The racial bias in pulse oximeters

[DukeTrout]

Missed on both.

First, older models were actually better about this. Look at the Masimo pulse oximeters from the mid-90’s to the mid-00s, before Covidien forced them to go cheap or die.

Second, as long as you’re using an approved brand as a backup, there’s nothing unethical about testing a new one. Gold standard would be to compare an approved pulse-ox monitor, the new one, and blood gas. That should also greatly decrease the number of test subjects required.

@Dioptase1

I’m not sure LED price is reason. The price of LED’s and sensors is ridiculously low compared to a few decades ago when these first came out. So is processing power. And the improvement is going to be very small at best. The real cost is going to be regulatory.

I wasn’t thinking of the LED cost. A big chunk of the cost would be in redesign of both the sensor assemblies (cables and connectors are not cheap) and the capital equipment. Regulatory could be exceptionally cheap, relatively speaking. If FDA allowed expedited review, the regulatory cost could be as little as $250k. You’re thinking of a PMA, which wouldn’t be required for these devices.

@Dioptase1

As for sunset dates, that’s a bad idea. It’s not like pulse oximeters are bad.

That’s going to be the one way to make sure that manufacturers do the work and hospitals get the new equipment. Otherwise, just don’t bother. And did you read the article? 12% critical failure rate in a sub population is fucking horrible. It’s unacceptable. The dollars are irrelevant. If the equipment is feeding bad data to clinicians, it has to go.

@Dioptase1

They just aren’t as good as we’d like.

That’s really cynical to say that when we’re talking about life-and-death decisions related to POC, dude.

ETA: A great example of how sun setting works is the ongoing effort to get external defibrillator pads from their old 510k approvals up to PMA approvals. FDA made it a priority and set an expedited review process and a sunset date. They are in year 4 of a 5 year process. Next year, any defibrillator pads that weren’t approved through a PMA cannot be sold. With the expedited review, the PMA process is costing about 1/10 of what it would usually cost a company.

[atl]

Pulse oximeters better than not bad, they were a complete revolution. One interesting factoid about that impact is that in 1980 or so, ahead of the introduction of pulse oximetry, the only malpractice insurance more expensive than anesthesiology was OB/GYN. If that’s the best proxy for risk (and it’s probably pretty good), by 1990 or so, anesthesiology was roughly middle-of-the-pack in there with ENTs.

[jerwin]

@DukeTrout

Missed on both.

First, older models were actually better about this. Look at the Masimo pulse oximeters from the mid-90’s to the mid-00s, before Covidien forced them to go cheap or die.

I thiought I said as much.

@DukeTrout

Second, as long as you’re using an approved brand as a backup, there’s nothing unethical about testing a new one. Gold standard would be to compare an approved pulse-ox monitor, the new one, and blood gas. That should also greatly decrease the number of test subjects required.

My source was

The Accuracy of 6 Inexpensive Pulse Oximeters Not Cleared... : Anesthesia...

But those aren’t medically certified devices.

[DukeTrout]

The original post was talking about ICU-level SpO2 monitors. Comparing them to the cheap consumer ones is like comparing a Ducati to a kid’s tricycle.

[Dioptase1]

Good point, I assumed a PMA was required. I assumed some of the older but more accurate versions were a different branch of technology (catheter). Getting a 510(k) can be cheap. Development costs are still high but not too bad for product improvement.

Another point I missed was this was ICU class pulse ox. Yeah, don’t want to cut corners there. Regular pulse ox is intended as a useful tool, but not definitive. ICU and surgical are another game. I was thinking of less critical uses where pretty good is good enough. My bad.

Given what you said and what jerwin posted, it looks like Covidien tried to push out a downmarket product (cheaper tech, but better than the alternatives) that instead is getting used upmarket. That’s a marketing and sales snafu (emphasis on the fu) combined with cost pressures at hospitals is leading to less capable devices dominating the market.

I’m still a firm believer against sunset registration. Too blunt a tool (that’s a long conversation itself). I’d go with standards more if I didn’t see how Europe and China are making an absolute mess of them. Paradoxically, you can get more situations like this. I’m dealing with that now of meeting several international standards vs making the safest product possible. I kind of like the US FDA’s pragmatic approach of letting us use standards OR use methodologies we can prove lead to safer products.

No, I think this is going to have to be clinician driven. And this paper is a start. It can’t stop with just saying “hey, here’s racial bias” but “hey, this product is more biased / less biased leasing to these results” Laying out the clinical and economic reason to use product A vs product B. Sounds cynical, but that’s what hospitals pay attention to.

Again, before anyone objects to cost/benefit reasoning, it was more more haphazard before being codified by the ACA and is SOP for single payer. I develop medical products that are sold in countries that are communist, socialist, single payer, military rule, purely private pay, capitalist, and hybrids and they all do cost benefit.

Ugh, sorry, too much caffeine and working from home. Haven’t had a chance to discuss like this in the office enough. Sorry everyone. You get the results of my technical bulimia.

[DukeTrout]

No worries from my side. I’m in much the same boat.

My turn to be cynical: I don’t think a “free market” solution will work for this (regarding it being “clinician-driven”). Docs just don’t have much pull with purchasing anymore. They save their battles for the tools that they use directly. SpO2 isn’t going to meet that threshold. This is the kind of thing that would have to be championed by one of three entities: a professional society; a huge hospital system like Kaiser or Acension; or the FDA. Of those three, only the last has the juice to make it universal across all critical care situations.

If the FDA just raises the bar, that usually means much higher cost for manufacturers. Strangely, when they go for the full mandate, they usually soften the blow by easing requirements to meet the mandate. They also usually close all loopholes so that hospitals have to comply with the change-over, and is one of the few times they will coordinate with CMS so that reimbursement moves in lockstep with the regulatory mandate. That prevents hospitals from just carrying the old equipment and finding a gray-label alternative.

If I had to pick one approach, the mandate seems to be the more industry-friendly, as strange as it sounds.

[Dioptase1]

@DukeTrout

Docs just don’t have much pull with purchasing anymore.

I 100% agree. But not what I was suggesting. It’s clinicians that need to be the ones to bring this information to the attention of bigger entities like you suggest. And in a way that grabs their attention. Racial bias … meh (sad, but true). Articulated results in terms of lives or money, Granny get your gun!

I’m not against mandates, especially when clearly called for. Just very cautious. I have seen setting hurdles wrong backfire. I have seen projects get canceled because they didn’t meet some internal safety metric, despite being safer than whats on the market. I could see that happen with pulse oximeters very easily because of good intentions. Mandates/standards/etc can be like evil genies: You will get what you asked for, not necessary what you wanted (Dr Deming noted this back in the 1940s). So extreme care is needed.

Existing manufacturers like it when the FDA raises the bar. Pretty easy to guess why. I’ve heard the last few years of expedited PMA, but that seems to be just time, not cost. The same clinical data seems to be required. Not all that helpful. But 510(k)s with clinical data seem to be a bit more accepted. As are some other 510(k) paths.

It’s all weird for me now, since I develop mostly for outside the USA. Yet we keep looking at registering in the USA because it’s faster, more rigorous, and can accelerate registration in other countries. So I need to stay up to date better.

[Flossaluzitarin]

If it’s a health band, it’s not internal (safe presumption? Not so, ye diabetics say?) so there’s no need for the 510K, though evidently more than one color of green light might promise percent-accurate data.

Jerwin> [Data showing mo’ melanin mo’ scatter, save for wavelength at 1/20 that scale.]
Well…snap.

Dioptase1> [more…]
NIR II and 530nm, 610nm devices to the rescue? They look down 10 points SpO2 and the health band starts topical 2-photon dye (preferably with a quick tell on heme?) Crazy aperture [sticks 100 mm lens watch on tattooed patient]? PulseOx are bad: there are so many more signals more important than the beep on those things, and they…they oft. beep with no music while people get variously ashy.

DukeTrout> Like comparing a Ducati to a kids’ tricycle.

[Googles it. Yep $300 kid tricycles by Ducati.] Be hale, masked motorsports fans?

The possibility of Asian engineers.

I’m not drinking chocolate milk and that made me snort chocolate milk out my nose. Pardon the ectoplasm.

[anon29537550]

I have a question. As you all gave probably noticed, I have absolutely no engineering background at all. I do have years of experience in PICU’s and ambulatory peds. In the ICU setting, at least, is there a way to get an initial correlation between pulse ox and ABG, set the sensitivity on the machine and be good to go? Obviously, no such adjustment knob exists currently, but is there a technical reason it could not exist?

[Dioptase1]

From a technology point of view, it’s easy enough. But the bigger hurdle (and one that affects any ‘calibration’ setting) is human error. Risk analysis is required for device development and this would get flagged. What are the odds someone will forget to input or enter input incorrectly? Regulatory bodies require that this be addressed. All too often, the control is just taken away.

It leads to what I call the Fisher-Pricing of medical equipment.

[anon29537550]

I could see that. Sigh.

[Dioptase1]

I wouldn’t give up hope entirely. Manufacturers might be constrained, but clinicians have more leeway. If the correlation is possible, then there’s nothing stopping them from taping a laminated chart to the pulse ox machine that lists correction factors. i.e. A table with left column pulse ox, top row as ABG. Just read off the correction factor for this patient.

IIRC, most beds have a dry erase board nearby with info, so write the correction there. It can only help when interpreting the pulse ox number. Dave reads 2 points high, so keep an eye on his numbers vs Susan reads 3 points low, so don’t panic. Not ideal, but has the advantage of improvement right away. And staff will quickly figure out which pulse oximeters are the good ones.

[anon29537550]

That is an excellent, low tech, not needing approval kind of fix. If we could gather a large enough database, I’m certain some form of nomogram could be generated. Unfortunately, as you know, variance on these machines tends to get larger as patients get sicker. Lower PaO2, poorer perfusion, more VQ mismatch. Nonetheless, even a rough-and-ready sort of correction might help. As i said earlier, I am forced to wonder how much of a role this plays in excess mortality among PoC. Sats reading 92, no biggie, looks good. Unless it is actually 85. Delayed recognition, delayed intervention, poorer outcome. Along with all the other hurdles they face, but maybe we can address this one in the very short term.

[anon55609254]

That’s definitely a theoretical solution, but I agree with @Dioptase1’s comments about user error. I wonder if the pulse oximeter could somehow detect the patient’s skin coloring and adjust itself.

[DukeTrout]

From what I understand, the earlier Masimo machines did not generate this much error. I don’t know if they used a third wavelength, stronger light sources, a better algorithm, or some combination thereof, but I don’t think it’s a technical challenge so much as an economic and user compliance problem.