How we test fairness across patient groups

Average performance can hide uneven results. We break our results down by age, sex, ethnicity and deprivation, and we'll publish what we find.

Published

A system can perform well on average and still work less well for some groups of patients. In a country where healthy life expectancy differs by about 20 years between the most and least deprived areas (ONS (opens in new tab)), that matters.

What we look at

When we evaluate the platform, we don't stop at one overall figure. We break the results down by:

  • age
  • sex
  • ethnicity
  • deprivation

We look for differences in how often the platform flags something, how often it abstains, and how often independent reviewers agree with its output.

Where it happens

Fairness testing runs through each stage of our evidence plan:

  1. Clinician-written test cases, designed to cover a range of patients and presentations.
  2. Validation on real, approved NHS data, with independent reviewers scoring the results.
  3. Evaluation in NHS services, where we can see how the platform works in practice.

What we do with differences

If we find a gap, we investigate it before a clinical area goes further. Sometimes the cause is the data; sometimes it's the guidance itself, or how a pathway is applied. Either way, the finding is recorded in our clinical safety process.

Publishing results

We'll publish what we find, including what didn't work. Read more about our approach on Evidence & fairness.

← All insights