Hacker News Reader

Comment by brbsix | Hacker News Reader

There appear to be two different sorts of "disruption":

1) There's disruption that's seen as impractical or insanely difficult (e.g. Tesla, SpaceX). Or technically feasible but with significant social or legal obstables (e.g. AirBnB, Uber).

2) Then there's another category, disruption that defies laws of physics. Kickstarter is replete with examples. This includes things like pocket-size underwater rebreaters, self-filling water bottles, and solar-powered motorcycles.

I'm no expert in the subject matter but Theranos tech seems to be in the latter.

View on HN

What law of physics makes what Theranos attempted to achieve impossible? They were still drawing blood from the patient. Thus, it seems theoretically possible for technology to overcome the challenges the approach presents.

The law of sample sizes? If I ask one person who they are going to vote for I can't tell you who is going to win the election.

Concentration is a global property which cannot be assessed locally in a fluid.

So all blood tests are useless?

Blood tests use a relatively large amount of blood to counter the problem. Also many blood tests are repeated multiple times before the doctor produces the final diagnosis. So multiple units of blood from a vein distributed over days vs a pinprick's worth of blood from a finger.

Also, blood found in a relatively large vein is much closer to the body average than that found in a capillary area, where much of the liquid is from extracellular space and not even coming from within blood vessels. Of course, extracellular space is filled from blood serum, but a lot of stuff has different concentrations in ECF.

Quoting from a comment on one of Derek Lowe's posts on the subject (http://blogs.sciencemag.org/pipeline/archives/2015/10/15/the...) (note that a large fraction of commenters are active chemistry, medical, or biology researchers):

> The blood you get from a prick will be a mixture of blood from small venules, small arterioles, capillaries and lymph vessels, mixed with whatever loose tissue and cells have been cut by the prick and carried over into the blood. No way you could ever get a reliable CBC from such a sample. Since the kind of general blood tests that this company is targeting are usually CBC + chem panel, even if you could get your chem results from the prick, you still need a normal blood draw for the CBC, so the competitive advantage that they are trying to push (quick sampling by patient at pharmacy, without need for phlebotomist) is lost. On top of that, the variability inherent in chem results from prick blood will be significant for several analytes, like CK and AST (present in muscle so could be raised from prick injury), glucose and lipids (different levels in venous or arterial blood or lymph), etc…

In short, while it might be possible to reliably use only µL of blood for tests, using µL of blood for dozens of tests seems questionable, and the chosen blood source is likely to be mostly worthless.

You underestimate the magic of machine learning! :0...kidding. Sort of. I suspect they thought there was a software solution to this problem (and for all I know there might be)

What? Please elaborate! I can't think of how software overcomes this problem.

I doubt it can but if you did a large number of blood tests with finger stick and a large number at the same time with a regular blood draw you could put that data into some machine learning and see if you could figure out an algorithm to correct the results. You'd probably need more data than just the final counts. Maybe some kind of imaging data? Maybe ions? It's tough to speculate. Machine learning learning is all about finding patterns. Maybe some pattern could be used to do the correction?

That is evidence of the problem being difficult, not a rigorous "perpetual motion"-level impossibility proof. I could write something similar about a proposal to have computers play chess, based on the absurd blowup in elementary operations (think 1940s) and logic gate resets needed to compute a good move.

> That is evidence of the problem being difficult, not a rigorous "perpetual motion"-level impossibility proof.

That the uncertainty produced by the source chosen exceeds the acceptable variability in the results is an impossibility proof for a testing method relying exclusively on a single sample from that source having acceptable accuracy, since the accuracy of the test result cannot exceed the accuracy of the source used in representing the underlying measure.

Only if you could know in advance that there was no way to filter the input to the point that there is a correlation between the the filtered input and the blood state, which is exactly the point under contention.

For comparison, there is tremendous variation in the sound waves that humans recognize as the word "apple", but there are still systems that can filter out the relevant noise and identify the critical dimensions of variation.

Unless their test involves technology that can lower the uncertainty to acceptable levels, e.g. being able to isolate the blood from certain vessels despite using a prick method.

Simply saying that the source is messy is not a proof that it's impossible to work around that.

The source is not merely messy. It's a tiny sample of a fluid which is not absolutely consistent throughout.

This is a fundamental problem. You might luck into roughly correct results most of the time, but the basic nature of statistics is against you.

Oh I don't disagree with you at all. But even based on what you said, it's not hard to imagine that it's possible to identify and test some previously unidentified component of said fluid that is consistent throughout.

My point is only that there is no hard theory decrying this to be impossible in the same way that, say, computer scientists can prove that a problem is NP-complete.

Solar panel roadways for latter

That falls clearly into the former category.

Extreme to the point of being nonsensical in its impracticality (cost of installation, cost of replacement, current PV efficiencies, etc), but not fundamentally impossible.

Well none of the examples for latter aren't technically impossible. Impossibility is ie. Perpetual machine.

Reply on news.ycombinator.com