3 ms·

On national news they mentioned 1% came back positive for people who were believed to be entirely healthy. Unfortunately it was anonymous so they could not get

On national news they mentioned 1% came back positive for people who were believed to be entirely healthy. Unfortunately it was anonymous so they could not get back in touch with the false positive patients. The spokesperson made it very clear its early days and they are looking to refine it further; they anticipate it to be a cheap and accessible 'first-call' test in regular cancer screening, not a definitive diagnosis.

carlmr 9y ago
According to https://www.cancer.org/cancer/cancer-basics/cancer-prevalence.html https://www.cancer.org/cancer/cancer-basics/cancer-prevalenc... about 15 million Americans have cancer. That is about 5% of the population. Given those numbers and assuming you have a positive result for a random person, that means according to Bayes: C being cancer, H healthy, P positive test P(C|P) = [P(P|C) * P(C)] / [P(P|C) * P(C) + P(P|H) * P(H)] = [ 0.7 * 0.05] / [ 0.7 * 0.05 + 0.01 * 0.95] = 0.035 / (0.035 + 0.0095) = 0.7865 That's pretty good. If you have a positive test result, in 4 out of 5 cases it actually is cancer. It gets a bit worse if you account for the fact that this test only works for a subset of the cancers. But this would be useful as a pre-screening.
- nonbel 9y ago
  Since this is a screening test, I don't think that 15 million number is correct for prevalence. You are using % of population diagnosed with cancer. The correct value would be the % of population with undiagnosed cancer.
  - carlmr 9y ago
    True. That is really difficult to obtain though. But assuming that most people are eventually diagnosed with cancer (even if just close to death), then a similar maybe a little lower population must have undiagnosed cancers. In the end you'll need a properly conducted study with random undiagnosed people to get reliable numbers though.
    - nonbel 9y ago
      It seems like pretty basic info to me. There should be some estimate out there somewhere, especially given the billions of dollars spent each year on studying cancer. A quick search didn't lead to anything promising though.

carlmr 9y ago

According to https://www.cancer.org/cancer/cancer-basics/cancer-prevalence.html https://www.cancer.org/cancer/cancer-basics/cancer-prevalenc... about 15 million Americans have cancer. That is about 5% of the population. Given those numbers and assuming you have a positive result for a random person, that means according to Bayes: C being cancer, H healthy, P positive test P(C|P) = [P(P|C) * P(C)] / [P(P|C) * P(C) + P(P|H) * P(H)] = [ 0.7 * 0.05] / [ 0.7 * 0.05 + 0.01 * 0.95] = 0.035 / (0.035 + 0.0095) = 0.7865 That's pretty good. If you have a positive test result, in 4 out of 5 cases it actually is cancer. It gets a bit worse if you account for the fact that this test only works for a subset of the cancers. But this would be useful as a pre-screening.

nonbel 9y ago
Since this is a screening test, I don't think that 15 million number is correct for prevalence. You are using % of population diagnosed with cancer. The correct value would be the % of population with undiagnosed cancer.
- carlmr 9y ago
  True. That is really difficult to obtain though. But assuming that most people are eventually diagnosed with cancer (even if just close to death), then a similar maybe a little lower population must have undiagnosed cancers. In the end you'll need a properly conducted study with random undiagnosed people to get reliable numbers though.
  - nonbel 9y ago
    It seems like pretty basic info to me. There should be some estimate out there somewhere, especially given the billions of dollars spent each year on studying cancer. A quick search didn't lead to anything promising though.