Worth noting that because of anonymity, they couldn't go back and check if those 7 people actually had cancer and just didn't know it.

Recently the woman who could smell Parkinson's Disease got one test wrong but when they rechecked they found that she was right, it was just undiagnosed in that person.

Smelling Parkinson's is fascinating -- thanks for dropping that nugget.

Read a little more about it --> Researchers identified 10 molecules that Parkinson's patients emit and could be used for early detection. Pretty cool. http://www.telegraph.co.uk/health-fitness/body/meet-woman-ca...

> because of anonymity, they couldn't go back

Is this so? I thought the original researchers would be able to follow-up if they had accounted for it in their consent forms. In other words, they surely have identifying patient information, and even though an outside team couldn't screen the same cohort, that original team could find those 7 patients and ask again in 2 years.

I'm pretty sure they would be able to - I worked with cancer trial data (just a student job, doing data entry) but all patients were completely anonymised to anyone working with the data but readily identifiable by a specific unique ID number, so higher-level trial managers and doctors could find patient information if needed.

It is common for the data to be anonymised. Especially if there is a third party involved (e.g. a lab producing sequencing data attached to sub-sets of clinical information, according to whatever contract or legal environment they are in).

But surely, as the other poster mentions, they'd be tracking patients by some unique ID and someone has the key. Otherwise, how do you correlate blood results with any other patient data?

I mean, they could go back to the provider of the clinical information with a unique ID and request follow-up. But people are still a) busy (clinical data may come from a hospital that has little interest in research; researchers want to publish and move on); b) extremely scared of making mistakes (to the point of delusion: wanting data/results to never change, even if its an improvement - because that implies there was something wrong in the past). Getting follow-up clinical data may also be out of scope of existing agreements.

It shouldn't be hard to make a statistical estimate of how many people have an undiscovered cancer of this kind?

How does that compare to the base rate of these cancers in the population? The false positive rate in a control group can be very low, but still overwhelming when applied to the general population.

That's a lot of false positives. If 4% of people actually have one of these cancers at any point in time (a number that seems quite high) that means that 20% of positive results would be false in the general population.

It's still extremely useful, even with those false positive rates. You will scare a few people, but they will just retest and maybe will improve their lifestyle.

Yeah, stats are hard :)

If the blood test becomes widespread you could end up with more false positives than true positives. So probably not useful as a routine test with this error rate.

That specificity implies one of these cancers apriori occurs in 5% of the tested population.

Is that about right taking into account pre-screening? We can identify a population that has a 5% chance of having one of these cancers?

If these represent 70% of cancers overall and ~40% of the US population will get a cancer diagnosis at some point then it seems likely you can find many populations with more than a 5% chance of having cancer.

OK, but if I have a 40% chance of getting a cancer diagnosis at some point in my life, and they run the test as part of my physical every year, then presumably my chance of having cancer at the time they run the test is much less than 5%.

That just means you should not give someone the test every year.

At 70 people have a much higher risk of cancer than 7. Long term smokers have higher risks than non smokers. Combine multiple risk factors and find groups with 5+% risk of these kinds of cancers is not that difficult.

Ah, missed that, thanks. I assumed it was in the paper itself.

a 1% false positive rate. ouch.