Inflammatory Bowel Disease and Pancreatic Cancer

A Scandinavian Register-based Cohort Study 1969-2017

Åsa H. Everhov; Rune Erichsen; Michael C. Sachs; Lars Pedersen; Jonas Halfvarson; Johan Askling; Anders Ekbom; Jonas F. Ludvigsson; Henrik Toft Sørensen; Ola Olén


Aliment Pharmacol Ther. 2020;52(1):143-154. 

In This Article


Study Design

We performed a cohort study where patients with IBD were compared to a matched reference population with respect to pancreatic cancer.

Setting and Data Sources

Denmark and Sweden are high-income countries with populations of 5.7[35] and 10.1 million in 2017[36] respectively. In both countries, healthcare is tax-funded with universal access to care.[37] The personal identity number assigned to all residents in each country allows for linkage of registers containing national data on demographics, morbidity, mortality and histopathology with virtually no loss to follow-up (Table S2). Patients with IBD are typically handled by gastroenterologists in hospital-based outpatient clinics.


Patients With IBD. We used the International Classification of Disease (ICD) codes (Table S3) in the National Patient Registers (Table S2) to identify incident cases with IBD (in Denmark between January 1979 and December 2011 and in Sweden between January 1969 and December 2017). We requested either ≥2 records of IBD in the National Patient Registers[38]or ≥1 record of IBD in the National Patient Registers plus a colorectal biopsy record suggestive of IBD from the Swedish ESPRESSO Biopsy Register or the Danish Pathology Register.[39] The IBD subtype definition was based on the first two diagnostic listings, and patients with listings of both Crohn's disease and ulcerative colitis, or a listing of IBD unclassified (IBD-U) were defined as IBD-U, and included in the IBD group.[40]

We defined PSC according to ICD-codes in the National Patient Registers (≥1 listing) (Table S3).

Matched Reference Individuals. For each IBD patient, we identified up to 10 reference individuals in the National Population Registers[41] and matched them by sex, age, calendar year and place of residence. The matched reference individuals had to be alive and free of IBD at the start of follow-up of the index patient, and stopped contributing person-time as reference individuals if and when they were diagnosed with IBD.

Pancreatic Cancer

Data on diagnosis of exocrine pancreatic cancer and cancer stage were retrieved from the Cancer Registers (Table S4), and pancreatic cancer death from the Causes of Death Registers. The proportion of patients and reference individuals undergoing pancreatic resections (indicating potentially curable disease) was estimated using procedure codes for pancreatic surgery (Table S5).

Statistical Methods

Time at Risk. To avoid immortal time bias, follow-up started on the date of the second diagnostic record of IBD in a patient register or the date when a patient had first accumulated one diagnostic record of IBD and one record of a colorectal biopsy suggestive of IBD. Stratification for age and year of IBD onset was based on the date of the first diagnostic record. Follow-up ended with death, emigration or end of follow-up (31 December 2011 in Denmark and 31 December 2017 in Sweden), whichever came first.

Incidence Rates and Hazard Ratios. We report crude incidence rates (number of pancreatic cancer diagnoses by person-time at risk) but also country-specific incidence rates standardised to the Nordic population in the year 2000.[42] We calculated hazard ratios (HRs) with 95% confidence intervals (CIs) to quantify the association between IBD case status and pancreatic cancer diagnosis using Cox regression adjusted for age at IBD diagnosis, sex, year of IBD diagnosis and place of residence. We additionally calculated incidence rates and HRs stratified by country, sex, year of IBD diagnosis and age at IBD diagnosis in categories. In the stratum-specific estimates of incidence rates and HRs for patients with PSC, follow-up started when the patient fulfilled criteria for both IBD and PSC.

Multi-state Model

To investigate the associations between IBD and the occurrence of different competing events during follow-up, we specified a multi-state model with the following states: initial (start of follow-up), pancreatic cancer, pancreatic cancer death and death from other causes. Individuals started in the initial state, and were considered censored at the date of emigration, or at the end of the study period. We estimated the cause-specific cumulative incidence rates for the transitions: initial to pancreatic cancer, initial to pancreatic cancer death, initial to other death, pancreatic cancer to pancreatic cancer death, pancreatic cancer to other death, using the Aalen-Johansen estimator.[43] We computed adjusted transition HRs for each of those transitions using a series of Cox regression models, adjusted for the same covariates as above and additionally adjusting for tumour stage for the transitions from pancreatic cancer. Finally, we assessed the time-varying association between IBD and each of the transition hazards using the semi-parametric additive hazards model[44] as implemented in the timereg R package.[45] These models allowed for a time-varying effect of IBD on each transition hazard and were adjusted for time-invariant effects of sex, age at diagnosis, year of IBD diagnosis and country.

Quantitative Bias Analysis. We conducted a sensitivity analysis for the HRs of IBD for pancreatic cancer to unmeasured confounding due to smoking using the quantitative bias analysis.[46] The figures display the estimated HRs that would have been observed (y-axis), if smoking were measured and had a particular association with the exposure and the outcome.

R statistical software (version 3.6.1, R Foundation for Statistical Computing, Vienna, Austria) and the survival package (version 2.38, Therneau, T (2015), were used for the analyses.