Clinicopathologic relevance of EpCAM and CD44 in pancreatic cancer: insights from a meta-analysis

Abstract

Recent evidence suggests that EpCAM and CD44 could serve as diagnosis or prognosis markers in pancreatic cancer (PC). In this meta-analysis, we evaluated their associations with clinicopathologic features. Specifically, we compared immunohistochemical-positive and -negative PC patients for T stage (T3-T4 vs. T1-T2), N stage (N1 vs. N0), M stage (M1 vs. M0), tumor grade (well/moderately vs. poorly differentiated), UICC Stage (III, IV vs. I, II), and overall survival (OS). The diagnostic meta-analysis was performed analysing the pooled sensitivity and specificity and evaluating overall accuracy to indicate the diagnostic efficacy of the markers. The protocol of this systematic review and meta-analysis was registered on the PROSPERO website under the registration number of CRD42024568390. A systematic search of PubMed, Scopus, and ISI Web of Science was conducted on January 30th, 2025. The statistical analysis was performed using the Review Manager 5.4 software and R language (R package Mada and Metafor). The quality of the studies included was assessed using the Newcastle-Ottawa scale and the QUADAS-2 tool. Data from relevant studies were independently screened and extracted using Rayyan, by at least two authors. A total of 19 studies were eligible (9 studies for EpCAM, 9 studies for CD44, and 2 studies for both EpCAM and CD44), comprising a total of 1370 patients. The diagnostic meta-analysis demonstrated moderate accuracy for EpCAM (AUC, 95% CI of 0.802, 0.69–0.96). A statistically significant association was found for CD44 expression and T-status (OR = 2.04, 95%CI = 1.18–3.51), or N-stage (OR = 2.68, 95%CI = 1.86–3.85), or TNM stage (OR = 3.79, 95%CI = 2.14–6.71). CD44v6 overexpression predicted worse OS (HR = 2.33, p < 0.00001), while EpCAM + CD44 + co-expression was prognostic (HR = 2.02, p = 0.02). Heterogeneity was not observed among the studies included, but further research is warranted to better understand the clinical implications of these markers’ positivity in PC diagnosis and prognosis.

Introduction

Pancreatic cancer (PC) is among the most aggressive and challenging malignancies, with a dismal prognosis due to late diagnosis, rapid progression, and resistance to therapy. By 2030, PC-related mortality is projected to exceed that of colorectal cancer. Both molecular and genetic alterations have been considered for PC diagnosis or prognosis hoping to achieve significant changes or specific mutations that might aid protein products of mutated genes [1].

Recent research highlights the role of cancer stem cells (CSCs) in driving tumor growth, metastasis, and treatment resistance. This type of cell has mechanisms that regulate self-renewal and differentiation, which are typically disrupted, leading to unchecked self-renewal and an overproduction of CSCs [2]. Furthermore, CSCs exhibit abnormal differentiation processes that produce early tumor cells, which multiply to constitute most of the tumor mass [3]. Different tumor histologic types express key markers such as CD24, CD44, CD133, CD166, and EpCAM. Even though their use in PC has been suggested over time, there is no consensus regarding the potential use of CSCs and their implication in tumor diagnosis or prognosis [4].

Experimental models have suggested that CSC may be nonresponsive to chemotherapy, mostly due to their capacity for DNA repair, ABC transporter expression, and intrinsic detoxifying mechanisms [5]. PC is known for its drug resistance and studies on tumor cell lines have suggested that multidrug resistance may develop and cause therapeutic failure. Along with other types of cancer putative CSCs have been identified in PC mostly based on the expression of three surface markers CD 44, CD24, and CD326 or EpCAM (epithelial-specific antigen, ESA) [6]. Most of the studies available so far on CSC and PC are focused on CD44 and EpCAM. CD44, a transmembrane glycoprotein, acts as a receptor for the extracellular matrix, playing a pivotal role in Epithelial-to-Mesenchymal Transition (EMT), activating ETM-related signaling pathways and is a downstream target of the Wnt/β-catenin pathway, leading to the downregulation of epithelial markers (e.g., E-cadherin) and upregulation of mesenchymal markers (e.g., N-cadherin, vimentin). Its expression has been linked to a more aggressive progression and the occurrence of metastases in PC [6–8]. EpCAM is often overexpressed in various epithelial cancers, including PC, and plays a role in adhesion, proliferation, differentiation, and signaling [9]. EpCAM consists of an extracellular region harboring an EGF-like domain and a thyroglobulin type-1 domain, followed by a single transmembrane helix and a short cytoplasmic tail. Its extracellular region mediates calcium-independent homophilic cell–cell adhesion, contributing to the maintenance of epithelial integrity [10]. Additionally, EpCAM undergoes regulated intramembrane proteolysis, releasing an intracellular domain that translocates to the nucleus and participates in transcriptional regulation of genes involved in cell proliferation and differentiation [11, 12]. These dual roles in adhesion and signaling underline EpCAM’s involvement in tumor growth and progression.

Our objective is to provide an update on available studies of CD44 and EpCAM immunohistochemistry expression and their clinical significance in terms of diagnosis and prognosis of PC. Although their role is currently a subject of debate with contradictory results, their potential impact on disease behavior remains significant. Therefore, we conducted a meta-analysis to provide more insights into their potential as targets in future clinical trials.

Materials and methods

Study selection

Our research was based on the Preferred Reported Items for Systematic Reviews and Meta-analyses (PRISMA) checklist (Supplementary Table) [13]. The protocol of this systematic review and meta-analysis was registered on the PROSPERO website under the registration number of CRD42024568390. We included the articles that met the following criteria: (1) observational studies with PC adult human subjects quantifying EpCAM or CD44 or (EpCAM and CD44), (2) exploring the correlation between CSC markers expression and clinicopathological characteristics of the PC patients, (3) histologically confirmed pancreatic ductal adenocarcinoma, (4) clear definition of immunostaining of CSCs markers. Exclusion criteria were: (1) tumor types other than pancreatic cancer; (2) patients who had undergone a perioperative or neoadjuvant chemo- or radiotherapy; (3) studies as case reports, letters, systematic reviews, abstracts, and animal or in vitro studies.

Search strategy

The search algorithm included the following Boolean search words: ((„Ep-CAM” OR „EPCAM” OR „epithelial cell adhesion molecule” OR „ESA” OR „epithelial cell adhesion molecule”) AND („pancreatic cancer” OR „pancreatic ductal adenocarcinoma”)) OR ((„CD44”) AND („pancreatic cancer” OR „pancreatic ductal adenocarcinoma”)) OR ((„Ep-CAM” OR „EPCAM” OR „epithelial cell adhesion molecule” OR „ESA” OR „epithelial cell adhesion molecule”) AND („CD44”) AND („pancreatic cancer” OR „pancreatic ductal adenocarcinoma”)) using PubMed, Scopus, and ISI Web of Science databases, from inception until January 30th, 2025. The reference lists of the retrieved publications were hand-searched for additional relevant articles. Study screening and selection were performed using Rayyan, a web-based tool designed for systematic reviews and meta-analyses [14]. The process involved: blinded screening (each reviewer independently categorized records as ‘Include,’ ‘Exclude,’ or ‘Maybe,’ with conflicts automatically flagged by Rayyan), and full-text review (screened abstracts were advanced to full-text assessment if marked ‘Include’ or ‘Maybe’ by either reviewer). Two investigators (B.S.U. and A.F.) independently selected the articles using Rayyan and, in case of differences, a third investigator (A.T.-S.) was consulted.

Data extraction

Two investigators (B.S.U. and C-C.V.) independently extracted the TP, FP, FN, and TN rates from the studies we could include in the diagnostic meta-analysis. The number of patients with EpCAM positive and EpCAM negative regarding the characteristics: sex (male/female), grade, pT stage, pN stage, M stage, grade, UICC stage, recurrence, was also extracted. Another outcome assessed in our meta-analysis was overall survival (OS). The disagreements between the two investigators were settled by discussion till an agreement was reached with the third investigator (A.T.-S.). Additionally, we extracted other information from the literature, including the authors’ details, publication year, nation, number of patients, and further information.

Bias and quality assessment

The publication bias was evaluated using the funnel plot. The quality of the included studies was assessed by two authors (B.S.U. and C-C.V.) independently using the Newcastle-Ottawa scale [15]. The QUADAS-2 tool was used to evaluate the studies included in the diagnostic meta-analysis.

Statistical analysis

We performed both a diagnostic meta-analysis and a conventional meta-analysis using Review Manager 5 [16] (RevMan 5, Version 5.4.1, Cochrane Collaboration, 2020) and the R programming language (employing the Mada [17] and Metafor [18] packages, Version 4.1, R Foundation, Vienna, Austria).

The pooled sensitivity and specificity of the included studies were calculated through statistical analysis, with results presented in forest plots. To evaluate the overall predictive performance, a summary receiver operating characteristic (SROC) curve was generated. The area under the ROC curve (AUC) and the normalized partial AUC (calculated within the range of observed false-positive rates and extrapolated across the full ROC space) was determined. An AUC exceeding 0.90 was interpreted as indicative of excellent diagnostic efficacy, while values between 0.80 and 0.90 suggested good diagnostic performance [19]. Diagnostic tests with higher AUC values (approaching 1) were deemed superior, whereas values close to 0.5 indicated minimal diagnostic utility. Additionally, the pooled diagnostic odds ratio (DOR) was estimated to describe the relative likelihood of a positive test result in individuals with PC compared to those without the disease. Correlations between sensitivity and false-positive rates were computed alongside their respective 95% confidence intervals (CIs), allowing for the delineation of a predictive region for future sensitivity-specificity pairs [20]. To assess heterogeneity across studies, the inconsistency index (I²) was calculated, with values exceeding 75% indicating substantial heterogeneity [21]. A bivariate random-effects model was applied in cases of significant heterogeneity; otherwise, a fixed-effects model was utilized.

The standard meta-analysis employed the odds ratio (OR) as the primary effect measure for dichotomous outcomes, such as sex, TNM stage, histology, UICC stage, and recurrence rates. For time-to-event outcomes, such as OS, hazard ratios (HR) were analyzed. In instances where HRs were not directly reported, survival data were extracted from Kaplan-Meier curves using WebPlot Digitizer (Version 4.7, Austin, Texas, USA), as outlined by Tierney et al. [22].

Statistical significance was defined as a P-value < 0.05, with 95% confidence intervals excluding the null value of 1. Heterogeneity was evaluated using the I² statistic, supplemented by the Q-test. To identify outliers and influential studies, studentized residuals and Cook’s distances were analyzed. Studies with studentized residuals exceeding the Bonferroni-corrected threshold (100 × [1–0.05/(2 × k)] percentile of a standard normal distribution, where k represents the number of studies) were flagged as potential outliers. Influential studies were identified if Cook’s distances exceeded the median value plus six times the interquartile range. Funnel plot asymmetry was assessed using rank correlation and regression tests, with the standard error of observed outcomes serving as the independent variable.

Results

Electronic search results and study characteristics

The meta-analysis included 19 studies investigating the clinicopathological relevance of EpCAM and CD44, and their co-expression in PC. The identification and selection of studies for this meta-analysis followed the PRISMA guidelines. For EpCAM, a total of 643 records were identified from databases, with 279 duplicates removed, as shown in Fig. 1. After screening 364 records, 213 were excluded due to irrelevance (e.g., studies on cells, mice, or other syndromes), and 142 were excluded during eligibility assessment (e.g., focus on other cancer types or lack of EpCAM data), resulting in 9 studies included in the meta-analysis [12, 23–30]. For CD44, 1,435 records were identified, with 667 duplicates removed. Screening of 798 records led to the exclusion of 705 records for similar reasons, and 84 were excluded during eligibility assessment, leaving 9 studies for analysis [29, 31–38]. For co-expression of EpCAM and CD44, 148 records were identified, with 79 duplicates removed. After screening 70 records and excluding 53, 15 were excluded during eligibility assessment, resulting in 2 studies being included [39, 40].

Fig. 1.

Study flow PRISMA diagram of the study selection process

These studies were published between 1998 and 2022 and comprise a total of 1370 patients, as shown in Table 1. The studies were conducted across 8 countries, with the majority originating from Asia (Japan, China, Taiwan), followed by Europe (Germany, Austria, Norway) and North America (USA, Mexico). All studies were retrospective in design. The studies focused on the expression of EpCAM (9 studies), CD44 (9 studies), and co-expression of EpCAM and CD44 (2 studies). The sample sizes ranged from 17 patients [29] to 157 patients [28]. The male-to-female ratio across studies was approximately 1.5:1, with males being more frequently represented. Most studies used a scoring system to quantify biomarker expression. One study [33] utilized TissueQuest software for CD44 quantification. Follow-up durations varied significantly across studies, ranging from 1 month to 145 months. The median follow-up periods were reported in several studies, with the longest median follow-up being 45 months [40]. Missing follow-up (measured from the date of surgery or the first medical consultation) was observed in six studies [24, 25, 29, 30, 34, 38]. Nine studies evaluated EpCAM expression with a total of 671 patients. Nine studies evaluated CD44 expression with a total of 532 patients (one study evaluated separately both EpCAM and CD44 expression [29]). Only two studies evaluated EpCAM and CD44 with a total of 239 patients. Newcastle-Ottawa scale (NOS) evaluates selection (4 stars), comparability (2 stars), and outcome (3 stars) of the studies. Studies scoring ≥ 7 were deemed high-quality.

Table 1.

Main characteristics of studies in the meta-analysis

Study (First name, year)	Country	No of patients	Male/ female	Study design	Quantification method	Follow-up	NOS
		Studies included in the meta-analysis regarding expression of EpCAM
Akita 2011 [23]	Japan	95	51/44	Cohort retrospective	Scoring system	median = 17.6mo range = 4.3-145mo	9
Bunger 2012 [24]	Germany	30	13/17	Cohort retrospective	Scoring system	-	8
Fong 2008 [25]	Austria	153	83/70	Cohort retrospective	Scoring System	-	8
Fong 2014 [26]	Austria	88	53/35	Retrospective	Scoring system	range = 1-88mo	8
Kure 2012 [27]	Japan	105	61/44	Retrospective	Scoring System	mean = 17.2mo range = 0.4-153.1mo	9
Meng 2015 [28]	China	157	99/55	Retrospective	Scoring system	87mo	8
Mizukami 2014 [29]	Japan	17	8/9	Retrospective	Scoring System	-	8
Ramirez 2022 [12]	Mexic	25	15/10	Retrospective	Scoring system	range = 1-33mo	9
Yonaiyama 2013 [30]	Japan	51	34/17	Retrospective	Scoring System	-	8
		Studies included in the meta-analysis regarding expression of CD44
Chen 2014 [31]	China	109	71/38	Retrospective	Scoring System	median = 13mo range = 3–46	9
Gotoda 1998 [32]	Japan	42	27/15	Retrospective	Scoring System	mean = 25mo range = 3-72mo	9
Hou 2014 [33]	Taiwan	96	63/33	Retrospective	TissueQuest software	median = 3.7y range = 1.9-12.5y	9
Immervoll 2011 [34]	Norway	51	-	Retrospective	Scoring System	-	8
Li 2014 [35]	China	67	43/24	Retrospective	Scoring System	median = 39mo range = 4-46mo	9
Li 2015 [36]	China	48	30/18	Retrospective	Scoring System	median = 39mo	9
Mizukami 2014 [29]	Japan	17	8/9	Retrospective	Scoring System	-	9
Xiaoping 2015 [37]	China	48	30/18	Retrospective	Scoring System	median = 39mo	9
Zhou 2013 [38]	China	54	27/27	Retrospective	Scoring System	-	9
		Studies included in the meta-analysis regarding co-expression of EpCAM and CD44
Askan 2021 [39]	USA	93	60/33	Retrospective	Scoring System	median = 20mo	9
Bao-Qing 2017 [40]	China	146	90/56	Retrospective	Scoring System	median = 45mo max = 87mo	9

NOS, Newcastle-Ottawa scale. NOS scores range from 0–9, with higher scores (≥ 7) indicating better quality

Quality assessment of selected literature

The QUADAS-2 tool was used to evaluate the risk of bias and applicability concerns for the four studies included in our diagnostic meta-analysis. The studies have unclear or high methodological limitations, particularly in flow/timing because no clear information was provided. With only four studies, sensitivity analyses, excluding high-risk studies, are underpowered to detect significant bias effects. The results are presented in Fig. 2.

Fig. 2.

The QUADAS analysis with bias and applicability concerns across the included studies

Results of the diagnostic meta-analysis

EpCAM

For this analysis, four studies, reporting 326 patients, were included. The utilization of a forest plot for sensitivity and specificity facilitated the evaluation of heterogeneity among the individual aspects of test accuracy. The studies found different sensitivities (χ² = 29.61, df = 3, p < 0.0001) ranging from 0.52 to 0.88. The pooled sensitivity was 0.71 (95% CI, 0.51–0.85). Similarly, differences in specificities were noted among the studies (χ² = 49.20, df = 3, p < 0.0001), ranging from 0.2 to 0.88, yielding a pooled specificity of 0.41 (95% CI, 0.14–0.74). The χ² tests suggested significant heterogeneity of sensitivities and specificities, a finding corroborated by the corresponding forest plots (Fig. 3). The correlation between sensitivities and false positive rates was low (rho = -0.136, 95%CI, -0.97 to 0.95), determined due to the small number of studies.

Fig. 3.

A. The forest plot for sensitivity. B. The forest plot for specificity. Sensitivity and specificity are reported as mean (95% confidence limits)

The analysis revealed a considerable diagnostic efficacy, with an AUC (95% CI) of 0.802 (0.69–0.96) and a partial AUC (95% CI) was 0.78 (0.65–0.95). There was no significant heterogeneity observed among the studies regarding their accuracy indicated by Tau² = 2.78, Cochran’s Q = 124.12, I² = 14.17%, and p = 0.312. DOR (95% CI) was 0.89 (0.64–1.25) with lnDOR − 0.11 (95%CI, -0.45 to 0.22), as depicted in Fig. 4B.

Fig. 4.

A. SROC curve; X-axis: False Positive Rate (1-Specificity); Y-axis: Sensitivity (True Positive Rate). B. Forest plot for univariate meta-analysis using the diagnostic odds ratio logDOR [95% CI]

Correlation between markers expression and T stage

EpCAM

A total of 4 studies were included in the analysis, involving 354 patients. As demonstrated in Fig. 5A, no statistically significant correlation was found between the positivity of EpCAM and pT stage (T3 + T4 vs. T1 + T2) with OR of 1.09 (95% CI 0.56–2.14) and p = 0.79. Due to low heterogeneity (I² = 0%, p = 0.67), a fixed-effects model was fitted to the data. Upon examination of the studentized residuals, it was noted that one study (Akita 2011) had a value exceeding ± 2.4977, suggesting potential outlier status within this model’s context. However, according to Cook’s distances, none of the studies were deemed overly influential. Neither the rank correlation nor the regression test revealed any funnel plot asymmetry (p = 0.7500 and p = 0.2536, respectively).

Fig. 5.

A. Forest plot on the association between EpCAM positivity and T stage (T3 + T4 vs. T1 + T2). B. Forest plot on the association between CD44 positivity and T stage (T3 + T4 vs. T1 + T2)

CD44

A statistically significant association was found for CD44 expression and T-status (OR = 2.04, 95%CI = 1.18–3.51), as shown in Fig. 5B. A total of 4 studies were included in the analysis (144 patients with CD-44 positive and 97 with CD44-negative). The estimated average OR based on fixed-effects model ranged from 1.13 to 4.85 and the average outcome differed significantly from zero (z = 2.57, p = 0.01). According to the Q-test, there was no significant amount of heterogeneity in the true outcomes (Q(3) = 5.75, p = 0.12, I² = 48%). Despite I²=48% for CD44 and T stage, a fixed-effects model was chosen due to limited study numbers (k = 4), which reduces reliability of random-effects estimates due to imprecise tau². The regression test indicated funnel plot asymmetry (p = 0.021), but not the rank correlation (p = 1.0).

Correlation between EpCAM and N

EpCAM

The analysis comprised a total of 4 studies involving 354 patients. As illustrated in Fig. 6A, no statistically significant correlation was observed between EpCAM positivity and pN stage (N1 vs. N0), with an odds ratio (OR) of 0.97 (95% CI 0.60–1.59) and p = 0.91. Due to low heterogeneity (I² = 19%, p = 0.30), a fixed-effects model was applied. Examination of the studentized residuals revealed that none of the studies had a value exceeding ± 2.4977, indicating no presence of outliers within this model’s context. Additionally, according to the Cook’s distances, none of the studies were deemed overly influential. Furthermore, neither the rank correlation nor the regression test revealed any funnel plot asymmetry (p = 0.3333 and p = 0.2803, respectively).

Fig. 6.

A. Forest plot on the association between EpCAM positivity and N stage (yes vs. no). B. Forest plot on the association between CD44 positivity and N stage (yes vs. no)

CD44

A statistically significant association was found for CD44 expression and lymph node status (OR = 2.68, 95%CI = 1.86–3.85), as shown in Fig. 6B. A total of 9 studies were included in the analysis (314 patients with CD-44 positive and 271 with CD44-negative). The estimated average OR based on fixed-effects model ranged from 0.55 to 8.00 and the average outcome differed significantly from zero (z = 5.32, p < 0.00001). According to the Q-test, there was no significant amount of heterogeneity in the true outcomes (Q(8) = 11.14, p = 0.19, I² = 28%). Neither the rank correlation nor the regression test indicated any funnel plot asymmetry (p = 0.275 and p = 0.221, respectively).

Correlation between biomarkers and M

EpCAM

A total of 2 studies were included in the analysis, involving 112 patients. As demonstrated in Fig. 7, no statistically significant correlation was found between positivity of EpCAM and pM stage (M + vs. M-) with OR of 0.37 (95% CI 0.12–1.19) and p = 0.10. A fixed-effects model was fitted to the data because of the low heterogeneity (I² = 0%, p = 0.86). Funnel plot asymmetry was not performed because of the small number of studies.

Fig. 7.

Forest plot on the association between EpCAM positivity and M stage (yes vs. no)

Correlation between biomarkers and grade

EpCAM

A total of six studies, encompassing 529 patients, were included in the analysis. As illustrated in Fig. 8A, no statistically significant correlation was observed between EpCAM positivity and histology (well and moderately differentiated vs. poorly differentiated), with an OR of 0.95 (95% CI, 0.61–1.49) and p of 0.84. We performed a fixed-effect model (I² = 25%, p = 0.24). Examination of the studentized residuals revealed that none of the studies exhibited values larger than ± 2.6383, indicating an absence of outliers within this model. Based on Cook’s distances, none of the studies were deemed overly influential. Furthermore, both rank correlation and regression tests detected no asymmetry in the funnel plot (p = 1.0000 and p = 0.9940, respectively).

Fig. 8.

A. Forest plot on the association between EpCAM positivity and grade (G3 vs. G1 + G2). B. Forest plot on the association between CD44 positivity and grade (G3 vs. G1 + G2)

CD44

No association was found between CD44 expression and tumour differentiation (OR = 1.20, 95%CI = 0.78–1.86), as shown in Fig. 8B. A total of 7 studies were included in the analysis (261 patients with CD-44 positive and 234 with CD44-negative). The estimated average OR based on fixed-effects model ranged from 0.52 to 3.13 and the average outcome did not differ significantly from zero (z = 0.88, p = 0.38). According to the Q-test, there was no significant amount of heterogeneity in the true outcomes (Q(6) = 7.49, p = 0.28, I² = 20%). Neither the rank correlation nor the regression test indicated any funnel plot asymmetry (p = 1.0 and p = 0.543, respectively).

Correlation between biomarkers and UICC stage III, IV vs. I, II

EpCAM

A total of three studies, involving 206 patients, were included in the analysis. As demonstrated in Fig. 9A, no statistically significant correlation was found between EpCAM positivity and UICC stage (III + IV vs. I + II), with an OR of 0.90 (95% CI, 0.39–2.09) and p of 0.81. A fixed-effects model was fitted to the data due to low heterogeneity (I² = 7%, p = 0.34). Examination of the studentized residuals revealed that none of the studies exhibited values larger than ± 2.3940, indicating an absence of outliers in the context of this model. According to Cook’s distances, none of the studies could be considered overly influential. Moreover, neither the rank correlation nor the regression test detected any funnel plot asymmetry (p = 1.0000 and p = 0.1438, respectively).

Fig. 9.

A. Forest plot on the association between EpCAM positivity and UICC stage (III + IV vs. I + II). B. A. Forest plot on the association between CD44 positivity and UICC stage (III + IV vs. I + II)

CD44

Overexpression of CD44 was statistically significantly correlated with TNM stage (OR = 1.28, 95%CI = 0.69–1.87, p < 0.001), as shown in Fig. 9B. A total of 6 studies were included in the analysis (236 patients with CD-44 positive and 233 with CD44-negative). The estimated average OR based on fixed-effects model ranged from 0.31 to 2.45 and the average outcome differed significantly from zero (z = 4.57, p < 0.00001). According to the Q-test, there was no significant amount of heterogeneity in the true outcomes (Q(5) = 4.57, p = 0.54, I² = 0%). Neither the rank correlation nor the regression test indicated any funnel plot asymmetry (p = 0.72 and p = 0.69, respectively).

Correlation between markers expression and OS

EpCAM

Five studies, involving 530 patients, were included in the meta-analysis assessing the HR for OS between patients who were EpCAM positive vs. EpCAM negative. A fixed-effect model was employed due to the absence of heterogeneity among the included studies (I² = 0%, p = 0.90). There was no observed correlation between OS and EpCAM positivity (HR = 1.04, 95% CI, 0.84–1.28, p = 0.72), as shown in Fig. 10A.

Fig. 10.

A. Hazard ratio for OS for patients with EpCAM positive vs. EpCAM negative. B. Hazard ratio for OS for patients with CD44s positive vs. CDs negative. C. Hazard ratio for OS for patients with CD44v6 positive vs. CD44v6 negative. D. Hazard ratio for OS for patients with CD44+/EpCAM- vs. CD44+/EpCAM-. E. Hazard ratio for OS for patients with CD44-/EpCAM- vs. CD44-/EpCAM-. F. Hazard ratio for OS for patients with EpCAM+/CD44 + vs. EpCAM+/ CD44-. G. Hazard ratio for OS for patients with EpCAM-/CD44 + vs. EpCAM-/ CD44-

CD44s

High expression of CD44s is not a predictor for OS, as shown in Fig. 10B. Five studies, involving 530 patients, were included in the meta-analysis assessing the HR for OS between patients who were CD44s positive vs. CD44s negative. A random-effect model was employed due to the presence of heterogeneity among the included studies (I² = 82%, p = 0.0002). There was no observed correlation between OS and CD44s positivity (HR = 1.11, 95% CI, 0.60–2.04, p = 0.74).

CD44v6

Patients with high expression of CD44v6 had significantly worse OS (HR = 2.33, 95% CI, 1.71–3.16, p < 0.00001), as shown in Fig. 10C. Four studies, involving 530 patients, were included in the meta-analysis assessing the HR for OS between patients who were CD44v6 positive vs. CD44v6 negative. A fixed-effect model was employed due to the absence of heterogeneity among the included studies (I² = 30%, p = 0.23).

EpCAM and CD44s

From the four combinations related to the expression of CD44s and EpCAM (Fig. 10D and G), in EpCAM positive patients, the co-expression of positive CD44s was considered a predictor of worse OS (HR = 2.02, 95%CI, 1.09–3.75, p = 0.02). Only two studies were included in the meta-analysis, performing a fixed-effect model due to the absence of heterogeneity among them (p > 0.05).

Discussions

In this meta-analysis, we focused on the relationship between the expression of specific CSC, CD44, EpCAM, and PC pathologic features. While CD44 and EpCAM have been related to the prognosis of different types of cancer, their role in PC is still under debate, as they may open new therapeutic avenues. Our results point out that EpCAM does not correlate with the clinicopathologic characteristics of PC patients, whereas CD44 may have a prognosis pathway concerning the TNM stage. Moreover, CD44v6 was highly correlated with OS and may be considered as a prognostic marker for PC.

EpCAM is a transmembrane glycoprotein that plays a relevant role in adhesion and signaling. It is commonly overexpressed in various epithelial cancers. In pancreatic cancer, EpCAM plays a critical role in regulating epithelial–mesenchymal transition (EMT), maintaining cancer stem cell (CSC) properties, and modulating key oncogenic signaling pathways. EpCAM overexpression has been associated with a partial EMT phenotype, supporting tumor cell plasticity and metastasis, likely through modulation of EMT transcription factors such as ZEB1 and Snail [41]. It also serves as a surface marker for pancreatic CSCs, which are implicated in tumor initiation, therapy resistance, and recurrence [42]. Mechanistically, EpCAM participates in Wnt/β-catenin signaling via its intracellular domain (EpICD), which upon regulated intramembrane proteolysis translocates to the nucleus and interacts with β-catenin/Lef-1 complexes, promoting the transcription of genes involved in self-renewal and proliferation [43].

Elevated EpCAM expression has been associated with tumor aggressiveness, metastasis, and poor prognosis in several cancer types, making it a potential biomarker for disease prognosis and therapeutic targeting. In PC, studies have shown variable expression levels of EpCAM, with some indicating high expression correlating with advanced disease stage and poor prognosis, while others suggest conflicting results. The correlation between EpCAM expression and differentiation grade, T stage, lymph node involvement, and vascular invasion in PC remains an area of active investigation. EpCAM is constitutively expressed in the normal epithelium but exhibits variable overexpression across carcinomas. Generally, EpCAM is considered to have a dual perspective. It acts as an oncogene with its overexpression associated with decreased survival and as a tumor-suppressive protein that may lead to an invasion inhibition. Thus, loss of EpCAM expression may be related to increased migratory potential, and its overexpression may be associated with improved survival. Both properties may be linked to controversial results available for PC and our meta-analysis results reveal no direct correlation with the clinicopathologic features.

CD44, also a transmembrane protein [44], has been linked to various human cell types, from immune to cancer cells [45]. It may be found on the cell surface, and it consists of an ectodomain, a stem region (divided into standard and variable regions), a transmembrane region and an intracellular region [46, 47].

The CD44 gene has been found to generate two splice isoforms, CD44s and CD44v with both exhibiting overlapping or distinct functions in cancer. Typically, CD44s is related to tumor development [48] and progression [49], whereas CD44v3 and CD44v6 are associated with invasiveness [50] and chemoresistance [51]. The distinct or overlapping roles of CD44s and CD44v may vary depending on the tumor type. Furthermore, varying induction conditions, along with the alternative splicing of CD44 isoforms, can result in different characteristics being expressed in tumor cells [52–54]. The role of CD44 in PC has been studied in the past years and several correlations were observed especially regarding lymph node invasion [55, 56]. An interplay may also be found between cancer-associated fibroblasts (CAFs) and the enrichment of stem cells in PC. A recent study revealed that CAFs may interact with tumor cells through a specific axis such as osteopontin or secreted phosphoprotein 1 (OPN/SPP1) - CD44. Silencing this step could lead to a decrease in various stem cells, CSC markers as well as tumor sphere ability [47].

We performed a diagnostic meta-analysis for EpCAM IHC expression, and our results point out that it may be considered a moderate marker to assess PC tissue. This finding may help future studies to provide a rather immunological approach in PC with specific targeted antibodies. On the other hand, when we tried to correlate EpCAM expression with clinicopathologic features our results showed that available studies have not reached a positive correlation. The lack of EpCAM-clinicopathologic correlations may reflect tumor heterogeneity or threshold effects in IHC scoring. However, other meta-analyses showed various results. For example, when considering gastric cancer, larger tumor size, lymph node metastasis, and worse prognosis were correlated with EpCAM overexpression [57]. Furthermore, EpCAM was also considered a worsening factor for hepatocellular patients with a correlation of poorer differentiation as well as high alfafetoprotein levels [58]. However, for lung cancer, the results from 14 studies showed that EpCAM did not correlate with available studies on matters of prognostic, but rather it may be used as a diagnostic indicator when compared with healthy individuals [59].

When correlating EpCAM with the presence of metastasis in PC patients, even though we did not obtain a relevant statistical result, there was a tendency to correlation between EpCAM and its presence. While cancer stem cells may often toggle between mesenchymal and epithelial states, EpCAM expression may facilitate re-epithelialization at the metastatic site, aiding colonization. Moreover, PC tumor heterogeneity is more than relevant for the metastatic stage with possible variability between patients in terms of molecular subtypes. Thus, dynamic EpCAM expression contributes to both metastatic competence and phenotypic switching, key features of heterogeneous tumors.

As confirmed by previous studies we achieved a significant correlation between CD44 and T and N stage. The overexpression in PC suggests its role in carcinogenesis as well as tumor progression, processes that might be mediated by deregulation of the epithelial to mesenchymal transition. Moreover, a study that focused on CD44 expression in early recurrence after PC resection, showed its correlation, thus confirming its role in the tumorigenesis process [60]. CD44 also correlated with the TNM stage, when comparing T1 + T2 vs. T3 + T4, with significant overexpression in advanced stages, thus suggesting a rather important prognostic role. On the other hand, no correlation was found with the degree of tumor differentiation.

In terms of OS, we analyzed both CSC expressions, separately or combined, and also by using two different isoforms of CD44. While EpCAM, CD44s, and EpCAM-CD44- vs. EpCAM-CD44 + did not show any correlation, a slight tendency was observed for CD44 + EpCAM- vs. CD44 + EpCAM + and CD44-EpCAM- vs. CD44-EpCAM+. However, we obtained a significant correlation between OS and the isoform CD44v6 as well as the combined score of EpCAM + CD44+. CD44v6 may play a critical role in tumor aggressiveness by promoting cell motility and invasion, primarily through its interaction with c-MET signaling and cytosckeletal components [61]. Additionally, CD44v6 has been implicated in resistance to gemcitabine and radiotherapy, potentially via the upregulation of anti-apoptotic signaling and drug efflux mechanisms [62]. Co-expression of EpCAM + and CD44 + may define a more refined CSC population with enhanced tumorigenic and metastatic potential. This may also be a strong candidate for future clinical studies with stratification based on double-marker expression. To our knowledge, these are the first published results that confirm their relationship with PC through a meta-analysis.

Our work has some specific limitations. On the one hand, there are limitations to only two CSCs, while other studies included more possible markers. We chose this path based on the included articles in the PRISMA diagram. The EpCAM/CD44 co-expression analysis included only two studies; thus, findings require validation in larger cohorts. Also, the quantification staining in included studies as well as a cut-off value was not well described. The retrospective design of included studies may introduce selection bias. Furthermore, heterogeneity in immunohistochemical (IHC) methodologies and cutoff values between studies limit the comparability of results.

Conclusion

Our results point out that CD44 expression correlates with the T and N stages as well as when comparing T1 + T2 vs. T3 + T4. Even though some studies suggested a possible correlation of EpCAM as a prognostic factor, we only encountered a certain tendency of EpCAM- with the presence of metastasis. However, when combined with CD44+, EpCAM + correlated with OS. Likewise, we showed that isoform CD44v6 expression directly correlated with OS and could be considered as a potential marker for PC prognosis.

Author contributions

All authors contributed to the review conception and design. Bogdan Silviu Ungureanu, Adina Turcu-Stiolica, Cristin-Constantin Vere, and Andrei Fierut were involved in data collection. Bogdan Silviu Ungureanu, Adina Turcu-Stiolica, and Andrei Fierut completed the quality assessment. Adina Turcu-Stiolica performed data analysis. All authors contributed to the analysis and interpretation of data, drafting of the article, revision of the article and final approval of the version to be published.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding authors.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.