Intracystic glucose is superior to carcinoembryonic antigen in diagnosing mucinous pancreatic cysts in genomic era: Systematic review and meta-analysis

Abstract

BACKGROUND

A differential diagnosis between mucinous and non-mucinous pancreatic cysts is critical for clinical decision making, as mucinous cysts have malignant potential. While intracystic carcinoembryonic antigen (CEA) remains widely used and molecular markers hold promise, intracystic glucose (IG) has emerged as a promising, easily assessable, and economical alternative.

AIM

To investigate the diagnostic performances of IG, alone or compared to CEA.

METHODS

A systematic review and meta-analysis was conducted until April 2025. Pooled estimates of diagnostic metrics and their comparisons were calculated using a random effects model. A bivariate summary receiver operating characteristic model was used. Subgroup analyses and risks of bias assessment were also conducted.

RESULTS

Sixteen studies (1398 patients) were included. Compared with CEA, IG showed higher pooled sensitivity (95% vs 57%) and accuracy (90% vs 72%), with a lower negative likelihood ratio (LR- 0.08 vs 0.46), whereas CEA demonstrated higher specificity (97% vs 85%) and a higher positive likelihood ratio (LR+ 11 vs 6). The area under the receiver operating characteristic was 0.96 for IG and 0.82 for CEA. Pairwise analysis confirmed the superiority of IG in terms of sensitivity [odd ratio (OR): 9.36] and accuracy (OR: 3.35), while specificity favored CEA (OR: 0.43). Specificity of CEA was influenced by many factors, even by the definitive diagnostic method (histology/cytology vs other modalities, 93% vs 98%, P = 0.02).

CONCLUSION

This study reinforced the current recommendations supporting the use of IG as a first-line rule-out biomarker for mucinous pancreatic cysts due to its high sensitivity, while CEA may retain a complementary role as a rule-in test in clinical scenarios where higher specificity is required, such as surgical referral.

Key Words: Pancreas; Pancreatic cyst; Mucinous cyst; Intracystic glucose; Intracystic carcinoembryonic antigen; Genetic test

Core Tip: Accurate diagnosis of pancreatic cysts is fundamental for patient management. Intracystic glucose (IG), a simple and cost-effective marker, showed promising results compared to carcinoembryonic antigen (CEA) and emerging molecular markers. Nonetheless, low quality evidences and weak recommendation have been reported in the guidelines. In this study, sixteen studies showed glucose had higher accuracy [odd ratio (OR): 3.35] and sensitivity (OR: 9.36) than CEA, while CEA had greater specificity (OR: 0.43). These findings support the use of IG as a methodologically stable biomarker, particularly suitable as a rule-out test for mucinous cysts. Therefore, this study reinforces the role of IG as a first-line diagnostic biomarker in identifying mucinous pancreatic cysts. Physicians can easily use IG in identifying mucinous cysts with greater certainty.

INTRODUCTION

Differentiation among pancreatic cystic lesions (PCLs), particularly mucinous and non-mucinous types, is clinically crucial as it directly influences therapeutic decisions and follow-up strategies[1,2]. Mucinous cysts, such as intraductal papillary mucinous neoplasms (IPMNs) and mucinous cystic neoplasms (MCNs), carry the risk of progression to malignancy and require close surveillance or surgical resection. In contrast, non-mucinous lesions, such as serous cystadenomas or pseudocysts, are typically benign, managed conservatively, or even discharged from active surveillance due to their non-existent malignant potential. Early and accurate characterization is essential to avoid both overtreatment and delayed management of high-risk cases[2-4]. Carcinoembryonic antigen (CEA) has long been widely used as a key intracystic biomarker of mucinous cysts[5]. However, its standard cut-off of 192 ng/mL offers high specificity (approximately 93%) at the expense of limited sensitivity (approximately 63%)[6-9]. On the other hand, intracystic glucose (IG) has gained attention as a promising alternative, with multiple studies reporting high diagnostic accuracy coupled with practical advantages such as low cost, simplicity, and point-of-care feasibility[10-14]. Although previous meta-analyses have examined the diagnostic performance of IG[15,16], they have been outdated by a substantial number of new studies[17-24]. Meanwhile, European Society of Gastrointestinal Endoscopy guidelines have conditionally recommended IG over intracystic CEA based on low-quality evidence[25]. While molecular biomarkers and next-generation sequencing (NGS) panels in cyst fluid analysis (e.g., PancreaSeq) are gaining increasing attention[26], they are expensive and have limited accessibility[26-28]. Therefore, biochemical markers such as IG continue to represent essential tools in the diagnostic workup of PCLs. In this context, an updated systematic review and meta-analysis was conducted to evaluate the diagnostic performance of IG, also compared to CEA, in differentiating mucinous from non-mucinous pancreatic cysts and to strengthen the evidence base for clinical decision-making and future guideline updates.

MATERIALS AND METHODS

Search strategy

The systematic review protocol adhered to the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines for meta-analyses of observational studies (Supplementary Table 1)[29]. A systematic search of the literature was performed on MEDLINE via PubMed, EMBASE via Ovid, Scopus, and Cochrane Library, covering all studies published up to April 2025. Only English language studies involving human subjects were included in the search. The search strategy combined terms related to “pancreatic cysts”, “intracystic glucose”, “carcinoembryonic antigen”, “CEA”, “glucose levels”, and “cystic fluid”. The full search strategy is available in Supplementary material (section “Search strategy”). The reference lists of included articles and previous reviews were manually screened for additional studies.

Eligibility criteria

Studies were considered eligible for inclusion if: (1) They enrolled adult patients (aged 18 years or older) with PCLs; (2) Reported data on the diagnostic performance, such as sensitivity, specificity, or accuracy, of IG, either alone or in combination with CEA, in differentiating mucinous from non-mucinous cysts; and (3) Used an appropriate gold standard. Acceptable gold standards included histopathological examination of surgical samples, cytology, expert consensus, or validated imaging criteria. For the purpose of this review, we established a predefined hierarchy of reference standards. Acceptable reference standards included: (1) Surgical histopathology (highest level); (2) Cytology from endoscopic ultrasound fine needle aspiration (EUS-FNA) or surgical specimens; and (3) Composite or consensus diagnoses based on imaging, clinical, biochemical findings and/or multidisciplinary review. When multiple reference standards were available within a study, surgical histology was prioritized whenever possible. Studies were excluded if they were case reports, editorials, reviews, or abstracts without a full text. Additional exclusion criteria included studies with fewer than 10 patients, those involving pediatric populations, animal or in vitro studies, and studies that did not assess IG as index tests. In cases where multiple publications reported overlapping patient cohorts, only the most complete or recent study was included in the analysis. The target condition was defined as mucinous PCLs (including IPMNs and MCNs) vs non-mucinous cysts (such as serous cystadenomas, pseudocysts and other non-mucinous entities).

All studies meeting the inclusion criteria were retained regardless of methodological quality. Study quality was formally assessed using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool and explored through predefined subgroup and sensitivity analyses rather than through post hoc exclusion, in order to avoid selection bias.

Study selection and data extraction

Two authors (Giacomo Emanuele Maria Rizzo and Giuseppe Infantino) independently performed the screening of titles and abstracts, the revision of full text studies and data extraction. Any discrepancies were resolved through consensus and the involvement of a third author (Gabriele Capurso). Patient and cyst characteristics were extracted, such as the number and type of cysts, mean cyst size, and the prevalence of diabetes among participants, when available. For each included study, we also collected detailed information on study characteristics, including the year of publication, study design (prospective or retrospective), country, sample size, and whether the study was conducted in a single center or across multiple centers. Regarding the two markers, we reported whether IG, alone or with CEA were evaluated, as well as the testing method used (glucometer or laboratory-based) and threshold values applied. Data on diagnostic outcomes were extracted. The values of true positive, false positive, true negative, and false negative were extracted when available or calculated when not specifically reported in the study. The study selection process is summarized in a PRISMA flow diagram (Supplementary Figure 1).

Quality assessment

Each included study was assessed for methodological quality using the QUADAS-2 tool[30]. Based on this scoring system, each study may achieve high, unclear or low risk of bias.

Statistical analysis

A bivariate meta-analysis of sensitivity and specificity was performed to evaluate diagnostic performance of IG and CEA. First, pooled estimates of accuracy, sensitivity, and specificity, along with 95%CI, were calculated using a random effects model (DerSimonian and Laird[31]) for the meta-analysis of proportions. Then, estimates representing the global test performance of both IG and CEA were displayed for the parameters of both the bivariate model[32] and the summary receiver operating characteristic (SROC) model[33] including hierarchical summary receiver operating characteristic (HSROC) model[33]. The following estimates for the summary point were displayed: Sensitivity, specificity, area under the receiver operating characteristic (AUROC), diagnostic odds ratio (DOR), positive likelihood ratio (LR+), negative likelihood ratio (LR-), and inverse of the negative likelihood ratio (1/LR-). A bivariate boxplot was used to demonstrate the degree of interdependence of sensitivity and specificity for both IG and CEA to further investigate heterogeneity, including the identification of outliers. A pairwise meta-analysis for each diagnostic parameter (accuracy, sensitivity, and specificity) was performed among the studies with paired data. In addition, subgroup analyses were performed evaluating significant differences using an interaction test (Cochran’s Q test). For evaluating the clinical relevance of the two markers, a Fagan nomogram was created[34]. This tool helps contextualize the findings within Bayes’ theorem by illustrating changes in disease probability before and after testing. Specifically, the Fagan nomogram serves as a visual method to estimate how a diagnostic test result influences the likelihood of disease in a patient. Heterogeneity was assessed using the I² statistic and was further explored through predefined subgroup analyses, sensitivity analyses, and bivariate modeling, allowing assessment of the impact of study-level characteristics on diagnostic performance. Doi-plot analysis, which is a combination with funnel and Galbraith plots[35], was used to assess publication bias in the meta-analyses of proportion. Publication bias was assessed qualitatively and quantitatively using the Luis Furuya-Kanamori (LFK) index[36]. Additionally, symmetrical evaluation of funnel plots and the Egger test[37] were used to identify publication bias in the paired meta-analyses. All analyses were conducted using STATA statistical software (Statistics and Data Science, version 18, College Station, TX 77845, United States).

RESULTS

Study selection and quality assessment

Ultimately, 16 studies evaluating IG levels[10-14,17-22,24,38-40], fulfilled all inclusion criteria and were included in the meta-analysis. One study reported two distinct cohorts that individually fulfilled eligibility criteria; therefore, this was included separately in the analysis, resulting in a total of 17 cohorts[21]. Ten studies (62.5%) were rated as low risk of bias and six (37.5%) as high risk of bias (Supplementary Table 2).

Study and patient characteristics

A total number of 1398 patients were included, with sample sizes ranging from 31[14] to 169[20]. Ten studies were single-center[10-14,18-20,24,39,40], conducted in Europe[12,17-21,23,38-40] or North America[10,11,13,14,22]. Eight studies were prospective[10,12,17,18,20,24,39,40]. The diagnostic gold standard for classifying cysts as mucinous or non-mucinous varied among studies (Table 1). Six studies[10,11,14,22,24,38] used exclusively histology or cytology, which are considered high-quality reference standards. The percentage of cysts confirmed through surgical pathology ranged from 7.7% to 100%, whereas the proportion of malignant lesions ranged from 1.98% to 38.5%. The cut-off values for IG ranged widely (from ≤ 41 mg/dL to < 87 mg/dL), although most studies adopted a threshold of 50 mg/dL. CEA cutoff values were generally consistent (≥ 192 ng/mL), with minor variations in a few studies. All the included cohorts (n = 17) evaluated IG, 15 of them, also reported CEA values. Among studies reporting both markers, twelve[10,12-14,18-20,22-24,38,40] evaluated both on the same set of cysts (Supplementary Table 3). All studies reported the acquisition of fluid samples via EUS-FNA, with 4 studies[10,11,14,24] including also surgically resected specimens.

Table 1 Characteristics and diagnostic performance of included studies evaluating intracystic glucose and carcinoembryonic antigen for differentiating mucinous from non-mucinous pancreatic cysts.

Ref.	Study design	Country	Patients, n	Male (%)	Mean age, years	MC, n	NMC, n	Mean lesion size, mm	Cysts undergoing surgery (%)	Malignant lesio, n (%)	Diagnostic methods for identifying mucinous cysts
Carr et al[10], 2018	Prospective, single-center	United States	153	30.1	NR	106	47	30.50	100	10.45	Surgical pathology
Ribaldone et al[12], 2020	Prospective, single-center	Italy	56	35.7	62.6 ± 14.4	31	25	43.9	23.2	5.35	Cytology, nCLE or surgical pathology, imaging and consensus
Simons-Linares et al[13], 2020	Retrospective, single-center	United States	113	37	66.9 ± 13.5	73	40	28	18	3.7	Combination of EUS, imaging, fluid analysis, cytology
Bruni et al[17], 2024	Prospective, multicenter	Italy and Sweden	50	42	70	33	17	22	22	18.9	Surgical pathology, cytology, imaging features or consensus
Gyimesi et al[18], 2024	Prospective, single-center	Hungary	91	30.8	65.6	52	39	32.3	14.3	8.8	Surgical pathology, cytology, imaging features
Ribeiro et al[19], 2024	Retrospetive, single-center	Portugal	78	38.5	64	44	34	31	NR	NR	Surgical pathology, EUS-TTNB, consensus
Rossi et al[20], 2024	Prospective, single-center	Italy	169	37.3	62.49	108	61	40.18 ± 19.37	7.7	NR	Surgical pathology, cytology, imaging features or consensus
Faias et al[38], 2020	Retrospective multicenter	Portugal	82	41	61.3 ± 14.8	38	44	38.5 ± 20.4	14.63	15.85	Surgical pathology or EUS cytology
Zikos et al[11], 2015	Retrospective, single-center	United States	65	46.2	60.6	42	23	35.3	100	20	Surgical pathology
Smith et al[22], 2022	Retrospective, multicenter	United States	93	45.2	65	59	34	38.3 ± 23.1	48.4	25.8	Histology (surgery, FNB, TTNB)
Williet et al[23], 2023	Retrospective, multicenter	France	81	49.9	61.8	46	35	33	14.8	11.1	Histology, surgical pathology, imaging, nCLE
Gorris et al[40], 2023	Prospective, single-center	Amsterdam	63	51	65	33	30	NR	57	3.17	Histopathology, cytology, imaging
Sinha et al[24], 2024	Prospective, single-center	India	100	38	60.8	54	46	NR	NR	3.00	Histology or EUS cytology
Zamir et al[39], 2022	Prospective, single-center	Israel	101	41.6	67.46 ± 11.86	77	24	NR	3.1	1.98	Expert consensus (clinical + imaging + cytology ± histology)
Noia et al[21], 2021 (D)	Retrospective, multicenter	Spain, Argentina	40	52.5	68.8± 11.7	26	14	35.2 ± 23.4	NR	NR	Histology, consensus (clinical, imaging, cytology)
Noia et al[21], 2021 (V)	Retrospective, multicenter	Spain, Argentina	32	34.3	63.7 ± 13.2	22	10	31.9 ± 17.3	NR	NR	Histology, consensus (clinical, imaging, cytology)
Park et al[14], 20131	Retrospective, single-center	United States	31	71.4	62	22	9	30	54	34.28	Histology or positive cytology

¹Data from the cohort of patients with both glucose and carcinoembryonic antigen measurement.

D: Derivation cohort; V: Validation cohort; MC: Mucinous cysts; NMC: Non-mucinous cysts; nCLE: Needle-based confocal laser endomicroscopy; EUS: Endoscopic ultrasound; FNB: Fine needle biopsy; EUS-TTNB: Endoscopic ultrasound-guided through-the-needle biopsy; TTNB: Through-the-needle biopsy; NR: Not reported.

Diagnostic accuracy of IG

Overall, the pooled accuracy of IG was 90% (95%CI: 89%-92%; I² = 9%) (Figure 1A). The pooled sensitivity of IG for identifying mucinous cysts was 95% (95%CI: 93%-97%; I² = 99.9%; Figure 1B). The pooled specificity was 84.7% (95%CI: 79.9%-89.5%; I² = 63.2%) (Figure 1C). The bivariate and SROC analysis confirmed the strong diagnostic performance of IG, with sensitivity of 93.5% (95%CI: 91.3%-95.2%), specificity of 84.1% (95%CI: 77.7%-88.9%), AUROC: 0.96 (0.93-0.97), DOR was 75.99 (95%CI: 48.7-118.58), LR+ was 5.87 (95%CI: 4.17-8.28), LR- was 0.077 (95%CI: 0.058-0.103) (Figure 2A and Supplementary Figure 2). The HSROC with cut-off as covariate showed the variability among studies (Figure 2B and Supplementary Figure 3). The excellent AUROC of 0.96 reflects the strong overall discriminative ability of IG across different diagnostic thresholds.

Figure 1 Forest plots showing pooled diagnostic performance of intracystic glucose. A: Accuracy showing pooled estimates (diamonds) 0.90 (95%CI: 0.89-0.92, I² = 9%); B: Sensitivity showing pooled estimates of 0.95 (95%CI: 0.93-0.97, I² = 99.87%); C: Specificity showing pooled estimates of 0.85 (95%CI: 0.80-0.90, I² = 63.20%). CI: Confidence interval.

Figure 2 Bivariate and summary receiver operating characteristic analysis confirmed the strong diagnostic performance of intracystic glucose. A: Bivariate model with summary receiver operating characteristic (SROC) curve for intracystic glucose (IG) in the diagnosis of mucinous pancreatic cysts. Results in the legend are reflected in the location of the summary point (in the upper-left quadrant of the receiver operating characteristic space), and in the shape of the 95% confidence and prediction regions, which were narrow, indicating a minimal threshold effect and consistent results across studies; B: Hierarchical summary receiver operating characteristic (HSROC) considering cut-off thresholds for each study as covariate: Colored dots represent individual studies, with different colors indicating the IG cut-off used in each study (see legend). Dashed horizontal and vertical lines indicate the 95%CI for sensitivity and 1 - specificity, respectively. The solid black curve represents the HSROC curve fitted across all cut-offs. SROC: Summary receiver operating characteristic; IG: Intracystic glucose; HSROC: Hierarchical summary receiver operating characteristic.

Diagnostic accuracy of CEA

Overall, the pooled accuracy of CEA was 72% (95%CI: 69%-76%; I² = 44.5%) (Figure 3A). The pooled sensitivity was 57% (95%CI: 52%-63%; I² = 57.4%) when evaluating the performance of CEA in identifying mucinous cysts, with estimates varying substantially (range: 41%-77%) (Figure 3B). The pooled specificity was 97% (95%CI: 95%-99%; I² = 99.7%) (Figure 3C). The bivariate and SROC analysis confirmed the strong diagnostic performance of CEA, with sensitivity of 56.9% (95%CI: 51.59-62.01), specificity of 94.7% (95%CI: 91.5-96.7), AUROC: 0.82 (0.79-0.85), DOR was 23.54 (95%CI: 14.88- 37.25), LR+ was 10.72 (95%CI: 6.87-16.73), LR- was 0.46 (95%CI: 0.41-0.51) (Figure 3D and Supplementary Figure 4). Although CEA showed high specificity, its lower AUROC (0.82) indicates a reduced overall discriminative performance compared with IG.

Figure 3 Forest plots showing pooled diagnostic performance of carcinoembryonic antigen. A: Accuracy, with pooled estimate (diamond) 0.72 (95%CI: 0.69-0.76, I² = 44.53%); B: Sensitivity, with pooled estimate 057 (95%CI: 0.52-0.63, I² = 57.43%); C: Specificity, with pooled estimate 097 (95%CI: 0.95-0.99, I² = 79.73%); D: Summary receiver operating characteristic curve for carcinoembryonic antigen in the diagnosis of mucinous pancreatic cysts. Results showed in the legend are supported by the distribution of the plots, and the confidence region was narrow, but the prediction region was broader along the sensitivity axis, suggesting variability in diagnostic performance across studies. CEA: Carcinoembryonic antigen; CI: Confidence interval.

Comparison of IG and CEA

Across 12 studies evaluating both IG and CEA (n = 1110), IG diagnosed mucinous cysts with 3.35 fold greater accuracy than CEA (95%CI: 2.34-4.80; I² = 92.9%) (Figure 4A). The sensitivity was also in favor of IG, showing a OR of 9.36 (95%CI: 6.72-13.06; I² = 0%) (Figure 4B), while the specificity had an OR of 0.43 (95%CI: 0.26-0.70; I² = 0%) (Figure 4C). From a clinical perspective, the markedly higher sensitivity of IG (95% vs 57% for CEA) translates into a substantially lower risk of false-negative results, reducing the likelihood of misclassifying mucinous cysts as non-mucinous. Conversely, the higher specificity of CEA (97% vs 85% for IG) minimizes false-positive diagnoses, which may be particularly relevant in surgical decision-making.

Figure 4 Forest plots showing the comparative diagnostic performance (pooled odds ratio) of intracystic glucose vs carcinoembryonic antigen. A: Accuracy, with pooled odds ratio (OR): 3.35 (95%CI: 2.34-4.80, I² = 92.87%); B: Sensitivity, with pooled OR: 9.36 (95%CI: 6.72-13.06, I² = 0%); C: Specificity, with pooled OR: 0.43 (95%CI: 0.26-0.70, I² = 0%). CEA: Carcinoembryonic antigen; CI: Confidence interval; OR: Odds ratio.

Sensitivity and subgroup analyses

Heterogeneity was first explored using a bivariate boxplot for both IG and CEA. Both the plots for IG and CEA showed some outliers graphically, so a sensitivity analysis without outliers was performed (Supplementary Table 4), showing no significant differences in any of the analyses regarding sensitivity or specificity. In subgroup analysis, the accuracy and sensitivity of IG measurement remained stable regardless of the study design, center type, sample collection method, standard reference for diagnosis or assay type (Supplementary Figures 5 and 6). In contrast, specificity showed significant variation according to the sampling technique. Specifically, it was higher when cyst fluid was collected exclusively via EUS-FNA as compared to combination of EUS-FNA and surgery (89% vs 72%; P = 0.02; Supplementary Figure 7). For CEA, most of the subgroup comparisons showed no statistically significant differences in sensitivity or accuracy (Supplementary Figures 8 and 9). However, specificity for CEA was higher in prospective studies than in retrospective studies (98% vs 94%; P = 0.04). Other differences on specificity were found based on the geographical site (Europe 99% vs Asia 96% vs America 93%, P = 0.05), samples obtained exclusively via EUS-FNA (98% vs 93%; P = 0.04), and risk of bias (low 88% vs moderate 93% vs high 99%, P = 0.01; Supplementary Figure 10). Additionally, a sensitivity analysis restricted to studies using histology or cytology showed higher sensitivity and lower specificity of CEA in studies with surgical or cytological diagnosis compared to other modalities for diagnosis (sensitivity 65% vs 51%, P = 0.001; specificity 93% vs 98%, P = 0.02). No differences were seen in IG performances according to the latter sensitivity analysis (Supplementary Figure 11).

Clinical utility

The Fagan plots (Figure 5) were used to assess the effectiveness of diagnostic testing in differentiating between mucinous and nonmucinous PCLs. When using IG testing with a pretest probability of 60.6% for a mucinous lesion (calculated by pooled analysis from the included studies), the probability of the patient having a mucinous lesion after a positive test rises to roughly 90%. In comparison, CEA testing yields a post-test probability of about 94%. Conversely, when both IG and CEA results were negative according to their respective cut-off, the chance of a mucinous lesion drops significantly to approximately 11% with IG testing and 41% with CEA. These findings further support a complementary clinical role of the two biomarkers. IG, owing to its very low negative likelihood ratio, is particularly suitable as a first-line rule-out test for mucinous cysts, whereas CEA, with a higher positive likelihood ratio, may be useful as a rule-in test in selected cases.

Figure 5 Fagan plots. A and B: Fagan nomograms illustrating the impact of intracystic glucose (IG) (A) and carcinoembryonic antigen (CEA) (B) on post-test probability, assuming a pre-test probability of 60.6%. The solid red arrows represent the shift in probability after a positive test, while dashed grey arrows represent the shift after a negative test, increasing post-test probability to 90% and 94% for positive IG and CEA, respectively, and reducing it to 11% and 41% for negative results. The legend reports the assumed prior probability (Prior Prob), the positive and negative likelihood ratios (LR_Positive and LR_Negative), and the corresponding post-test probabilities for positive and negative test results (Post_Prob_Pos and Post_Prob_Neg). CEA: Carcinoembryonic antigen.

Assessment of publication bias

Publication bias affected sensitivity for IG and specificity for CEA, suggesting possible overestimation of their diagnostic performance due to selective reporting. Funnel plots and Egger’s test confirmed publication bias for accuracy when comparing IG and CEA, but not for sensitivity or specificity. In Supplementary Figure 12A-C, Doi plots for IG show LFK indices of 2.43, -3.5, and 0.98, respectively, indicating publication bias or small-study effects for sensitivity and specificity (Supplementary Figure 12A and B), but not for accuracy (Supplementary Figure 12C). In Supplementary Figure 12D-F, Doi plots for CEA show LFK indices of 1.79, -2.10, and 0.07, indicating publication bias or small-study effects for sensitivity and specificity (Supplementary Figure 12D and E), but not for accuracy (Supplementary Figure 12F). In Supplementary Figure 13A, the Funnel plot for accuracy suggests publication bias, whereas in Supplementary Figure 13B and C, the Funnel plots for sensitivity and specificity suggest absence of relevant publication bias.

DISCUSSION

The differential diagnosis between mucinous and non-mucinous PCLs remains a significant clinical challenge as it directly influences therapeutic decisions, including surgical vs follow-up strategies. Indeed, in surgical series, the rate of patients with PCLs misdiagnosed as mucinous in the preoperative work-up is as high as 30%-40%[10,41]. In recent years, molecular analysis of pancreatic cyst fluid has gained increasing relevance in the classification of PCLs, owing to its ability to detect characteristic genetic mutations associated with mucinous cysts such as KRAS and GNAS, as well as VHL mutations in serous cystadenomas[28,42-44]. Therefore, a robust revisiting of the role of intracystic markers as IG and CEA remains warranted to strengthen the current evidence and support future updates of clinical practice guidelines. In this context, the present study provides the most up-to-date, robust, and comprehensive analysis of the diagnostic performance of IG, even compared to CEA, including the largest number of studies (n = 16) and patients (n = 1398). Our pooled results for each test confirmed that IG has high accuracy (90%), sensitivity (95%), and specificity (85%), indicating a strong performance in accurately identifying both positive and negative cases. Of note, this performance is similar to that calculated for testing intracystic fluid for KRAS and GNAS mutations[26]. Anyway, the studies included in this meta-analysis employed a wide range of cut-off values, varying from < 40 mg/dL to < 87 mg/dL. Specifically, 9 studies used a threshold of 50 mg/dL[10-12,17-19,24,38,40], while only 3[13,23] adopted a lower cut-off of < 41.8 mg/dL. The HSROC curve showed that lower cut-offs tend to reduce specificity, while higher thresholds may compromise sensitivity, as expected. The graphical balance point appears optimal around mid-range cut-offs (50 mg/dL), where both metrics are reasonably preserved. We further investigated the global test performance using a bivariate model (SROC), which demonstrated an excellent discriminatory ability with an AUROC of 0.96 and a high DOR (75.99). Furthermore, the LR+ of 5.87 indicates that a positive test result significantly increases the probability of having the disease, while the LR- of 0.077 confirms that a negative result strongly reduces the probability of disease. These findings support IG as a sensitive and methodologically robust biomarker that is particularly useful as a rule-out test for mucinous cysts. In contrast, CEA showed a significantly lower pooled accuracy of 72% (vs 90% of IG) and sensitivity of 57% (vs 96% of IG), yet with a higher specificity of 97% (vs 85% of IG). Bivariate models suggested a strong association between positive CEA results and mucinous cysts (AUROC = 0.82, DOR = 23.54), even if lower than IG. A direct comparison between the two markers among the studies (n = 12) with paired data (n = 1110) showed results in favour of IG compared to CEA for accuracy (OR: 3.35) and sensitivity (OR: 9.36), while it was in favour of CEA when evaluating specificity (OR: 0.43). This finding strengthens and confirms the robustness of IG.

In comparison with the meta-analysis by Guzmán-Calderón et al[45], the present study stands out for its larger sample size (16 studies vs 6 studies; 1398 patients vs 506 patients), more rigorous methodology, and more comprehensive analytical approach. Similarly, compared with the meta-analysis by Mohan et al[15], which included seven studies (566 patients) and reported 90.1% sensitivity and 85.3% specificity, our study offers greater statistical power, detailed subgroup analysis, and includes only fully published studies. McCarty et al[16] included eight studies reporting 94% accuracy, 91% sensitivity, and 86% specificity for IG, while our meta-analysis included twice as many studies as McCarty’s meta-analysis. Furthermore, our Fagan plot for the clinical utility of these test was based on a pretest probability which was calculated based on the data from the included studies (60.6%), while the authors from the latter meta-analysis reported an empiric pretest probability (40%), so our findings result more applicable in clinical context. Additionally, Mohan et al[15] and Guzmán-Calderón et al[45] could not perform a quantitative assessment of publication bias due to the limited number of included studies, while our study explored this aspect through Doi plots and LFK index for single arm pooled analysis and funnel plot plus Egger’s test for the comparative results, identifying potentially significant asymmetry.

The methodological quality of the included studies in this meta-analysis was overall high: Ten studies (62.5%) had low risk of bias according to the CHADAS-2 tool. Despite that, heterogeneity (I²) was high for most of the analyses, so a deep exploration of variables potentially influencing the results through subgroup analyses was performed, further reinforcing the conclusions. The substantial heterogeneity observed in several pooled estimates likely reflects important methodological and clinical differences among the included studies. First, the reference standards used to classify PCLs varied considerably across studies, ranging from surgical histopathology to cytology or composite diagnoses based on imaging and multidisciplinary consensus. Because histology is available only in a subset of patients undergoing surgery, many studies relied on composite diagnostic criteria, which may introduce misclassification and contribute to variability in specificity estimates, particularly for CEA. Then, variability in diagnostic thresholds for IG across studies may have influenced diagnostic performance, as discussed previously. Differences in thresholds can influence the balance between sensitivity and specificity and therefore contribute to heterogeneity across studies.

Importantly, a bivariate model was used to identify outliers that could have influenced the results, and the subsequent sensitivity analysis excluding these outliers did not find differences in the performances of the two markers, without materially changing the pooled estimates, suggesting that the overall conclusions of this meta-analysis remain robust despite the observed heterogeneity. Furthermore, the accuracy of both IG and CEA was not influenced by any of the subgroup analyses just as the sensitivity of IG remained remarkably stable across all subgroups. Notably, subgroup analyses did not reveal any significant differences between the two methods for measuring IG: Point-of-care glucometry and laboratory-based analysis. These results suggest that the diagnostic performance of IG is independent of the measurement technique, further confirming its robustness and broad clinical applicability considering also that point-of-care glucometry represents a practical and immediate solution allowing on-site assessment during EUS-FNA[21,39]. In addition, the specificity of IG was significantly higher in samples obtained exclusively through EUS-FNA than in studies that included both surgical and EUS-FNA fluid acquisition (89% vs 72%; P = 0.02), underscoring the importance of sampling techniques as EUS-FNA for providing cleaner and more targeted aspirates than mixed techniques (e.g., surgical acquisition), which may compromise specimen integrity by contamination. Therefore, differences in cyst fluid acquisition techniques may also play a role, potentially affecting biomarker concentrations and diagnostic accuracy. Notably, subgroup analyses revealed that the standard reference for diagnosing mucinous cysts (histology by surgery or cytology) did not influence performances of IG, while it made significant differences in sensitivity and specificity of CEA, resulting in higher sensitivity and lower specificity compared to other modalities of diagnosis [65% vs 51% (P = 0.001) and 93% vs 98% (P = 0.02), respectively]. In addition, the use of composite or consensus reference standards in some studies, although reflective of real-world practice, may have introduced misclassification and contributed to heterogeneity in specificity, particularly for CEA. The latter result suggest that CEA needs to be more deeply evaluated in future studies, since its performances can be significantly different when histological diagnosis is the reference standard, as also reported in a large study on 394 cysts confirmed by histology concluding that CEA at the threshold of 192 ng/mL does not appear as specific as previously reported[46].

Therefore, novel biomarkers as genomic alterations have been explored to reduce the misdiagnosis in an era in which available biomarkers are not enough. The PancreaSeq applies a 74-gene DNA + RNA NGS panel to EUS-FNA cyst fluid and evaluates multiple genomic alterations in a single assay. A combination of alterations from this panel made a score, which identified IPMNs, MCNs and intraductal oncocytic papillary neoplasms with 90% of sensitivity and 100% of specificity when it was ≥ 3[47]. The inclusion of proteomic markers improves the diagnostic accuracy of pre-operative assessments of PCLs, so combining RNA and proteomic analyses with NGS could significantly enhance the accuracy of diagnosing mucinous cysts[48]. Limitations in the use of NGS on cystic fluid include intrinsic features such as epithelial atrophy and reduced cell turnover, which lower sensitivity by leading to decreased epithelial thickness and fewer cells entering the fluid[28,49]. Comparing multiple diagnostic strategies showed that KRAS mutation alone demonstrated a sensitivity of 55% and specificity of 93%, GNAS mutation 39% and 99%, while the combination of KRAS + GNAS reached a sensitivity of 77% and specificity of 93%[50]. Although these molecular assays have the advantage of excellent specificity and can provide additional information, their widespread clinical implementation is limited by high costs, need for specialized laboratories, restricted availability outside academic and research centers and lack of standardized specific testing panels[51]. In contrast, IG testing offers several advantages: It is inexpensive, rapid, and easily performed either at the bedside with a glucometer or in a laboratory setting. Therefore, even if future reductions in the cost and availability of molecular tests may favor their broader use, IG is expected to remain the preferred first-line test due to its diagnostic performance. However, its lower specificity compared with molecular markers suggests that these two approaches may be complementary rather than competitive, with IG serving as an initial screening tool and molecular testing reserved for cases requiring further risk stratification.

The strengths of the present study are its broad dealing with a clinically relevant topic, the rigorous methodology, the comparison of the markers, the exploration of publication bias, and the attempt to consider several factors that might have caused heterogeneity, as the standard reference (histology/cytology) for diagnosing mucinous cysts. Conversely, the limitations of this study must be acknowledged. A proportion of the included studies (n = 8, 50%) had a retrospective design, which may introduce selection bias related to patient referral patterns and cyst sampling, potentially limiting the generalizability of the findings to broader clinical settings. In addition, diagnostic thresholds for IG varied across studies, ranging from < 40 mg/dL to < 87 mg/dL, although the majority of studies (9 of 16) used a cut-off of 50 mg/dL. Differences in reference standards, which have been already discussed, were also present, as only six studies relied exclusively on histology or cytology, while others used composite clinical or imaging diagnoses. These factors may have contributed to heterogeneity in pooled estimates. The very high heterogeneity observed in the specificity of CEA (I² ≈ 99%) likely reflects substantial variability across studies in reference standards, diagnostic thresholds, and the distribution of cyst subtypes. In particular, differences in the use of histology or cytology vs composite or consensus diagnoses may have contributed to inconsistent specificity estimates. The performance of CEA is therefore limited by the inclusion of studies in which IG was used as the index test, while excluding those evaluating CEA alone, due to the aim of this meta-analysis being to focus on IG and its comparison with CEA using paired data. Additionally, some studies lacked detailed information on clinical characteristics or follow-up, limiting their ability to perform more stratified analyses.

CONCLUSION

This updated meta-analysis confirms that IG is a more accurate biomarker for differentiating mucinous from non-mucinous pancreatic cysts compared to CEA due to its high diagnostic performance, reinforcing current guideline recommendations. Although CEA maintains high specificity and remains clinically useful as a rule-in test, its diagnostic metrics underperform compared to IG and are further influenced by many factors, including the diagnostic reference standard (histology/cytology vs other modalities). These findings support the use of IG as a methodologically stable biomarker, particularly suitable as a rule-out test for mucinous cysts. Eventually, even if molecular analyses have the potential to become more affordable and widely accessible in the future, IG is likely to keep a central role in clinical practice because of its rapid turnaround, simplicity of use, and reliable diagnostic performance with genetic testing employed in selected cases. Given its rapid, low-cost measurement and high sensitivity, IG may serve as a first-line diagnostic test, with CEA reserved for selected cases where high specificity is required, such as surgical decision-making.

ACKNOWLEDGEMENTS

All members of the 10^th course of Pancreas 2000 working group.