Journal Club: DETECT-A and Cancerguard (Exact Sciences)

Background

The CancerSEEK test, the MCED test used in DETECT-A, was an early version of what became Cancerguard. The CancerSEEK concept was first described in a landmark 2018 paper published in Science. That study was a case-control analysis of 1,005 patients with known, nonmetastatic cancers of eight types (ovary, liver, stomach, pancreas, esophagus, colorectum, lung, and breast) and 812 healthy controls. The median sensitivity of the test among the eight cancer types was 70%, and the sensitivities ranged from 69 - 98% for the five cancer types lacking any standard screening (ovarian, liver, stomach, pancreatic, and esophageal). Stage I sensitivity was 43%. Specificity exceeded 99%, with 7 of 812 controls testing positive. As is the case with other case-control studies (the same type of study as GRAIL’s CCGA), including patients with known cancer inflates apparent sensitivity, and the 55% prevalence of cancer in the study population bears no resemblance to the (lower) prevalence in a screening population. However, the study was intended as a proof of concept.

Study Design

DETECT-A (Detecting cancers Earlier Through Elective mutation-based blood Collection and Testing) is part of the foundation for Exact Sciences' MCED program and enrolled 10,006 women aged 65 to 75 with no personal history of cancer through the Geisinger Health System in Pennsylvania and New Jersey between 2017 and 2019. Participants had high adherence to standard-of-care screening. The study reportedly only included women to enrich for ovarian cancer. The baseline test, an early version of CancerSEEK, analyzed mutations in defined regions of 16 cancer-associated genes in cfDNA and cancer-associated protein biomarkers, such as CA-125, CEA, and CA 19-9 (among others). As mentioned in the post “How MCED Testing Works”, this is different from Galleri, which is based on analyzing methylation patterns.

The study used a two-step testing process to enhance specificity. A positive baseline test triggered a repeat confirmatory test for confirmation (specifically to determine if the abnormal biomarker seen in the first test was persistently abnormal and to exclude tests deemed incorrectly positive due to CHIP, a phenomenon described in the post, “How MCED Testing Works”). If the repeat confirmatory test was also positive, the participant underwent diagnostic PET-CT imaging.

Of the 9,911 women who completed baseline testing, 490 had an initially positive blood test. Of those, only 134 were confirmed positive on second testing. The investigators reported that 60% of non-confirmed cases were attributable to CHIP.

All participants, regardless of test result, were encouraged to continue standard-of-care cancer screening throughout the study, and each was followed for at least 12 months.

Results

During the study period, 96 cancers were diagnosed among the 9,911 participants who completed baseline testing. The MCED test was the first to identify just 26 of those cancers, of which 17 (65%) were early-stage. Of the 26 MCED-detected cancers, 15 underwent confirmatory PET-CT imaging, and 9 were surgically resected. An additional 24 cancers were detected through standard-of-care cancer screening methods, and 46 were not detected by either approach (I assume by other pathways like symptoms, incidental findings, and the like).

The performance metrics for the MCED test, incorporating both the baseline and confirmation blood-based testing steps, were as follows: sensitivity was 27.1%, specificity was 98.9%, PPV was 19.4%, negative predictive value was 99.3%, and the number needed to screen to detect one cancer was 381.

What I Think

The good news:

1. Of the 50 cancers detected by either MCED testing or guideline-recommended screening (colonoscopy, mammography, etc.), 26 were identified solely through MCED testing. So, adding the MCED test doubled the number of detected cancers when added to guideline-recommended screening (albeit at small absolute numbers). This lends some credence to the idea that MCED tests won't simply duplicate what existing screening methods already catch. Notably, the MCED test detected six ovarian cancers (of the seven detected in the study), a type of cancer without a guideline-recommended screening test for people without risk factors.
2. Guideline-recommended screening adherence was maintained. One concern about MCED testing is the thought that a "no cancer detected" result might make people less likely to get their guideline-recommended cancer screening tests. DETECT-A tracked this explicitly and found no evidence of reduced adherence to standard screening among participants who received the blood test, regardless of the test result.

Several limitations are present and are worth discussing.

1. The design of the study was that of a case-control study (similar to GRAIL’s CCGA study) and not a prospective interventional trial (like PATHFINDER and PATHFINDER 2). Performance metrics might have been different had this test been applied to a screening population.
2. A sensitivity of 27.1% is low by any conventional screening metric. The study authors explicitly acknowledged this and noted that the baseline test was an early version of CancerSEEK (designed before the machine learning methods described in the 2018 proof-of-concept paper were finalized).
3. CHIP was a major confounder. Sixty percent of initially positive tests that were not confirmed as positive on repeat testing were attributable to CHIP. This is a known challenge for mutation-based cfDNA assays, particularly in older populations where the prevalence of CHIP can exceed 10%.
4. DETECT-A enrolled women aged 65 to 75 from a single health system in Pennsylvania and New Jersey, which is not representative of the population who would likely order an MCED test. So, generalizability to a broader screening population is uncertain.

What Came After

Since DETECT-A was published, Exact Sciences has evolved its platform.

Ascertaining Serial Cancer patients to Enable New Diagnostic 2 (ASCEND-2) was a case-control study that bridged DETECT-A to Cancerguard. It enrolled over 11,000 participants across 151 sites in the US and Europe. The study's primary purpose was to refine the classifier algorithm and determine the biomarkers for the final design of Cancerguard. The first analysis, presented at the American Association for Cancer Research (AACR) conference in April 2024, evaluated a test utilizing DNA methylation and protein biomarkers in 6,354 participants. In the poster that was presented, across 21 cancer types, overall sensitivity was 50.9%, and specificity was 98.5%. Sensitivity for cancers without standard-of-care screening was 54.8%, and sensitivity for the six aggressive cancers with the shortest five-year survival (pancreatic, esophageal, liver, lung, stomach, and ovarian) was 63.7%. A second analysis, presented at the AACR Special Conference on Liquid Biopsy in November 2024, evaluated a three-biomarker-class approach that added a DNA mutation reflex step to the methylation and protein-based approach. The overall sensitivity was 52.9% (19.1% for stage I, 42.2% for stage II, 72.3% for stage III, and 89.9% for stage IV), and specificity was 98.5%.

Exact Sciences has reported data for the commercial Cancerguard test, which launched in September 2025, from development and validation cohorts that are distinct from those in the ASCEND-2 conference presentations. These data appear in the Cancerguard Clinician Brochure. According to this brochure, in the development cohort of 590 patients with cancer and 2,434 non-cancer controls (excluding breast and prostate), Cancerguard achieved 64.1% sensitivity and 97.4% specificity. The brochure also notes that in an independent validation cohort (223 cancer patients, 800 controls), sensitivity was 55.6% with the same specificity (97.4%).

Notably, the ASCEND-2 presentations used a specificity threshold of 98.5% while the commercial Cancerguard test operates at 97.4%. This is a deliberate tradeoff that captures more cancers at the cost of more false positives. The higher sensitivity of the commercial test relative to the ASCEND-2 presentations reflects, at least in part, this lower specificity threshold. Additionally, there is a difference between the development cohort sensitivity (64.1%) and the validation cohort sensitivity (55.6%), which is a reminder that headline numbers from development data can overestimate what an independent validation shows. Further, and perhaps most importantly, none of these numbers comes from a prospective screening population. Both the ASCEND-2 data and the commercial test performance data are from case-control designs, where participants were enrolled because they either had a known cancer diagnosis or did not. Performance metrics from case-control studies routinely overestimate what is observed in true screening populations.

Exact Sciences is currently building registrational-quality evidence through the ongoing Falcon Registry (NCT06589310), which is a multi-site, prospective study conducted under an FDA-authorized Investigational Device Exemption. It is enrolling up to 25,000 participants aged 50 to 80 with no history of cancer within the prior three years. The primary study sites are Baylor Scott & White in Texas and Endeavor Health in Illinois. Participants in the Falcon Registry study undergo annual MCED testing for three years with two additional years of follow-up data collection. The study also includes a matched comparator cohort of up to 50,000 patients receiving "standard-of-care clinical management" without the MCED test, providing a control group that many studies evaluating MCED tests have lacked in the prospective setting (like PATHFINDER and PATHFINDER 2). The study's objectives span from test performance to the psychological impact of MCED testing on participants. The first patient was enrolled in August 2024.