Testing PPG Across Skin Tones: A Practical Plan

Concept render

Rate each participant on the 10-level Monk Skin Tone scale, ideally with a colorimeter reading too, and recruit enough people in every tone group to reach the precision you want. FDA's January 2025 draft for pulse oximeters offers a reference bar of 150 people. Compute your own group size from your measured error spread.

If your wearable uses green-light PPG for heart rate, someone will ask how it performs across skin tones. This guide helps a hardware PM or validation lead plan that study: which scale to use, how to size each group, and what to report. It is an engineering summary of general wellness testing and is not a medical device guide. It contains no Duobay test result and no price.

What do the published sources say?

Every row below comes from a primary source we read on October 31, 2026. The FDA document is about medical pulse oximeters, so read it as a reference for how a regulator describes a diverse test group.

One result needs plain wording. The Bent study found no significant link between skin tone and heart-rate error. It did find differences between devices and between activities. That does not settle the question for your sensor, because about 9 people per tone is a small group. The point of your own study is to find out for your board.

Which scale should you use?

The FDA draft uses the 10-level Monk Skin Tone scale, rated by an observer, and also asks for a colorimetry measurement (ITA). It notes that the scale is not a proxy for race or ethnicity. The Bent study used the six Fitzpatrick types. We could not fetch the Monk scale's own site, so we rely on the FDA description of it.

Practical consequence: pick one scale as your primary label, record the other if you want to compare with earlier studies, and add an instrument reading if you can. Say in your report who rated each person.

How many people per tone group?

The FDA draft says sample size should be the maximum of the sizes needed for adequate power on each success criterion, with 80% or more recommended. That depends on how accurate the device is and how variable the data are. So the number comes from your own data.

A simple way to size a group when you want to estimate the mean error within a half-width:

n = (z × σ / d)²

z = 1.96 for 95% confidence
σ = SD of heart-rate error, in bpm   (your measurement)
d = allowed half-width of the estimate, in bpm   (your requirement)
n = participants per tone group, rounded up

Bland-Altman 95% limits of agreement = mean difference ± 1.96 × SD of differences

Worked example, illustrative arithmetic only. Assume σ = 5 bpm and d = 2 bpm. Then n = (1.96 × 5 / 2)² = 4.9² = 24.01, which rounds up to 25 participants per tone group. Both inputs are assumptions we picked to show the steps. They are not measured values, and the result does not come from any study.

What a small group buys. Rearrange the formula to d = z × σ / √n. With 9 per tone, d = 1.96 × σ / 3, about 0.65 × σ. If σ were 5 bpm, that is about 3.3 bpm, but σ is a variable you must measure. The takeaway is the ratio: a group of about 9 gives a half-width of roughly two thirds of one standard deviation.

The FDA group shares add a second constraint. With 150 people and at least 25% in each of three groups, each group holds at least 37.5 people, so 38 when rounded up to whole people. If you want to match the FDA draft's reference bar, compare your computed n with 38 per group and note which one you chose and why.

What should the report contain?

Keep the wording to general wellness. A result about heart-rate readings does not support a diagnostic or medical statement.

What are the common traps?

These follow from the sources above. They come from engineering reasoning. None describes an incident at any company.

• Treating the FDA numbers as a requirement for your product. The 150-person, three-group design is a non-binding draft for medical pulse oximeters. A wellness heart-rate wearable is a different product. Use it as a reference bar for how a regulator describes a diverse cohort, and say so in your own documents.
• Assuming the effect before you measure it. The one wearable study in our source list found no significant link between skin tone and heart-rate error, and found significant differences between devices and activities. Writing "degrades on dark skin" as established fact has no primary source here, and so does writing "works on all skin tones".
• Reading "no significant difference" as "no difference". About 9 people per tone gives a wide margin on the group mean. With the formula below, the half-width is about 0.65 times sigma. A test that small can miss an effect smaller than that.
• Pooling everyone into one number. A single average over all participants can look fine while one group is off. Report per group, then report the pool.
• Skipping the instrument. A rater score is subjective. The FDA draft pairs the MST rating with a colorimetry measurement. Without one, you cannot show your rating was consistent.
• Counting people instead of paired measurements. If each person gives many readings, those readings are not independent. Decide in advance whether your n counts people or readings, and say which.

What should you measure on your own board?

No published source gives the error spread of your module in your enclosure. The vendor did not publish it, so it must be measured on your board. We have not measured it for you, and no result appears here. Set the pass line for each row yourself from your product requirements, and record the reason for it.

The pass line is yours to set by product requirement. Write it down, with the basis for it, before you run the first participant.

What should you ask before you start?

• What claim do we want to make, and what half-width d does that claim need?
• Where does our sigma come from? Has anyone measured it on our board, or are we borrowing a number?
• Which scale will we use to rate participants, and will we add an instrument measurement?
• How many people per tone group can we actually recruit, and is the group count set before we start?
• Are we counting people or paired measurements, and have we written that down?
• Which activities, body sites and ages are inside the test, and which are outside?
• Who writes the pass line, when, and where is the reason recorded?
• What will our public wording say about skin tone if the result is mixed?
• Does every piece of marketing copy stay inside general wellness, with no medical or diagnostic claim?

If you want help planning a validation protocol for a health wearable, tell us about your device. You can also read about our health wearable development and how the gates structure test work.

Sources

All pages accessed October 31, 2026. Documents change, so re-read them before you rely on them.

• FDA, Pulse Oximeters for Medical Purposes: draft guidance (January 2025): https://www.fda.gov/media/184896/download (accessed 2026-10-31)
• FDA, guidance document page and status: https://www.fda.gov/regulatory-information/search-fda-guidance-documents/pulse-oximeters-medical-purposes-non-clinical-and-clinical-performance-testing-labeling-and (accessed 2026-10-31)
• Bent et al. 2020, npj Digital Medicine, full text: https://pmc.ncbi.nlm.nih.gov/articles/PMC7010823/ (accessed 2026-10-31)