The Diagnostic Accuracy of Claudin-4 Immunochemistry in Differentiating Metastatic Carcinomas From Mesothelial Processes in Serous Effusion Cytology: A Systematic Review and Meta-analysis Open Access

This study followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines for reporting systematic reviews and meta-analyses of diagnostic test accuracy studies.^18,19  The PubMed, Scopus, and Cochrane Library databases were searched for available articles, using the following search string: “claudin-4” AND (pleural OR peritoneal OR ascitic OR pericardial OR effusion* OR mesothelioma* OR mesothelial). The latest date of search was October 9, 2023. We did not apply any filters regarding the publication date or article type. All references were inserted into the Paperpile reference tool, and duplicates were subsequently removed.

Study selection was performed independently by 3 authors (M.K., M.A., I.P.N.) in a blinded fashion, and all disagreements were resolved with a consensus among 4 authors (M.K., M.A., A.I., I.P.N.). As a first step, screening of titles and abstracts was done with the Rayyan tool (https://rayyan.ai/)²⁰ ; this was followed by a full-text evaluation of all eligible articles derived from the first step. Original studies reporting claudin-4 immunochemistry findings in serous effusion cytology were included. The following exclusion criteria were applied: reviews, letters, commentaries, conference abstracts, case reports and small case series of fewer than 50 cases in total, articles in languages other than English, in vitro and animal studies, studies reporting results in body sites other than serous effusions, studies comprising overlapping cases with larger series published from the same center, studies reporting results on cancer types other than carcinomas (eg, sarcomas or melanomas), and studies with claudin-4 levels measured with ancillary techniques other than immunochemistry (eg, polymerase chain reaction) or material other than cytology (eg, pleural biopsies or resection specimens).

Data extraction was performed independently by 2 authors (M.K., I.P.N.) in a blinded fashion, and all disagreements were solved with a consensus among 3 authors (M.K., A.I., I.P.N.). Data on the following variables were extracted from all eligible articles: first author and publication year, country, study period, study type, sample type used for immunochemistry (eg, smears, liquid-based cytology, or cell blocks), type of metastatic carcinomas included (eg, non–small cell lung cancer), effusion location (pleural, pericardial, or peritoneal), total number of carcinoma cases and controls (RMCs or mesothelioma) analyzed, claudin-4 antibody type (eg, 3E2C1), cutoff used to interpret claudin-4 immunochemistry, number of cases with a positive claudin-4 immunochemistry result in carcinomas and controls, and number of cases with a negative claudin-4 immunochemistry result in carcinomas and controls. Data regarding malignancies other than carcinomas and mesotheliomas, for instance melanomas or sarcomas, were not extracted. Carcinoma and control cases with a positive claudin-4 result were considered as true positive and false positive, respectively. Likewise, carcinoma and control cases with a negative claudin-4 result were considered as false negative and true negative, respectively.

Study quality assessment was performed by using the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) tool on the following 4 domains: study selection, index test, reference test, and flow and timing (Supplemental Figure 1, see the supplemental digital content containing 1 table and 5 figures at https://meridian.allenpress.com/aplm in the April 2025 table of contents).²¹ 

Statistical analysis was performed with R, version 4.3.1 (R Foundation for Statistical Computing). For each study, the univariate sensitivity and specificity were calculated and jointly pooled with the random effects bivariate binomial model, as proposed by Chu and Cole.²²  The model was applied to estimate the hierarchical summary receiver operating characteristic (ROC) parameters and to draw the summary ROC plot, using the equivalence equations as presented by Harbord et al.²³  To estimate the pooled diagnostic odds ratio (DOR)—representing the odds ratio of a positive result indicating a carcinoma rather than a mesothelial process—a random effects model was applied, using the inverse variance method, the restricted maximum-likelihood estimator for the between-study variance τ², and a continuity correction of 0.5 in studies with zero or unity cell frequencies. Subgroup analyses were additionally performed for the following variables: control type, effusion type, year of publication, and immunochemistry cutoff. For one study where the study population could not be completely reconstructed from the aggregated data,²⁴  a sensitivity analysis was performed by using a conservative and an optimistic scenario. Lastly, a funnel plot was designed to investigate potential publication bias, according to the method proposed by Deeks et al²⁵  for systematic reviews of diagnostic test accuracy.

The flowchart of this study is shown in Figure 1. The initial database search resulted in 187 studies. Following the removal of duplicates, 124 studies were inserted into the Rayyan tool (https://rayyan.ai/)²⁰  and screened in a title-abstract fashion, resulting in 30 studies for full-text evaluation. Of these, 16 studies were further excluded (Figure 1), leaving 14 studies for data extraction.

Fourteen studies were included in this review, published between 2011 and 2023 (Table 1). Of these, 3 were performed in the United States^13,24,26 ; 2 in Italy^27,28 ; and 1 each in Thailand,²⁹  Egypt,³⁰  Brazil,³¹  Spain,³²  Australia,³³  South Korea,³⁴  India,³⁵  Iran,³⁶  and Japan.³⁷  Claudin-4 immunochemistry was performed on cell blocks in 13 studies and on smears in 1 study.²⁸  While all 14 studies used the 3E2C1 clone for claudin-4 immunochemistry, the cutoff for interpreting its results substantially differed among them. Lastly, 7 studies used RMCs only^{29–32,34–36}  and 1 used mesothelioma only²⁶  as their control, whereas the rest used both RMCs and mesotheliomas.

Supplemental Table 1 shows the data used for the diagnostic test accuracy meta-analysis. Figure 2 shows the forest plot of the univariate sensitivity and specificity, and Figure 3 shows the summary ROC plot of this study. Claudin-4 immunochemistry exhibited a high diagnostic accuracy to detect metastatic carcinoma in serous effusion cytology. The pooled sensitivity and specificity were 98.02% (95% CI, 93.96%–99.37%) and 99.72% (95% CI, 97.36%–99.97%), respectively. In addition, the pooled DOR was 1660.5 (95% CI, 760.0–3627.8), showing no evidence of heterogeneity (I² = 0%, τ² = 0; Figure 4). Likewise, no evidence for publication bias was found in this study (P = .18; Figure 5).

Claudin-4 immunochemistry was positive in most cases with metastatic carcinoma included in this systematic review (Table 2). For instance, 353 of 355 lung adenocarcinomas (99.44%), 155 of 156 breast carcinomas (99.36%), 115 of 116 gastric carcinomas (99.14%), and 50 of 52 colorectal carcinomas (96.15%) were positive. Notably, only 3 of 6 (50%) of the total hepatocellular carcinoma (HCC) cases were claudin-4 positive (Table 2). Regarding controls, staining was present in only 1 of 266 mesothelioma³³  and 14 of 531 RMC^30,31,34  cases included in this review.

Subsequently, to investigate the diagnostic accuracy of claudin-4 immunochemistry in differentiating metastatic carcinomas from RMCs or mesotheliomas separately, analysis was performed for both scenarios. The summary ROC plots are shown in Supplemental Figure 2, while the forest plots of specificity are shown in Figure 6. When RMCs were used as a control (Supplemental Figure 2, A), the pooled sensitivity and specificity were 97.90% (95% CI, 93.39%–99.35%) and 99.76% (95% CI, 95.55%–99.99%), respectively. In the scenario where mesotheliomas were used as a control, the pooled sensitivity and specificity were 97.83% (95% CI, 94.66%–99.13%) and 99.93% (95% CI, 67.52%–100%), respectively (Supplemental Figure 2, B).

Subgroup analyses were also performed for the effusion type, year of publication, and immunochemistry cutoff and did not reveal any significant subgroup differences (Supplemental Figures 3 through 5). Notably, regarding the immunochemistry cutoff, we did not see great DOR variations between the included studies (Supplemental Figure 5). When any intensity and percentage of claudin-4 staining was used (Supplemental Figure 5), the pooled DOR was 1073.2 (95% CI, 253.6–4542.0).

Lastly, we performed a sensitivity analysis in one of the included studies where the study population could not be completely reconstructed from the aggregated data.²⁴  More specifically, this study presented data from 211 effusion and 18 fine-needle aspiration (FNA) cases (claudin-4 immunochemistry diagnostic performance on FNA cytology was not part of our review question). Whereas the data from 14 of 18 FNAs diagnosed as mesotheliomas or RMCs were readily excluded from the analysis, this was not the case for 4 of 18 FNAs diagnosed as carcinomas. For this reason, we ran a sensitivity analysis where the 4 cases were subtracted either from the true-positive (assuming they were all claudin-4 negative; conservative scenario) or false-negative (assuming they were all claudin-4 positive; optimistic scenario) cell of the initial 2 × 2 table of this study (Supplemental Table 1). Pooled sensitivity and specificity in both conservative (98.02% [95% CI, 93.95%–99.37%] and 99.72% [95% CI, 97.36%–99.97%]) and optimistic (98.12% [95% CI, 94.29%–99.40%) and 99.72% (95% CI, 97.33%–99.97%)] scenarios revealed that these 4 FNA cases had a negligible impact on the pooled meta-analysis results.

This meta-analysis showed that claudin-4 immunochemistry exhibits a high diagnostic accuracy in terms of distinguishing metastatic carcinomas from mesothelial cell populations (RMCs and mesotheliomas) in serous effusion cytology. More specifically, the pooled sensitivity and specificity were calculated to be 98.02% (95% CI, 93.96%–99.37%) and 99.72% (95% CI, 97.36%–99.97%), respectively. Apart from claudin-4, other immunomarkers are often applied in routine practice to detect metastatic carcinoma in malignant serous effusions, including MOC-31, B72.3, and BerEP4.^3,9,10  Notably, claudin-4 has previously been shown to exhibit a higher diagnostic accuracy than MOC-31, B72-3, and BerEP4 in effusions,^13,33,37  yet no meta-analysis has been performed to date.

Our review revealed that claudin-4 is a sensitive pan-carcinoma marker, highlighting the presence of metastatic carcinomas derived from various primary sites such as the lung, breast, gastrointestinal system, and ovaries (Table 2). Of interest, there was a difference regarding claudin-4 immunopositivity in lung carcinomas when all histologic subtypes were considered (465 of 483; 96.27%), compared to adenocarcinomas alone (353 of 355; 99.44%). This could be explained by a potentially lower claudin-4–positive rate in histologic subtypes other than adenocarcinoma. For instance, 5 of 20 metastatic small cell carcinomas (25%) in one included study were claudin-4 negative.³³  However, all 4 tested small cell carcinoma cases from 2 other studies^28,31  were positive. Regarding metastatic lung squamous cell carcinomas, Najjar et al²⁴  reported 1 of their 2 included cases as claudin-4 negative, whereas claudin-4 showed positivity in all 4 cases tested in the studies of Bernardi et al²⁷  and Lonardi et al.²⁸  Nevertheless, in a few other research studies using tissue rather than cytology samples, claudin-4 staining has been reported as negative in a substantial number of squamous cell or sarcomatoid carcinomas.^38–41  Notably, a metastatic ovarian small cell carcinoma of the hypercalcemic type was claudin-4 negative in one of the included studies²⁸ ; this finding is consistent with the results of another study,⁴²  where staining was absent in all 10 cases tested. Regarding renal cell carcinomas, one study addressed that claudin-4 staining performed on tissue sections was positive in 91% of the tested renal cell carcinomas; of these, 28 of 33 clear cell carcinomas (85%) were positive.⁴³  In our review, 15 of 16 (93.75%) of the included renal cell carcinomas were positive (Table 2). HCC and adrenocortical carcinomas have often been reported to be claudin-4 negative^44–46 ; in our review, staining was absent in 3 of 6 (50%) of the total HCC cases (Table 2), while no metastatic adrenocortical carcinoma case was tested. In contrast to carcinomas, claudin-4 shows negativity in melanomas and most sarcomas.^33,39,42  For instance, all 8 metastatic melanoma cases were negative in one of the included studies.³³ 

Apart from being a sensitive immunochemical biomarker of metastatic carcinomas in serous effusion cytology, claudin-4 was also shown to be a specific biomarker. Only 1 of 266 of the included mesothelioma³³  and 14 of 531 of the RMC^30,31,34  cases were claudin-4 positive. Eleven of the positive RMC cases came from a single study,³⁴  testing a total of 40 RMC cases; in 8 cases, the immunomarker showed positivity in 1% to 5% of the target cells, and in 3 cases, in 6% to 25% of the target cells, whereas in 29 cases it showed negativity. In the scenario where RMCs were used as a control in our meta-analysis, the pooled sensitivity and specificity of claudin-4 immunomarker were 97.90% (95% CI, 93.39%–99.35%) and 99.76% (95% CI, 95.55–99.99%), respectively (Supplemental Figure 2, A). When mesotheliomas were used as a control, the pooled sensitivity and specificity were 97.83% (95% CI, 94.66%–99.13%) and 99.93% (95% CI, 67.52%–100%) (Supplemental Figure 2, B). Of interest, in a recent study by Naso and Churg³⁸  using tissue microarrays, claudin-4 staining was negative in all epithelioid (n = 68) and sarcomatoid (n = 31) mesotheliomas tested.

In routine practice, cytologists often apply 2 epithelial and 2 mesothelial markers to differentiate carcinomas from mesotheliomas or RMCs.^12,13  Our study showed that claudin-4 may be used as a single carcinoma marker in effusions, given its high diagnostic accuracy. Notably, Bernardi et al²⁷  previously proposed the use of the “Brescia panel,” composed of the claudin-4 and BAP-1 immunostains, to differentiate among metastatic carcinomas, mesotheliomas, and RMCs with high accuracy in histology and cytology preparations.

To reduce the possibility of several biases (eg, publication or reporting bias)^47,48  and similar to other meta-analyses relevant to cytopathology,^49,50  we only included studies with more than 50 cases in our protocol. Claudin-4 immunochemistry was shown to have a high diagnostic accuracy in this meta-analysis, while no statistical evidence of publication bias (P = .18) was found. However, this study is not without limitations. Firstly, all included studies were observational and of retrospective nature. In the quality assessment (Supplemental Figure 1), most studies showed a high or unclear risk of bias in the “study selection” domain. For instance, many used a case-control design, which is known to artificially increase diagnostic accuracy.⁵¹  In addition, patients enrolled in most studies did not receive the same reference standard (eg, clinical and radiologic data for RMCs and histology for malignancies), having a high bias risk in the “flow and timing” domain of the QUADAS-2 tool (Supplemental Figure 1). Furthermore, the meta-analysis model to pool the DOR estimated τ² to be 0, leading to the common effect model. This could result in CIs that are too narrow, as clinical heterogeneity between studies cannot be excluded. The included studies also used various immunochemistry cutoffs, yet we did not see great DOR variations between the included studies (Supplemental Figure 5). Lastly, only studies written in the English language were included in this review.

In conclusion, this is the first meta-analysis regarding the diagnostic accuracy of claudin-4 immunochemistry in serous effusion cytology. Claudin-4 may be used as a single pan-carcinoma immunomarker to differentiate carcinomas from mesotheliomas or RMCs in effusions. However, although we report a high pooled sensitivity and specificity of this biomarker, caution is needed given the limitations of this study. Consequently, future studies with a prospective cohort design could strengthen the reported findings.