Comparative diagnostic accuracy of twenty blood-based biomarkers for pancreatic cancer: a network meta-analysis

Pancreatic cancer (PC) is a highly lethal malignancy with a poor prognosis, largely due to late-stage diagnosis. Although CA19-9 is the most widely used serum biomarker, its suboptimal sensitivity and specificity limit its clinical utility. Hundreds of biomarkers have been proposed, but no network meta-analysis has systematically compared their diagnostic performance. We systematically searched PubMed, Embase, Web of Science, and Cochrane Library for studies evaluating blood-based biomarkers for PC diagnosis. We selected the 20 most commonly reported biomarkers and extracted sensitivity, specificity, and other diagnostic data from 185 eligible studies. A Bayesian ANOVA model was employed to pool and compare sensitivity and specificity, and to calculate the dominant index. The Frequentist approach was used to calculate SUCRA for sensitivity and specificity and to validate the stability of the rankings. Subgroup and sensitivity analyses were also conducted. Among the 20 biomarkers, the Bayesian ANOVA model identified IMP3, S100P, and CA195 as the top three based on the dominant index. S100P had the highest sensitivity ranking, while p53 had the highest specificity ranking. In the Frequentist analysis, based on SUCRA for sensitivity, the top three biomarkers were S100P, Span-1, and MIC-1; based on SUCRA for specificity, the top three were p53, IMP3, and CA72-4. Subgroup analyses across different study designs, control groups, and geographic regions yielded results consistent with the overall analysis. IMP3 and S100P demonstrate high diagnostic accuracy and are promising blood-based biomarkers for PC. Given the limitations of CA19-9, these markers warrant further investigation in prospective studies to confirm their clinical utility.