PancreaSure and PanMETAI Show High Accuracy for Pancreatic Cancer Detection
PancreaSure detected advanced PDAC with 87.9% sensitivity and 97.7% specificity, outperforming CA19-9 alone. PanMETAI, an AI metabolomics model, achieved AUC 0.99 in training and 0.93 in external validation for early PDAC detection.
Two blood-based tests for pancreatic ductal adenocarcinoma (PDAC) demonstrated strong diagnostic performance in new studies: the PancreaSure five-biomarker panel reliably detected advanced disease, and the PanMETAI algorithm achieved high accuracy for early-stage detection. PDAC accounts for approximately 90% of pancreatic cancers and has a 13% five-year survival rate, largely because most patients are diagnosed only after the disease has progressed. The standard serological marker CA19-9 has limited utility as a standalone diagnostic marker because it is elevated in various cancers and nonmalignant conditions.
The PancreaSure assay is a five-biomarker serum signature based on TIMP1, ICAM1, CTSD, THBS1, and CA 19-9. It was originally developed to detect stage I and II PDAC in high-risk surveillance populations, including individuals with pathogenic germline variants, family history of PDAC, and/or mucinous pancreatic cysts. In a retrospective, blinded study of 619 serum samples—224 treatment-naïve patients with stage III or IV PDAC and 395 non–high-risk controls—PancreaSure distinguished advanced PDAC from non–high-risk controls with 87.9% sensitivity and 97.7% specificity. Sensitivity was 86.7% in stage III disease and 89.1% in stage IV disease. The assay showed higher sensitivity than CA 19-9 alone in stage III and IV PDAC (87.9% vs 77.2%), with similar specificity. Compared with earlier validation cohorts, PancreaSure detected stage III and IV disease with significantly higher sensitivity than stage I and II disease: 87.9% versus 77.3%. Across stages I to IV combined, sensitivity was 80.6%, though the confidence interval was wide because data from different cohorts were pooled. Specificity was 97.7% in non–high-risk controls and 92.2% in high-risk controls. The study was published in JCO Oncology Advances on June 9, 2026. PancreaSure is not intended to replace imaging or biopsy, and it is not recommended for patients who already have imaging evidence of advanced pancreatic cancer, but it may be administered to symptomatic patients, with the caveat that results alone should not establish a diagnosis.
In a separate study, researchers developed PanMETAI, a high-performance tabular foundation model for PDAC detection using ¹H NMR-based metabolomics and machine learning. The algorithm integrates serum metabolomic profiles—including small-molecule metabolites and lipoproteins—with clinical and biochemical parameters (age, CA19-9) and Activin A. The framework was derived from 902 participants, including 424 high-risk controls and 478 PDAC cases. In a Taiwanese training and validation cohort, PanMETAI achieved an AUC of 0.99 (95% CI 0.98–0.99). In a Lithuanian external validation cohort of 322 participants, the model yielded an AUC of 0.93 (95% CI 0.90–0.95). The model identified key signature patterns that improve early-stage (I/II) PDAC diagnosis and performed well with small sample sizes (n=50). The study's findings indicate the approach offers a rapid, accurate, and non-invasive tool for early PDAC detection.
The PancreaSure study had limitations, including its retrospective design and the limited availability of complete demographic and clinical data, particularly for controls, whose clinical information was limited beyond negative HIV, HBsAg, and HCV status.