Slashdot Mirror
Blood Test of 4 Biomarkers Predicts Death Within 5 Years
retroworks writes "The NHS and the Daily Telegraph report on two studies (original and repeat duplicating results) in Estonia and Finland which predict whether an apparently healthy human will likely die within 5 years. The four biomarkers that appeared to determine risk of mortality in the next five years were: alpha-1-acid glycoprotein – a protein that is raised during infection and inflammation; albumin – a protein that carries vital nutrients, hormones and proteins in the bloodstream; very-low-density lipoprotein (VLDL) particle size – usually known for being 'very bad' cholesterol; and citrate – a compound that is an essential part of the body's metabolism. Researchers found that people in the top 20% of the summary score range were 19 times more at risk of dying in the next five years than people in the lowest 20%."
The NHS's summary of the news points out that "the implications of such a test are unclear. As this was an observational study, it can only show an association between the biomarkers and risk of death. It does not predict what the underlying cause of death would be for an individual and does not therefore provide an answer in terms of treatment."
This should be very helpful to the insurance industry in determining risk.
The links to the Telegraph and the NHS are both bad.
Genius is one percent inspiration and 99 percent perspiration, which is why engineers sometimes smell really bad.
If a person's age is over 105 years, they're somewhere around 99% likely to die within the next 5 years.
He tried to kill me with a forklift!
Didn't Heinlein write a short story about something similar? "Life-Line" I think it was called.
Hoist Number One and Number Six.
First, correlation will not tell you causes. Second, correlation does not necessary make individual outcome predictions possible.
For example, out of population that have bad scores on this test mortality may be 19 times higher, but for any given individual it does not necessary means they are going to die in 5 years, or event that they are significantly more likely to die in 5 years.
I think high levels of cyanide are also a good predictor of death. And that's just one example.
19 times more likely to die than without the protein, well this is very conclusive of death within 5 years this could easily be influenced by lifestyle choices, I see no limitations on diet or activities how do we know that the participants weren't all taking up extreme skydiving or eating fried chicken every night.
Two of the three links in the summary are broken.
When I read "predicts death within five years," I inferred a "to," that is, I expected that the study predicted when a person would die to within a margin of error of five years (death clock), which would be a much bigger deal than what they actually did.
OK so the markers indicate a state that normally leads to death. Now we can detect it. Does that mean we can now do fix for the detectable condition and those deaths now become avoidable?
I suppose this is some sort of organ failure early warning system. What it definitly is not is some sort of quantum state indicating that the body is about to undergo soul removal.
Death is a quite rare thing; ignoring age and other factors, the probability of someone to die within five years is less than 5%. Even when you belong to the top 20% in terms of risk, the probability of death is just 15%, so you're much more likely to be alive than dead after this time. And for what it's worth, the biomarkers are strongly correlated with other factors like "does this person have cancer?", so that in the end the authors say that their new model is just 4% better than previously used models.
OS Reviews: Free and Open Source Software
Just glad TFA doesn't try to sensationalize this, though who knows what the media might try to do with it.
"The implications of such a test are unclear. As this was an observational study, it can only show an association between the biomarkers and risk of death. It does not predict what the underlying cause of death would be for an individual and does not therefore provide an answer in terms of treatment."
if your physician decides to drop some papers onto the floor as a ditraction, you're doomed.
correlation imply causation of not finding an insurance.
Everyone at or over the age of five has died, or will die, not within, but EXACTLY in five years, eventually.
What would be trippy would be if this was regardless of the type of death - accidental, murder, etc.
-Styopa
People who take the test may wish to load up on life insurance.
Insurers, on the other hand, will want to use these tests and DENY to offer insurance, unless the premium goes up 20 times.
This could easily destroy the life insurance industry as it is today.
Of the top 10 causes of death in the world, according to the WHO, ischemic heart disease and stroke kill more people than the other 8 combined. Doctors already knew that bad cholesterol (ldl/vldl) and inflamation were the leading cause of both of those and have known for decades. The study doesn't show anything that wasn't already known and just adds sensationalism, probably to get increased funding.
Running to be the first to find a cure for the symptoms, instead of the percieved problem. And charge you more for not taking it, when it is marketed. 5 in 1! Profit (for them)! paleo-GATTACA.
The study doesn't "predict death within 5 years" it doesn't even predict death for those with the biomarkers. All it says is people who had higher levels of the biomarkers exhibited a greater risk of dying within 5 years than those with lower levels; according the TFA the study didn't even claim a causal relationship between the markers and a cause of death. Of course, a headline that reads "Study should some people have a higher risk of dying in 5 years than others..." wouldn't be as catchy.
I'm a consultant - I convert gibberish into cash-flow.
Honestly, after the p-value article, why is this crap still being published? p value was said to mean its worth a second look, but NOT imply anything else
Second, notice its not in PLOS one. Wonder why? Oh right, they require all data to be public, so you can't "use a model" that just happens to make the results you were looking for.
Lastly, percentages e.g. 19 times more likely! See http://xkcd.com/1252/ Without the baseline, this 19 times more likely is utterly useless. If their "average case" had a 0.000000000001% chance of death, 19 times that would be 0.000000000019% Thats still pretty low. It reeks of numbers manipulation in an attempt for publicity and funding. If the baseline was something reasonably high, like say 1%, and it jumped to 19%, sure that's quite significant! However, were that the case, it would be far more exciting to say that, than simply 19 times, and they would have done so. My guess is my examples are hyperbole, and the actual is probably closer to 0.1 with those markers, their modeling, number fudging, etc, 1.9%. Still not an accurate predictor of mortality, and basically useless.
It's already pretty crazy the number of blood and urine samples insurance companies collect before they issue you insurance. There is no way they will pass up on incorporating this technology into their tests.
So if I'm reading correctly, this test tells us that having a body full of inflamation and bad cholesterol might kill you quickly. Thanks a lot guys. Treat the underlying cause(s) of those conditions and get back to us. If the subject keeps eating burgers, drinking up a storm and going to brothels, well... the mere fact that we have a chemical test to determine he's killing himself doesn't bother me. You could probably tell just by looking.
If I were suicidal I'd show you how to predict within a few hundred ms.
I disagree, but we can have this debate when a tests is released that is at least 99.9% accurate. This test is nowhere near that. Being 20x more likely to die is not a guarantee. What if the percentage of people expected to die within 5 years is 0.01? Even a 100 folder increase in risk would mean that fewer that 1 in 100 are actually expected to die. And don't even get me started on outliers and mitigating factors.
- Why? Yes! Because that website said you were going to die at 3pm, didn't it?
- Did it? I can't remember.
- How could you forget? It clearly stated that you were going to die, today. At precisely 3:00. Unless it was tomorrow. But no, it's today, at 3:00.
http://howlonghaveyougot.com/
Some phones are meant to have a weedy vibrate setting.
Scarce, scared, scarred, sacred... -Col. Bruce Hampton
Why would correlation!=causation get anything but -1 pedantic ass mods?
People keep repeating that so much on this board that it is negatively affecting their ability to critically think. When dealing with statistics, sometimes cause isn't important, just predicting an outcome with a high degree of certainty.
Just shut your mouth, and think about the conversation at hand differently.
These 4 markers are wildly disjoint. Clearly not one cause involved here. The causes of them individually are pretty well known. But if they can predict something with a high degree of certainty, what does that imply? Who can benefit? Who can get hurt?
I'm curious how these people were classified as 'apparently healthy'. It sounds like these biomarkers were all associated with various health conditions. Did these people have undiagnosed health problems that would have been discovered with a general checkup or did this indicate the presence of problems that would have been otherwise undetected?
I stole this Sig
Honestly, after the p-value article, why is this crap still being published? p value was said to mean its worth a second look, but NOT imply anything else
Second, notice its not in PLOS one. Wonder why? Oh right, they require all data to be public, so you can't "use a model" that just happens to make the results you were looking for.
Lastly, percentages e.g. 19 times more likely! See http://xkcd.com/1252/ Without the baseline, this 19 times more likely is utterly useless. If their "average case" had a 0.000000000001% chance of death, 19 times that would be 0.000000000019% Thats still pretty low. It reeks of numbers manipulation in an attempt for publicity and funding. If the baseline was something reasonably high, like say 1%, and it jumped to 19%, sure that's quite significant! However, were that the case, it would be far more exciting to say that, than simply 19 times, and they would have done so. My guess is my examples are hyperbole, and the actual is probably closer to 0.1 with those markers, their modeling, number fudging, etc, 1.9%. Still not an accurate predictor of mortality, and basically useless.
Fortunately there's a paper linked to in the summary that answers your concerns.
The 5-y mortality for persons with a biomarker score within the highest quintile was 19 times higher than for those in the lowest quintile (288 versus 15 deaths during 5 y, corresponding to 15.3% versus 0.8%). Individuals within the highest quintile were further differentiated in terms of their short-term probability of dying according to their biomarker score percentiles: 23% of the individuals with a biomarker score within the highest percentile had died within the first year of follow-up (23 out of 99), and the estimated 5-y mortality was 49% (Figure 5B).
I'm not gonna run the numbers but 288 vs 15 is probably outside of most p-values.
Also note this was a replication of another study, once could be publication bias, but replication raises the odds you're looking at something real.
23% first year mortality for the highest percentile group?? That's definitely something worth writing home about (and you might want to send a will along with it).
I stole this Sig