Why We're Validating Our Screening on Every Skin Tone

This month we began a formal validation study at Anubhav Life Care. The short description: we are measuring how well SamaClip's painless screen agrees with hospital-grade reference instruments, in a representative group of the people we actually serve. The longer description is about a choice we made in how to recruit, and why it matters.

The problem with how devices usually get validated

A device gets validated by testing it against a trusted reference and seeing how closely they agree. Simple enough. The hidden weakness is in who gets tested. If a device is validated mostly on lighter-skinned people, its reported accuracy is really only an accuracy for lighter-skinned people — even though it gets sold and used on everyone. The numbers look reassuring while quietly not applying to a large share of the population.

For optical measurements like oxygen saturation and non-invasive haemoglobin, which read light through the skin, this is not a small detail. Skin pigment affects how light behaves. A validation that skips darker skin tones is, for our community, a validation of the wrong population.

What we are doing differently

We are deliberately recruiting across the Monk Skin Tone categories 4 to 10, including the darkest tones. That range is underrepresented in nearly every major calibration dataset the field relies on. We are recruiting through the same diagnostic centre the surrounding community already uses, not through an artificial lab population, so the people in the study look like the people the device is for.

The study is cross-sectional, run at an NABL-accredited facility, and compares the screen against reference-standard instruments for each parameter. We registered an observational study protocol for it. And critically, when we report results, we intend to publish performance broken down by skin-tone group rather than collapsing everything into one average that could hide a failure on darker skin.

Why publish the breakdown

A single average accuracy number is exactly the thing that lets skin-tone bias hide. A device can look accurate "overall" while being meaningfully worse for the people who most need it to work. The only honest way to claim a device works across skin tones is to show its error bounds for each group, including the darkest, and let the breakdown speak.

That is a higher bar than a single headline figure, and we are setting it for ourselves on purpose. If we are going to screen a population that is mostly medium-to-dark-skinned, the burden is on us to prove the screen works for them specifically.

What comes next

The study runs through the end of the year. We are not going to pre-announce results we do not have yet. When the data is in and analysed, we will publish what it shows — the agreement with reference instruments, the per-skin-tone performance, the honest picture, including any limitations. That is the right order: do the work, then report it.