New blood test could detect more than 20 types of cancer

Crucially, 99.4 per cent cases identified as cancer were correctly spotted - meaning just 0.6 per cent of cases were misdiagnoses of healthy patients.

Re: New blood test could detect more than 20 types of cancer

#5

Post by petilon » Sun, Sep 29, 2019, 3:12 PM UTC

Original press release that is the basis for the story: https://www.dana-farber.org/newsroom/news-releases/2019/new-...

Re: New blood test could detect more than 20 types of cancer

#6

Post by doctoring » Sun, Sep 29, 2019, 3:15 PM UTC

In addition to the press release someone posted, if you’re interested in some of the (high level) science behind this, I’d encourage you to read the ESMO presentation slides this news comes from: https://grail.com/wp-content/uploads/ESMO_2019_Oxnard_CCGA2_...

I don’t think the article mentions it, but the study is funded by Grail, which is commercializing a variety of related technologies.

Re: New blood test could detect more than 20 types of cancer

#7

Post by kurthr » Sun, Sep 29, 2019, 4:27 PM UTC

PhantomGremlin wrote: post #4

aszantu wrote: post #2

no mentioning of false positives

It does mention this, but doesn't use those exact two words: Crucially, 99.4 per cent cases identified as cancer were correctly spotted - meaning just 0.6 per cent of cases were misdiagnoses of healthy patients.

By the birthday paradox, that means 11% of healthy people will be misdiagnosed for one form of cancer on the 20 test panel.

That's going to require some in person discussion and review... maybe they can handle that with the NHS, but I think the US patient would end up spending an extra $1000+ due to required human contact, beyond any test cost.

Re: New blood test could detect more than 20 types of cancer

#8

Post by burke » Sun, Sep 29, 2019, 4:35 PM UTC

kurthr wrote: post #7

PhantomGremlin wrote: post #4

Earlier quoted context omitted.

It does mention this, but doesn't use those exact two words: Crucially, 99.4 per cent cases identified as cancer were correctly spotted - meaning just 0.6 per cent of cases were misdiagnoses of healthy patients.

By the birthday paradox, that means 11% of healthy people will be misdiagnosed for one form of cancer on the 20 test panel. That's going to require some in person discussion and review... maybe they can handle that with the NHS, but I think the US patient would end up spending an extra $1000+ due to required human contact, beyond any test cost.

It seems quite obvious to me that the 0.6% is already accounting for there being multiple tests, and even if not, confounding factors are likely not completely independent per test.

Re: New blood test could detect more than 20 types of cancer

#9

Post by ethhics » Sun, Sep 29, 2019, 4:36 PM UTC

From [0], the types of cancer they can screen make up at least 81.4% of new cancer diagnoses, which means that with their 99.4% true positive rate, this test should be able to correctly identify ~80% of new cancer incidences. If I'm doing my math right, that's really impressive.

[0]: https://www.wcrf.org/dietandcancer/cancer-trends/worldwide-c...

Re: New blood test could detect more than 20 types of cancer

#10

Post by sandworm101 » Sun, Sep 29, 2019, 4:40 PM UTC

ethhics wrote: post #9

From [0], the types of cancer they can screen make up at least 81.4% of new cancer diagnoses, which means that with their 99.4% true positive rate, this test should be able to correctly identify ~80% of new cancer incidences. If I'm doing my math right, that's really impressive. [0]: https://www.wcrf.org/dietandcancer/cancer-trends/worldwide-c...

It seems to have a great false positive rate, but a not-good false negative rate. It misses the bulk of stage one cancers. So its utility for early diagnosis (the important bit) is marginal at this time.