Slashdot Mirror
Promising Blood Test for Alzheimer's
The online edition of the British journal Nature Medicine has a study of a blood test for Alzheimer's disease, developed at Stanford. The test lights up if 18 specific molecules are present in a blood sample. Using samples of stored blood, the test proved 90% accurate in identifying people who had been diagnosed with the disease by other methods. It was also 87% accurate in distinguishing samples from people who do not have Alzheimer's but exhibit some other form of dementia. The numbers of samples involved were small — SFGate's writeup has some details. The Mercury News's article says the test's developers want to begin selling it to laboratories in 2008, for which FDA approval would not be required. They hope to get FDA approval for general use by 2009.
What the article fails to point out is the real benefit to getting early diagnosis for Alzheimer's. If people could be diagnosed earlier, they could get better care and avoid accidents.
Interesting question. Should we require elected politicians to meet certain levels of health, and mental capacity? I think that might be a good thing, but it could set a dangerous president for the rest of society. I certainly wouldn't want to be denied a job due to my Alzheimer's, heart disease, or cancer risk factors.
Well.. maybe. Or Maybe not. But Definitely not sort of.
I term this reverse confirmation bias: if many people have tried and failed, it must be impossible.
... until it wasn't.
But what credit is there to that? Many were the claims to transmute lead into gold. What proved impossible by chemical means was by no means impossible within the framework of the right technology. I think you need to study the "Four Colour Corollary". This theorem states that the truth or falsity of the theorem is entirely independent of the number of bozos who publish unfounded and incorrect speculations disguised as purported proofs. Furthermore, we still don't have a proof that could possibly have been discovered before the computer era, so the deck was stacked towards impossible
The same thing happened within the field of AI. This still annoys me. A lot of grand claims were put forward in the 1960s, and it all fell far short of what was promised. Nevertheless, there has been an unbroken stream of solid and important results, if not yet worth writing home about. Weren't the smart people silently expecting it to play out this way all along?
I feel the statistical results are the most important:
http://www.ucl.ac.uk/media/library/robotillusions
And there recent is progress even in the long discredited field of automatic proof:
http://www.maa.org/devlin/devlin_01_05.html
Guess what? Computers are now checking computerized proofs. Does this series converge, or not?
As for this new blood test, the human genome was sequenced a scant seven years ago, the explosive shock wave of proteomics is expanding almost at the limiting wave velocity, and we are now beginning to disentagle some of the fundamental neurochemistry involved. If there are any correlates in the blood whatsoever, it would be shocking to not find them at the present time, or in very short order.
Concerning percentage prediction rates, have we learned nothing? If you have a population of size N which you wish to classify into two distinct groups, given prior p and (1-p), the information required to achieve this is N * H(p), using Shannon's information measure. If this test provides any additional information beyond the prior, one can formally determine the ratio of the unknown information this test provides. If the test is worthless, the ratio will be zero. If the test is perfect, the ratio will be one. If the ratio comes out negative, you just assume the water goes the other direction (by metaphor with electrochemistry), and substitute the absolute value.
The interesting term is the cross entropy between what the experts can determine and what this test can determine. If the cross entropy is 100%, then either test gets you to exactly the same place, and it will probable come down to a matter of economics, which the cheaper approach prevailing. If the cross entropy is significantly less than 100%, then one will likely employ both tests, possibly using the cheaper test to screen the more expensive test, depending on tolerance rates for false negatives and false positives.
Given that they have included 18 elements in this test given a small positive sample size (they don't state their negative sample size), it's almost certain that some of these 18 factors are bogus, and will be eliminated as the sample size increases. If this test is bogus, the factors remaining will dwindle to zero, as the predictive rate also dwindles to nothingness. If the test is fundamentally predictive (to some ratio of the information content) as the bogus factors are pared out, the predictive ratio will likely improve by some marginal amount, maybe enough to be worth doing, maybe not.
In the 1970s one could make easy sport of predicting that any given claimant of the "four colour proof" was wrong and pat yourself on the back for an unbroken chain of confirmations. Great work: you've managed to predict that the world is full of de
But once they can identify who is going to have dementia or Alzheimer's, they can start figuring out why and then find real treatments. So even though it would suck to know you were going to eventually drift off into the nether it is an important step.
"All those moments, will be lost in time...like tears in rain..."
This is sort of my field too although I am not a doctor but a basic researcher. So there could be three things valuable about this test:
1. It may be able to give diagnosis earlier.
2. It may be cheaper and faster than current screening.
3. When combined with current techniques it may add a few percentages to diagnostic precision. If it is cheap, it will be worth it for that alone.
From my perspective, #1 is key. We need to find the cause of Alzheimer's and many people (including myself) think that plaques are a red herring, a symptom not the cause.
So if we can find a test to screen for the earliest stages of disease then finding the root cause may be easier.