Evaluating the prognostic efficacy of biomarkers in pancreatic cyst fluid

Abstract

Evaluating the prognostic significance of biomarkers in pancreatic cyst fluid, accessed through endoscopic ultrasound-guided fine-needle aspiration, is essential for improving the clinical management of pancreatic cysts. This review synthesizes the evidence from studies published on the field in the last years, focusing on the accuracy and clinical utility of biomarkers such as carcinoembryonic antigen, intracystic glucose, and novel genetic markers including DNA mutation analysis. Our findings indicate that elevated carcinoembryonic antigen levels and decreased intra-cystic glucose levels are strongly associated with mucinous cysts which carry a higher malignancy risk, while DNA mutation analysis has shown increased predictive accuracy for identifying malignant transformations. Integrating these biomarkers with imaging techniques enhances risk stratification and can significantly influence therapeutic decisions. The review highlights the need for standardization of biomarker assays and further validation of biomarker panels to refine their prognostic value in clinical settings, ultimately aiding in the tailored management of patients with pancreatic cysts.

Key Words: Pancreatic cysts; Cyst fluid biomarkers; Endoscopic ultrasound; Prognostic evaluation; Moucinous cysts

Core Tip: The analysis of pancreatic cyst fluid obtained via endoscopic ultrasound-guided fine-needle aspiration provides critical diagnostic and prognostic information. Conventional biomarkers, such as carcinoembryonic antigen and intra-cystic glucose, which can be offered as a point-of-care assay, remain valuable tools for distinguishing mucinous from non-mucinous cysts. However, neither reliably differentiates malignant from benign mucinous cysts. While emerging genetic markers represent a promising advancement, fine-needle biopsy remains the most reliable method for accurately characterizing cysts with malignant potential Continued progresses in molecular diagnostics and integrated clinical models are essential to optimize the management of pancreatic cystic lesions.

INTRODUCTION

Pancreatic cystic lesions, once considered rare, are now increasingly detected due to the widespread use of advanced cross-sectional imaging techniques. Up to 20% of individuals - particularly among the elderly population - are incidentally found to harbor pancreatic cysts[1,2]. While the majority of these lesions follow an indolent course, a subset carries malignant potential, underscoring the critical need for accurate risk stratification[3]. The primary clinical challenge upon diagnosis is to distinguish cysts that require surveillance or preemptive intervention from those that do not, with the goal of preventing progression to pancreatic ductal adenocarcinoma - a malignancy associated with poor survival outcomes, even in surgically treated patients[4].

Benign lesions, such as pseudocysts (PC) and serous cystadenomas (SCA), account for approximately one-quarter of incidentally discovered cysts. In contrast, mucinous cysts (MC) - including intraductal papillary mucinous neoplasms (IPMNs) and mucinous cystic neoplasms (MCNs) - constitute nearly half of all cases and are linked with a significantly increased risk of malignant transformation[5]. Less commonly, malignant cystic lesions such as solid pseudopapillary neoplasms and cystic pancreatic neuroendocrine tumors may also be encountered[6,7].

Accurate identification of MC is essential, as they are the primary recognized precursors to pancreatic cancer. Although retrospective studies have suggested that up to 15% of pancreatic cysts may harbor malignancy, data from the Surveillance, Epidemiology, and End Results database indicate a lower overall malignant transformation rate of 0.24%, based on the broad assumption that all pancreatic cancers arise from cystic precursors[8]. Nonetheless, both IPMNs and MCNs are well-established as precancerous lesions[9]. For IPMNs, the risk of malignant transformation ranges from 22% for branch-duct lesions to 68% for main-duct lesions. The 10-year cumulative incidence of high-grade dysplasia (HGD) or carcinoma is estimated at 7.77% for low-risk and 24.68% for high-risk lesions[8,10-12]. Although MCNs have a comparatively lower risk - approximately 15% - this remains clinically significant and necessitates vigilant surveillance and, when appropriate, surgical resection[13].

Imaging remains the cornerstone of pancreatic cyst diagnosis and can provide vital, non-invasive information about the malignant potential of MC[14]. Specific imaging features - particularly when using contrast agents - such as cyst size greater than 3 cm in IPMNs or 4 cm in MCNs, abrupt changes in size, pancreatic duct dilation, enhancing mural nodules > 0.5 mm, and thickened or enhancing cyst walls - have been correlated with high-risk MC. These features are especially relevant once a mucinous diagnosis has already been established[15]. A meta-analysis has highlighted that magnetic resonance imaging/magnetic resonance cholangiopancreatography, particularly diffusion-weighted imaging, can more accurately differentiate between high- and low-risk IPMNs (LR-IPMNs) for malignant transformation[16]. Nonetheless, imaging modalities typically display suboptimal accuracy in the initial characterization of cystic lesions[17-20].

As emphasized in recent European and American guidelines, traditional diagnostic modalities such as imaging and cytology have demonstrated limited sensitivity and specificity for accurately classifying pancreatic cysts[4,5]. Consequently, adjunctive techniques such as cyst fluid analysis have gained increasing importance. Endoscopic ultrasound-guided (EUS) fine-needle aspiration (FNA) allows for the collection of cystic fluid for both biochemical and molecular evaluation, providing deeper insights into cyst pathology. Conventional biomarkers - such as cyst fluid carcinoembryonic antigen (CEA), intracystic glucose, and amylase - alongside serum carbohydrate antigen 19-9, have been widely studied, though with inconsistent diagnostic reliability.

Recent advances in molecular profiling have identified key genetic alterations associated with cystic neoplasms. Mutations in oncogenes such as KRAS and GNAS, as well as in tumor suppressor genes including TP53, SMAD4, CDKN2A, and PIK3CA, offer promising avenues for improving diagnostic accuracy and risk stratification. Integrating molecular biomarkers into clinical decision-making could revolutionize the management of pancreatic cystic lesions by enabling more targeted surveillance strategies and earlier therapeutic intervention.

This article aims to critically assess the prognostic utility of cyst fluid biomarkers -particularly emerging genetic markers identified via EUS-FNA - and to explore their evolving role in managing pancreatic cystic neoplasms.

CEA

Cystic CEA is the first cystic biomarker widely utilized for the differentiation of mucinous from non-mucinous pancreatic cysts. It continues to be the cornerstone of cystic fluid analysis in the differential diagnosis of pancreatic cysts. However, despite its utility, cyst CEA levels are unable to reliably distinguish malignant from non-malignant MC, as demonstrated by older studies and meta-analyses[21,22]. Furthermore, up to 30% of IPMNs exhibit low CEA levels, and it cannot differentiate IPMNs from MCNs[23].

Various studies have proposed different CEA cut-off values in an effort to differentiate mucinous from non-MC. Generally, higher CEA cut-offs increase specificity, at the cost of reduced sensitivity, while lower cut-offs improve sensitivity, but sacrifice specificity. A landmark pooled analysis by van der Waaij et al[24], published two decades ago, concluded that a CEA cut-off of 800 ng/mL achieved a specificity of 98% for mucinous cyst diagnosis, though with a sensitivity of only 48%. Conversely, a cut-off of < 5 ng/mL could identify SCA or pancreatic PC with a sensitivity of 50%, specificity of 95%, positive predictive value of 94%, negative predictive value of 55%, and overall accuracy of 67%.

A meta-analysis by Thosani et al[25] reported a pooled sensitivity of 63%[95% confidence interval (CI): 59%-67%] and a specificity of 88% (95%CI: 83%-91%) for cystic CEA, although it did not provide data on specific CEA cut-offs. However, it did report a positive likelihood ratio of 4.37 for a CEA level > 192 ng/mL. A similar specificity (93%) and sensitivity (54%) were also observed in another large meta-analysis[26].

The most widely accepted CEA cut-off for differentiating mucinous from non-MC is > 192 ng/mL, which has been adopted by several guidelines for the surveillance and management of pancreatic cysts[4,5,23]. This cut-off was initially proposed by Brugge et al[27], who demonstrated a sensitivity of 75%, specificity of 84%, and an accuracy of 79% for distinguishing between mucinous and non-mucinous cystic lesions. Similarly, a study by Park et al[28] found that a CEA cut-off of > 200 ng/mL, a value very close to 192 ng/mL, resulted in a sensitivity of 60%, specificity of 93%, and diagnostic accuracy of 72%.

However, a decade after Brugge’s original study, data questioned the diagnostic accuracy of the > 192 ng/mL cut-off. The same study showed that a lower cut-off of 105 ng/mL performed better, with a sensitivity of 70% vs 61% and specificity of 77% vs 63%, respectively. This cut-off also yielded an area under the receiver operating characteristic curve of 0.77 (95%CI: 0.71-0.84; P < 0.01), while the 192 ng/mL cut-off misdiagnosed 39% of MC[29]. Furthermore, a study published the same year suggested that CEA levels in cyst fluid may fluctuate in up to 20% of patients during serial measurements, without corresponding changes in imaging findings. This raises concerns about the clinical relevance of isolated CEA level increases, particularly in the context of stable EUS examinations[30].

A more recent study has proposed a CEA cut-off of > 100 ng/mL, which showed a 100% negative predictive value (NPV) in differentiating LR-IPMNs from low-risk MCNs and high-risk IPMNs. The area under the receiver operating characteristic curve (AUROC) for differentiating LR-IPMNs from low-risk MCNs was 0.930 (95%CI: 0.5-0.8; P < 0.001), and for differentiating LR-IPMNs from high-risk IPMNs, it was 0.921 (95%CI: 0.823-1.000; P < 0.001)[31]. Another recent study reported suboptimal sensitivity (43.5%) using a > 192 ng/mL cut-off, although with an increased specificity (98.3%) and an accuracy of 63.3%. This study proposed an optimal CEA cut-off of > 2.5 ng/mL for differentiating mucinous pancreatic cystic neoplasms, with a sensitivity of 95.4%, specificity of 82%, accuracy of 90.5%, and an excellent AUROC of 0.94[32]. Additionally, a large study by Kwan et al[33], which included a substantial number of patients, found that the 192 ng/mL threshold led to suboptimal sensitivity (56%) and specificity (78%). The study recommended adopting a higher CEA threshold of 250 ng/mL to achieve a specificity of 85% in differentiating mucinous from non-MC.

In conclusion, although CEA remains a useful biomarker for differentiating mucinous from non-mucinous pancreatic cysts, as supported by the latest Kyoto guidelines[23], its interpretation must be approached with caution. While extreme values (either high or low) can provide meaningful diagnostic direction, values in between should be evaluated on a case-by-case basis, with clinicians avoiding sole reliance on them (Table 1).

Table 1 Characteristics of studies evaluating the diagnostic efficacy of cystic carcinoembryonic antigen and intracystic glycose for mucinous pancreatic cysts.

Ref.	Type of study	Biomarker assessed	Patients included	Cut-off	Sensitivity	Secificity	AUROC
van der Waaij et al[24], 2005	Pooled analysis	CEA	450	< 5 ng/mL	50% (for PC, SCA)	95% (for PC, SCA)	-
				> 800 ng/mL	48%	98%
Thosani et al[25], 2010	Meta-analysis	CEA	376	-	63% (pooled)	88% (pooled)	0.89
Thornton et al[26], 2013	Meta-analysis	CEA	1438	-	63% (pooled)	88% (pooled)	-
Brugge et al[27], 2004	Multicenter study -prospective collection of data	CEA	341	192 ng/mL	73%	84%	0.79
Park et al[28], 2011	Single-center retrospective	CEA	124	200 ng/mL	60%	93%	0.89
Gaddam et al[29], 2015	Multicenter study, retrospective study	CEA	226	105 ng/mL	70%	63%	0.77
				192 ng/mL	61%	77%
Köker et al[31], 2021	Retrospective study	CEA	466	100ng/mL	-	-	0.930 (for differentiating LR-IPMNs from LR-MCNs)
							0.921 (for differentiating LR-IPMNs from HR-IPMNs)
kwan et al[33], 2024	Single center, retrospective	CEA	1169	20 ng/mL	89%	64%	0.80
				192 ng/mL	56%	78%
Rossi et al[32], 2024	Prospective observational multicenter study	CEA	50	192 ng/mL	55.6%	87.5%	0.65
		Intracystic glucose (glucometer)		50 mg/dL	93.2%	76.5%	0.74
Carr et al[36], 2018	Single center study -prospective collection of data	CEA	153	192 ng/mL	58%	96%	0.92
		Intracystic glucose (glucometer)		50 mg/dL	92%	87%	0.91
Gyimesi et al[38], 2024	Single center study -prospective collection of data	CEA	91	192 ng/mL	67.5%	97.5%	-
		Intracystic glucose (glucometer)		50 mg/dL	94.2%	81.3%	-
Gorris et al[39], 2023	Single center study -prospective collection of data	CEA	63	20 ng/mL, 192 ng/mL	80%, 50%	62%, 93%	-
		Intracystic glucose glucometer		50 mg/dL	100%	45%
		Laboratory		50 mg/dL	100%	60%
Smith et al[41], 2022	Multicenter - prospectively maintained database	CEA	93	192 ng/mL	62.7%	88.2%	0.81
		Intracystic glucose (laboratory)		25 mg/dL	88.1%	91.2%	0.96
Ribeiro et al[42], 2024	Single center, retrospective study	CEA	78	192 ng/mL	55.6%	87.5%	-
		Intracystic glucose		50 mg/dL	93.2%	76.5%	-
		Glucometer laboratory		-	-	-	0.870, 0.912
Williet et al[43], 2023	Multicenter study, retrospective	CEA	121	192 ng/mL	41.7%	96.9%	0.812
		Intracystic glucose (laboratory)		41.8 mg/dL	95.3%	91.2%	0.936
McCarty et al[44], 2021	Meta-analysis	CEA	609	-	56% (pooled)	96% (pooled)	-
		Intracystic glucose			91% (pooled)	86% (pooled)
Faias et al[45], 2021	Meta-analysis	CEA	5286	-	67% (pooled)	80% (pooled)	0.79
		Intracystic glucose	460		91% (pooled)	75% (pooled)	0.95
Zikos et al[35], 2015	Single center study -prospective collection of data	CEA	67	192 ng/mL	77%	83%	-
		Intracystic glucose	-	50 mg/dL	-	-	-
		Laboratory	-	-	95%	57%	-
		Glucometer	-	-	88%	78%	-
Faias et al[37], 2020	Single center study -prospective collection of data	CEA	82	192 ng/Ml	72%	96%	0.842
		Intracystic glucose (glucometer)		50 mg/dL	89%	86%	0.860
Simons-Linares et al[40], 2020	Single center study -prospective collection of data	CEA	113	192 ng/mL	50%	92%	0.78
		Intracystic glucose		41 mg/dL	88%	97%	0.95
		Laboratory		21 mg/dL	92%	92%	-
Guzmán-Calderón et al[46], 2022	Meta-analysis	CEA	506	-	61% (pooled)	93% (pooled)	0.861 (pooled SROC)
		Intracystic glucose			91% (pooled)	85% (pooled)	0.959 (pooled SROC)
Mohan et al[47], 2022	Meta-analysis	Intracystic glucose	566	50mg/dL	90.1%	85.3%	-
		All methods of measurement		-	90.5% (pooled)	88% (pooled)
		Glucometer		-	89.5%	83.9%

AUROC: Area under the receiver operating characteristic curve; CEA: Carcinoembryonic antigen; PC: Pseudocysts; SCA: Serous cystadenoma; LR: Low risk; HR: High risk; IPMNs: Intraductal papillary mucinous neoplasms; MCNs: Mucinous cystic neoplasms; SROC: Summary receiver operating characteristic.

CYSTIC GLUCOSE

The suboptimal diagnostic accuracy of CEA in distinguishing mucinous from non-mucinous pancreatic cysts has prompted researchers to explore alternative biomarkers with improved discriminative capabilities. In 2013, Park et al[34] identified low intracystic glucose levels as a potential diagnostic marker for mucinous pancreatic cysts.

Multiple studies have evaluated the diagnostic performance of intracystic glucose levels and proposed specific cut-off values. Three reports using a threshold of < 50 mg/dL reported sensitivities and specificities ranging from 89% to 95% and 78% to 87%, respectively[35-37]. In the study by Carr et al[36], low intracystic glucose levels demonstrated an AUROC comparable to that of cyst fluid CEA (0.91 vs 0.92). Using a cut-off of > 192 ng/dL, CEA showed a sensitivity of 58%, specificity of 96%, and overall diagnostic accuracy of 69%, whereas intracystic glucose achieved a diagnostic accuracy of 90%. Furthermore, two more recent, small, single-center studies validated these findings, concluding that intracystic glucose demonstrated higher sensitivity and NPV compared to CEA, although with slightly lower specificity[38,39].

More recent studies have proposed various glucose cut-off levels, though all remain close to the 50 mg/dL mark. In a 2020 study by Simons-Linares et al[40], an intracystic glucose level ≤ 41 mg/dL yielded both sensitivity and specificity of 92%, along with a positive predictive value of 96%, a NPV of 86%, and an AUROC of 0.95 - supporting the superior diagnostic performance of glucose over CEA. Another study involving patients with histologically confirmed pancreatic mucinous neoplasms concluded that glucose had surpassed CEA as the most effective cyst fluid biomarker. In that study, glucose demonstrated an AUROC of 0.96, compared to 0.81 for CEA (difference: 0.145, P = 0.003), with an optimal cut-off of ≤ 25 mg/dL yielding a sensitivity of 88.1% and specificity of 91.2%[41].

A very recent study reported that a glucose cut-off of < 58 mg/dL achieved a sensitivity of 92.6%, specificity of 95.1%, and overall accuracy of 93.5%, with an excellent AUROC of 0.98[32]. Another contemporary study found that on-site glucose measurement effectively distinguished mucinous from non-mucinous pancreatic cystic lesions, outperforming CEA (AUROC: 0.870 for on-site glucose vs 0.806 for CEA), and demonstrated strong correlation between on-site and laboratory glucose values (P = 0.919)[42]. Similarly, a relatively small French multicenter study showed that a cut-off of 41.8 mg/dL outperformed CEA for discriminating mucinous from non-MC, with an AUROC of 93.6% (vs 81.2% for CEA), higher sensitivity (95.3% vs 41.7%), but slightly lower specificity (91.2% vs 96.9%)[43].

Several meta-analyses have reinforced the diagnostic utility of intracystic glucose in differentiating mucinous from non-mucinous pancreatic cysts. A 2021 meta-analysis found that the pooled sensitivity of cyst fluid glucose was significantly higher than that of CEA (91% vs 56%, P < 0.001), as was overall diagnostic accuracy (94% vs 85%, P < 0.001), despite a numerically lower specificity (86% vs 96%, P > 0.05)[44]. Additionally, combining glucose with CEA did not improve diagnostic performance compared to glucose alone (accuracy: 97% vs 94%, P > 0.05)[44]. Another 2021 meta-analysis also concluded that glucose outperformed CEA, reporting higher sensitivity (90% vs 67%), similar specificity (82% vs 80%), and a greater AUROC (0.96 vs 0.79)[45].

In 2022, Guzmán-Calderón et al[46] reported pooled sensitivity and specificity values for intracystic glucose in diagnosing mucinous pancreatic cysts of 91% and 85%, respectively. That same year, Mohan et al[47] presented a meta-analysis yielding pooled sensitivity and specificity of 90.5% and 88%, respectively, for low intracystic glucose. Importantly, point-of-care testing with glucometers also demonstrated high diagnostic accuracy, with sensitivity and specificity of 89.5% and 83.9%, respectively. They further calculated that a glucose cut-off of 50 mg/dL had sensitivity and specificity of 90.1% and 85.3%, respectively[47].

In conclusion, low intracystic glucose levels provide strong sensitivity, specificity, and diagnostic accuracy for identifying mucinous pancreatic cysts. Moreover, glucose testing is simple, cost-effective, and feasible as a point-of-care test using standard glucometers. However, combining glucose measurements with cyst fluid CEA does not appear to enhance diagnostic performance. Importantly, glucose levels cannot distinguish between malignant and non-malignant MC. It should also be noted that PC may present with low glucose concentrations. Therefore, glucose results must always be interpreted within the proper clinical context, considering the patient’s history and the radiologic features of the cystic lesion, which can help differentiate between PC and true neoplastic cysts Table 1.

Cystic amylase

Cyst fluid amylase is a useful marker when attempting to exclude the presence of a pseudocyst. Relatively older data from a pooled analysis of 12 studies demonstrated that amylase levels below 250 IU/L could exclude a pseudocyst in 98% of cases[24]. However, a more recent retrospective study questioned the utility of the 250 IU/L cut-off. While amylase levels were significantly higher in PC compared to non-pseudocystic lesions (P = 0.009), 54% of non-inflammatory pancreatic cystic lesions also exhibited amylase levels above 250 IU/L. Although this cut-off yielded a sensitivity of 89%, it showed a specificity of only 54% for identifying PC. Interestingly, the same study found that malignant MC had significantly lower amylase levels compared to benign MC (P = 0.0008)[28].

Another small trial published subsequently reported that cyst fluid amylase levels were significantly lower in mucinous neoplasms, and an optimal cut-off of 6800 IU/mL yielded an accuracy of 69% in distinguishing PC from mucinous neoplasms[48]. In contrast, a study by Soyer et al[49] found no significant difference in amylase levels among benign non-mucinous, benign mucinous, and malignant cysts.

In conclusion, although more recent high-quality data are lacking and existing studies present conflicting results, cyst fluid amylase remains a helpful tool for ruling out pancreatic PC. Nevertheless, clinicians should always interpret amylase levels in the context of the patient's clinical history and the radiologic characteristics of the cystic lesion.

Key genetic markers in pancreatic cyst fluid

KRAS and GNAS mutations: Mutations in the KRAS and GNAS genes have emerged as pivotal biomarkers in evaluating pancreatic cystic lesions, particularly in identifying MC and assessing their malignant potential[50]. KRAS mutations, especially at codon 12, are commonly observed in IPMNs and MCNs, while GNAS mutations, particularly at codon 201, are highly specific for IPMNs.

The diagnostic value of molecular analysis of pancreatic cyst fluid was first demonstrated in 2005 by a prospective study from Khalid et al[51], which showed that malignant cysts had significantly higher CEA levels (P = 0.034), poorer nucleic acid quality (P = 0.009), and a greater number of mutations (P = 0.002) compared to premalignant lesions. The combination of a KRAS mutation followed by allelic loss yielded the most predictive molecular signature, with a sensitivity of 91% and specificity of 93% for malignancy. In 2009, Sawhney et al[52] retrospectively evaluated CEA and molecular markers for differentiating mucinous from non- MC, finding poor agreement between CEA and molecular markers (κ = 0.2), and low concordance between CEA and DNA quantity, KRAS mutations, or allelic imbalance.

Subsequent studies supported these findings. In a later study by Khalid et al[53], KRAS mutations were highly specific for MC (96%, odds ratio = 20.9). In 2013, Nikiforova et al[54], in a cohort of 603 patients, found KRAS mutations to be highly specific (100%) but only moderately sensitive (54%) for mucinous differentiation. Sensitivity varied by cyst type - 67% for IPMNs but only 14% for MCNs. Notably, although 53% of aspirates in Nikiforova et al’s study[54] were inadequate for cytology, molecular testing was feasible in 98% of cases. Similar conclusions were reached by Singhi et al[55], who confirmed the high diagnostic value of combined GNAS and KRAS analysis for IPMNs but emphasized the limited sensitivity for MCNs, highlighting the need for additional markers.

In 2015, Winner et al[56] evaluated KRAS mutations, loss of heterozygosity at tumor suppressor loci, and DNA content. Although molecular markers improved the diagnostic yield of EUS-FNA, they were less accurate in identifying MC than CEA levels alone[56]. More recent, albeit smaller, studies have supported the role of GNAS in improving mucinous cyst identification[57,58].

Focusing on IPMNs, the 2019 ZYSTEUS biomarker study explored the utility of next-generation sequencing (NGS) in distinguishing IPMNs from PC. In a cohort of 22 patients, all IPMN cases harbored at least one KRAS or GNAS mutation, whereas none were found in PC. The cellular fraction of cyst fluid provided higher DNA yields and greater mutation diversity[59].

A meta-analysis by McCarty et al[60] including 785 Lesions demonstrated that combined KRAS and GNAS mutation detection significantly outperformed CEA in diagnosing IPMNs and mucinous cystic lesions. The pooled sensitivity, specificity, and diagnostic accuracy for KRAS + GNAS mutations in IPMNs were 94%, 91%, and 97%, respectively, all significantly higher than for CEA (P < 0.001). While sensitivity and specificity for diagnosing MCNs were similar between the two modalities, the overall diagnostic accuracy was significantly higher for KRAS + GNAS (97% vs 89%; P < 0.001). Similarly, a meta-analysis by Pflüger et al[61] analyzing 42 studies found that KRAS and/or GNAS mutations identified MCNs with 79% sensitivity and 98% specificity, outperforming CEA (58% sensitivity, 87% specificity). In five studies involving 1269 cysts, KRAS mutation testing correctly identified about half of MCNs, with 99% specificity.

Several researchers have investigated combining molecular markers with CEA. Khalid et al[53] reported that KRAS mutations increased the sensitivity of CEA from 67% to 84%. In 2016, Jones et al[62] showed that NGS reclassified cyst etiology in cases with normal CEA levels, demonstrating greater sensitivity (86%) than CEA (57%) for mucinous differentiation, although CEA remained more specific (100%). The detection of KRAS, GNAS, TP53, CDKN2A, and SMAD4 mutations further facilitated identification of high-risk or malignant cysts.

In 2017, Kadayifci et al[63] evaluated the diagnostic benefit of adding GNAS to KRAS and CEA analysis. Among 197 patients, adding GNAS to KRAS modestly increased accuracy (from 76.6% to 79.1%), but significantly boosted diagnostic accuracy when combined with CEA (from 66.4% to 80.7%). The triple combination yielded the highest accuracy (86.2%).

In a 2024 study, Belfrage et al[64] used a 50-gene NGS panel alongside imaging and CEA to evaluate pancreatic cysts requiring surgery. NGS identified KRAS and/or GNAS mutations in 53% of MC, with a sensitivity of 53% and specificity of 92%, outperforming cytology (33% sensitivity, 100% specificity) and CEA alone (53% sensitivity, 89% specificity). Combined NGS and CEA improved sensitivity and specificity to 78% and 87%, respectively. Notably, HGD cysts lacked “worrisome” mutations, whereas 80% of malignant lesions had mutations associated with advanced disease. NGS had a higher diagnostic accuracy (area under the curve = 0.777) than CEA (area under the curve = 0.631), influencing surgical decisions in nearly half of cases.

Beyond diagnostic classification, cyst fluid DNA analysis also holds prognostic value. In 2013, Rockacy et al[65] identified four independent risk factors for malignancy or progression: Presence of a solid component, symptomatic presentation, cyst size > 3 cm, and the presence of a KRAS mutation. In 2015, Al-Haddad et al[66] applied integrated molecular pathology, which combines genetic, clinical, radiologic, and cytologic data, and found it superior to the 2012 Sendai guidelines in malignancy prediction[66,67]. Hu et al[68] emphasized the diagnostic value of KRAS in MCNs and suggested GNAS mutations may predict better overall and recurrence-free survival in invasive IPMN cases. Additional studies have since reinforced the prognostic relevance of genetic markers in risk stratification[69-74].

In conclusion, the collective findings underscore the value of KRAS and GNAS mutational analysis in the preoperative evaluation of pancreatic cysts. Their incorporation into diagnostic workflows enhances diagnostic precision and improves risk stratification, ultimately guiding more informed clinical decision-making (Table 2).

Table 2 Molecular marker studies in pancreatic cyst fluid.

Ref.	Type of study	Molecular marker	Patients	Sensitivity	Specificity	Accuracy	Conclusions
Khalid et al[51], 2005	Prospective single-center diagnostic study	KRAS codon 12, allelic loss, CEA, DNA quality	36	91%	93%	92.0%	KRAS followed by allelic loss is highly predictive for malignancy
Sawhney et al[52], 2009	Retrospective EUS-FNA study	KRAS, CEA, DNA quantity, allelic imbalance	100	82% (CEA), 77% (Mol)	-	-	CEA and molecular markers had poor concordance; combining improved sensitivity to 100%
Khalid et al[53], 2009	Multicenter prospective PANDA study	KRAS, DNA yield, LOH amplitude	113	-	KRAS 96%	-	Allelic loss and DNA quantity predicted malignancy
Nikiforova et al[54], 2013	Large retrospective cohort	KRAS codon 12	603	67% (IPMN), 14% (MCN)	100%	77%	KRAS highly specific for mucinous cysts, low sensitivity for MCNs
Singhi et al[55], 2014	Prospective surgical validation	KRAS, GNAS (codon 201)	91	65%	100%	82.5%	Highly specific for IPMNs; limited sensitivity for MCNs
Winner et al[56], 2015	Retrospective cohort study	KRAS, LOH, DNA content	56	50%	96%	73.0%	Molecular markers increased diagnostic yield, but less accurate than CEA
Rockacy et al[65], 2013	Retrospective prognostic study	KRAS	113	-	-		KRAS associated with poor clinical outcomes
Al-Haddad et al[66], 2015	Prospective multicenter IMP study	IMP (KRAS, TP53 + cytology/clinical data)	492	-	-	-	IMP outperformed Sendai guidelines in risk stratification
Springer et al[69], 2015	Prospective classifier study	KRAS, GNAS, TP53, SMAD4, LOH, BRAF	130	90%-100%	92%-98%	95%	Molecular algorithm reduced unnecessary surgeries by 91%
Jones et al[62], 2016	Prospective reclassification study	KRAS, GNAS, TP53, CDKN2A, SMAD4	92	86% (NGS)	100% (CEA)	93%	NGS reclassified cysts with normal CEA; detected high-risk lesions
Singhi et al[70], 2016	Clinicopathologic accuracy study	KRAS, GNAS, TP53, VHL, PTEN	225	100%	100%	100%	Outperformed AGA guidelines in detecting advanced neoplasia
Kadayifci et al[63], 2017	Retrospective validation study	CEA, KRAS, GNAS	197	86.2% (triple)	-	-	GNAS improved accuracy when added to CEA and KRAS
Rosenbaum et al[71], 2017	Retrospective cytology correlation	KRAS, GNAS, TP53, SMAD4	113	75% (cytology), 46% (late mutations)	-	-	NGS added value in identifying malignant cysts
Singhi et al[76], 2018	Large prospective validation	KRAS, GNAS, TP53, PTEN, PIK3CA	626	89%	100%	94.5%	High accuracy in cyst classification and risk stratification
Volckmar et al[59], 2019	Prospective biomarker study	KRAS, GNAS via NGS in cell fraction	22	100% (IPMN)	100% vs pseudocysts	100%	Mutations distinguished IPMNs from pseudocysts
Farrell et al[72], 2019	Cohort analysis with imaging features	KRAS, LOH, DNA amount	478	-	-	-	≥ 2 DNA abnormalities increased malignancy risk in cysts with worrisome features
Springer et al[75], 2019	Multimodal AI tool (CompCyst)	CompCyst (clinical + KRAS, GNAS, TP53)	862	-	-	-	Significantly improved diagnostic accuracy and reduced overtreatment
Laquière et al[73], 2019	Pilot NGS concordance study	KRAS, GNAS, TP53	17	78%	62%-100%	-	PCF and tissue mutations were concordant in 88% of cases
Ren et al[57], 2021	Prospective molecular reclassification study	KRAS, GNAS, BRAF	108	88.5%	100%	94.3%	KRAS-negative mucinous cysts had alternative BRAF pathway mutations
McCarty et al[44], 2021	Meta-analysis	KRAS, GNAS	785	94%	91%	92.5%	KRA+ GNAS significantly outperformed CEA in mucinous cyst diagnosis
Herranz Pérez et al[58], 2021	Routine practice molecular study	KRAS, GNAS	-	-	-	-	Feasibility and utility confirmed in real-world setting
Pflüger et al[61], 2023	Meta-analysis (42 studies)	TP53, SMAD4, CDKN2A, VHL	666	9%-42%	95%-99%	69.5%	High specificity for HGD; VHL for SCA discrimination
Hata et al[74], 2023	Prospective epigenetic biomarker study	TBX15, SOX17 methylation	70	69.6%	90%	79.8%	Methylation identified HGD in IPMNs
Nikiforova et al[77], 2023	PancreaSeq GC validation (DNA/RNA NGS)	74-gene panel + CEACAM5 mRNA	185	95% (precursor), 82% (HGD)	100%	97.5%	PancreaSeq GC outperformed imaging/cytology in neoplasia detection
Paniccia et al[78], 2023	Multicenter real-time NGS registry	PancreaSeq 22-gene panel	1933	93%	95%	94%	NGS with cytology guided clinical decisions
Hu et al[68], 2024	Review on progression markers	KRAS, GNAS (prognostic)	-	-	-	-	GNAS may be linked to improved recurrence-free survival
Belfrage et al[64], 2024	Prospective diagnostic accuracy study	50-gene NGS, KRAS, GNAS, TP53 + CEA	97	78% (combo)	87%	82.5%	NGS + CEA improved classification and influenced surgical decisions

CEA: Carcinoembryonic antigen; EUS-FNA: Endoscopic ultrasound-guided fine-needle aspiration; AI: Artificial Intelligence; Mol: Molecular; PANDA: Pancreatic cyst DNA analysis; LOH: Loss of heterozygosity; MCN: Mucinous cystic neoplasms; IPMN: Intraductal papillary mucinous neoplasms; IMP: Integrated molecular pathology; NGS: Next-generation sequencing; HGD: High-grade dysplasia; AGA: American Gastroenterological Association; GC: Genomic classifier.

CDKN2A, PIK3CA, SMAD4, TP53 and other mutations: Besides KRAS and GNAS mutations, other genetic markers - such as CDKN2A, PIK3CA, SMAD4, and TP53 - have been evaluated for their diagnostic and prognostic value in patients with pancreatic cysts. Pflüger et al[61] reviewed two studies involving 666 cysts and demonstrated that mutations in these genes within cyst fluid are associated with a high risk of HGD or pancreatic ductal adenocarcinoma. Although these markers showed high specificity (95%-98%), their sensitivity remained relatively low (9%-42%). Additionally, VHL mutations were found to be highly specific (99%) for identifying SCAs, helping distinguish them from MC[75]. Therefore, the absence of such mutations does not rule out the presence of HGD or pancreatic ductal adenocarcinoma, highlighting the need for continued surveillance in certain cases[61,63,72].

Springer et al[69] analyzed cyst fluid from 130 resected pancreatic cystic neoplasms and assessed mutations in a panel of genes including BRAF, CDKN2A, CTNNB1, GNAS, KRAS, NRAS, PIK3CA, RNF43, SMAD4, TP53, and VHL. By combining these molecular markers with clinical features, they developed a classification system with 90%-100% sensitivity and 92%-98% specificity, which could potentially reduce unnecessary surgeries by up to 91%[69]. In 2019, Springer et al[75] advanced this concept further by creating a multimodal diagnostic tool, CompCyst, which integrates clinical, imaging, genetic, and biochemical data. CompCyst significantly improved diagnostic accuracy and could have prevented more than half of unnecessary surgical interventions.

In a large prospective analysis of 626 cyst fluid samples from 595 patients, Singhi et al[76] found that KRAS and GNAS mutations were present in 49% of cases, yielding 89% sensitivity and 100% specificity for diagnosing IPMNs and MCNs. Moreover, combining these mutations with TP53, PIK3CA, and PTEN mutation testing achieved high accuracy for cyst classification and malignancy risk stratification, outperforming conventional imaging findings such as ductal dilatation and mural nodules[76].

In 2023, Nikiforova et al[77] introduced the PancreaSeq Genomic Classifier, a DNA/RNA-based NGS assay utilizing a 74-gene panel. This test assessed mutations, gene fusions, expression profiles, and CEACAM5 mRNA levels. In both the training (n = 108) and validation (n = 77) cohorts, PancreaSeq Genomic Classifier achieved 95% sensitivity and 100% specificity for identifying cystic precursor neoplasms, and 82% sensitivity and 100% specificity for detecting HGD or early carcinoma. Notably, it outperformed conventional clinical, imaging, and cytologic criteria, boosting guideline sensitivity by over 10% without compromising specificity[77].

A recent prospective, multi-institutional study by Paniccia et al[78], published in Gastroenterology, evaluated the PancreaSeq 22-gene panel in 933 patients across 31 centers. Mutations in the mitogen-activated protein kinase pathway and GNAS had a sensitivity of 90% and specificity of 100% for detecting MC. When combined with TP53, SMAD4, CTNNB1, and mTOR pathway mutations, the panel achieved 88% sensitivity and 98% specificity for detecting advanced neoplasia. Adding cytopathologic evaluation further increased sensitivity to 93% while maintaining a specificity of 95%. Importantly, the study also identified actionable genomic alterations such as BRAF fusions and ERBB2 amplifications, which could guide personalized treatment decisions[78].

The incorporation of extended molecular profiling - including high-risk mutations beyond KRAS and GNAS - significantly enhances the diagnostic and prognostic assessment of pancreatic cystic lesions. Emerging tools like PancreaSeq and CompCyst, which integrate genetic, clinical, and imaging data, offer a more accurate and personalized approach to patient management. These advancements hold the potential to reduce unnecessary surgeries, improve early detection of malignancy, and ultimately refine the decision-making process in the care of patients with pancreatic cysts (Table 2).

Summary of evidence

EUS-FNA continues to be a cornerstone in the evaluation of pancreatic cystic lesions, particularly in distinguishing mucinous from non-MC. Commercially available cyst fluid biomarkers - such as CEA, intracystic glucose, and amylase - provide valuable diagnostic information when interpreted alongside imaging features and clinical history. However, no definitive cut-off value for CEA, intracystic glucose, or any combination thereof can reliably classify a cyst as mucinous. Clinicians should therefore interpret these biomarkers with clinical judgment.

Intracystic glucose levels below 50 mg/dL, as supported by several studies, offer superior sensitivity for identifying MC compared to a CEA level > 192 ng/mL - a threshold also endorsed by several clinical guidelines. Particularly, glucose values below 25 mg/dL are strongly suggestive of MC. Conversely, CEA levels < 192 ng/mL, and especially substantially lower levels, have demonstrated a specificity exceeding 95% in numerous studies and should primarily be used to rule out MC.

Despite the inclusion of molecular analyses - such as testing for KRAS and GNAS mutations - EUS-FNA alone does not achieve the diagnostic accuracy of newer modalities like EUS-guided through-the-needle biopsy and needle-based confocal laser endomicroscopy[79]. Molecular characterization of cyst fluid represents a significant advancement, offering deeper insights into neoplastic potential. Specifically, the detection of genetic alterations - such as mutations in KRAS, GNAS, and tumor suppressor or oncogenic pathway genes (TP53, SMAD4, PIK3CA, PTEN) - has shown considerable promise in risk stratification for malignancy[80].

While molecular markers display excellent specificity, their sensitivity remains limited; absence of detectable mutations does not exclude HGD or invasive carcinoma[81]. Additionally, cost and availability concerns limit their routine use in clinical practice. Consequently, due to these practical limitations, the application of molecular testing remains best suited to research settings (Table 3).

Table 3 Diagnostic ability of various cystic fluid biomarkers.

Molecular marker	Diagnostic ability of mucinous cysts	AUC for mucinous cysts diagnosis (range)	SENS for mucinous cysts diagnosis (range)	SPEC for mucinous cysts diagnosis (range)	Diagnostic ability discriminating malignant from non -malignant cysts	Simplicity of use/availability
Cystic CEA	Moderate	0.633-0924	48%-95.4%	82%-98%	-	Moderate
Cystic glucose	Moderate	0.870-0980	89%-95%	78%-95.1%	-	Good
Cystic amylase	-	-	-	-	Inadequate	Moderate
KRAS	Moderate	-	61%	99%	Inadequate	Inadequate
GNAS	Moderate	-	39%-90%	100%	Inadequate	Inadequate
KRAS + GNAS	Moderate	-	79%	98%	Inadequate	Inadequate

CEA: Carcinoembryonic antigen; AUC: Area under the curve; SENS: Sensitivity; SPEC: Specificity.

Cytological evaluation continues to play a critical role in distinguishing benign from malignant pancreatic cysts. However, its utility is frequently compromised by low cellular yield, with adequacy rates as low as 34% reported in some studies[82]. Recent data suggest that EUS-guided fine-needle biopsy (FNB) - which utilizes a 22-gauge Franseen-tip needle with cutting edges, as opposed to the 20-22 - gauge needles used in FNA - achieves higher diagnostic yield. FNB has demonstrated improved sensitivity (76.6%) and specificity (98.9%) for malignancy detection, albeit with a modestly increased risk of adverse events such as pancreatitis[79,83,84].

CONCLUSION

In conclusion, although EUS-FNA combined with conventional cyst fluid biomarkers and molecular testing remains a vital component in managing pancreatic cysts, it has not yet fully supplanted the superior diagnostic accuracy of FNB or cytology for detecting malignancy. There is a critical need to develop and validate novel, minimally invasive blood-based or cystic fluid biomarkers capable of reliably assessing malignant potential. Future research should focus on large, prospective studies that integrate molecular, radiologic, and clinical parameters into comprehensive diagnostic algorithms. Such advancements could enhance risk stratification, reduce unnecessary surgical procedures, and ultimately improve patient outcomes in the management of pancreatic cystic lesions.