The racial bias in pulse oximeters

Originally published at:


This makes sense given how pulse oximeters work. I’m not really sure what a good solution would be either.


Perhaps a settable option on the meter for higher illumination. But the nerd (in me) says to take the absorption spectrum of oxyhemoglobin and subtract off the spectrum of melanin to see if there’s a more likely light frequency to calibrate to (perhaps to include a couple of different frequencies) [shrug]


it needs to scan skin pigment and then calibrate its O2 sensor for the conditions.

nobody seems to know how to use a tri-corder any more


It’s a real shame that there’s some linguistic nuance available for talking about the people affected by this problem. (You don’t have to be Black, you can also be “darker than white”)…

But there doesnt seem to be a way in english to distinguish between the racial construct of “white” and the actual tint value of human skin. No one gets these blood ox readings while wearing clown makeup: why can’t the language reflect reality?

Edit: something more like this is what Im thinking is needed


Well, shit. I have never seen this info before. I have to rethink how to interpret pulse ox’s. To the research batcave!


Did you miss a ‘not’ in there somewhere?

It is not a shame that there is linguistic nuance. There is a lot of linguistic nuance available. What’s a shame in this context is the risk that using such nuanced linguistic terms may be misapprehended and end up with the user being accused of using racistly discriminatory language.

Remember to take off your batgloves!


right but that adds cost and introduces the problem of “well we have a sensor that works well for people with darker skin but its more expensive than the unit that only works well for light skinned people.” Its technically sound but a political minefield.

1 Like

Well, upon quick literature search, this seems to be a problem facing adult patients and docs, not so much in little ones. I feel a little bit less bad for my ignorance. However, this has been reported in respectable medical journals as far back as 1998 that I came across. Wonder what mysterious factor might be at play that kept it from being more widely known, and fixed. Yup, the world may never know. (So much /s)

This was about a 10 min literature search. It’s not buried info, I guess just not seen as important. Sometimes I think my profession needs an enema.


Yes! You both understand what I’m getting at, and also express it better than I did.

1 Like

I can’t speak to the current crop of pulse oximeters out there, but my Dad was an engineer intimately involved in the earliest commercialization of pulse oximetry in the early 1980s and I know that they did indeed test some of the earliest ones on black people back then. I don’t recall the solutions or adjustments made – although it is a real engineering issue, dealing with various absorption of visible and IR light given different pigmentation – but it wasn’t because they didn’t think about it.

I have rather a vivid recollection of my Dad bringing this very instrument (Nellcor’s first Beta) over to our neighbor’s house (the father quite fair skinned Black, the mother fairly dark) and comparing the two of them, the Dad was himself a biology professor at Cal, long before it went to production. I’ll ask him about what they did back then to adjust for it (and I don’t recall) but it certainly wasn’t for a lack of thinking about it.

Okay, just talked to my Dad about this. While this may be different in the newer models of pulse oximeters which measure a far broader array of gases, in its earliest form, pulse oximetry engineers were quite concerned about this very thing. Not only as a question of making sure it worked on different complexions, but also making sure it worked on jaundiced premature infants.

As it happens, pulse oximetry works by measuring on the ratio of the absorption of IR and red/visible light – and since melanin absorbs red and infrared equally, melanin levels don’t really affect the fundamental science of pulse oximetry. Dad noted that other applications of oximetry that don’t necessary have the advantage of the increase in the volume of arterials (just upstream of the capillaries) such as applications with surgical margins, etc., need to do something that’s not necessarily a ratiometric calculation (which cancels out a lot of these issues with skin coloration) but rather trying to get some absolute measurement of oxygen levels.

TL;DR: PulseOx in its basic form doesn’t have a problem with melanin, because the ratio between IR and Red/Visible absorption doesn’t change very much, and it operates fundamentally as a measurement of a ratio not absolute values. (But newer blood gas monitoring may be creating some errors).*

The technology has drifted since back then, but the overall point’s is an important one – variations in biology require careful engineering tweaks – but it wasn’t some original sin here of not having it in mind, at least not pulse oximetry in the earliest generations.


The instruction pamphlet might include a full color printed card that let the user match a panel to their skin, and input a calibration adjustment.

It would be more expensive for the printing cost, and maybe slightly more complicated if the device needs to remember more than one setting. But the deal killer is the social cost in admitting that it’s not a “one size fits all” unit. Look how long it took to get race-neutral band-aids!


The pantone review you added makes sense but probably needs to be of fingers only, given we are talking about pulse oximeters.
My fingers are definitely a different pantone shade to most parts of the rest of my body.

But I guess even generalised shades would help, if a pulse oximeter could be thus calibrated. I won’t hold my breath, though. (Pun intended.) :wink:




I have to add that this affect tanned people, and as you couls see above people could get a very heavy tan.

One would think that doctors in Africa (or South America) would have found out something like that a while ago, especially because simple pulse oximeters are so cheap.

So why is there not a single word in this study, or in the one from 2005 how people in Africa deal with this? I can only assume that nobody bothered asking them. And I find that pretty strange, especially when people are in the business of researching racial bias.


Or, they would just experience the instruments as being somewhat less reliable than do doctors in almost-exclusively-light-skinned countries, and treat the results accordingly.

It may be that this method of measurement simply doesn’t work as well with darker skin – which would kind of make sense, since melanin is there to stop light passing through the skin – but that in itself isn’t necessarily a problem. The problem arises if you establish thresholds for significance based on conditions that don’t apply to the actual people you’re testing.


A simple switch with 3 or 4 positions for selecting skin color would probably make a huge difference.

Here’s a project from Canada that tries combining a simple and cheap sensor with a mobile phone to bring cheaper pulse oximeters to developing countries. It appears to be part of WHO’s Patient Safety Pulse Oximetry Project.

In that setting, the compensating for skin color would not add production cost because it could be handled entirely in software. The user could simply chose skin color from a selection of colors, or the app might detect skin color from a photo of the patient’s finger, using the (known) color of the sensor for calibration.


True. I’d still have expected that somebody at least made an effort at getting the perspective of the people who would have a lot first-hand experience about that.

1 Like

I’d think the ability to calibrate based of skin color? or a more robust method engineered/tested against a wider variety of skin tones?

From FDA’s 2013 guidance on pre-market testing of pulse oximeters, it looks like at least some thought went into the issue:

Your data should include 200 or more data points (paired observations: pulse oximeter, co-oximeter). These data should be distributed as described in Annex EE.2.3.4(g). Your study should have subjects with a range of skin pigmentations, including at least 2 darkly pigmented subjects or 15% of your subject pool, whichever is larger.

One problem is that this is for 510(k) devices, where they just have to be “substantially equivalent” to some other device on the market in 1976 when devices first became regulated by FDA (the compromise Congress made with the medical device industry when they wrote the law). Thus you have “device drift,” where A is substantially equivalent to B, which is SE to C, which is . . . you get the idea. Supposedly every device has to go through the testing quoted, but it seems to me to be very inadequate to the job.