This poorly analysed review concluded that cytokeratin stain is reasonably accurate for differentiating Barrett's oesophagus from contiguous tissues. These conclusions are not supported by the data presented and the review suffered from a number of methodological weaknesses, including a limited literature search and failure to adequately assess study quality.

To review the accuracy of cytokeratin (CK) stain to differentiate Barrett's oesophagus (BE) from contiguous tissues including tissues of the gastric cardia, corpus or antrum, with or without intestinal metaplasia (IM).

PubMed was searched from January 1983 to June 2005 for full papers published in English; the search terms were reported. The reference lists of included studies were screened for additional relevant studies.

Diagnostic accuracy studies were eligible for inclusion. Both prospective and retrospective studies were included.

Studies that aimed to assess tissue differentiation using CK stain were eligible for inclusion. The included studies assessed CK 7 and CK 20.

Studies in which BE was defined endoscopically were eligible for inclusion. Reference standards for non-BE groups were not specified. The histological staining methods used to define BE in the included studies were high iron diamine, Alcian Blue and haematoxylin/eosin stains.

Studies had to include one group of patients with BE and a second with a potentially contiguous tissue condition. The BE groups included patients with long-segment BE (LSBE), short-segment BE (SSBE) and undefined BE. The control groups included patients with cardia IM, fundic IM, gastric IM, gastric IMGA, Z-line IM, corpus IM, antrum IM, normal cardia, normal antrum, normal fundus and squamous mucosa.

Studies had to report sufficient data to calculate the sensitivity and specificity to be included in the review. The outcomes reported in the review were the sensitivity, specificity, and positive and negative predictive values.

Two reviewers independently assessed studies for relevance; any disagreements were resolved through referral to a third reviewer.

A formal validity assessment was not undertaken but a number ofmethodological features, including prospective/retrospective sampling, blinding, reference standard and sample size, were discussed. The authors did not report how many reviewers performed this evaluation. The authors did not state how the data were extracted for thereview, or how many reviewers performed the data extraction. The sensitivity, specificity, and positive and negative predictive values were calculated if these values were not reported.

The authors did not state how the data were extracted for thereview, or how many reviewers performed the data extraction. The sensitivity, specificity, and positive and negative predictive values were calculated if these values were not reported.

A narrative synthesis was presented. The studies were grouped according to the control group to which BE specimens were compared. The studies were classified as being positive or negative depending upon whether a statistically significant difference in CK staining pattern was found between BE and non-BE specimens.

Differences between the studies were discussed and effects of study design and quality were investigated.

Sixteen studies reporting 46 comparisons were included in the review (1,521 patients, 1,556 specimens). Some studies reported data separately for different subtypes of BE, for different observers, and separately for patients who were positive and negative for Helicobacter pylori.

Only 4 studies were prospective. Nine studies were reported to have used a blinded pathologist to interpret samples. The blinded studies showed higher, less heterogeneous sensitivity (range: 81 to 100%) and specificity (range: 61 to 100%) than the non-blinded studies.

BE and cardia IM (12 studies, 18 comparisons).

The sensitivity ranged from 10 to 100% and the specificity from 22 to 100%. For studies that reported data for LSBE (8 studies), the sensitivity ranged from 17 to 100%. For studies that reported data for SSBE (5 studies), the sensitivity ranged from 10 to 95%.

BE and corpus IM (3 studies, 3 comparisons).

The sensitivity ranged from 10 to 94% and the specificity from 96 to 100%. The study with the lowest sensitivity was a study of SSBE.

BE and antrum IM (6 studies, 9 comparisons).

The sensitivity ranged from 17 to 94% and the specificity from 71 to 100%. All but two of the comparisons evaluated LSBE or BE length was not stated. For the three comparisons that assessed SSBE, the sensitivity ranged from 30 to 94%.

BE and gastric IM (7 studies, 11 comparisons).

The sensitivity ranged from 27 to 100% and the specificity from 85 to 100%. For studies that reported data for LSBE (4 studies), the sensitivity ranged from 94 to 100%. For studies that reported data for SSBE (4 studies), the sensitivity ranged from 27 to 95%.

BE and non-metaplastic tissues (2 studies, 3 comparisons).

The sensitivity ranged from 45 to 100% and the specificity from 15 to 45%. One study did not report on BE segment length, while the other reported data separately for LSBE and SSBE. This study found that LSBE was less sensitive (45%) than SSBE (81%).

Immunostaining for CK 7 or CK 20 is reasonably accurate for differentiating BE, especially LSBE, from antrum IM, fundus IM and noncardiac gastric IM. The accuracy of CK in distinguishing BE, especially SSBE, from gastric cardia IM is more conflicting.

The review addressed a focused question that was supported by clearly defined inclusion criteria. The literature search was limited to one electronic database and studies published in English. It is therefore likely that relevant studies have been missed and the review may be subject to language and publication bias. Appropriate steps were taken to minimise errors and bias when assessing studies for inclusion, but the authors did not report whether such steps were also taken during the other stages of the review. A formal quality assessment was not undertaken, although the reviewers discussed a small number of methodological features. A more detailed assessment of the quality of the included studies using established criteria would have helped to determine the reliability of these studies.

Some study details were tabulated but these were somewhat confusing as the columns were not clearly labelled and abbreviations were not defined. The synthesis of the results was not appropriate for diagnostic data. The authors summarised the number of studies that showed a significant difference in staining patterns for samples with and without BE and then only discussed accuracy data for these studies. A more appropriate analysis would have focused on accuracy data for all studies and investigated possible explanations for the very large variations seen across the studies, especially in estimates of sensitivity. The authors' conclusions are not supported by the data presented, which show very large variations in accuracy across the included studies.

Practice: The authors did not state any implications for practice.

Research: The authors stated that further studies are required to examine the utility of CK staining in distinguishing SSBE from cardia IM. Future studies should use a diagnostic cohort design and enrol consecutive patients.

NIH, grant number R21 DK 067366-02; VA HSRD, award RCD00-013-2.

This record is a structured abstract written by CRD reviewers. The original has met a set of quality criteria. Since September 1996 abstracts have been sent to authors for comment. Additional factual information is incorporated into the record. Noted as [A:....].