What is ground truth: home meter or lab venipuncture?

[pem]

"In the INR range below 2.0, 90% of the allowable difference between results from the POC (point of care = meter) system and the reference (lab result) shall be +/- 0.5%

In the INR range or 2.0 to 4.5, 90% of the allowable differences between results from the POC system and the reference result shall be +/- 30%"

Protimenow - you have pointed me to a quite interesting finding. I was putting together a table of values based on my predictive model for the convenience of anyone who is interested in seeing what the converted values look like but who don't want to do the actual calculations. In generating the conversion table (see below), I discovered that the percent error in the meter value based on my model seems to fit almost exactly with the error values spelled out in your technical bulletin. See what you think!

InRatio2 INR Predicted Lab INR % Error
2.0 1.9 4%
2.1 2.0 5%
2.2 2.1 7%
2.3 2.1 9%
2.4 2.2 10%
2.5 2.2 12%
2.6 2.3 13%
2.7 2.4 15%
2.8 2.4 16%
2.9 2.5 17%
3.0 2.5 19%
3.1 2.6 20%
3.2 2.6 21%
3.3 2.7 22%
3.4 2.8 23%
3.5 2.8 25%
3.6 2.9 26%
3.7 2.9 27%
3.8 3.0 28%
3.9 3.0 29%
4.0 3.1 30%
4.1 3.1 31%
4.2 3.2 32%
4.3 3.2 33%
4.4 3.3 34%
4.5 3.3 35%
4.6 3.4 36%
4.7 3.4 37%
4.8 3.5 37%
4.9 3.5 38%
5.0 3.6 39%


In other words, Alere is properly characterizing an error model that is consistent with my predictive model. Or to put it another way, there seems to be nothing about my predictive model that conflicts with Alere's claims about their meter! Given the side-by-side result with the CoaguChek, I wonder if this finding transcends InRatio monitors and if it is actually pervasive among point-of-care meters.

Any thoughts?

Best,
pem

 

[Protimenow]

Alere e-mailed me the bulletin. It's probably on their website - perhaps in an area for 'professionals.' It's Technical Bulletin 109.

Yes, I miscalculated 30% of 3.8 -- this seemed to be way out of range, and for some reason I calculated just 3%. There was no distinction between errors at an INR of 2, and an INR of 4.5. (+/- 30% for an INR of 4.5 is SCARY - a variance of +/- 1.35 points.) From 4.6-6.0, 'no performance criteria' were listed. At some point, obviously, you've just got to get lab tests.

There was also a mentioned goal of <10% difference between a lay person's INR with the meter and that of a professional's INR of the same blood (or taken at the same time).

 

[Protimenow]

One other thing -- I'm not sure if it was in this long thread or somewhere else, but someone said that he(she?) compares the results of his meter against that of the lab by having a drop of blood put onto the strip before it goes to the lab for testing.

This sounds like a good idea - but isn't. The meters are designed to use capillary blood - which includes some levels of clotting proteins that aren't in venous blood. The result you'll get will probably NOT be accurate because of this. The only meter that can accurately measure venous blood is the ProTime - which instructs the person doing a finger stick to discard the first drop of blood (which contains the extra clotting factors) and use the second and later drops of blood.

Unless you're using that blood with a ProTime meter, you're probably asking for an incorrect INR from the meter.

 

[pem]

. The meters are designed to use capillary blood - which includes some levels of clotting proteins that aren't in venous blood.

Good information - thanks!

Pem

 

[ElectLive]

However-according to InRatio Technical Bulletin 109 (they sent me a copy), the Clinical and Laboratory Standards Institute (CLSI) calls for the following "minimum acceptable accuracy" for 'Professional Operators:'

"In the INR range below 2.0, 90% of the allowable difference between results from the POC (point of care = meter) system and the reference (lab result) shall be +/- 0.5%

In the INR range or 2.0 to 4.5, 90% of the allowable differences between results from the POC system and the reference result shall be +/- 30%" ...

I hope that we haven't lost too many readers here...

Well, I'm definintely more into numbers than most, but I was admittedly struggling to keep interest here. Then, I read that quote, and I was astonished. 30%?!?!

So, I was very curious about all of this after reading that, and while I haven't been able to track down the technical bulletin, I did find a related Precision/Accuracy document on the INRatio2 that mentions the same minimum 90% - 30% correlation. Even better, for me at least, I found a difference plot graphic illustrating this relationship of INRatio2 results to lab values:

Now, I'm an outsider looking in, but as a possible user in the future, those results just seem to me at least to be slightly unsettling. Sure, you could interpret it several ways, focus on the middle and not the periphery for instance, but still, is this really the best degree of accuracy that can be expected? Well, I guess that's a rhetorical question, since yes, those are the official INRatio2 results. But anyway, just a little surprising, never having researched any of this, I was expecting a little more accuracy. I guess it is what it is, though, nothing is perfect.

 

[pem]

Thanks a lot, ElectLive, for posting these findings. Could you please provide a link to the source material for this? Is it from an Alere document or an independent study?

What's most striking to me about the plot you provided is that my personal data correlate much better than the study data (less variability), but with a strong bias.

So either my meter data or reference (lab) data or both do not align with this study's findings.

Thanks again,
Pem

 

[ElectLive]

It's not a direct link, unfortunately, since it opens directly to pdf download, but click on the "Documentation" tab here: http://alere.com/EN_US/products/alere-inratio-2-pt-inr-monitoring-system-poc-test/index.jsp then it's the bottom right one: "Evaluation of the Precision and Accuracy..." etc. I wasn't really interested in the Precision component, but you'll probably find that pretty informative too.

 

[Protimenow]

Wow.

The 30% figure is a bit frightening to me. If a reported INR of 2.6 (which I got within minutes of my stroke) can be 30% off, this could mean that my ACTUAL, Lab-tested INR, could have been as low as 1.8 (2.6 - .78). FWIW - I've allowed my INR to drop lower, believing that my range was 2.0+. If I trusted the meter (I did), this could mean that my INR was actually far below 2.0.

Although I don't go bananas if I LEGITIMATELY record a reading below 2.0 - I follow Duke Clinics guidelines, increase my dose, test regularly until I'm in range -- but I didn't count on being so far below 2.0 for so long.

OTOH - if the values can be up to 30% ABOVE the InRatio result, an 'in range' 3.5 could actually be more than a full point higher.

This wide range, as noted, is outside the 90% required average, and may actually call into question the clinical usefulness of the InRatio monitor. If 'in range' can be far out of range - perhaps even dangerously so - is the meter an effective tool for accurately handling anticoagulation? (I may just go back to my harder to use, clunkier, older ProTime 3, which uses optics and a reagent with a value of 1.0 to actually see the clot, rather than relying on impedance in the blood sample).

 

[pem]

Thanks a lot, ElectLive, for posting these findings. Could you please provide a link to the source material for this? Is it from an Alere document or an independent study?

What's most striking to me about the plot you provided is that my personal data correlate much better than the study data (less variability), but with a strong bias.

So either my meter data or reference (lab) data or both do not align with this study's findings.

Thanks again,
Pem

So here's a diff plot of my own InRatio2 data using the lab results as the Reference INR:



pem

 

[pem]

Thanks a lot, ElectLive, for posting these findings. Could you please provide a link to the source material for this? Is it from an Alere document or an independent study?

What's most striking to me about the plot you provided is that my personal data correlate much better than the study data (less variability), but with a strong bias.

So either my meter data or reference (lab) data or both do not align with this study's findings.

Thanks again,
Pem

And here's a diff plot based on my Model2 predictions showing that the model is indeed consistent with the error margin indicated by Alere (the predicted data points fall within the 30% arrow margins):



pem

 

[pem]

So here's a diff plot of my own InRatio2 data using the lab results as the Reference INR:

View attachment 9041

pem

There are fewer data points, but it looks like, in contrast to the Alere results posted by ElectLive, in my case only about 75% (6 out of 8) data points fall within the 30% error margin.

Best,
pem

 

[Protimenow]

30% is kind of troubling--especially if the values are HIGH for high INRs and LOW for low INRs. I'm especially concerned about the lower values.

I wonder if the CoaguChek XS has a similar range.

This kind of makes me want to go back to a meter that actually times CLOTTING, and not some electrical phenomena associated with it. (If I had the money, I'd re-invest in Protime cuvettes, or get a new meter that I hope to write more about if I'm provided with the opportunity to test it).

Unfortunately, Roche and Alere seem to be driving the testing bus and it may be hard to create a third lane for a new, probably more accurate, meter.

 

[pem]

(If I had the money, I'd re-invest in Protime cuvettes, or get a new meter that I hope to write more about if I'm provided with the opportunity to test it).

I think we can help each other here. Maybe we can't buy you another protime machine, but if we pool our efforts and report parallel test results we can start to more accurately characterize the operating specifications of these meters.

Best,
pem

 

[Protimenow]

I already HAVE the ProTime machine -- I just need some strips. And parallel test results from machines in addition to the InRatio would be helpful in establishing whether or not other machines may have similar bias.

 

[Protimenow]

I think we can help each other here. Maybe we can't buy you another protime machine, but if we pool our efforts and report parallel test results we can start to more accurately characterize the operating specifications of these meters.

Best,
pem

I'd certainly like testing on different meters - and with blood draws. I have a somewhat different idea.

I'd like to validate that the INR reported by your meters actually spikes when you take Quercetin + C. I have HUNDREDS of these pills. I would be happy to send some to those who ask -- on the assumption that you'll take your INR before starting the pills, then after taking them (one a day) for a few days. It'll be interesting to see if there's any change in reported INRs. It would also, of course, be good to get a blood draw when you've taken Quercetin for a few days to see if you can confirm a gap between blood draw and meter. (Yes, I realize that my 3.8 was within the prediction model's 3.1 in the lab, but I also had a similar .8 gap when my meter said 2.6 -- and this is beyond the predictions).

If you'd like me to send some Quercetin (and, perhaps, even some 21 gauge lancets), just send me a private message. I don't know how many will respond, but I'll check my mail regularly and try to keep the mailbox empty enough for more requests.

 

[pem]

fall within the 30% arrow margins

And by "arrow margins", of course I meant "error margins"

 

[Protimenow]

Latest update -- my InRatio 2 gave me a 3.7. An hour later, at an anticoagulation clinic, their Hemochron got a 3.2.

This is not inconsistent with an apparent .5-.6 bias in the InRatio readings, related to high intake of Quercetin. (Although, I guess, it's also fairly consistent with the 30% error reported and with the predictive model for InRatio results).

I'm stopping the Quercetin + C and will test in a few days. I wouldn't be surprised if my reported INR is closer to the clinic's 3.2.

 

[Freddie]

Latest update -- my InRatio 2 gave me a 3.7. An hour later, at an anticoagulation clinic, their Hemochron got a 3.2.

This is not inconsistent with an apparent .5-.6 bias in the InRatio readings, related to high intake of Quercetin. (Although, I guess, it's also fairly consistent with the 30% error reported and with the predictive model for InRatio results).

I'm stopping the Quercetin + C and will test in a few days. I wouldn't be surprised if my reported INR is closer to the clinic's 3.2.

If you stop both at the Quercetin + C at the same time, you won't know which one is causing the increase. Quit the Quercetin for two weeks and see what happens, but stick with the "C", as many members take it every day and have no issues with it.

 

[Protimenow]

The Quercetin+C is a single pill.

In the past, I've taken Vitamin C -- as much as 2 or 3 grams/day -- with no influence on my INR.

 

[pem]

Latest update -- my InRatio 2 gave me a 3.7. An hour later, at an anticoagulation clinic, their Hemochron got a 3.2.

Thanks for posting your meter/lab results! From an InRatio2 INR of 3.7 My model2 would have predicted a lab value of 2.9.

I may add your data and newmitral's data to my own and then update the model accordingly.

Best,
Pem