In my July 2022 blogatorial, I covered the GRAIL blood test (for 50 types of cancer) that had been promoted locally by Mercy Hospital OKC as well as the entire Mercy system. The implication was strong — “We now offer….” which might have meant they had arranged a discount with bulk orders, but it was NOT a statement of exclusivity. Any physician could order the test. In fact, within a 20-mile radius of Mercy Hospital, there were 22 providers signed up to be GRAIL users.
The GRAIL test is a juggernaut backed by Bill Gates, Jeff Bezos, etc., to the tune of billions of dollars (read that July 2022 piece for details). Problem is, it is very difficult to prove a test like this when there are 5 performance characteristics to be applied to 50 cancers, giving 250 data points to evaluate. But when the Big Announcement came, the numbers floating about were not good — that is, the most important characteristic of Sensitivity (percentage of cancers detected) was mediocre at best, while for some cancers (like breast) the results were terrible (all while heavy promotion was underway).
What the circulating tumorDNA (ctDNA) could do quite well was Specificity (fewer false positives) but that’s not what saves lives. Specificity allows for test feasibility, and their number was near 100% for most of the cancer types. Great! Fewer false-positives, more efficient, patient-friendly. But the raw data presented way back in 2022 revealed a twisting of reality that made the test look better than it really was. First of all, much of the reporting was done with advanced stage disease (that’s NOT screening). Take away the Stage III and IV cases, and the Sensitivity for most cancer types plummets. So, rather than admit the poor Sensitivity, the marketing department loaded up using a performance characteristic of “Accuracy” which, in testing parlance does NOT mean Accuracy like you’d think. It’s a formula that blends Sensitivity and Specificity. So, if you have an excellent Specificity of 100% and a bad Sensitivity of 50%, you can blend the two parameters and come up with 75% Accuracy, which is OK, but not great. Much of the marketing material about GRAIL does just that, i.e., rely on Accuracy. But when it comes to measuring saved lives, Sensitivity dominates as the key parameter. If a cohort of patient has 100 cancers, and your test under study can detect 90 of them, then great! 90% Sensitivity! (though you still have to prove fewer breast cancer deaths). Bottom line: it was suspicious in 2022 that GRAIL was not getting spectacular results, and the question was “Does it have any role in patient care today? Specifically, will it save lives, or at least lower the number of advanced cancers?”
Sure enough, when put to the test by the NHS of the United Kingdom, the GRAIL test did not save lives overall, at least when judged by “fewer advanced cases of cancer.” This is a surrogate for saved lives. Now, after the company stock crashed, they went to work and they’ve identified trends of efficacy, noting a core group of 12 cancer types where there might be benefit (the official results, though, are a failure to meet the original primary endpoints). So, by focusing on great Specificity and hiding lackluster Sensitivity, the company shot itself in the foot (though I suspect the test will get better and better over time….especially in those cancers where there are no other proven screening methods).
But what about breast cancer? For some reason, circulating tDNA does not allow reliable early detection of breast cancer. This should be headline news in the breast cancer community, but it’s hardly mentioned. Some companies using this ctDNA approach have dropped breast cancer from their lists of detectable cancers. Oddly enough, Cologuard lists breast cancer in their overview of cancers detected, but with a subtle asterisk denoting “not indicated.” Go figure. Breast cancer and prostate cancer are the only cancer types that are market by the asterisk.
In the PATHFINDER study (35,878 women), one has to go to the online supplements and use a magnifying lens to find out what happened in the 14 cases of breast cancer that were all discovered by mammography. Using GRAIL, NOT A SINGLE BREAST CANCER WAS IDENTIFIED by the blood test. All 14 were found by mammography. Incidentally, please note that the blood test research where I’ve been involved since 1993 is aimed at finding breast cancer EARLIER than mammography, detectable by Ultrasound or more certainly, by MRI. The very purpose of blood test research, in my view, has always been to use blood test results in tandem with mammography, and if the mammograms are negative (esp. if dense tissue), then a positive blood test should prompt Ultrasound at a minimum, and preferably MRI. GRAIL (and other companies using the same ctDNA) doesn’t have a chance in breast cancer — it can’t even find the cancers found on mammography, while our goal is to improve detection beyond mammography.
I’ve spent a lot of time on this topic lately (July 2022 and January 2026), as clinicians are perplexed about this form of testing, and many are confused by the 5 performance characteristics I discussed in detail in the January 2026 blog. Bottom line — by focusing on a single cancer type (like PSA for prostate cancer), we generate tons of data that has to be analyzed closely. Over the years, I’ve submitted 10,000 samples to researchers who are focused on Sensitivity using completely different approaches than GRAIL et al. The company I’ve worked with since 2017 as a clinical consultant is Syantra, Inc. (Calgary, Alberta) where scientists are looking at changes in biomarkers that are NOT generated by cancer cells, but are the changes that occur in immune cells…in response to the breast cancer. It’s a different approach, but getting far better results than GRAIL. If you are interested, here’s a PR piece that was written recently about the company:
Early Breast Cancer 