Health care is increasingly being transformed by AI, with technologies such as medical scan analysis and cancer risk prediction tools enabling faster diagnosis and treatment. Yet, alongside these advancements comes a critical challenge: how do we translate powerful AI methods into safe, trustworthy, and sustainable clinical tools? To address this, UHN’s Cancer Digital Intelligence (CDI), a research and innovation program within the Princess Margaret Cancer Centre, developed AFFRM-AI, a framework for the fair and responsible use of AI tools.

Bridging a Critical Gap: AFFRM-AI

CDI aims to improve cancer care through technology and AI. As such, it has been actively researching the use of AI in medicine since the technology’s early development. One goal of CDI is to promote a culture of fair and responsible AI and take the steps necessary to mitigate bias in AI solutions caused by social, data-related, methodological or algorithmic factors. Despite the growth of AI use, there has been a lack of accessible and actionable resources on the fair and safe use of AI in clinical settings. 

To address this, CDI assembled a 16-member working group composed of leading experts from various institutions in related fields including AI, bias, ethics, privacy, education and medicine to develop a guidance document.

The CDI team developed, A Framework for Fair and Responsible Machine Learning and AI (AFFRM-AI) that outlines best practices and recommendations for two key goals:  

  1. Encouraging responsible, safe, and compassionate AI development and deployment within UHN clinics 
  1. Safeguarding patients and care providers from biased and inequitable AI predictions.  

AFFRM-AI provides actionable guidance across four key stages: (1) problem identification and study design, (2) model development and training, (3) silent deployment—testing real-world performance observationally without influencing clinical decisions—and clinical evaluation, and (4) operational deployment and ongoing monitoring. Within each stage, the framework provides concrete recommendations, reflective questions, and documentation prompts addressing how data are used and approved, data readiness and availability, how fairness is measured, selection of appropriate outcome measures, and more.

Designed to adapt to local contexts, AFFRM-AI enables multidisciplinary teams—including clinicians, data scientists, privacy experts, ethicists, and operational leaders—to share a common language and set clear expectations for building safe AI systems that are grounded in equity, fairness, scientific rigor, and transparency.

Driving Responsible AI at UHN and Beyond

The AFFRM-AI framework and the methodology behind its development have been published in The Lancet Digital Health.  The team is now focused on cultivating engagement within the UHN community, including working with the Research Ethics Board (REB) and the AI Deployment team, among others, to encourage organization-wide adoption of standards for equitable and responsible AI use.

By providing concrete, operational guidance, AFFRM-AI aims to put responsible AI into practice and protect patients. Initiatives such as this position UHN at the forefront of responsible innovation, ensuring that powerful AI methods and progress translate into safe, trustworthy, and sustainable clinical tools.

For more information and to view the Framework, see here.

The AFFRM- AI Framework was prepared by Mattea Welch, Benjamin Grant, and Christopher Deutschman, in consultation with Clare McElcheran, Adam Badzynski, Jennifer A.H. Bell, Andrew Hope, Robert C. Grant, Tran Truong, Kelly Lane, Patti Leake, Divya Sharma, Ian Stedman, Mike Lovas, Ale Berlin, Jeremy Petch, Benjamin Haibe-Kains, and James A. Anderson.

Work on AFFRM-AI was partly supported by the Associated Medical Services (AMS) Healthcare, CDI and The Princess Margaret Cancer Foundation.

See the manuscript for additional acknowledgements and competing interests.

Welch ML, Grant B, Deutschman C, McElcheran C, Badzynski A, Bell JAH, Hope A, Grant RC, Truong T, Lane K, Leake P, Sharma D, Stedman I, Lovas M, Petch J, Berlin A, Haibe-Kains B, Anderson JA. A practical framework for operationalising responsible and equitable artificial intelligence in health care: tackling bias, inequity, and implementation challenges. Lancet Digit Health. 2026 Mar;8(3):100957. doi: 10.1016/j.landig.2025.100957. Epub 2026 Mar 20.