Comparative diagnostic performance of endoscopic ultrasound, MRI, and CT for preoperative assessment of pancreatic cancer: a Bayesian network meta-analysis under a graded reference standard.

objectiveWithin the framework of a graded reference standard, this study systematically compared the diagnostic performance and certainty of evidence of multiple preoperative imaging modalities—including endoscopic ultrasound, magnetic resonance imaging, and computed tomography—for pancreatic cancer. Based on these findings, a relative ranking of diagnostic effectiveness was established to provide evidence-based guidance for optimizing clinical diagnostic pathways.

methodsA systematic search was performed in PubMed, Embase, Web of Science, and the Cochrane Library from January 2000 to July 12, 2025, to identify prospective diagnostic studies. Data were extracted from 2 × 2 contingency tables (true positive, false positive, false negative, true negative). A Bayesian bivariate random-effects model was applied to conduct the Bayesian network meta-analysis, estimating pooled sensitivity, specificity, diagnostic odds ratio, and Youden’s index with corresponding 95% credible intervals. Diagnostic performance ranking was assessed using the surface under the cumulative ranking curve and cumulative ranking probabilities. Consistency between direct and indirect evidence was evaluated through node-splitting analysis. Transitivity and robustness were examined by relative sensitivity and relative specificity models, leave-one-out sensitivity analysis, summary receiver operating characteristic curves, and regression tests for publication bias. Finally, the certainty of evidence for key comparisons was graded and summarized using the GRADE framework.

resultsA total of 10 prospective diagnostic studies involving 623 patients and 8 imaging modalities were included: magnetic resonance cholangiopancreatography, contrast-enhanced endoscopic ultrasound, endoscopic ultrasound-guided biopsy, multidetector contrast-enhanced computed tomography pancreatic protocol, endoscopic retrograde cholangiopancreatography with intraductal ultrasound, transabdominal ultrasound, manganese dipyridoxyl diphosphate–enhanced magnetic resonance imaging, and fluorodeoxyglucose positron emission tomography combined with multidetector computed tomography dynamic perfusion contrast-enhancement. (1) Ranking of performance: Contrast-enhanced endoscopic ultrasound ranked highest, followed by magnetic resonance cholangiopancreatography and endoscopic ultrasound-guided biopsy; multidetector contrast-enhanced computed tomography pancreatic protocol and transabdominal ultrasound were of intermediate level; endoscopic retrograde cholangiopancreatography with intraductal ultrasound and fluorodeoxyglucose positron emission tomography combined with multidetector computed tomography dynamic perfusion contrast-enhancement ranked lower; manganese dipyridoxyl diphosphate–enhanced magnetic resonance imaging was supported by limited evidence. (2) Key indicators: Contrast-enhanced endoscopic ultrasound showed the best specificity and diagnostic odds ratio; magnetic resonance cholangiopancreatography and endoscopic ultrasound-guided biopsy demonstrated higher sensitivity; Youden’s index was consistent with these trends. (3) Consistency and transitivity: Most comparisons satisfied the consistency assumption, with marginal inconsistency observed in a few contrasts; overall study design and reference standards supported transitivity. (4) Robustness and bias: Relative sensitivity and relative specificity models, leave-one-out sensitivity analysis, and summary receiver operating characteristic curves all indicated stable results; The funnel plot and Egger’s regression test did not indicate significant publication bias. (5) Quality of evidence: The certainty of evidence for most comparisons was moderate to high; downgrading was mainly due to limited sample size and local inconsistency; no very low-certainty evidence was identified.

conclusionThe results demonstrated a clear hierarchy of diagnostic performance among imaging modalities in the preoperative evaluation of pancreatic cancer. Contrast-enhanced endoscopic ultrasound and magnetic resonance cholangiopancreatography showed the highest overall diagnostic effectiveness, while multidetector contrast-enhanced computed tomography pancreatic protocol remains the cornerstone for staging and vascular assessment. In clinical practice, standardized multidetector contrast-enhanced computed tomography pancreatic protocol can serve as the foundation, with stepwise integration of magnetic resonance cholangiopancreatography and endoscopic ultrasound (including contrast-enhanced) tailored to specific clinical contexts. Such multimodal integration enables cross-validation across geometric features, microvascular perfusion, and stromal–cellular architecture, thereby enhancing diagnostic precision and interpretability. Future studies should focus on multicenter, prospective, head-to-head trials, promoting international harmonization of imaging parameters and interpretation criteria, and further exploring the integration of multimodal imaging with artificial intelligence as well as cost-effectiveness evaluation, in order to optimize diagnostic pathways and improve patient outcomes.