If this study is correct, and blood drops have a high variability in certain important clinical features, then, no matter how accurate the instruments are, it's impossible to actually get a clean measurement from low volumes of blood.
This is one of the most obvious tests that should come to mind for anyone who wishes to create a blood diagnostic company. How could Theranos not know about this?
More importantly how did their sophisticated investors not know this? This issue has been known in the industry for years.
The whole Theranos fiasco would not matter except for the damage it does to other biotech startups - when sentiment turns it does not matter if what you are doing is real or not, everyone is frozen out.
I highly doubt their sophisticated investors are any more clued in than their board / executive management, which included prominent military figures, industrial complex powerbrokers, politicos, but zero or nearly zero medical expertise.
I need to be a little less subtle with my snark I think. If there ever was a company you could have avoid investing in by looking at their board it is Theranos [1].
Go figure, right? "Awarded the Nobel Peace Prize" - and also generally considered guilty of war crimes if we're to use the rules of Nuremberg.
Oh, and Bechtel (nuclear, defense), Cargill (hedge funds on the back of deforesting rain forests to grow more soy), antitrust lawyers, Marine Corps General, ex-Armed Services Committee Senators, a Secretary of Defense, a Navy admiral - oh, and an epidemiologist and surgeon. Very unusual leadership.
That must be the understatement of the year. The board looks like the board you would put together if you were planning a coup of some central american country.
The study states that they got accurate measurements once they started taking 5 drops of blood which is less than 0.1 ml. A blood sample could vary from 3-10 ml according to the first thing I found on google. So it seems like all they have to do is take a few more drops to be accurate (which will still be an order of magnitude from the volume of traditional blood tests).
To get 100 µl out you need more than a finger prick. Once you go this way you might as well cannulate. There might be some special cases where you can get 100 µl but not 3 mL, but this market is not going to support the weight of a unicorn.
If he is, he's generally right. You get 1.5, maybe 3 drops out before it seals back up. Finger lancets are very tiny needles, even smaller than those used for insulin injections.
I worked in diabetes where finger stick tests are the lingua franca. 3 microliter sample sizes were common about 10 years ago for home use systems using relatively fine lancets. Using something just a bit larger would pretty easily get you 5 drops of blood.
Additionally important is the fact that the minimum required volume varies based on the feature that is being measured. This seems incongruent with Theranos' focus on a few standardized sample containers.
I'm sure they do/did, and hope(d) that they could work around it in some way. Unfortunately for them, it sounds like they're going to have to pay the piper before they wanted to.
The best they can do is measure the sample with zero measurement noise - but that's no panacea if the variation is in the sample itself, and not the measurement.
It's not so much the volume that is the problem from what I understand it is the way in which it is drawn. So if that same volume were drawn from a vein then the variability would be substantially less.
> How could Theranos not know about this?
Who says they did not know about this? Maybe they did but it did not agree with their plans, maybe they thought they could overcome it.
I've found even glucose meters to be wildly inaccurate. There is often a 20% difference in readings between two meters moments apart, and thats enough to have consequences for how your care is managed.
It's pretty heteroskedastic, which practically makes the errors not tooo bad. For example, say my wife is low (< 70), in that case the error rate will be small. That's where precision matters. Now suppose she's 200+, in that case the errors will be large but it doesn't really impact how to take care of it. In terms of error, What worries me much more is a server at a restaurant giving my wife a regular coke rather than a diet coke or vice versa.
Where smaller blood is really beneficial is that it requires a less painful finger prick.
Hmm, it's not clear to me what the benefit of that would be in terms of information (a transformation). It's univariate data, it's been a bit since I looked at it but a histogram showed readings were lognormal / weibull / gamma distributed (point is asymmetric, right-skewed), I believe.
In terms of inference, it's kind of mind-bending to think of a "population" of blood and each person having a specific variation.
I've noticed this as well in rodents. Take 2 consecutive drops and the readings can be as much as 60 mg/dl off. It's not a problem for us because we take duplicate measures, and if they are too far off, we can remeasure. An individual with one meter and only one drop must be relying very heavily on a pretty inaccurate measurement.
Fingerpricks provide capillary blood mixed with tissue fluid.
A venupuncture provides venous blood.
The normal ranges and the error bars are different for finger prick blood.Virtually data and research on the normal ranges and ranges in various diseases has been done on venous blood.
The thousands of journal articles that come out each week are done using results of venous blood.For instance, see this article from this week's NEJM http://www.nejm.org/doi/full/10.1056/NEJMoa1504720?query=fea... The glycated Hb readings referred to in this article are automatically taken to be on venous blood samples, as are the fasting glucose levels.
Your physician relies on readings from venous blood.
His training in medical school was on normal and pathological ranges in venous blood samples.
Physicians are thrown a bit when given pathology readings from other countries ( mmol versus the old fashioned American mgs/litre are confusing. if the pathology printout is in German or Arabic, it adds even to the fun ). It breaks the flow of rapid data analysis while the physician breaks out bc https://www.gnu.org/software/bc/ or some other calculator to convert units.
Add in the confusion and inaccuracy of fingerprick samples, and the potential for even more error, increases a bit.
My wife's a nurse and she said if the test comes out high or low they always do a blood draw to double check. So this doesn't seem that ground breaking of a study.
This is one of the most obvious tests that should come to mind for anyone who wishes to create a blood diagnostic company. How could Theranos not know about this?