This Blood Test Caught 87% of Early Pancreatic Cancers

Subtitle: A City of Hope–led Nature Medicine study of nearly 1,800 people validates PANXEON — an investigational liquid biopsy that fuses circulating and exosomal microRNAs with CA19-9 into one AI risk score for stage I–II pancreatic ductal adenocarcinoma.

Pancreatic ductal adenocarcinoma (PDAC) is the malignancy clinicians dread for a brutally simple reason: by the time most people feel sick, the window for cure has already narrowed. Roughly 90% of tumors are found only after the cancer has advanced. Five-year survival sits near 13–14% overall and drops to about 3% once disease is distant. Localized disease still only reaches about 44% five-year survival — better, but nowhere near the stage-shift miracles of breast or prostate screening.

What medicine has lacked is a clinically relevant blood test for early-stage PDAC. CA19-9, the familiar protein marker, helps manage known disease but falters as a standalone early screen. Circulating tumor DNA shines after diagnosis; it has not solved asymptomatic detection.

On September 16, 2026, Nature Medicine published a large, international answer to that gap. Led by Ajay Goel at City of Hope, with first author Caiming Xu and a multinational clinical consortium, the paper introduces PANXEON — PANcreatic cancer eXosome Early detectiON (DOI 10.1038/s41591-026-04625-x; ClinicalTrials.gov NCT06388967). City of Hope’s public release, Healio coverage the same day, and Medical News Today follow-up on September 17 put the shareable claim in one line: a blood draw that can flag early pancreatic cancer — and, in some cases, a precancerous state — with numbers strong enough to matter.

Three signals, one score

PANXEON does not bet on a single biomarker. It measures three biological channels in the same plasma sample:

Circulating microRNAs floating freely in blood
Exosomal microRNAs packaged inside tiny extracellular vesicles shed by cells
CA19-9, the protein already used in pancreatic oncology

A machine-learning model then collapses those measurements into a composite risk score. Outreach framing calls it the first investigational assay to combine all three classes in one test. The peer-reviewed microRNA panel is a multi-miRNA signature — reported as a 13-miRNA set spanning free and vesicle-borne species — validated against other gastrointestinal cancers and found to show minimal cross-reactivity.

That design choice is the scientific punchline. Most prior blood tests told one chapter of the story. PANXEON tries to read the chapter where the chapters overlap.

The numbers that traveled

The study is an international, multicenter, observational, prospective biomarker analysis involving people with and without PDAC across the United States, Europe, and Asia. Nature Medicine reports 1,785 individuals in the abstract; Healio’s clinical write-up cites 1,757 plasma samples from 1,649 unique participants — a cohort that deliberately includes not only healthy-ish “low-risk” controls but also the messy real-world groups that actually get monitored: familial/inherited risk, pancreatic cysts, and chronic pancreatitis.

Key testing-cohort results:

MicroRNA signature alone: AUROC 88.6%; 83.8% sensitivity for early-stage PDAC
MicroRNA + CA19-9 (PANXEON composite): 86.8% sensitivity for stage I–II PDAC
False-positive rates: 3.2% in low-risk controls; 15.6% in high-risk controls
High-grade dysplasia in high-risk pancreatic cysts (“stage 0” territory): 64.3% detection

In a small longitudinal slice of 19 patients, microRNA signature levels fell during neoadjuvant chemotherapy and after surgery, then rose before recurrence — a hint that the same biology might one day track treatment response, not only first detection.

City of Hope rounded the stage I–II hit rate to “87%” in public messaging. That is the number that will travel on Sunday night. The peer-reviewed figure behind it is 86.8%.

What this is — and what it is not

It is not an approved population screening test. It is not a replacement for CT, MRI, or endoscopic ultrasound. Goel’s own framing is careful: PANXEON could help triage who among high-risk people needs deeper evaluation, not swap out imaging.

Independent perspective matters here. In Healio’s same-day commentary, Cleveland Clinic oncologist Suneel Kamath called the results exciting — and then listed the hard constraints. For a rare disease (~70,000 U.S. cases per year), even high sensitivity can yield a disappointing positive predictive value. He argued clinical utility may need mid-90s sensitivity, flagged possible overfitting (training ROC curves near 97–98%), noted the counterintuitive pattern that false positives were higher in high-risk controls, and warned that cost could run several times a CT scan. Localized PDAC’s 44% five-year survival also means early detection alone will not “cure” the disease without better therapy. Honest Sunday science keeps that entire paragraph next to the headline.

What it is: the strongest recent peer-reviewed demonstration that a multi-analyte liquid biopsy — free miRNAs + exosomal miRNAs + CA19-9, fused by AI — can detect stage I–II PDAC at clinically interesting sensitivity while still being stress-tested in the very populations that would use such a test. Primary paper: Xu, Goel, et al., Nature Medicine (2026). Public anchors: City of Hope / EurekAlert (September 16), Healio (September 16), Medical News Today (September 17).

Who might benefit first — and why cysts matter

The first users are unlikely to be everyone at an annual physical. They are people already living under a pancreatic shadow: inherited risk, family history, cysts, chronic pancreatitis. Today’s surveillance is often invasive, expensive, and only partly helpful at catching cancer early. A blood score that enriches for who needs the next scan could shrink that burden.

The 64.3% sensitivity for high-grade dysplasia inside high-risk cysts is the sleeper statistic. Catching advanced precancer before invasion is the stage shift before the stage shift — a chance to intervene while the biology is still “almost.” That number is lower than the early-cancer figure, and Kamath’s caveat about needing further optimization applies doubly here. It is still a signal that PANXEON’s biology is not blind to the pre-malignant corridor.

Disclosure for editors: Goel is a paid consultant to Biocartis, which holds an option on PANXEON rights. Conflict statements belong in the same breath as the hope.

Why tonight’s shareable fact sticks

Pancreatic cancer kills by arriving late. PANXEON’s claim is not magic. It is multi-signal blood biology plus machine learning, validated across continents, published in Nature Medicine, and already being argued over in the oncology press on the right questions: sensitivity versus PPV, training versus testing, cost versus CT, detection versus cure.

For tonight’s volume, keep the sentence tight. Nearly nine in ten early-stage tumors flagged in the testing cohort. False positives low in low-risk blood — higher where risk is real. A precancerous cyst state spotted more than six times in ten. Not a clinic-ready screen yet. Not Virtual Biotech’s AI drug design. A different fight on the same disease: find it while surgery and new therapies still have a chance.

Sources

Xu, C., Mannucci, A., Han, H., et al., & Goel, A. (2026). Liquid biopsy for early detection of pancreatic ductal adenocarcinoma. Nature Medicine. https://doi.org/10.1038/s41591-026-04625-x
City of Hope / EurekAlert! (2026, September 16). New study: Blood test shows promise for detecting pancreatic cancer at its earliest stages. https://www.eurekalert.org/news-releases/1143614
Leiser, M. / Healio. (2026, September 16). Blood test may detect pancreatic cancer earlier, allow for ‘meaningful intervention’. https://www.healio.com/news/hematology-oncology/20260916/blood-test-may-detect-pancreatic-cancer-earlier-allow-for-meaningful-intervention
Morales-Brown, P. / Medical News Today. (2026, September 17). Pancreatic cancer: New blood test may lead to earlier detection. https://www.medicalnewstoday.com/articles/ai-enhanced-biomarker-blood-test-may-detect-pancreatic-cancer-earlier
ClinicalTrials.gov. Pancreatic Cancer Detection Consortium (NCT06388967). https://clinicaltrials.gov/study/NCT06388967