Pancreatic cystic neoplasms present a diagnostic challenge in determining which lesions are malignant or carry a high malignant potential. Cyst fluid carcinoembryonic antigen (CEA) is traditionally used to distinguish mucinous from non-mucinous pancreatic cysts, with limited value in identifying malignancy. However, in clinical practice, markedly elevated CEA levels are occasionally observed in patients with invasive cystic neoplasms. This study investigates the association between extremely high cyst fluid CEA concentrations and the presence of underlying malignancy in pancreatic cysts.

We retrospectively analyzed pancreatic cyst fluid CEA concentrations in patients with known cyst outcomes. A total of 177 patients who underwent EUS-FNA between 2022 and 2024 for the evaluation of pancreatic cysts were included.Malignancy was established by expert cytopathology. Non-malignant lesions included low-grade intraductal papillary mucinous neoplasms, pseudocysts, serous cystadenomas, and lymphoepithelial cysts. We compared CEA levels between non-malignant and malignant groups and constructed a receiver operating characteristic (ROC) curve using malignant cytology (Yes/No) as the reference standard. The area under the curve (AUC) was calculated (accounting for tied values in CEA measurements), and we examined sensitivity and specificity at a high CEA cutoff. Finally, a univariable logistic regression was performed with log-transformed CEA (ln[CEA]) as the predictor of malignancy, reporting the odds ratio (OR) and model performance metrics.

A total of 177 pancreatic cysts were included (122 non-malignant, 55 malignant by cytology). Cyst fluid CEA levels ranged widely in both groups. The median CEA was 18 ng/mL in non-malignant cysts versus 800 ng/mL in malignant cysts, although there was notable variance and overlap between groups (some malignant cysts exhibited only modest CEA elevations, and occasional benign lesions had moderately increased CEA). ROC analysis yielded an AUC of 0.873 for CEA in predicting malignant cytology. Using an exemplary high cutoff (CEA >1500 ng/mL), sensitivity was 78.8% and specificity 94.3% for identifying malignancy (based on cross-tabulation of the dichotomized CEA). In a logistic regression model, ln(CEA) was a significant independent predictor of malignancy (OR ≈ 1.97 per unit increase, p < 0.001). This single-predictor model achieved an overall diagnostic accuracy of 82.4%, with a Nagelkerke R² of 0.504, indicating that approximately half of the variance in cyst malignancy status was explained by fluid CEA alone.

Extremely elevated pancreatic cyst fluid CEA levels are associated with malignancy, yet the overlap with benign lesions limits CEA’s role as a standalone diagnostic marker. These findings support CEA as a valuable, though imperfect, tool in malignancy risk stratification. Clinicians may tailor interpretation of CEA thresholds based on the desired balance between sensitivity and specificity in the context of surgical decision-making. Our retrospective analysis suggests that markedly elevated CEA levels (in the thousands of ng/mL) are indicative of an invasive mucinous cyst and may aid in identifying high-risk cysts that warrant more aggressive management. Therefore, cyst fluid CEA results should be interpreted in conjunction with cytology, imaging, and clinical findings rather than in isolation.