Do the multi-cancer blood tests work yet? What the evidence actually shows

The pitch for multi-cancer early detection tests, the best known of which is Galleri, is close to irresistible: one blood draw, screened for a signal shed by dozens of cancers at once, many of which have no screening test of their own. The science behind them is real. Whether they are ready to be used as screening tools is a separate question, and the honest answer from the published evidence is: not yet, and possibly not for the cancers where they would help most.

How the tests work

These tests, often abbreviated MCED, analyse cell-free DNA — fragments of DNA shed into the bloodstream. Rather than reading the DNA sequence for mutations, the leading test looks at methylation patterns, chemical marks on the DNA that differ between healthy and cancerous tissue, and uses machine learning to decide whether a cancer signal is present and, if so, to predict where in the body it came from [s1]. That second trick, predicting the cancer signal origin, is what could make a positive result actionable rather than merely alarming.

What the accuracy numbers say

The pivotal validation came from the Circulating Cell-free Genome Atlas study. In an independent set of 4,077 participants, the test's specificity for a cancer signal was 99.5 percent [s1] — meaning false alarms were rare, about 5 in 1,000 people without cancer. That is the test's strongest number, and it matters, because a screening test applied to healthy populations must almost never cry wolf.

Sensitivity — the share of real cancers the test catches — is where the picture gets more sobering. Overall sensitivity across all cancers and stages was 51.5 percent, so the test missed roughly half of the cancers present [s1]. And sensitivity climbed steeply with stage: 16.8 percent for stage I, 40.4 percent for stage II, 77.0 percent for stage III and 90.1 percent for stage IV [s1]. In other words, the test is good at detecting advanced cancer and poor at detecting the earliest-stage disease — which is precisely the disease that early detection is supposed to catch. For a pre-specified group of 12 cancers that account for about two-thirds of US cancer deaths, sensitivity across stages I to III was 67.6 percent [s1]. When it did flag a cancer, it predicted the origin correctly 88.7 percent of the time [s1].

What happens when you actually use one

Accuracy in a case-control study is not the same as usefulness in practice. The PATHFINDER study gave the test to 6,662 adults aged 50 or older in real clinics [s2]. A cancer signal was detected in 92 people, about 1.4 percent [s2]. Of those, 35 turned out to have cancer — true positives — and 57 did not, false positives [s2]. So even with the test's high specificity, well over half of the people who got a positive result did not have cancer, a direct consequence of screening for rare events. Those 57 people underwent diagnostic workups, and the study measured how long that took: a median of 79 days to resolution [s2]. PATHFINDER's own authors framed it as supporting feasibility while underscoring the need for further research into the test's clinical utility [s2]. The study was funded by GRAIL, the test's maker [s2].

The question the evidence has not answered

The decisive question for any screening test is not whether it finds cancers but whether using it helps people live longer or better — and no completed randomised trial has yet shown that an MCED test reduces cancer deaths. That is the same trap ovarian-cancer screening fell into: a test can shift diagnoses earlier, or find more cancers, without saving lives, if it mostly catches cancers that were never going to be fatal or advanced cancers it cannot change. Large randomised trials designed to measure that outcome are underway, and until they report, the central claim — that these tests improve outcomes — remains unproven rather than disproven.

What this means for a reader

Multi-cancer blood tests are a genuine scientific advance with a real weakness: they are best at detecting cancer that is already advanced and weakest at detecting the early-stage disease that screening is meant to find [s1]. They are currently offered in some settings as an addition to standard screening, not a replacement for it, and a positive result triggers a diagnostic hunt that often ends in no cancer [s2]. Anyone considering one should understand both the appeal and the unanswered mortality question, and discuss it with a clinician rather than treating a normal result as an all-clear or a positive result as a diagnosis. Established screening tests — for breast, cervical, colorectal and, for eligible smokers, lung cancer — have the randomised evidence these newer tests are still gathering.

Sources

• [s1] Clinical validation of a targeted methylation-based multi-cancer early detection test (CCGA), Annals of Oncology, 2021-06-24
• [s2] Blood-based tests for multicancer early detection (PATHFINDER), The Lancet, 2023-10-01