Interpretable extracellular vesicle long RNA framework for noninvasive pancreatic cancer diagnosis: A multi-omics artificial intelligence-driven liquid biopsy paradigm

Table 1 Comparative diagnostic performance of biomarker strategies for pancreatic ductal adenocarcinoma

Biomarker/method	AUC	Sensitivity	Specificity	PPV (30% prevalence)	Key limitations
CA19-9 alone	0.70-0.85[4]	70%-80%	70%-90%	50%-79%	Low early-stage sensitivity; false positives in pancreatitis
EV-miRNA panels	0.85-0.92[42]	75%-85%	80%-90%	64%-79%	Limited specificity; cancer-type overlap
ctDNA mutation	0.65-0.80[12]	50%-70%	85%-95%	59%-84%	Low sensitivity in early disease; requires high-depth sequencing
ECD-itMLF (EV-lRNA)	0.9698[22]	About 90%	About 95%	About 89%	Requires validation in jaundice/diabetes; method standardization needed
Multi-omics integration	0.90-0.95[47]	80%-90%	85%-95%	70%-84%	Complexity; cost; technical standardization

AUC: Area under the curve; PPV: Positive predictive value; EV: Extracellular vesicle; ctDNA: Circulating tumor DNA; miRNA: MicroRNAs; CA19-9: Carbohydrate antigen 19-9; EV-lRNA: Extracellular vesicle long RNA.