Cancer incidence and factors associated with malignancies in coeliac disease during long‐term follow‐up

The increased risk of malignancies in coeliac disease is well recognised, but data on incidence and predisposing factors are scarce. Our aim was to determine cancer incidence and its prognostic factors in coeliac disease during a long‐term follow‐up.


| INTRODUC TI ON
Diarrhoea and malabsorption are well-known symptoms in coeliac disease, but presenting symptoms are diverse. Coeliac patients are increasingly diagnosed due to extraintestinal manifestations such as dermatitis herpetiformis, arthralgia and neurological symptoms or detected by screening risk groups with mild or even no coeliac symptoms at all, especially subjects with family history of coeliac disease or with autoimmune diseases. 1 Coeliac disease is provoked by ingestion of gluten-containing cereals in genetically predisposed individuals. The resulting small intestinal mucosal damage recovers on a life-long gluten-free diet. 2 According to serological screening studies, coeliac disease affects approximately 1%-2% of population in Western countries. 3 Diagnosed coeliac disease has been associated with increased risk of malignancies, especially with non-Hodgkin lymphoma and adenocarcinoma of the gastrointestinal tract. [4][5][6][7][8][9][10][11][12][13] Male gender, older age at coeliac disease diagnosis and classical coeliac disease phenotype with diarrhoea and weight loss have been suggested as possible risk factors, especially for developing refractory coeliac disease and subsequently non-Hodgkin lymphoma. [14][15][16][17][18] Conversely, the overall risk of malignancies in screen-detected coeliac disease has not been reported to be increased in most studies but the data are controversial. 12,19,20 Childhood coeliac disease has not been associated with an increased risk of lymphoma or other malignancies 5,10 although studies on this subject are scarce.
Reports on whether gluten-free diet has a protective effect against malignancies are inconsistent. 16,21,22 Otherwise, data on risk factors for coeliac disease-associated malignancies are mostly lacking.
The aim of our study was to assess the relative risk and risk factors of malignancies in coeliac disease, focussing on coeliac disease phenotype at the time of diagnosis. We scrutinised a well-defined cohort of coeliac patients collected prospectively between 1960 and 2000 with a long-term follow-up of altogether 37 years.

| Study population
All patients diagnosed with coeliac disease at the Tampere University Hospital catchment area between January 1960 and December 2000 were prospectively included in the cohort. Approximately 16% of the Finnish population reside in Tampere University Hospital catchment area. The prevalence of coeliac disease in the area is slightly higher (0.7%) than that in Finland in general (0.6%). 23 The diagnoses were based on histological verification according to approved guidelines; duodenal villous atrophy and crypt hyperplasia (compatible with Marsh-Oberhuber 3 morphology 24,25 ) in coeliac disease, and granular IgA deposits of the dermal papillae of the skin in direct immunofluorescence assay in dermatitis herpetiformis. The histological evaluations mainly took place in our department and ambiguous findings were scrutinised in multidisciplinary meetings.
The date of duodenal or skin biopsy was set as the date of diagnosis. For those dermatitis herpetiformis patients clinically diagnosed before immunofluorescence assay became available in the 1980s, but whose diagnosis in each case was confirmed with the test later when it became available, the date of diagnosis was taken to be the date of clinical diagnosis. Duodenal biopsy was performed to 87% of dermatitis herpetiformis patients and 32% of them had normal villous architecture.

| Review of medical files and stratification
The medical records of the cohort were reviewed retrospectively in 2015 and individual data on patients, coeliac disease characteristics and the use of gluten-free diet were obtained when available. Eighty-eight records of deceased patients were no longer found. The cohort was stratified (Table 1) according to sex, age at (<20, 20-39, 40-59 and ≥60 years) and decade of (≤ 1980, 1981-1990, 1991-2000)  Screen detected group (n = 153) contained individuals with coeliac disease in the family or with an autoimmune disease (eg type 1 diabetes, autoimmune thyroiditis, arthralgias or IgA nephropathy). The whole cohort was stratified according to degree of small bowel villous atrophy at diagnosis. The damage was rated to be severe in cases of total or subtotal villous atrophy (corresponding Marsh-Oberhuber 3b-c), mild in case of partial villous atrophy (corresponding Marsh-Oberhuber 3a) and normal when no villous atrophy was found (corresponding Marsh 0-2). The response to gluten-free diet was considered to be achieved when villous atrophy was alleviated in follow-up biopsies. Patients showing no histological villous improvement as well as patients not adhering to gluten-free diet at all were assigned to the 'no response' group.
In case of multiple control biopsies during the follow-up, the response was determined according to the best villous recovery achieved. The occurrence of non-Hodgkin lymphomas was evaluated case by case in relation to presenting symptoms.

| Malignancy data
The Finnish Cancer Registry maintains national registry of all cancers diagnosed since 1953. Since 1961 it has been compulsory for physicians and pathologists to report cancers diagnosed and the registry covers more than 98% of all diagnosed malignancies. 26

| Statistical analysis
Concerning cancer incidence this was a prospective follow-up study while the risk factors for malignancy were analysed in a retrospective manner. The follow-up went from 1 January 1978 for patients who received their diagnoses before 1978 and for the rest from the date of diagnosis. The follow-up continued until death, date of emigration, or 31 December 2014, whichever occurred first. All cancers that were found before coeliac disease diagnosis was established were excluded. The expected numbers of malignancies were calculated by multiplying the observed person-years of follow-up by the incidence of each malignant disease in the respective sex, age and calendar period in the population of the Tampere University Hospital catchment area. The ratios of observed-to-expected cancers, the standardised incidence ratios (SIRs), were calculated for different malignancies altogether and according to stratifications and exact 95% confidence intervals (CI) were defined assuming that the numbers observed followed Poisson distribution. Basal cell carcinoma and precancerous lesions of breast, cervix and ovary were not included in the analysis of overall risk of malignancy. In stratified analyses, patients with missing data were excluded from analyses in each strata.

| RE SULTS
Basic characteristics of the cohort are presented in Table 1  In stratified analysis, no potential risk factors were associated with the overall risk of malignancy (

| D ISCUSS I ON
In this follow-up study of coeliac disease patients followed up for a median of 24 years, the overall cancer incidence was not increased in any phenotype based on presenting symptoms or at any time  8,27 ; in one study only the postdiagnostic period was included in the risk-estimate 27 and in another the confidence intervals were relatively wide. 8 Our risk-estimate for non-Hodgkin lymphoma was within the range (two to sevenfold) reported previously. [8][9][10][11][12]14,[27][28][29][30] During the follow-up, the risk declined and disappeared after ≥15 years of follow-up. It can be speculated that good compliance with gluten-free diet had an impact to the better results, even though we were not able to show the positive effect of the diet in the present survey. We were not able to evaluate adherence to the diet in individual cases, but as many as 91% of our patients achieved response in duodenal control biopsy indicating that the overall compliance was good, which again indicates that gluten-free diet would have a protective effect against malignancy. It is noteworthy that compliance or response has not been reported in the many other studies. 4,5,7,10 The data on response were missing in 522 patients, mostly because control biopsies were performed after transmission of follow-up to primary health care. It can be speculated whether these patients were engaged to coeliac disease treatment and how did it affect the estimation of cancer risk. However, no difference was observed in the cancer rate between patients with (13.6%) and without (13.4%) data on response. Furthermore, it is important to note that the overall cancer incidence was not increased despite the increases in risk of non-Hodgkin lymphoma and gastrointestinal cancers. This is possibly explained by the low absolute

Risk factor
Any malignancy a Abbreviation: CI, confidence interval. a Basal cell carcinoma excluded from the analysis of overall risk of malignancy.

Non-Hodgkin lymphoma
b Non-responder = no villous improvement in control biopsy or non-compliant to gluten-free diet.
TA B L E 4 Standardised incidence ratios (SIRs) with 95% confidence intervals (CI) for malignancies, non-Hodgkin lymphomas and gastrointestinal malignancies among coeliac disease (CD) patients according to potential risk factors numbers of these cancers in conjunction with the decreased risk of breast cancer. . Consistent with the present study, anaemia at diagnosis did not increase the risk of LPDs in these two studies. 14,16 The relative risk of non-Hodgkin lymphoma was slightly higher in dermatitis herpetiformis patients than in the other phenotypes.
In contrast, in Swedish 10 and British 5 studies, the risk was two to threefold higher in coeliac disease than in dermatitis herpetiformis.
In the present series, the SIR of non-Hodgkin lymphoma was clearly highest among dermatitis herpetiformis patients diagnosed before 1981. The need for strict gluten-free diet in the treatment of dermatitis herpetiformis gained recognition after this, in the 1980s, which might explain the observed excess risk of non-Hodgkin lymphoma. Supporting this, Finnish dermatitis herpetiformis patients who contracted non-Hodgkin lymphoma were shown to adhere to gluten-free diet significantly less strictly than the patients without lymphoma. 31 The elevated risk of gastrointestinal cancer was mainly due to excess of stomach and liver cancers in women. Moreover, the relative risk of stomach cancer was slightly increased among dermatitis herpetiformis patients. A recent review on the association of autoimmune diseases and stomach cancer reported a similar finding (RR 1.7) for dermatitis herpetiformis, 32 but other studies focussing specifically on the risk of malignancies in dermatitis herpetiformis found no increased risk. 5,10,33 Nevertheless, the number of these cancers observed was small in our cohort and no precise conclusions can thus be drawn. This issue requires further studies. Interestingly, opposite to the recent Swedish study, 34 we did not observe any small bowel adenocarcinomas.
We found a decreased risk of breast cancer; a similar conclusion has also been reported in several earlier studies. 5,7,10,11,27,28,35  The main strength of our study was the large and prospectively assembled cohort of coeliac patients followed up for almost 40 years, allowing thus a plausible estimation of long-term cancer risk in coeliac disease. At the time our cohort was compiled, it included all coeliac disease cases diagnosed with a spectrum of clinical phenotypes in the Tampere University Hospital catchment area.
Most of the diagnoses were established in our centre but follow-up was transferred to primary health care already 1 year after the diagnosis. Therefore, we believe that this series gives a more reliable picture of the true risk of malignancy than some earlier studies.
We were able to obtain detailed individual information enabling assessment of the risk rate of malignancies occurring in different coeliac disease phenotypes. By contrast, other studies have mainly evaluated the malignancy risk in general, not focussing on different presentations. 7,10,11,27 This study also had limitations. The data on coeliac disease characteristics were drawn from patient records, and all the desired information could not be retrieved. Moreover, it was not possible to assess the duration of coeliac disease symptoms prior to diagnosis and therefore the association between coeliac disease and the 30 excluded cancers could not be defined. Longitudinal assessment of the persistence of strict gluten-free diet was not possible. However, the observed response rate in duodenal control biopsies in our study was 91% indicating good adherence to gluten-free diet. The response rate is also in line with those reported in other Finnish studies assessing compliance with gluten-free diet and villous recovery. 2 Moreover, long-term compliance with gluten-free diet has been excellent in our Tampere coeliac disease and dermatitis herpetiformis cohorts->93% 38 and 98% 39 respectively. These percentages involve by and large the same patient series as the present study. We had no information on confounding risk factors for malignancies such as alcohol and tobacco consumption.
In conclusion, we showed that the overall risk of malignancy was not increased in any coeliac disease phenotype even in long term. As to non-Hodgkin lymphoma, the risk was increased, but the excess declined over time. No significant risk was observed in patients with diagnosis of coeliac disease in childhood, with screen-detected coeliac disease or in those suffering from malabsorption at the time of diagnosis. Our findings suggest that coeliac disease-associated risk of malignancies is not increased in general, but variations in coeliac disease presentation may have an impact on prognosis. In order to focus surveillance to coeliac patients with an elevated risk of malignancy, prospective studies assessing the risk in different coeliac disease phenotypes are warranted.

ACK N OWLEG EM ENTS
Declaration of personal interests: None.

AUTH O R S H I P S TATEM ENT
Guarantor of the article: Katri Kaukinen.
Specific author contributions: Inka Koskinen involved in study design, project coordination, review of the medical files, statistical analyses, interpretation of the data and writing of the article; Kaisa Hervonen and Timo Reunala involved in data collection and revision of the manuscript; Eero Pukkala involved in study design, statistical analyses and revision of the manuscript; Katri Kaukinen and Pekka Collin involved in study design, interpretation of the data and revision of the manuscript.
All the authors have read and approved the final manuscript.

E TH I C S S TATEM ENT
The authors confirm that the ethical policies of the journal, as noted in the journal's author guidelines page, have been adhered to and the appropriate data permits have been received. According to the Finnish legislation, no informed consent or approval by an ethics committee is required for registry-based studies when study populations are not contacted.

PE E R R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1002/ygh2.446.

DATA AVA I L A B I L I T Y S TAT E M E N T
Research data are not shared.