Why This Study Matters
In the United States, only about 14% of all cancers are currently detected through guideline-recommended screening tests, and more than 70% of cancer deaths occur from cancers for which no guideline-recommended screening exists at all — a figure that rises above 80% once real-world adherence to existing screening is taken into account. Blood-based multi-cancer early detection (MCED) testing has emerged as a way to screen for many cancer types, including those that currently have no recommended screening pathway, from a single blood draw.
The MCED test evaluated here analyzes methylation patterns on cell-free DNA (cfDNA) fragments in peripheral blood to detect a shared cancer signal and predict a cancer signal origin (CSO) that guides where clinicians should look. PATHFINDER 2 is the largest prospective, interventional evaluation of this specific test to date, designed to characterize its real-world performance and safety in an intended-use population of adults 50 and older with no clinical suspicion of cancer.
Key Findings
Study Design
PATHFINDER 2 (NCT05155605) was a prospective, interventional, single-arm cohort study of the Galleri MCED test (GRAIL, Inc.), conducted at 32 clinical sites in the United States and Canada. Unlike a therapeutic trial, this was a diagnostic-performance study with no comparator arm and no randomization — every enrolled participant received the same blood-based MCED test, and outcomes were tracked against clinical cancer diagnoses over 12 months of follow-up.
The study had co-primary objectives: characterizing the performance of the MCED test (positive predictive value, negative predictive value, sensitivity, specificity, false-positive rate, likelihood ratios, cancer signal origin prediction accuracy, and number needed to screen) and characterizing the safety of the diagnostic workup triggered by a positive result (type and number of invasive procedures and adverse events). All endpoints were prespecified in the statistical analysis plan and analyzed descriptively; the study was not designed around formal hypothesis testing or a pass/fail success criterion.
| Element | Detail |
|---|---|
| Participants enrolled | 35,878 |
| Clinical sites | 32 (US & Canada) |
| Design | Single-arm (no randomization; not applicable to this diagnostic-performance study) |
| Intervention | 1 blood draw (no drug dosing; a single MCED blood test) |
This study evaluated a diagnostic blood test rather than a therapeutic agent, so conventional randomization and dosing statistics do not apply in the usual sense; the design fields above are adapted accordingly rather than left blank.
Patient Population
Enrollment ran from December 8, 2021 to July 10, 2024. Eligible participants were adults aged 50 years or older without clinical suspicion of cancer and without a cancer diagnosis or cancer treatment within the prior 3 years. Of 35,878 enrolled participants, 35,350 (98.5%) were analyzable. The performance analysis set comprised 32,007 participants with a completed 12-month cancer status assessment, and the safety analysis set comprised 35,335 participants who underwent one or more diagnostic evaluations after their MCED result.
Median age was 64.0 years (Q1–Q3: 58.0–70.0). The population was 56.2% female and 43.8% male; 80.1% self-reported as White, 8.8% as Black or African American, and 5.9% as Asian. Most participants (69.3%) were never-smokers, and 58.8% had a bachelor's degree or higher.
Primary Endpoint Results
Of 32,007 participants in the performance analysis set, 287 (0.9%) had a positive MCED test result. Of those, 173 were true positives (a cancer was subsequently diagnosed) and 114 were false positives. This yielded a cancer detection rate of 0.54% (95% CI 0.47–0.63%), a positive predictive value of 60.3% (95% CI 54.5–65.8%), and a negative predictive value of 99.2% (95% CI 99.1–99.3%). Specificity was 99.64% (95% CI 99.57–99.70%), with a corresponding false-positive rate of 0.36% (95% CI 0.30–0.43%).
Twelve-month episode sensitivity across all cancer types was 39.3% (95% CI 34.9–44.0%). The positive likelihood ratio was 108.9 (95% CI 87.6–135.2) and the negative likelihood ratio was 0.61 (95% CI 0.56–0.66). The number needed to screen to detect one cancer was 185 (95% CI 159–215). Cancer signal origin (CSO) prediction accuracy was 87.3% (95% CI 81.5–91.5%) for the top prediction alone, and 91.3% (95% CI 86.2–94.7%) when either the first or second predicted origin was counted as correct.
| Metric (Performance Analysis Set, N = 32,007) | Result (95% CI) |
|---|---|
| Cancer detection rate | 0.54% (0.47–0.63%) |
| Positive predictive value (PPV) | 60.3% (54.5–65.8%) |
| Negative predictive value (NPV) | 99.2% (99.1–99.3%) |
| Specificity | 99.64% (99.57–99.70%) |
| False-positive rate | 0.36% (0.30–0.43%) |
| 12-month episode sensitivity, all cancers | 39.3% (34.9–44.0%) |
| Positive likelihood ratio | 108.9 (87.6–135.2) |
| Negative likelihood ratio | 0.61 (0.56–0.66) |
| Number needed to screen | 185 (159–215) |
| CSO1 prediction accuracy | 87.3% (81.5–91.5%) |
| CSO1-or-CSO2 prediction accuracy | 91.3% (86.2–94.7%) |
Subgroup Analyses
| Subgroup | Episode Sensitivity (95% CI) |
|---|---|
| All cancer types | 39.3% (34.9–44.0%) |
| 12 cancers responsible for two-thirds of US cancer deaths | 69.8% (62.8–76.0%) |
| Six aggressive cancers with low 5-year survival | 66.2% (55.1–75.8%) |
Beyond these prespecified statistical subgroups, the source article also reports — without accompanying point estimates in the main text — that performance varied qualitatively by age, sex and smoking status: cancer detection rate increased with age and was higher in male participants and those with a smoking history, and the 80-years-and-older age group had lower PPV, specificity and positive likelihood ratio than other age groups. Exact figures for this age/sex/smoking comparison are not specified in the source and are not estimated here.
Safety Profile
Among 35,335 participants in the safety analysis set, 290 underwent diagnostic evaluation following a positive MCED result, and 213 of them (0.6% of the full safety-analysis set) had at least one invasive procedure, for a total of 295 invasive procedures. Most diagnostic evaluations (1,694 of 1,989, 85.2%) were noninvasive, and the invasive procedures that did occur were mainly nonsurgical (267 of 295, 90.5%). The median number of invasive procedures per participant with a positive test was 1.0 (Q1–Q3: 0.0–1.0).
During the diagnostic-evaluation period, 9 of 290 participants (3.1%) experienced a total of 10 adverse events. Five were not study-related (including one serious event, hyponatremia); the other five were study-related and occurred only in participants ultimately diagnosed with cancer, and none of these five was serious. Across all 35,850 consented participants with cleaned data, 56 (0.2%) experienced one or more study-related adverse events, none serious, most occurring in the phlebotomy setting.
| Measure | Definition | Result |
|---|---|---|
| A | Participants with a positive test and an invasive procedure / all safety-set participants | 0.6% (213/35,335) |
| B | Invasive procedures among all positive-test participants / all safety-set participants | 0.01 (295/35,335) |
| C | Participants with a positive test and an invasive procedure / participants with a positive test | 73.4% (213/290) |
| D | Invasive procedures / participants with a positive test | 1.02 (295/290) |
No comparative safety data by treatment arm applies here, since this was a single-arm diagnostic study rather than a randomized comparison of treatments.
Interpretation and Broader Context
The study authors frame the positive predictive value observed here as substantially higher than that reported for existing single-cancer screening tests, and argue that the combination of high specificity and CSO-guided workup limited unnecessary diagnostic procedures relative to what might be expected from a less targeted test.
PATHFINDER 2 provided evidence that the MCED test detected tumor-derived cfDNA in an intended-use population without clinical suspicion of cancer and demonstrated robust performance characteristics and a favorable safety profile.
— Discussion, PATHFINDER 2 publication, Nature Medicine (2026) (verbatim; "cfDNA" = cell-free DNA)
The authors are explicit that episode sensitivity in this single-arm design, where cancer status is unknown at testing, is not directly comparable to sensitivity estimates from case-control studies, where cancer status is already known. They also caution that a negative MCED result should not be interpreted as an absence of cancer, and that participants were instructed to continue with all guideline-recommended cancer screening regardless of their MCED result.
Future Directions
Participants are being followed for a total of 3 years to assess cancer status and screening utilization annually. The authors state plainly that randomized trials and longer follow-up will be needed to determine the clinical utility of MCED testing — that is, whether earlier detection through this test translates into improved outcomes such as reduced late-stage diagnoses or mortality, which this single-arm, 12-month analysis was not designed to establish.
Limitations
PATHFINDER 2 is a single-arm, non-randomized cohort study with no control arm, so it cannot determine whether use of the MCED test itself improves outcomes relative to guideline-recommended screening alone — only 12-month test performance and safety were assessed. The prespecified sensitivity subgroups aside, the reported differences in performance by age, sex and smoking status are qualitative, without point estimates in the main text, and should be read as descriptive rather than confirmatory. Episode sensitivity in this prospective single-arm design is explicitly stated by the authors to not be directly comparable to sensitivity figures from case-control studies. Median follow-up was limited to 12 months at this data cutoff, and the authors call for further research through the full 3-year follow-up and for randomized trials before MCED testing's effect on late-stage diagnosis or mortality can be established.
In a prospective, single-arm study of 35,878 adults with no clinical suspicion of cancer, a blood-based multi-cancer early detection test returned a positive predictive value of 60.3%, specificity of 99.64%, and 12-month episode sensitivity of 39.3% across all cancer types, with very low rates of invasive follow-up procedures and no serious study-related adverse events. Because PATHFINDER 2 has no comparator arm and only 12 months of follow-up, it establishes test performance and safety rather than clinical benefit; whether MCED screening reduces late-stage diagnoses or cancer mortality is the question the authors say randomized trials and longer follow-up are still needed to answer.