Meta-analysis: coeliac disease and the risk of all-cause mortality, any malignancy and lymphoid malignancy

Authors

  • M. Tio,

    1. The Whiteley-Martin Research Centre, The Discipline of Surgery, The University of Sydney, Sydney Medical School, Nepean, 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, Nepean, 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, Nepean, 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. P. Collins.

Correspondence to:

Associate Professor 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: guy.eslick@sydney.edu.au

Summary

Background

Coeliac disease has been associated with an increased risk of mortality and malignancy. However, the strength of this association is conflicting among different studies.

Aim

To perform a systematic review and quantitative meta-analysis to determine the risk of all-cause mortality, any malignancy and lymphoid malignancy in coeliac disease patients.

Methods

Four electronic databases (Medline, PubMed, Embase and Current Contents Connect) were searched to 4 January 2012, with no language restrictions. From 8698 citations identified, a total of 17 studies met our inclusion criteria.

Results

The all-cause mortality meta-analysis showed an increased risk for all-cause mortality in coeliac patients [odds ratio (OR) 1.24; 95% confidence interval (CI) 1.19–1.30]. A subgroup analysis showed that patients identified by positive serology alone were also at an increased risk of all-cause mortality (OR 1.16; 95% CI 1.02–1.31). The non-Hodgkin lymphoma (NHL) meta-analysis showed an increased risk for NHL in coeliac patients (OR 2.61; 95% CI 2.04–3.33). A subgroup analysis showed that patients identified by positive serology alone were also at an increased risk of NHL (OR 2.55; 95% CI 1.02–6.36). The T-cell non-Hodgkin lymphoma (TNHL) meta-analysis showed an increased risk of TNHL (OR 15.84; 95% CI 7.85–31.94). The any malignancy meta-analysis showed no increased risk (OR 1.07; 95% CI 0.89–1.29).

Conclusions

Patients with coeliac disease are at an increased risk of mortality and non-Hodgkin lymphoma, particularly T-cell non-Hodgkin lymphoma; they do not have an increased risk of any malignancy overall. Serologically defined patients with coeliac disease have an elevated risk of mortality and non-Hodgkin lymphoma.

Introduction

Coeliac disease is a common autoimmune disorder, with a prevalence approaching 1% in the United States.[1, 2] Although it is classically associated with malabsorption and the attendant complications that can arise from the malabsorptive state, it has also been associated with an increased risk of lymphoid malignancies.[3] This association is particularly clear in the case of a specific non-Hodgkin lymphoma (NHL) subtype, enteropathy-associated T cell lymphoma (EATL), which has been long established as a complication of coeliac disease.[4]

The magnitude of the risk of other forms of lymphoid malignancies is much less clear. Data on lymphoid malignancy risk other than NHL are scant, and the risk estimates for the more commonly reported NHL vary significantly.[5] Similarly, whether coeliac disease increases the overall risk of malignancy or not remains unknown, with large discrepancies in reported estimates.[6, 7] Coeliac disease has also been associated with an increased risk of all-cause mortality, although again, the estimates vary widely, ranging from no association to a 3.9-fold increased risk.[2, 8]

Given this uncertainty, and as no systematic evaluation of these risks has been performed to date, we carried out a systematic review and quantitative meta-analysis of published studies on coeliac disease that reported risk estimates for either all-cause mortality, any malignancy or any lymphoid malignancy.

Methods

Search strategy

We followed the Meta-Analysis of Observational Studies in Epidemiology (MOOSE) guidelines where possible in performing our systematic review.[9] Relevant articles were identified by one reviewer (M.T.) by systematically searching through MEDLINE (from 1950), PubMed (from 1946), EMBASE (from 1949) and Current Contents Connect (from 1998) through to 4 January 2012. The search used the terms ‘coeliac disease’ OR ‘coeliac disease’ AND ‘cancer’ OR ‘neoplasms’ OR ‘lymphoma’ OR ‘death’ OR ‘mortality’ OR ‘survival’. The search terms used 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

To be included, eligible studies needed to: (i) have a study design of either a cohort or case control; (ii) report the risk of all-cause mortality, any specific mortality, any malignancy or any lymphoid malignancy in coeliac patients; (iii) report the risk point estimate as an odds ratio (OR), hazard rate or relative risk; (iv) report the 95% confidence interval (CI) for the point estimate; (v) use an internal comparison when calculating the risk estimate. Studies were excluded if they did not meet the inclusion criteria. We chose to only include studies that used an internal comparison, as there is evidence suggesting that observational studies using external comparisons may not estimate the risk as accurately as those that use an internal comparison.[10]

Data extraction

One reviewer (M.T.) performed the data extraction via a standardised data extraction form. Information was extracted on the publication year, study design, number of cases, number of controls, total sample size, temporal direction, population type, country, continent, economic development, response rate, case control matching, mean age, number of adjusted variables, years of follow up, method of coeliac diagnosis and the risk estimates on mortality, overall malignancy and lymphoid malignancy subtypes. Quality of the studies was not assessed. Authors were not contacted if data were missing. Adjusted ratios were extracted in preference to non-adjusted ratios. Where more than one adjusted ratio was reported, the ratio with the highest number of adjusted variables was chosen.

Statistical analysis

Pooled estimates of the OR and 95% CI for coeliac disease and the risk of mortality, NHL, T cell non-Hodgkin lymphoma (TNHL), and any malignancy were calculated using the random-effects model of DerSimmonian and Laird.[11] Heterogeneity was assessed using the I2 statistic, which determines the proportion of variability across studies that is due to heterogeneity as opposed to sampling error.[12] Sensitivity analyses were performed when statistically significantly heterogeneity was detected. Subgroup analyses stratified by either study design or by diagnosis were also performed.

Publication bias was assessed using Egger's regression model.[13] If publication bias was detected, the additional publication bias methods consisting of the fail-safe number method and the trim-and-fill method were employed to quantify the effect of the bias. The fail-safe number method calculates the number of unpublished studies needed to convert the observed result to statistical non-significance at the alpha level of significance P < 0.05 level. Publication bias is considered to be an issue if the fail-safe number is <5n + 10, where n is the number of studies included in the meta-analysis.[14] The trim-and-fill method simulates unpublished studies in the meta-analysis to calculate a new pooled odds ratio, which is then compared with the original pooled odds ratio. If the new pooled OR is similar to the original pooled odds ratio, this indicates that publication bias has little effect on the meta-analysis results. Results were regarded as statistically significant if P < 0.05. All analyses were done with Comprehensive Meta-analysis (version 2.0; Biostat, Englewood, NJ, USA).

Results

From 8698 citations screened by our search, we identified a total of 17 observational studies that met our inclusion criteria (Figure 1). Table 1 shows selected characteristics of the identified studies.[2, 8, 15-29] All studies examined populations from developed Western countries, with 12 studies examining populations from 10 European countries and the remaining five studies examining the USA population. In terms of study design, nine were case control studies and eight were cohort studies.

Figure 1.

Study selection flowchart.

Table 1. Studies included in the systematic review
SourceStudy typeCountryCasesTotal sizePopulation derivationPopulationMajor outcomes measuredCoeliac or cancer casesCoeliac diagnosis method
  1. AGA, antigliadin antibody; CLL, chronic lymphocytic leukaemia; celikey-tTG, celikey tissue-transglutaminase antibody; DLBL, diffuse large cell lymphoma; EMA, endomysial antibody; Eu-tTG, Eurospital tissue-transglutaminase antibody; HL, Hodgkin lymphoma; MM, multiple myeloma; NHL, non-Hodgkin lymphoma; TNHL, T-cell non-Hodgkin lymphoma; tTG, tissue-transglutaminase antibody.

Anderson et al.[15]Case controlUSA33 721156 252PopulationAdults aged 67–99 with medicare coverage without HIV infectionNHL, TNHL, DLBL, CLL, MMCancerMedicare claim (medical record)
Ansell et al.[16]Case controlUK7911663PopulationAdults aged 18–69NHL, TNHL, DLBL, CLL, MMCancerPrimary care records (medical record)
Brown et al.[17]CohortUSA46414 501 578HospitalAdult male veterans of either black or white ethnicityMMCancerHospital discharge record (medical record)
Canavan et al.[29]CohortUK877527PopulationAdults aged 45–76All-cause Mortality, Cancer Mortality, Cardiovascular MortalityCoeliacEMA
Catassi et al.[18]Case controlItaly6536373Hospital cases; population controlsOver 20 without AIDS diagnosisNHL, TNHLCancerPatient interview, or positive EMA plus positive histopathology
Delco et al.[19]Nested case controlUSA4583150HospitalUS veteransLymphomaCoeliacHospital discharge record (medical record)
Elfstrom et al. 1a[28]CohortSweden28 989173 209PopulationSwedish residents without lymphoproliferative malignancy prior to study entryNHL, TNHL, HL, CLL, MM, lymphoproliferative malignancyCoeliacHistopathology
Elfstrom et al. 1b[28]CohortSweden371118 449PopulationSwedish residents without lymphoproliferative malignancy prior to study entryNHL, TNHL, HL, CLL, MM, Lymphoproliferative malignancyCoeliacAGA or tTG or EMA
Farre et al.[20]Case controlSpain298549HospitalSubjects without HIV infection or systemic immunosuppresive treatmentLymphomaCancerPatient interview, or EMA and/or histopathology
Gao et al.[21]Case controlSweden37 869187 609PopulationUnselectedNHL, HL, CLLCancerMedical record
Godfrey et al.[8]Nested case controlUSA127381PopulationAdults aged over 50All-cause mortality, cancer mortality, any malignancyCoeliactTG plus EMA
Lohi et al. 1a[22]CohortFinland746987PopulationAdults aged 30–99, without previously diagnosed and treated coeliac disease or dermatitis herpetiformisAll-cause mortality, cancer mortality, cardiovascular mortalityCoeliacEu-TTG plus EMA
Lohi et al. 1b[22]CohortFinland2046987PopulationAdults aged 30–99, without previously diagnosed and treated coeliac disease or dermatitis herpetiformisAll-cause mortality, cancer mortality, cardiovascular mortalityCoeliacEu-TTG plus Celikey-tTG
Lohi et al. 2a[23]CohortFinland736849PopulationAdults aged 30 and over, without any malignancy at the time of serum draw, and without previously diagnosed and treated coeliac disease or dermatitis herpetiformisNHL, any malignancy, lymphoproliferative diseaseCoeliacEu-tTG plus EMA
Lohi et al. 2b[23]CohortFinland2026849PopulationAdults aged 30 and over, without any malignancy at the time of serum draw, and without previously diagnosed and treated coeliac disease or dermatitis herpetiformisNHL, any malignancy, lymphoproliferative diseaseCoeliacEu-tTG  + plus Celikey-tTG
Ludvigsson et al. 1a[24]CohortSweden29 096257 728PopulationUnselectedAll-cause mortality, cancer mortality, cardiovascular mortalityCoeliacHistopathology
Ludvigsson et al. 1b[24]CohortSweden3719232 351PopulationUnselectedAll-cause mortality, cancer mortality, cardiovascular mortalityCoeliacAGA or tTG or EMA
Mearin et al.[27]Case Controlthe Netherlands, Italy, UK, Spain, Ireland, Sweden, Finland, France, Poland, Yugoslavia118110 836Hospital cases; hospital and population controlsAdults over 18 without HIV or primary immunodeficiencyNHLCancerPrevious diagnosis or EMA plus histopathology
Rubio et al.[2]CohortUSA149133PopulationAdult malesMortalityCoeliactTG plus EMA
Smedby et al.[25]Case controlDenmark and Sweden30556242PopulationAdults aged 18–79, Danish or Swedish language speaker, without HIV infection, history of organ transplantation, or previous haematological malignancyNHL, TNHL, DLBLCancerPatient interview
West et al.[26]CohortUK473228 352PopulationUnselectedAll-cause mortality, any malignancy, lymphoproliferative diseaseCoeliacMedical Record

There was heterogeneity in the method of coeliac disease diagnosis, with six studies using medical records, one study using patient interview, three studies using patient interview or positive serology plus positive histopathology and two studies using positive histopathology. Latent or undiagnosed coeliac disease was used as a coeliac disease subtype diagnosis in seven studies, with two studies using positive eurospital tissue-transglutaminase antibodies (Eu-tTG) plus positive endomysial antibodies (EMA) or positive celikey tissue-transglutaminase antibodies (celikey-tTG), two studies using positive tTG plus positive EMA, two studies using positive tTG or positive EMA or positive antigliadin antibody and two studies using positive EMA alone.

To avoid the possibility of ascertainment bias, risk estimates that excluded an outcome diagnosis within the first year of risk factor diagnosis were used when available in preference to overall risk estimates. Figure 2 shows the pooled OR for all-cause mortality; Figure 3 shows the pooled OR for NHL; Figure 4 shows the pooled OR for TNHL; Figure 5 shows the pooled OR for any malignancy. Table 2 shows the pooled ORs for other lymphoid malignancy risks and cause specific mortality.

Figure 2.

Meta-analysis of the association between coeliac disease and all-cause mortality. Test for heterogeneity P = 0.44, I2 = 0.00%. Each study is shown by an odds ratio estimate with the corresponding 95% CI.

Figure 3.

Meta-analysis of the association between coeliac disease and non-hodgkin lymphoma. Test for heterogeneity P = 0.23, I2 = 23.3%. Each study is shown by an odds ratio estimate with the corresponding 95% CI.

Figure 4.

Meta-analysis of the association between coeliac disease and T-cell non-hodgkin lymphoma. Test for heterogeneity P = 0.06, I2 = 56.6. Each study is shown by an odds ratio estimate with the corresponding 95% CI.

Figure 5.

Meta-analysis of the association between coeliac disease and any malignancy. Test for heterogeneity P = 0.58, I2 = 0.00. Each study is shown by an odds ratio estimate with the corresponding 95% CI.

Table 2. Pooled estimates of cause specific mortality and lymphoid malignancies
OutcomePooled study referencesCasesTotal sizeRisk estimateHeterogeneity
Cardiovascular mortality22, 24, 2933 180275 9611.19 (95% CI 1.12–1.27)I2 = 0%, P = 0.79
Any malignancy mortality8, 22, 24, 2933 307276 3421.24 (95% CI 0.96–1.60)I2 = 51.8%, P = 0.07
Lymphoma19, 2075636991.99 (95% CI 0.29–13.58)I2 = 73.9%, P = 0.05
Lymphoproliferative disease/malignancy23, 26, 2837 707230 5702.53 (95% CI 1.59–4.04)I2 = 50.8%, P = 0.09
Hodgkin lymphoma21, 2837 312214 6922.01 (95% CI 1.01–4.01)I2 = 0%, P = 0.48
Diffuse large cell lymphoma15, 16, 2513 990140 5802.25 (95% CI 1.32–3.85)I2 = 0%, P = 0.62
Chronic lymphocytic leukaemia15, 21, 2851 984373 2430.80 (95% CI 0.46–1.38)I2 = 0%, P = 0.69
Multiple myeloma15, 17, 2846 8154 828 9521.26 (95% CI 0.83–1.90)I2 = 0%, P = 0.64

All-cause mortality

Six prospective studies[2, 8, 22, 24, 26, 29] (five cohort studies, one nested case control study) comprising of 38 053 cases with a total of 313 827 individuals were initially identified for the all-cause mortality meta-analysis. Coeliac patients were at an increased risk of all-cause mortality, with a pooled OR of 1.22 (95% CI, 1.08–1.38). There was significant heterogeneity (I2 = 59.4%, P = 0.02). A sensitivity analysis showed that one study[2] was contributing significantly to the heterogeneity, possibly due to an exceptionally long follow-up period of 45 years. In comparison, the mean follow-up of the five other studies was 13.4 years. Removal of this study did not significantly affect the pooled OR.

With removal of the heterogeneous study, Figure 2 shows the all-cause mortality meta-analysis of five prospective studies[8, 22, 24, 26, 29] comprising of 38 039 coeliac cases with a total of 304 694 individuals, indicating that coeliac patients are at an increased risk of all-cause mortality with a pooled OR of 1.24 (95% CI 1.19–1.30). There was no significant heterogeneity (I2 = 0.0%, P = 0.44), and no significant publication bias (Egger's test P = 0.81). Using risk estimates that included the first year after diagnosis gave a pooled OR of 1.29 (95% CI 1.17–1.42), with no significant heterogeneity (I2 = 48.7%, P = 0.07).

In a subgroup analysis of biopsy proven coeliac cases vs. other methods of diagnosis (serology or medical record), there was no substantial change to the mortality risk. The biopsy proven OR came from one study[24] which was 1.26 (95% CI 1.20–1.32), with the pooled OR of all other forms of diagnosis at 1.19 (95% CI 1.08–1.30), with no significant heterogeneity (I2 = 0.0%, P = 0.47). In a subgroup analysis of four studies[8, 22, 24, 29] using serological diagnosis only, the risk was decreased, but remained significant with an OR of 1.16 (95% CI 1.02–1.31). There was no significant heterogeneity (I2 = 7.9%, P = 0.36). In a subgroup analysis of study design, removal of the single nested case control study did not significantly change the results, with the pooled results of four cohort studies[22, 24, 26, 29] giving an OR of 1.25 (95% CI 1.20–1.30). There was no significant heterogeneity (I2 = 0.0%, P = 0.62).

Non-Hodgkin lymphoma

Eight studies[15, 16, 18, 21, 23, 25, 27, 28] (six case control studies and two cohort studies) comprising of 110 245 cancer or coeliac cases with a total of 538 493 individuals were identified for the non-Hodgkin lymphoma (NHL) meta-analysis. Coeliac patients were at significantly increased risk of NHL, with a pooled OR of 2.75 (95% CI 2.0–3.78). There was significant heterogeneity (I2 = 54.9%, P = 0.02). There was no publication bias (Egger's test P = 0.61). Sensitivity analysis showed that one study[21] contributed significantly to the heterogeneity. This study was the only case control study in the analysis that examined NHL cases diagnosed prior to 1993. While the overall OR was 5.35 (95% CI 3.56–8.06), when stratified according to the decade of diagnosis, the ORs were progressively higher than the earlier the diagnosis. For the period of 1995–2004, the OR was 3.84 (95% CI 2.28–6.45). As all of the other case control studies only recruited NHL diagnoses from 1993 or later, this OR was examined in the meta-analysis. This resulted in a pooled OR of 2.61 (95% CI 2.04–3.33), with no significant heterogeneity (I2 = 23.3%, P = 0.23) (Figure 3). There was no publication bias (Egger's test P = 0.77).

In a subgroup analysis stratified by study design, the removal of the two cohort studies did not substantially change the risk. The case–control only OR was 2.52 (95% CI 1.78–3.55), with no significant heterogeneity (I2 = 31.6%, P = 0.20), whereas the cohort only OR was 2.71 (95% 1.73–4.37) with no significant heterogeneity (I2 = 25.1%, P = 0.26).

In a subgroup analysis of two studies[23, 28] stratified by serological diagnosis only, the risk remained significantly elevated with an OR of 2.55 (95% CI 1.02–6.36), and with no significant heterogeneity (I2 = 44.2%, P = 0.17). The non-serological diagnosis OR of seven studies[15, 16, 18, 21, 25, 27, 28] was 2.65 (95% CI 2.06–3.41) with no significant heterogeneity (I2 = 24.6%, P = 0.24).

T-cell non-Hodgkin lymphoma

Five studies[15, 16, 18, 25, 28] (four case–control studies, one cohort study) comprising of 35 358 cancer cases with a total of 311 888 individuals were identified for the TNHL meta-analysis. The risk of TNHL was significantly elevated, with a pooled OR of 15.84 (95% CI 7.85–31.94) (Figure 4). There was no significant heterogeneity (I2 = 55.6%, P = 0.06). There was no publication bias (Egger's test P = 0.70). Sensitivity analysis with the removal of the study with the highest OR[28] (OR of 48) reduced the overall OR, but it still remained substantially elevated at 12.12 (95% CI 6.70–21.91), with no significant heterogeneity (I2 = 25.8%, P = 0.26).

Any malignancy

Three prospective studies[8, 23, 26] (two cohort studies, one nested case–control study) comprising of 5134 cancer or coeliac cases with a total of 35 582 individuals were identified for the any malignancy meta-analysis. There was no association between coeliac disease and the risk of any malignancy, with a pooled OR of 1.07 (CI 0.89–1.29) (Figure 5). There was no significant heterogeneity (I2 = 0%, P = 0.58). There was no publication bias (Egger's test P = 0.49).

Cause specific mortality and other lymphoid malignancies

As shown in Table 2, we found that coeliac patients had an increased risk of cardiovascular mortality 1.19 (95% CI 1.12–1.27), lymphoproliferative disease/malignancy 2.53 (95% CI 1.59–4.04), Hodgkin lymphoma 2.01 (95% CI 1.01–4.01) and diffuse large cell B-lymphoma 2.25 (95% CI 1.32–3.85). There was no increased risk of any malignancy mortality 1.24 (95% CI 0.96–1.60), lymphoma 1.99 (95% CI 0.29–13.58), chronic lymphocytic leukaemia 0.80 (95% CI 0.46–1.38) or multiple myeloma 1.26 (95% CI 0.83–1.90). However, all of these results are based on pooling together only two or three studies.

Discussion

Our systematic review and meta-analysis shows that coeliac patients are at an increased risk of mortality, and are at a substantially increased risk of NHL and TNHL. Serologically defined coeliac patients are at a similarly increased risk of both NHL and mortality. However, coeliac patients do not have an increased risk of any malignancy overall.

The all-cause mortality results were highly heterogeneous when every eligible study was combined. The sensitivity analysis revealed that the study by Rubio-Tapia et al.[2] was the main contributor to the heterogeneity, with a substantially higher OR when compared with the other studies: 3.9 (95% CI 2.0–7.5). We determined three hypotheses to explain the heterogeneity created by this study: (i) The study was a retrospective cohort study that analysed the stored serum of US Air Force personnel which had been drawn between 1948 and 1954. This gave the study a follow-up length of approximately 45 years, which was significantly longer than the average follow up of 13.4 years of the other studies. Death among the coeliac serology positive patients did not occur until an average of 29 years after the serum draw, indicating that the lower mortality risk found in the other studies may be due to insufficient follow-up time to detect mortality; (ii) The study used coeliac serology only to diagnose coeliac disease, and all of the coeliac disease patients remained undiagnosed throughout the follow-up period. Thus, the increased mortality rate may be due to the fact that the disease was not treated; (iii) The study had a very small sample size of cases, with only 14 coeliac positive patients. The very wide CI reflects this, and the high-risk estimate may simply be falsely elevated.

However, since all of the other studies that used coeliac serology as a diagnosis showed substantially lower mortality estimates compared to Rubio-Tapia et al.,2 it is less likely that the lack of treatment hypothesis is correct. It appears more likely that the long follow up and/or the small sample size explains the heterogeneity.

This raises the possibility that as the current pooled mortality risk estimate is based on an average of 13.4 years of follow up, this may in fact be an underestimate of the true risk. Further prospective studies with a longer follow-up are needed to determine if the mortality risk is higher than the current estimate. However, our estimate still needs to be interpreted with caution, as it is limited by the availability of only six studies to analyse, combined with the heterogeneity in coeliac disease diagnosis method between studies. Nonetheless, it is likely that our pooled result is the most reliable estimate currently available.

This increased risk of all-cause mortality in coeliac patients may be partly explained by our findings that coeliac patients have an increased risk of cardiovascular mortality. However, earlier studies using external comparisons have reported inconsistent findings,[30, 31] and as our cardiovascular mortality result is based on only three studies, more studies using internal comparisons are required to determine this association. Further studies examining specific mortality risks among coeliac patients may help guide future studies on whether any preventative measures beyond the gluten free diet could be taken to minimise mortality.

Similar to the initial mortality estimate, the initial NHL estimate also had significant heterogeneity when the overall odds ratios of each study were combined. However, restricting the single heterogeneous study to reporting an OR from the same time frame as the other studies removed the heterogeneity. This pooled estimate of 2.61 using studies with internal comparisons is substantially less than the bulk of the reported risk estimates from recent studies using external comparisons, with estimates ranging from 3.20[32] to 42.70.[7] Given the level of heterogeneity between the risks of NHL reported by studies using external comparisons, our non-heterogeneous pooled result probably represents the most accurate estimation currently available.

The TNHL pooled result showed a substantially elevated risk of 15.84 for coeliac cases. Unfortunately, it was not possible to determine the exact risk of EATL specifically. However, it is likely that a large proportion of this elevated risk is due to the increased risk of EATL. Although the heterogeneity did not reach statistical significance, the only cohort study by Elfstrom et al.[28] reported an OR of 48 for TNHL, more than twice of any of the case control studies. The removal of this study reduced the risk of TNHL, but that risk still remained substantially elevated, with an OR of 12.12. It is possible that the extremely high OR reported by Elfstrom et al. may be due to ascertainment bias. The OR for NHL overall was significantly reduced in this study when ascertainment bias was limited by excluding lymphoma diagnosis within the first 12 months of biopsy proven coeliac diagnosis, moving from an OR of 4.36 to an OR of 2.93. However, the results for TNHL reported only the overall risk, and did not report any results with the first 12 months excluded. This lack of controlling for ascertainment bias in the single largest study along with the small number of studies available to combine limits the reliability of our result. However, it still probably represents the most reliable estimate available on the risk of TNHL in coeliac patients.

No association between the risk of any malignancy and coeliac disease was found in our pooled risk estimate. This is consistent with most of the recently reported estimates derived from external comparisons,[6, 32-35] indicating that it is likely that no such association exists. It is not clear how an increased risk of any malignancy overall is not found when there is a clear increased risk of NHL, although one study[36] found a decreased risk of breast cancer in coeliac disease. If coeliac disease decreases the risk of certain non-lymphoid malignancies, it may explain why there is no increased risk of any malignancy overall. Further study into the possibility of decreased cancer risks other than NHL may be warranted. However, it is possible that the lack of increased risk is due to a lack of statistical power, given the low absolute risk of NHL.

The serology subgroup analysis found that serologically diagnosed coeliac patients had an increased risk of mortality and NHL compared with non-coeliac patients. These results suggest that serologically defined coeliac patients may be subject to a similar level of risk as coeliac patients diagnosed by other means. However, given that these results are only marginally significant, further research into clarifying any association between positive coeliac serology and NHL and mortality is warranted.

One issue in the interpretation of these results is the concern that positive serology may not be independently associated with either mortality or NHL, and that the results may simply reflect a selection bias with patients complaining of coeliac related symptoms being more likely to receive coeliac serology testing. However, both the design of the four studies involved in the mortality results and the design of the two studies involved in the NHL results indicate that any such bias is very unlikely.

For the mortality results, three of the studies[8, 22, 29] were prospective, population-based studies that acquired the serology results from serum which was initially collected for non-coeliac research purposes, whereas the fourth study[24] involved in the mortality result only included serologically defined patients if they had a positive coeliac serology and a negative biopsy. Similarly, for the NHL results, one study[23] was a prospective, population-based study that acquired the serology results from serum which was initially collected for non-coeliac research purposes, whereas the second study[28] only included serologically defined patients if they had a positive coeliac serology and a negative biopsy.

As only one study in the serology subgroup mortality analysis and one study in the serology subgroup NHL analysis used positive coeliac serology with negative biopsy results as the definition of serology positive coeliac disease, we were unable to determine if these patients have similar increased risks of mortality and NHL to patients with biopsy proven coeliac disease. Further studies stratifying patients with positive coeliac serology into biopsy positive and biopsy negative groups are needed.

An additional prospective coeliac serology-based study by Metzger et al.[37] was identified during our search that reported a mortality risk estimate much higher than all the other studies except for Rubio-Tapia et al., with a hazard ratio of 2.53. However, as the particular diagnostic method used was not considered reliable enough by the authors to constitute a diagnosis of latent/silent coeliac disease, the study did not comply with our inclusion criteria.

We were not able to determine from our systematic review and meta-analysis if compliance with a gluten-free diet had any effect on the risk of mortality and NHL in coeliac patients. There is evidence suggesting that the gluten-free diet decreases the risk of mortality[30] and malignancy[7]; however, some studies have not found this association.[38, 39] Future studies should include measurement of compliance with the gluten-free diet.

Our study had a number of strengths. We followed the MOOSE guidelines where possible in performing our systematic review. We performed a comprehensive search using multiple databases that was not limited by language. We only included studies with internal comparisons to ensure methodological rigour. Our study also suffered from a number of limitations. We did not search for unpublished studies. We were only able to find a small number of studies to combine, limiting our statistical power. In addition, although we found no publication bias, such a small number of studies limit the reliability of this result. Our studies were also heterogeneous with regard to study design, particularly with the method of coeliac disease diagnosis. Lastly, all of the study populations were derived from countries that are highly economically developed, and are predominantly made up of individuals of European ethnicity. This limits the applicability of our results to other populations.

Conclusion

Our systematic review and meta-analysis shows that coeliac patients are at an increased risk of mortality and non-Hodgkin lymphoma, but do not have an increased risk of any malignancy overall. Serologically defined coeliac patients also have an elevated risk of mortality and non-Hodgkin lymphoma. Our results are limited by the small number of studies available for analysis, and by heterogeneity in diagnostic methods. More studies using internal comparators are needed, particularly to more accurately determine the risk of NHL and mortality in both serologically defined and histologically defined coeliac patients.

Acknowledgement

Declaration of personal and funding interests: None.

Ancillary