, but they don't address why AI systems produce biased or inaccurate recommendations in the first place. Research on AI health recommendation systems can provide examples of this. I plucked the ones below from the OECD AI Incidents and Hazards Monitor and from PubMed: 1/ Training data comes disproportionately from a few geographic hubs and predominantly white, male populations. Models then perform poorly for everyone else. See: https://lnkd.in/eH5FJqJq 2/ Algorithms use proxies that embed existing inequities. Using healthcare spending as a measure of medical need disadvantages patients who historically faced barriers to care, even when they are sicker. See: https://lnkd.in/e2Gjz3Bv 3/ AI models can detect demographic characteristics from medical images and use these as shortcuts for predictions, leading to higher error rates for marginalized groups. See: https://lnkd.in/eqfcSxGZ 4/ Consequently, I think that patients with incomplete records due to inconsistent care access will get interpreted as lower risk, compounding existing gaps. (see: https://lnkd.in/e-Ye2Br3) Serving health equity through AI governance means shifting focus upstream. Some questions I keep coming back to: - Who curates training data, and whose experiences are included or excluded? - What accountability structures exist when AI recommendations cause harm? - How are affected communities and frontline health workers included in governance, not just technologists and regulators? - Are equity audits happening throughout the AI lifecycle, not just at deployment? AI is now infrastructure in health systems - or we're trying to build it as such. Diagnostic algorithms, triage systems, benefit eligibility tools, and misinformation detection platforms all inherit the biases built into their foundations. Governance frameworks that focus only on outputs will keep producing the same inequitable results.

💡 AI in healthcare isn’t one-size-fits-all, and Community Mental Health Centers have our own blueprint for success. ❗ Rolling out AI in a CMHC is different. We work where trust, privacy, and cultural responsiveness aren’t just important, they’re foundational. Our clients often face trauma, stigma, and systemic inequities, and our workflows are built around those realities. With the right approach, AI can: ✔️ Give clinicians back hours each week ✔️ Extend access to care 24/7 ✔️ Predict and prevent missed appointments ✔️ Support better outcomes for underserved populations 📌 What makes AI implementation in CMHCs unique: 🔹 Trust comes first. Transparent, opt-in AI builds confidence with clients navigating sensitive situations (Teen Vogue, 2023). 🔹 Regulatory depth. HIPAA plus 42 CFR Part 2 and state behavioral health rules demand thoughtful design (PsyHC Care, 2025). 🔹 Safety is non-negotiable. Oversight prevents risks like “AI psychosis” and ensures AI supports wellness (TIME, 2025). 🔹 EHR integration matters. Behavioral health systems require tailored, workflow-embedded AI (Springer, 2025). 💡 CIOs in CMHCs can set the standard by: ✔️ Starting with 2–3 high-impact, low-risk pilots. ✔️ Forming governance teams that include clinicians, compliance, and community voices. ✔️ Measuring what matters, access, equity, and satisfaction alongside efficiency. ✔️ Sharing results to help the whole field move forward. AI in CMHCs isn’t about following someone else’s playbook. It’s about building one that reflects our communities, our values, and our commitment to safe, equitable innovation, while still moving fast. What’s the biggest barrier you see to rolling out AI in community mental health, and how would you tackle it? #AI #BehavioralHealth #CIO #Leadership #Innovation #MentalHealthInnovation #AIforGood

I’m proud to be a co-author of one of the most comprehensive guidelines for implementing AI in healthcare. An Artificial Intelligence Code of Conduct for Health and Medicine: Essential Guidance for Aligned Action has just been published by the National Academy of Medicine. It presents a unifying AI Code of Conduct (AICC) framework designed to align the field around responsible development and application of AI and to catalyze collective action to ensure that AI’s transformative potential is realized. This isn’t a philosophical treatise. It's designed to be applied at every level of decision making, from boardroom to bedside and from innovation labs to reimbursement policies. The AICC is based upon six commitments: 1.    Advance humanity 2.    Ensure equity 3.    Engage impacted individuals 4.    Improve workforce well-being 5.    Monitor performance 6.    Innovate and learn   This represents an enormous and inclusive effort on the part of the healthcare community to offer guidelines for realizing the greatest benefit from a technology that is already transforming medicine. I hope you'll have time to have a look and send me some of your thoughts! The publication can be downloaded for free: https://lnkd.in/e6QivskV #HealthcareAI #AIAdoption #HealthTech #GenerativeAI #AIMedicine

Artificial intelligence (AI) is reshaping healthcare, emphasizing the importance of addressing biases that can worsen healthcare disparities. This review stresses the need to systematically detect and counter biases across the AI model lifecycle, from development to implementation and monitoring. The FDA's latest update reveals a rise in approvals for AI-driven medical devices, underscoring AI's expanding role in healthcare. This includes applications such as medical image analysis, health metric tracking, and outcome forecasting from Electronic Medical Records. Despite AI's benefits, biases within these models can result in unequal care distribution, underscoring the necessity for robust bias detection frameworks. Regulatory bodies like the European Commission, FDA, Health Canada, and WHO are enhancing efforts to establish stringent guidelines for AI development and deployment, ensuring fairness, equity, and transparency. This review examines various bias types in healthcare AI, delving into their sources and proposing mitigation strategies to promote fair and unbiased healthcare provision. #AI #Healthcare #BiasMitigation #Equity #Fairness #DigitalMedicine Citation: 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, 8(154). https://lnkd.in/euMDKfmc