Meta-analysis: Barrett's oesophagus and the risk of colonic tumours

Authors

  • J. Andrici,

    1. The Whiteley-Martin Research Centre, The Discipline of Surgery, The University of Sydney, Sydney Medical School, Penrith, NSW, Australia
    Search for more papers by this author
  • M. Tio,

    1. The Whiteley-Martin Research Centre, The Discipline of Surgery, The University of Sydney, Sydney Medical School, Penrith, NSW, Australia
    Search for more papers by this author
  • M. R. Cox,

    1. The Whiteley-Martin Research Centre, The Discipline of Surgery, The University of Sydney, Sydney Medical School, Penrith, NSW, Australia
    Search for more papers by this author
  • G. D. Eslick

    Corresponding author
    • The Whiteley-Martin Research Centre, The Discipline of Surgery, The University of Sydney, Sydney Medical School, Penrith, NSW, Australia
    Search for more papers by this author

  • As part of AP&T's peer-review process, a technical check of this meta-analysis was performed by Dr Y. Yuan.

Correspondence to:

Prof. G. D. Eslick, The Whiteley-Martin Research Centre, Discipline of Surgery, The University of Sydney, Nepean Hospital, Level 5, South Block, Penrith, NSW 2751, Australia

E-mail: eslickg@med.usyd.edu.au

Summary

Background

Barrett's oesophagus (BO) is a premalignant condition associated with oesophageal adenocarcinoma. Although speculation exists, it is currently unclear if BO is associated with an increased risk of colonic tumours.

Aim

To conduct a meta-analysis of studies reporting the prevalence of colonic tumours in patients with BO vs. controls and thus quantify the risk of colonic tumours associated with BO.

Methods

A search was conducted through Medline, PubMed, Embase, and Current Contents Connect to 7 October 2012. We calculated pooled odds ratios (OR) and 95% confidence intervals (CI) using a random-effects model for the risk of all colonic tumours associated with BO, as well as for the subgroups of colorectal cancer (CRC) and benign adenomatous tumours.

Results

In total, 11 studies, with 2580 BO cases, met our inclusion criteria. BO was associated with an increased risk of any colonic tumours (OR: 1.96; 95% CI: 1.56–2.46). BO was associated with an increased risk of benign adenomatous tumours (OR: 1.69; 95% CI: 1.20–2.39), as well as an increased risk of CRC (OR: 1.90; 95% CI: 1.35–2.67). No statistically significant heterogeneity was observed. Publication bias was not present.

Conclusions

Barrett's oesophagus was associated with an increased risk of both benign adenomatous colonic tumours and colorectal cancer. Barrett's oesophagus had a stronger association with colorectal cancer than with benign colonic tumours. Further prospective cohort studies are needed to confirm the relationship.

Introduction

Barrett's oesophagus (BO) is a premalignant condition associated with oesophageal adenocarcinoma.[1, 2] It involves the replacement of the normal squamous oesophageal lining by specialised or intestinal columnar epithelium.[3, 4] In North America, BO is diagnosed when endoscopically observed columnar metaplasia is confirmed to contain specialised intestinal epithelia (characterised by the presence of goblet cells) on histological examination.[5]

The possibility of an association between BO and an increased risk of colonic tumours was first raised by Sontag et al.[6] in 1985. This is a clinically significant question because if a relationship is found, it carries implications in terms of screening BO patients for colorectal cancer (CRC). Since then, several studies have reported conflicting results, and the association is not well established. A systematic review in 1995[7] showed a strong association; however, we questioned the authors' methodology, including the use of a synthetic control group constructed by the authors, which appears to have been used in calculating the risk estimates for the studies analysed. In addition, several new studies have been published since the time of the last systematic review reporting the prevalence of colonic tumours in patients with BO and therefore we considered it appropriate to perform a systematic review that also incorporated these studies.

Our aim was to perform a meta-analysis combining the results of studies reporting the prevalence of colonic tumours in BO vs. controls, and thus provide a quantitative estimate of the risk of colonic tumours associated with BO.

Materials and Methods

Study protocol

We followed the Meta-Analysis of Observational Studies in Epidemiology (MOOSE) guidelines,[8] where possible, in performing our systematic review. A systematic search was performed by two reviewers (J.A. and M.T.) through MEDLINE (from 1950), PubMed (from 1946), EMBASE (from 1949) and Current Contents Connect (from 1998) through to 7 October 2012, to identify relevant articles. The search used the terms ‘Barrett's Esophagus’ OR ‘Barrett's Oesophagus’ AND ‘colorectal cancer’ OR ‘colon cancer’ OR ‘rectal cancer’ OR ‘colonic tumours’ OR ‘colonic tumours’ OR ‘colonic neoplasms’, which were searched as text word and as exploded medical subject headings where possible. The reference lists of relevant articles were also searched for appropriate studies. No language restrictions were used in either the search or study selection. A search for unpublished literature was not performed.

Study selection

We included studies that met the following inclusion criteria: (i) the study examined the prevalence of either benign (adenomas) or malignant (CRC) colonic tumours, or both, in BO patients and controls, (ii) the cases were patients diagnosed with BO and the controls were patients without BO, (iii) the risk point estimate was reported as an odds ratio (OR), or the data were presented such that an OR could be calculated, (iv) the 95% confidence interval (CI) was reported, or the data were presented such that the CI could be calculated, (v) an internal comparison was used when calculating the risk estimate. We excluded studies that did not meet the inclusion criteria. Specifically, four studies[9-12] were excluded as they included patients with oesophageal adenocarcinoma to serve as either the cases or controls; three studies[13-15] were excluded as they used an external comparator group and standardised incidence ratios; and one study[16] was excluded as it used an external comparator group made up of groups of patients from previous studies. Studies were included or exclusion following consensus among three authors (J.A., M.T. and G.E.).

Data extraction

We performed the data extraction via a standardised data extraction form, collecting information on the publication year, study design, number of cases, number of controls, total sample size, temporal direction, population type, country, continent, economic development, case–control matching, mean age, number of adjusted variables, the risk estimates or data used to calculate the risk estimates, CIs or data used to calculate CIs, and the type of colonic tumour investigated (benign or malignant). Quality of the studies was not assessed and authors were not contacted for missing data. Adjusted ratios were extracted in preference to non-adjusted ratios; however, where ratios were not provided, unadjusted ORs and CIs were calculated. Where more than one adjusted ratio was reported, the ratio with the highest number of adjusted variables was selected. Where multiple risk estimates were available in the same study, for example, studies providing risk estimates for both malignant and benign tumours, they were included as separate risk estimates. Where studies provided only the risk ratio (RR) or hazard ratio (HR), we assumed that the RR or HR would be similar to the OR and thus the RRs and HRs provided were combined with the provided or calculated ORs.

Statistical analysis

We calculated pooled odds ratios and 95% confidence intervals for the effect of BO on the risk of any colonic tumours, as well as on benign colonic tumours and CRC, using a random-effects model.[17] For the relationship between BO and any colonic tumours, we performed subgroup analyses based on studies which adjusted for any variables and specifically for the important confounder of body mass index (BMI), as well as subgroup analyses by temporal study direction (prospective vs. retrospective). We also analysed the effect of the four[18-21] studies included in our analysis which were in abstract form, and for comparison, also performed an analysis of the three studies[13-15] which were excluded based on their use of an external comparator group.

Heterogeneity was tested with Cochran's Q statistic, with P < 0.10 indicating heterogeneity, and quantified the degree of heterogeneity using the I2 statistic, which represents the percentage of the total variability across studies, which is due to heterogeneity. I2 values of 25%, 50% and 75% corresponded to low, moderate and high degrees of heterogeneity respectively.[22] Publication bias was quantified using the Egger's regression model.[23] All analyses were performed with Comprehensive Meta-analysis (version 2.0).

Results

Study characteristics

From 1351 studies initially identified, 11[6, 18-21, 24-29] met our inclusion criteria (Figure 1), of which four[18-21] were abstracts. Selected characteristics of the included studies are presented in Table 1. The studies were all conducted in developed Western countries, with five studies examining European populations, and the remaining six studies examining North American populations. In terms of study design, one study was a prospective cohort study, and the remainder were retrospective studies. Sample sizes ranged from 96 to 15 093, and BO cases ranged from 32 to 1677. Overall, there were 2580 BO patients and 27 272 participants.

Table 1. Studies reporting prevalence of any colonic tumours in BO patients which were included in the meta-analysis
AuthorsYearTemporal direction (prospective or retrospectiveCountryCasesControlsAdjusted variablesCase-control matchingNumber of CasesTotal SizeColonic tumour investigated
  1. BO, Barrett's oesophagus; IBS, irritable bowel syndrome; CRC, colorectal cancer; GORD, gastro-oesophageal reflux disease; GI, gastrointestinal; BMI, body mass index; PPI, proton pump inhibitor; NSAID, nonsteroidal anti-inflammatory; GP, general practitioner; NA, not available.

Cauvin et al.[24] a1995RetrospectiveFranceBO patientsConsecutive patients with symptoms suggestive of IBSAge, gender, family history of CRC, rectal bleedingNone104641Benign (adenomas)
Cauvin et al.[24] b1995RetrospectiveFranceBO patientsConsecutive patients with symptoms suggestive of IBSAge, gender, family history of CRC, rectal bleedingNone104641Malignant
Elli et al.[18]2010RetrospectiveItalyBO patientsPatients who underwent esophagogastroduodenoscopy and colonoscopy.Not statedNANA1018Benign (colon polyps
Gerson et al.[25]2002RetrospectiveUSBO patients free from GORD symptoms, who were undergoing sigmoidoscopy for CRC screeningPatients free from GORD symptoms, who were undergoing sigmoidoscopy for CRC screeningAgeNone44110Benign (colon polyps)
Laitakari et al.[26]1995RetrospectiveFinlandBO patientsPatients referred to hospital for benign thyroid, inguinal hernia or hand surgery, who had been attending urological or vascular examinationsNoneNone7299Benign (adenomas)
Limburg et al.[19]1994RetrospectiveUSBO patientsPatients with peptic ulcer disease and gastric polypsAge, gender, time between first colonoscopy and oesophagogastro-duodenoscopy 1758922Malignant (CRC at any site)
Lyons et al.[20]1993RetrospectiveUSBO patientsPatients with gastro-oesophageal reflux with strictureNoneNone99153Both benign and malignant
Robertson et al.[27] a1989RetrospectiveUKBO patientsPatients clinically diagnosed IBSNoneAge and gender3296Benign (polyps and adenomas)
Robertson et al.[27] b1989RetrospectiveUKBO patientsPatients clinically diagnosed IBSNoneAge and gender3296Malignant
Robertson et al.[27] c1989RetrospectiveUKBO patientsPatients clinically diagnosed IBSNoneAge and gender3296Both benign (polyps and adenomas) and malignant
Rothstein et al.[21]1991RetrospectiveUSBO patientsPatients undergoing upper and lower endoscopic surveillance in the workup of an iron deficiency anaemiaNoneNone4499Benign polyps
Siersema et al.[28] a2006RetrospectiveUSBO patientsPatients who had had undergone upper GI endoscopy within 14 days of the corresponding caseAge, BMI, other malignancies, use of PPIs, use of aspirin/NSAIDS, alcohol consumption, smoking statusDate of endoscopy268536Both benign and malignant
Siersema et al.[28] b2006RetrospectiveUSBO patientsPatients who had had undergone upper GI endoscopy within 14 days of the corresponding caseAge, BMI, other malignancies, use of PPIs, use of aspirin/NSAIDS, alcohol consumption, smoking statusDate of endoscopy268536Malignant
Solaymani-Dodaran et al.[29]2004ProspectiveUKBO patientsPatients on the GP database with no restriction other than not having BEAge, gender, number of visits per year, smoking, alcohol and BMIAge, gender and GP practice167715 093Malignant
Sontag et al.[6] a1985RetrospectiveUSBO patientsPatients who underwent colonoscopy for occult blood in stool, weight loss, rectal bleeding, or abdominal painNoneNone65505Benign
Sontag et al.[6] b1985RetrospectiveUSBO patientsPatients who underwent colonoscopy for occult blood in stool, weight loss, rectal bleeding, or abdominal painNoneNone65505Malignant
Sontag et al.[6] c1985RetrospectiveUSBO patientsPatients who underwent colonoscopy for occult blood in stool, weight loss, rectal bleeding, or abdominal painNoneNone65505Both benign and malignant
Figure 1.

Study selection flowchart.

Any colonic tumours

Eleven studies[6, 18-21, 24-29] comprising of 2580 BO cases, reported an association between colonic tumours (either benign adenomas or CRC) and BO and were included in the analysis. We found an increased risk of colonic tumours in patients with BO, with pooled OR of 1.96 (95% CI: 1.56–2.46) (Figure 2). There was low heterogeneity, which was not statistically significant (I2 = 15%, P = 0.295). There was no publication bias (P = 0.520), and this was depicted visually on a funnel plot in Figure 3.

Figure 2.

Meta-analysis of BO and any colonic tumours.

Figure 3.

Funnel plot to assess publication bias.

We performed sensitivity analyses to assess whether adjustment for confounding variables changed the overall risk estimate. When looking at the five studies[19, 24, 25, 28, 29] which adjusted for any variables, the risk of colonic tumours was 1.91 (95% CI: 1.48–2.46), with no heterogeneity (I2 = 0%, P = 0.591). The six studies[6, 18, 20, 21, 27] that did not adjust for any variables showed a risk of colonic tumours of 2.05 (95% CI: 1.29–3.26), with moderate heterogeneity, which was not statistically significant (I2 = 45%, P = 0.103).

The risk of colonic tumours calculated from the two studies[28, 29] which adjusted for BMI was statistically significant (OR 1.89; 95% CI: 1.30–2.76), while that calculated from the remaining ten studies, which did not adjust for BMI, was 1.89 (95% CI: 1.30–2.76), with no heterogeneity (I2 = 0%, P = 0.379).

When analysing the ten retrospective studies,[6, 18-21, 24-28] the OR was 2.01 (95% CI: 1.58–2.54), with low heterogeneity, which was not statistically significant (I2 = 17%, P = 0.275). The only prospective cohort study had a risk estimate of 1.23 (95% CI: 0.44–3.43).

The pooled risk estimate after removing the four studies[18-21] in abstract form was 1.91 (95% CI: 1.34–2.70). There was moderate heterogeneity, which was not statistically significant (I2 = 34%, P = 0.161).

The meta-analysis including the three excluded studies[13-15] which used an external comparator group yielded a positive association with an OR of 1.67 (95% CI: 1.35–2.07), with moderate heterogeneity (I2 = 50%, P = 0.013). The risk estimate calculated from only the three excluded studies[13-15] was not significant (OR: 1.13, 95% CI: 0.63–2.02). There was a high degree of heterogeneity (I2 = 84%, P = 0.002).

Benign colonic tumours

Seven studies[6, 18, 21, 24-27] comprising of 361 BO cases and a total of 2568 individuals reported an association between benign colonic tumours and BO and were included in the benign colonic tumours meta-analysis. There was an increased risk of benign colonic tumours in patients with BO, with pooled OR of 1.69 (95% CI: 1.20–2.39) (Figure 4). We found low, statistically insignificant heterogeneity (I2 = 13%, P = 0.449).

Figure 4.

Meta-analysis of BO and benign colonic tumours.

Colorectal Cancer

Six studies[6, 19, 24, 27-29] comprising of 2321 BO cases and 25 793 individuals were included in the meta-analysis for CRC. We found an increased risk of CRC in patients with BO, with pooled OR of 1.90 (95% CI: 1.35–2.67) (Figure 5). There was no heterogeneity (I2 = 0%, P = 0.452).

Figure 5.

Meta-analysis of BO and malignant colonic tumours.

Discussion

The meta-analyses showed a statistically significant positive relationship between BO and colonic tumours. This association was stronger for CRC than for benign tumours, although an exact mechanism explaining the association between BO and colonic tumours has not yet been established.

A previous systematic review and meta-analysis by Howden and Hornung[7] found a strong association between BO with specialised columnar epithelium and CRC, with an OR of 8.71. We questioned these authors' use of a ‘comparison cohort’, which was constructed ‘from previously published studies of colorectal cancer screening in the general population’. In addition, the ORs which they used in the meta-analysis were not the ORs which were published in the original studies, and we were unsure as to how these numbers were obtained. In some instances, the ORs used by Howden and Hornung[7] were larger than the ORs published in the original studies by a factor of four to five. Since their publication, several investigators have disagreed with these large risk estimates.[16, 30, 31] Our meta-analysis includes four studies,[18, 25, 28, 29] which were published subsequent to the Howden and Hornung[7] study and therefore were not included in that study. Although like Howden and Hornung,[7] we also find a statistically significant association between BO and colonic tumours, there are major issues with the interpretation of our results.

First, only one of the studies included in the analysis, Solaymani-Dodaran et al.,[29] was a cohort study which looked at disease incidence over the study period, and it reported a positive but statistically insignificant association (OR: 1.23; 95% CI: 0.44–3.43). A sensitivity analysis excluding the Solaymani-Dodaran et al.[29] study yielded a risk estimate of 2.01 (95% CI: 1.58–2.54), with low heterogeneity. The remainder of the studies followed a retrospective study design and therefore were subject to the limitations and biases inherent in studies of that nature. This raises the question as to whether our overall positive result merely reflects biases inherent to retrospective studies, or whether it represents a real association. To ascertain whether a true association between BO and colonic tumours exists, there is a need for more large, prospective studies. The ideal study type would thus be a cohort study that recruited patients with BO and a control group representative of the general population with no colonic tumours at baseline, and observed the incidence of colonic tumours in the two groups over the study period. One problem with such a study would be the ethical issue of performing colonoscopies, which are invasive and carry risks of complications, on young asymptomatic individuals in the absence of any indications. Studies could, however, be carried out in patients aged over 50, in whom colonoscopies are recommended as a screening tool in many countries, including the USA.[32, 33] Prospective cohort studies would also be able to establish the existence of a temporal relationship, should one exist, between BO and colonic tumours, which would then have implications for screening.

Our study was also subject to the likely presence of confounders in the individual studies. While we used the adjusted ORs where available, six of the eleven studies did not adjust for confounders, or did not state if such adjustments had been made. This raises the possibility that any observed association could be due to confounders that have not been adjusted for. In the subgroup of four studies that reported adjustments for possible confounders, we still found a significant association between BO and colonic tumours, with no significant change from the unadjusted result. However, of those studies that used adjusted ORs, only two studies adjusted for BMI, which is known to be a risk factor for both BO,[34] and CRC,[35, 36] although the strength of the association between BMI and CRC varied with gender and cancer site in both Moghaddam et al.[35] and Larsson et al.[36] Additionally, some studies[37, 38] found the association of BMI with BO to be insignificant when adjusting for waist-to-hip ratio or waist circumference, which suggests that this association is driven mostly by central adiposity. Our subgroup analysis of the studies that adjusted for BMI (OR: 1.89) vs. those that did not adjust for BMI (OR: 1.80) did not show any appreciable difference. Solaymani-Dodaran et al.,[29] the only prospective cohort study in our meta-analysis, did not find a significant association between BO and CRC, but they reported a number of different risk estimates after adjusting for different factors. The risk estimate that included BMI was not significantly different from the risk estimate that excluded BMI, with the BMI adjusted risk at 1.23 (95% CI: 0.44–3.43), and the non-BMI adjusted risk at 1.14 (95% CI: 0.41–3.18), which suggests that in their study, adjustment for BMI did not play a significant role in the risk estimate.

There may also exist a referral or diagnostic bias with respect to BO and colonic tumours. This arises as both are gastrointestinal disorders and may be subject to investigation by the same physician upon the patient presenting with nonspecific gastrointestinal symptoms and signs such as anaemia, which may warrant both upper and lower gastrointestinal endoscopies. One study by Murphy et al.,[15] which looked at the risk of CRC associated with BO but was excluded from our analysis because it did not use an internal comparator group, found that the standardised incidence ratio of CRC rose progressively as the follow-up period approached the time of BE diagnosis, thus raising the possibility of diagnostic bias.

Even though we found no statistical heterogeneity in our meta-analyses, studies with differing designs and methodologies were included in the analysis. Specifically, we looked at one cohort study and ten retrospective studies, with four of the studies being in the form of abstracts. There existed also differences in the control groups between studies. Some of the patients were asymptomatic, while others were being investigated for gastrointestinal symptoms related to irritable bowel syndrome and even rectal bleeding. This may impact the risk estimates and add to the diagnostic bias discussed above, as well as contribute to heterogeneity.

Additionally, our meta-analysis only comprised eleven studies, with the subgroup analyses of CRC and benign adenomas comprising six and seven studies respectively. All but one of the studies were retrospective in nature. Adjusted ORs were only available in five studies, and of those, only two adjusted for the confounder of BMI.

For comparison, we performed an analysis based on the three studies[13-15] which used an external comparator group. The studies reported conflicting results, with de Jonge et al.[14] reporting a positive, statistically significant association, and neither Murphy et al.[15]nor Cook et al.[13] finding an association. The overall pooled risk estimate was positive, but not statistically significant. This analysis is only based on only three studies which reported conflicting results, so while it is difficult to make a meaningful definitive comment on the impact their exclusion had on our results, it does not appear to have greatly influenced our result. The analysis with these studies included yielded a risk estimate of 1.67, which, while lower than the risk estimate of 1.96 obtained with the studies excluded, is still positive and still statistically significant.

Our analysis included four studies which were published in abstract form, and to see if these had a significant impact on our overall results, we performed a subgroup analysis excluding the abstracts. No appreciable difference was found when removing these studies from the pooled risk estimate for any colonic tumours (OR: 1.91 with the abstracts removed vs. OR: 1.96), with the results maintaining statistical significance. This suggests that the abstracts did not have a large impact on the overall results.

Notwithstanding the current lack of an established mechanism to explain the relationship between BO and colonic tumours, the existence of a positive association between the two as ascertained by our meta-analysis warrants a call for more large cohort studies to elucidate whether the relationship is a real one and not a result of bias. Should the association then be shown to be real, this discovery would carry a number of important implications. First, an established association will warrant a search for common genetic or environmental risk factors as well as more studies in basic science to establish a mechanism for and thus provide a better understanding of the association. Secondly, it will raise the important clinical question as to whether BO patients should be regularly screened for CRC, which several investigators[19, 24, 25] have considered unwarranted due to inconclusive data, a view which we agree with at present. The CRC risk estimate in our study, 1.88 (95% CI: 1.32–2.68), is comparable to the increased risk of CRC in first degree relatives of patients with CRC of 2.24 (95% CI: 2.06–2.43), as reported in a recent meta-analysis by Butterworth et al.[39] Most of the studies included in that meta-analysis reported risk estimates between 1.5 and 4, which again is comparable to the association we found between BO and CRC. In addition, a more recent Italian study Castiglione et al.,[40] found the risk of CRC in first degree relatives of CRC patients to be 1.53 (95% CI: 1.27–1.83), also comparable to our risk estimate. If the risk estimates for CRC in patients with BO reflects a real relationship, this risk for CRC is similar to that found in first degree relatives of patients with CRC, and serious consideration may need to be given in the future to screening BO patients for CRC.

Our study had a number of strengths. The MOOSE guidelines were followed where possible. We performed a thorough search through four databases with no language restrictions. Studies using external comparators were excluded. The use of an internal control group is recognised as superior in terms of study design,[41-43] and by excluding studies that used external comparators, we added statistical rigour to our analysis. Additionally, we observed no statistically significant heterogeneity in any of our analyses, and publication bias was not present.

In summary, our results suggest that BO is associated with an increased risk of colonic tumours. The association was present for both benign and CRC, but was stronger for CRC. More prospective cohort studies adjusting for possible confounders are needed to further elucidate this relationship. At present, we recommend against screening BO patients for CRC due to the lack of robust prospective evidence supporting this association.

Authorship

Guarantor of the article: Juliana Andrici.

Author contributions: Juliana Andrici and Guy D. Eslick were responsible for the study concept and design. Juliana Andrici and Martin Tio handled acquisition of data. Juliana Andrici, Martin Tio and Guy D. Eslick were responsible for analysis and interpretation of data. Juliana Andrici drafted the manuscript. Juliana Andrici, Martin Tio, Guy D. Eslick and Michael Cox performed critical revision of the manuscript for important intellectual content. Juliana Andrici and Guy D. Eslick performed statistical analysis. Guy D. Eslick supervised the study. All authors approved the final version of the manuscript.

Acknowledgement

Declaration of personal and funding interests: None.

Ancillary