Journal Club: PATHFINDER 2 (GRAIL)

A note on sourcing: This journal club is based on a conference presentation (not a peer-reviewed publication). The data below come from the ESMO 2025 late-breaking abstract (LBA64), GRAIL's press releases, and the principal investigator's public comments. A full manuscript has not yet been published at the time of this book.

Study Design

PATHFINDER 2 was a prospective, multi-center, interventional study conducted at healthcare institutions across the US and Canada, spanning academic centers like Mayo Clinic, Duke, Oregon Health & Science University, and Weill Cornell, alongside community sites like Ochsner Health, Henry Ford, and Hoag Memorial. Between 2021 and the completion of enrollment, the investigators recruited 35,878 adults aged 50 or older with no clinical suspicion of cancer and no cancer diagnosis. Patients with a prior history of cancer were allowed if their treatment finished more than three years before enrollment.

The trial population was deliberately enriched for people 60-79 years old to increase the number of cancer events during the trial, and there were specific diversity targets for age, sex, and race/ethnicity. This was an important evolution from PATHFINDER, where the population was 91.7% White.

Each participant underwent a Galleri test in addition to whatever standard-of-care screening (per USPSTF A and B category recommendations for breast, cervical, colorectal, and lung) they were already eligible for. When Galleri returned a "cancer signal detected" result, participants underwent diagnostic evaluation guided by the predicted cancer signal origin. The primary objectives were safety and performance. A three-year follow-up is planned.

The ESMO presentation reported results from a pre-specified analysis of participants with at least 12 months of follow-up as of December 31, 2024.

1. Of the 35,878 enrolled participants, 23,161 were analyzable for test performance, and 25,114 were analyzable for safety. The remaining participants had not yet reached the 12-month follow-up mark at the data cutoff.

Results

In the analyzable cohort, Galleri detected a cancer signal in 216 of 23,161 participants (0.93%). This was consistent with the roughly 1% rate seen in PATHFINDER. Of those 216 positive results, 133 were confirmed to have cancer, yielding a PPV of 61.6% (95% CI 54.9--67.8%). That is a substantial jump from PATHFINDER’s 43.1% PPV (for the refined test version reported separately from the main cohort). The increase in PPV seen in PATHFINDER 2 doesn't seem to be related to an increase in prevalence, as the prevalence of cancers in PATHFINDER was 1.84% (122/6,621) compared to 1.42% in PATHFINDER 2 (329/23,161).

Specificity was 99.6%, translating to a false positive rate of 0.4%, which is similar to the specificity reported in PATHFINDER (99.5%).

The sensitivity was 40.4% (95% CI 35.3-45.8%) for all cancers and 73.7% (95% CI 65.6--80.4%) for a prespecified subgroup of "12 cancer types responsible for roughly two-thirds of cancer deaths in the US" (anus, bladder, colon/rectum, esophagus, head and neck, liver/bile duct, lung, lymphoma, ovary, pancreas, plasma cell neoplasm, and stomach). Interestingly, breast and prostate cancer, the most common cancer in women and men, respectively, are excluded from this list.

The first cancer signal origin prediction was accurate 91.7% of the time (95% CI 85.8-95.3%), up from 88% in the PATHFINDER analysis of the refined test. Diagnostic resolution took a median of 46 days (interquartile range 42--59 days), which was faster than PATHFINDER’s 79-day median.

Of the 133 cancers diagnosed in the study, 114 were new cancers, and 19 were recurrent cancers. Among the 114 new cancers, 53.5% were stage I-II and 69.3% were stage I-III, meaning the majority were found at stages where treatment intent is typically curative. Also, 75.2% of the cancers were types without common screening options.

Across all detection methods, 329 participants were diagnosed with cancer within 12 months. Of those, 200 had "screen-detected" cancers (133 by the MCED test, 20 by USPSTF A and B recommended screenings (breast, cervical, colorectal, and lung), and 47 by USPSTF C recommended screening (prostate)). The remaining 129 cancers were diagnosed through other means, such as symptoms and incidental findings.

What I Think

The good news:

1. The PPV improvement from 43.1% in PATHFINDER to 61.6% in PATHFINDER 2 changes the clinical calculus fundamentally. In PATHFINDER, a "cancer signal detected" result was not even better than a coin flip for whether or not someone had cancer. In PATHFINDER 2, the odds are now roughly 3:2 in favor of a true positive.
2. The sensitivity of 73.7% for the 12 pre-specified high-mortality cancers that don't have effective screening tests for the general population, such as pancreatic, liver, esophageal, and ovarian (among others) is worth highlighting. These are the cancers where finding anything represents a meaningful intervention compared to the status quo of waiting for symptoms.
3. The diagnostic workup data are encouraging. A median of 46 days from signal to diagnostic resolution, with only 0.6% of all participants requiring an invasive procedure, is an improvement compared to PATHFINDER. However, part of the improvement may be related to the diagnostic delay stemming from effects of the COVID era, when PATHFINDER enrolled patients.

That said, this dataset has meaningful limitations.

1. The full manuscript has not been published as of the time of this book. Conference presentations are optimized for impact, and the data that get highlighted in a late-breaking abstract at ESMO may not be as glossy as those published in a peer-reviewed manuscript. I have no reason to doubt the numbers, but without the complete dataset, including demographics, cancer type breakdown, and false negative characterization, it's hard to draw firm conclusions.
2. It can be assumed that breast and prostate cancers were excluded from the "12 cancer types responsible for roughly two-thirds of cancer deaths in the US" sensitivity figure of 73.7% because the performance of the test for breast and prostate cancers is poor relative to other types of cancers. On Galleri's website, they report the sensitivity of Galleri at 30.5% for breast cancer and 11.2% for prostate cancer, which seems to support this hypothesis. However, to me, this is only relevant for those making the case that MCED testing should replace all current cancer screening methods rather than serve as an adjunct (I'm in the adjunct camp).
3. An overall sensitivity of ~40% is not impressive. Additionally, as with CCGA and PATHFINDER, the test seems to be best at identifying patients with more advanced disease, which does not support the idea that the test is suited primarily to find early-stage cancers.

What Came After

In June 2026, GRAIL presented the registrational results from PATHFINDER 2 at the American Society of Clinical Oncology (ASCO) Annual Meeting (Late-Breaking Abstract LBA10509). These superseded the interim ESMO 2025 analysis described above, drawing on a larger analyzable cohort with longer follow-up.

The headline performance figures held up, with modest changes. There were 32,007 participants who were analyzable (up from 23,161 in the interim ESMO 2025 report). Of 287 participants with a “cancer signal detected” result (0.9%), 173 were diagnosed with cancer (151 new and 22 recurrent) within 12 months, for a PPV of 60.3% (the ESMO 2025 abstract reported 61.6%). Specificity was 99.6%, essentially unchanged. Sensitivity was 39.3% across all cancers (40.4% in the ESMO 2025 abstract) and 69.8% for the prespecified group of 12 cancers responsible for roughly two-thirds of US cancer deaths (73.7% in the ESMO 2025 abstract). Cancer signal origin accuracy was 91.3% (91.7% in the ESMO 2025 abstract). The median time from a positive result to diagnostic resolution was 48 days (46 days in the ESMO 2025 abstract).

Importantly, of the 151 new primary cancers the test found, 80 (53%) were stage I-II and 70.9% were stage I-III. Of the 80 test-detected cancers, 71% did not have a USPSTF A/B screening recommendation, which is (in my opinion) precisely the types of cancers MCED tests are meant to detect.

Overall, my read does not change much from the interim analysis. PATHFINDER 2 remains, in my opinion, an extremely clinically relevant performance dataset for Galleri, and the numbers from the ASCO 2026 abstract tell the same story as what was reported previously.