HEAL: A framework for health equity assessment of machine learning performance
Google just published a framework to measure whether AI healthcare tools actually reduce disparities—not just perform equally. Here's what they found in dermatology.
Why it matters
As AI healthcare deployments accelerate, Google's HEAL framework addresses a critical blind spot: most AI fairness measures don't account for pre-existing health disparities. This research-backed methodology lets companies assess whether their ML models actually improve outcomes for populations with the worst health outcomes—a distinction that matters for both ethics and regulatory compliance.
The key facts
6 to knowGoogle developed HEAL (Health Equity Assessment of machine Learning performance) framework published in The Lancet eClinicalMedicine
Dermatology case study: model trained on 29k cases, evaluated on 5,420 teledermatology cases enriched for diversity across age, sex, race/ethnicity
Framework measures model performance (top-3 agreement) against pre-existing health outcome disparities (YLLs, DALYs) across subpopulations
Health equity defined as fairness of opportunity—different from AI fairness which often targets equal performance across groups
Framework uses Pareto condition: model changes must maintain or improve outcomes for all subpopulations without worsening any group
Study acknowledges gap: framework assesses likelihood of equitable performance but not actual real-world outcome reduction across populations
Go to the source
Google Research Blogblog.research.google
Publisher excerpt: Posted by Mike Schaekermann, Research Scientist, Google Research, and Ivor Horn, Chief Health Equity Officer & Director, Google Core Health equity is a major societal concern worldwide with disparities having many causes. These sources include limitations in access to healthcare, differences in…
