There was a recent announcement of a stool test that can detect colon cancer. A company called Exact Sciences has developed this test, and presented some results at an American Association of Cancer Research meeting at the end of October. I wrote another blog entrya while back that talked about positive predictive value (PPV) in the context of mammography for breast cancer screening. I think it's worth discussing the numbers that Exact Sciences reported and what they imply.
They said that they correctly detected 85% of the cancers, which is the sensitivity. They also reported a specificity of 88%, which means they correctly identified 88% of the patients without cancer. So the test has a 12% false positive rate and a 15% false negative rate. So that means that 12 out of 100 patients with positive tests do not have cancer. And 15 of 100 negative tests do have cancer.
The false positive rate is not so big a deal. This is because Exact Sciences is envisioning this test as a first screening in which a patient with a positive test would then go for a confirmatory colonoscopy. But, to be clearer about how many people would need those follow-up colonoscopies you need to calculate the positive predictive value (PPV). The positive predictive value is the fraction of positive tests that actually have the disease. The incidence rate of colon cancer is roughly 50 per 100,000, or 0.05% (from SEER), per year. So based on the sensitivity and specificity, you'd catch 43 of the 50 with a positive test, and also 12% of the 99,950 who didn't have cancer. So the PPV is 43 / (43 + 11,194) = 0.36%. Which means you'd get 262 false positives for every true positive.
This might sound bad but it's actually quite good, because the alternative would be that all 100,000 people would have to get colonoscopies. Now only 262 of them will have to. That will save money and no doubt be a viewed as a good thing by people who don't like the whole colonoscopy experience.
The false negative rate is little more interesting. So if 100,00 people were tested, you'd get 87,963 negative results, of which 7 actually did have cancer. That doesn't sound so bad, but it means that 7 of the 50 people that had cancer in that group of 100,000 would not be identified as having cancer. And if this test from Exact Sciences is being used as a screening test, those people would just go home and assume they were fine. This doesn't sound so great.
Now the real thing to ask here is: How accurate is a colonoscopy? How many cancers does it miss? The tests that Exact Sciences ran were using colonoscopies as the way to determine who actually had cancer and who didn't. If colonoscopies were perfect then if you substituted Exact Sciences' test for colonoscopies then 15% of people who would have gotten a cancer diagnosis would go undiagnosed.
There doesn't appear to be a lot of data on this issue, since to assess accuracy of a test you need a true result to compare it to. There was a study several years ago that looked at people who died of colon cancer and how often they had colonoscopies. They found that 7% of people who died from colon cancer did have a colonoscopy prior to that, although the time period in which the colonoscopy occurred could have been as long as 10 years. So you can assume that those 7% had cancer but it was missed by the colonoscopy. Now I really don't want to be quoted as saying the false negative rate of colonoscopies is 7% based on his one study. I found a similar study that found about 3% of patients who were diagnosed with cancer had prior colonoscopies, in this case within 5 years. I suspect the true rate is hard to know exactly, but if you use these studies as a guide you could guess that the false negative rate for colonoscopies is somewhere in the range of 3-7%, which is definitely less than the 15% rate that Exact Sciences reports for their test.
I would be interested to see what the proposed cost if the Exact Sciences' test is. I would assume it would be cheaper than a colonoscopy, and it is certainly less invasive and presumably has fewer, if any, side effects. Given that, we get into yet another debate about how much it is worth to us, and society, to have the absolute best diagnostic tools. This might be an example of a substantially cheaper test, with fewer side effects, that does almost as good a job. How that will be viewed by the public and the healthcare community will be very interesting. If we want to rein in healthcare costs in this country we'll have to start making cost part of these equations, although I suspect some will not be happy about this.
