Slashdot Mirror
Consumer Genetic Tests May Have a Lot of False Positives (theverge.com)
A new study, published in the journal Genetics in Medicine, found that consumer genetic tests bring up a lot of false positives. "In this case, 40 percent of the results from the consumer tests were false positives," reports The Verge, noting that the findings "cover a very small sample size and don't show that consumer tests always have a 40 percent false positive rate." From the report: The research was done by scientists at Ambry Genetics, a medical laboratory in California. By looking through their own database, they found that 49 people had been referred to them because of some worrying results from their consumer genetic tests. Still, scientists at Ambry were able to confirm only 60 percent of the results when they compared the raw data from consumer tests with more thorough genetic tests done by themselves and other clinical laboratories. So, 40 percent of variants in a variety of genes reported in DTC raw data were false positives, meaning that they said a genetic variant was there when it wasn't. (Most of these turned out to be variants linked to cancer.) Additionally, the authors write, some variants classified as "increased risk" were not only classified as "benign" by clinical laboratories, but they were actually common variants.
Do you **REALLY** think these companies are going to put up a legal fight for you when your DNA is requested by the government?
I came here for the clever comments
There's a massive incentive to *find something* if you're paying to have a check done... if you get nothing reported then the value of the test seems to be nothing. If you are told to be "at risk" for some condition then you can go to your friends and tell them that you had no idea and that it was so worth it and that now you can take steps... and that perhaps they should get tested too. The best part is that if you're just reported to be "at risk", rather than an actual diagnosis, there's almost no need for accuracy. If you never develop the condition it will be far in the future, and/or you got 'lucky'.
So wait, does that mean ancestry results and services like âoe23andme.comâ are bogus? Essentially, the scientific results are trash now; questionable and thus worthless data??
Understand that the only useful ancestral data are the YDNA and mtdna haplogroups, and that health data is 90% for the researchers.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
The primary income for these businesses will be to sell the DNA info to insurance companies and LE. And they don't even need to get a 100% capture rate. As long as they get someone in your family they can make a judgement about your health risks when writing a policy. Or imagine being LE. With DNA from a crime. Even through you may not have the actual person in the database you will still be able to get a very short list of suspects.
Most "serious" genetic indicators are either quite rare, or their effect has already become apparent to consumer through other means. Common consumer genotyping tests test hundreds of thousands of SNPs. The rate of errors being at least 0.1-0.6% for these methods, there are bound to be hundreds of errors in a typical test result. People are not interested of benign errors, but are very interested on errors which would indicate a health-threatening condition, and this small fraction of test results gets lots of attention. IMHO, there's nothing strange on the fact that large portion of tests leading to a second round are false positives; it's how statistics and bias in such a situation happens to work.
When it comes to whole genome sequencing, both amount of data and error rates can be considerably higher; people just have to get used to the fact noise in big datasets causes strange effects...
By looking at people with worrying results, they are strongly filtering for people will results that are already likely to be false positives.
So that '40% false positive' number is nonsense.
It's not shit on this too much. Anyone who's had to work in any science related field, you're as good as your model and training data. To even make ancillary matches that maybe cause a bit of a 'concern' but later turn out to be benign or a false positive, um, sign me the fuck up? Especially if you're not in the health and welfare camp of "You don't know what you don't want to know". Time is on our side in improvement; of course we will, but I hope it's better for my children and not an ultimate target-on-the-back for pre-healthcare screening (which is already really is) and any future healthcare you plan to get. If accuracy is improved for the right reasons, absolutely, but humanity and greed tell me otherwise.
TFA is making it sound like we have world of mega hypochondriacs --- which probably isn't entirely out of the question with today's social media platform of "Oh everybody, on my way to another dicey check-up, fingers crossed " --- but by any measure, if there are genetic health concerns that run in my blood line, why not entertain it, especially if you want to be proactive and curb or neutralize it, if it's in your immediate control to change it? At least you had some concrete information as opposed to none.
I sent some of creimer's hair I found at a Panda Express and it came back as "walrus"!
CROFLOL!!!
Maybe they should have used a Dyson vacuum cleaner to clean up that hair!
I remember an example from a book on probability: You have a test with a 1% false positive rate and a 1% false negative rate. It tests for a condition present in 1% of the population. 10000 people take your test. 100 of them actually have the condition, so you get 99 true positives from that; the 9900 people without the condition produce 99 false positives. By the criteria in the summary, in your population of people who sought further testing, 50% were false positives, but that doesn't make it a bad test.
" By the criteria in the summary, in your population of people who sought further testing, 50% were false positives, but that doesn't make it a bad test."
So flipping a coin gets the same results?
No.
He is saying of the people that got a positive result, for certain genes, it was in fact a false positive. No information is provided about what percentage of people taking the test got a false negative or true negative. If a test is incredibly accurate at knowing if you are not at risk, that's still a valid test. Especially if it isn't too expensive or invasive to follow up on positive results to see if they were accurate.
You take the test, it says you're okay you go on with life.
You take the test, it says your're not okay, you get a second opinion.
When 50% of your test positives actually have the condition that is a 50% positive predictive value, which is pretty good.
eventually these tests will be ordered by doctors. And insurance companies have a nasty habit of not paying your doctor if the test comes up clean. After all, if the test was going to come up clean anyway why order it in the first place? It sucks, because it makes doctors hesitant to order tests (since they might not get paid).
Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/
"... 40 percent of the results from the consumer tests were false positives,"
Um, no. 40 percent of the "positive" results were false positives.
There's a big difference between "40 of every 100 results are false positives, the rest are true positives or are negative or inconclusive results" and "for every 3 true positives, there are 2 false positives."
Not sure why the original article even mentions Family Tree DNA (FTDNA) who do testing only for genealogical purposes. Their name appears only in the Introduction. None of their tests is recommended or (I believe) can be used for medical or health-related purposes. On the other hand, 23AndMe is a big DTC marketer of medical DNA testing and I would bet that most of the tests evaluated in this article were by 23AndMe.
Disclaimer - I have been testing my own and the DNA of others since 2007 with Family Tree DNA, and have no connection with the company other than as a satisfied customer and a volunteer administrator for three genealogical surname DNA projects (Foad, Huntsman, and Mugford).
I wish I could give you mod points, but I've already commented on this thread... you are exactly right though... without knowing how many tests were performed you know nothing, and can't make a judgement about sensitivity and specificity.
Which has more power: the hammer, or the anvil?
Three years ago, I had my DNA analyzed by 23andme. According to my raw SNP data, there's no evidence of the genes for delayed sleep phase disorder... but I unquestionably have it, and have since childhood (& probably before). I found seemingly blatant discrepancies in a few other reported SNPs that appear to diverge quite a bit from observed reality.
I can think of a few possibilities:
* Promethase incorrectly interpreted my raw data & reported SNPs with incorrect values.
* The lab didn't actually sequence those specific values... it reported them based on indirect markers elsewhere that didn't necessarily pan out in this case
* The sample was contaminated by the lab (unlikely?)
* The lab's test for the relevant SNP was inaccurate... or at least, a lot less precise than the raw results imply.
I'd like to eventually get my genome analyzed again... by a company that uses a different lab, with slightly different test... but reports the results in a form that can be directly compared (SNP-for-SNP) with the original (or at least, deterministically converted into equivalent SNP form) to find discrepancies & get an objective idea about the likely error rate for the first.
Any suggestions for whom to use for round 2?
I depends how the result of the test is used. If it jacks up insurance premiums then 50% false positives could be really bad. If it leads to a lifetime of worry it could be pretty terrible. But if all that happens is someone gets a more expensive, reliable test (cost amoritized over everyone who took the unreliable test) then it's probably not so bad.
const int one = 65536; (Silvermoon, Texture.cs)
SJW, n: "Someone I don't like, and by the way I'm a fuckwit" - AC
All these political pundits keep pushing Elizabeth Warren to get a genetic test to "prove" she has Native American ancestry. They reference web-pages by these DNA testing co's in their claims that its possible to test. First, the vendors' own pages often state that the tests are imperfect, and that they often cannot rule out ethnic links. They are better able to say one is "probably" related to a given group, but poor at ruling out a relationship to a group. They look for specific markers or patterns, but currently not the entire genome.
Besides, Elizabeth going through with such tests is feeding the trolls. According her, her relatives told her when she was young that the family has at least one Native American in their background, and there's no reason to question one's older relatives about that. It's rude to the family, in my opinion, to publicly question their word. I hope she doesn't give in to the trolls the way Obama gave into T about the long-form birth certificate.
I see no reason to make a big deal about it. It's just partisan tiddly wings (which both sides do, by the way). Don't even get me started about T's rude "Pocahontas" jokes in front of a Native American event. Very tacky.
Table-ized A.I.
Ooops Angelina Jolie!
wow, you are a great statistician
if you truly believe your unassailable math, we wish you to be a false positive in cancer tests for 50% of your visits to the doctor