Data sources

Since 1960 all citizens in Denmark have had a personal identification number recorded in the Civil Registry that holds records of age, death, and emigration. Data from The Danish Cancer Register (containing histologically verified cancers since 1943) provided information on any malignant disease; the Danish National Patient Register contains clinical diagnoses and surgical codes at discharge from public and private hospitals since 1976, and the National Birth Register information about births since 1973. The National Register of Medicinal Product Statistics contains details of redeemed prescriptions for hormonal contraception and was linked using the personal identification number as the key identifier. Data from the National Register of Medicinal Product Statistics is considered complete from January 1, 1995 and this was consequently the start date of this study. Data was stored and analyzed on servers at Statistics Denmark. The unique personal identification numbers were encrypted ensuring anonymity during the data analyses.

Study population

All women living in Denmark aged 15–49 years at January 1st, 1995, and those subsequently reaching age 15 years up to December 31st, 2014 were eligible for the study.

Hormonal contraception

From the National Register of Medicinal Product Statistics, we obtained information on the date of redeemed prescriptions, the specific Anatomical Therapeutic Chemical code, dose, number of packages, defined daily doses, and route of administration (tablet, patch, IUD, etc.). The use of hormonal contraception was updated daily, with women changing status at time of cessation or switch in type of hormonal contraception used. Gaps between prescriptions of less than 28 days were filled out prospectively.[18]

Hormonal contraceptive use was categorized as current or recent use (cessation within last six months), previous use (cessation more than six months ago) or never use. The levonorgestrel-intrauterine system (IUD) was used for four years, unless pregnancy or another hormonal product was prescribed before the IUD expired.

Confounding factors

The National Birth Register and Statistics Denmark provided information on potential confounders; parity (0, 1, 2, 3, 4 or >4), age at first birth among parous women (<25, 25–30, >30 years), smoking status among parous women (yes, no, unknown; available since 1991), body mass index among parous women (<18.5, 18.5–25, 25.1–30, >30, unknown; available since 2004) and completed or ongoing education (elementary school only, high school only, further education but not college/university, college/university education, university education with research qualifications, unknown). The Danish National Patient Register provided information on diagnosis of polycystic ovary syndrome (yes/no) or endometriosis (yes/no). All models were adjusted for these conditions since women with these diseases more often use hormonal contraception.

Analysis

Data was analyzed by Poisson regression using SAS statistical software, version 9.1 (SAS Institute, Inc., Cary, North Carolina) to calculate incidence rate ratios (referred to as relative risk) and 95% confidence intervals. Age was used as a timescale in the Poison regression analyses with 5-year age bands. The study population was followed until a first diagnosis of pancreatic cancer, death, emigration, 50 years of age, or end of follow-up on December 31, 2014, whichever came first. Women were censored permanently at time of cancer diagnosis, infertility treatment, or venous thrombosis. Women were censored temporarily during pregnancy and for six months after delivery.

The following time-dependent variables were updated daily: hormonal contraception variables, age, calendar year, parity, age at first birth, and education. The simple adjusted models included hormonal contraception, age, and calendar year. Calendar year was included to account for possible changes in the use of specific types of contraception and in pancreatic cancer incidence during the study period. The fully adjusted models included all the time-dependent and the time-constant variables; polycystic ovary syndrome (ever), and endometriosis (ever). Additional adjustment for body mass index and smoking status and the time of first birth was conducted among parous women with this information. The reference group in each analysis was women who had never used hormonal contraception at a given time during follow-up. Calculated risk ratios will be referred to as relative risks throughout the manuscript, this includes all tables.

A search of the PubMed database was performed where relevant articles were identified using the following search terms: “pancreatic cancer”, “pancreatic adenocarcinoma”, “pancreatic neoplasm” combined with “reproductive factors”, “oral contraception”, “exogenous hormone use”. Date of last search was on the 1^st of February 2018. A flow chart further describing the inclusion of studies can be found in the supporting information. (S1 Fig).

Approval

Data was made available through Statistics Denmark and approval obtained from the Danish Data Protection Agency and Health Data Board (previous National Board of Health). According to the Danish Research Ethics Committee Law (§ 8, section 3), ethical approval is not required for register-based studies. Data must be requested from the Danish National eHealth Authority due to legal restrictions on sharing data publicly. Data requests are sent through the application form at: https://sundhedsdatastyrelsen.dk/da/forskerservice/ansog-om-data (the website is in Danish).

Limitations

We were not able to adjust for alcohol consumption and dietary factors [19][22]. If women on hormonal contraception have higher alcohol consumption than never-users, we would have overestimated the risk of pancreas cancer associated with current or recent use of hormonal contraception. However, since our risk estimates were not adjusted for these factors, hormonal contraceptive use is unlikely to increase the risk of pancreas cancer; if anything, these factors might have biased a potential decreased risk with hormonal contraception use. We cannot be sure that residual confounding or unknown confounders could influence the results. If our results are biased it would most likely imply that we had failed to notice a protective effect of hormonal contraception which would be contradictory to other studies which have explored the same association.[23][11][10] Finally, redeemed prescriptions of medication are not necessarily taken. Repeat prescriptions account for of most of the exposure history, and thus this potential misclassification of exposure is unlikely to have influenced our results.

Strengths

Our study adds to the sparse evidence on the influence of newer hormonal contraceptive products on the pancreatic cancer risk in women of reproductive age. Our nationwide study is a large-scale prospective cohort study with 1.9 million Danish women followed for an average of more than ten years. We had complete follow-up until diagnosis of cancer, venous thrombosis, infertility treatment, emigration, death or end of study. The validity of pancreatic cancer diagnoses is high because the Danish Cancer Register cross-check their records with histology records to ensure both a high level of completeness and validity of diagnoses. [24,25][26]

The validity of exposure is also likely to be high in our study because recall bias has been eliminated. Information about redeemed hormonal contraception prescriptions is transferred electronically from the pharmacies using bar codes. Thus, we were able to account for switch in use and cessation of use. Our information about both exposure and relevant confounders was updated daily. Information about hormonal contraception and pancreatic cancer was recorded in the relevant national registers without the intention of exploring the association between hormonal contraceptive use and clinical outcomes known, making differential ascertainment of exposure and cancer incidence unlikely.

Our results were adjusted for age, year, educational attainment, number of births, endometriosis, and polycystic ovary syndrome. Additionally, for parous women, smoking and BMI. Adjustment for smoking and BMI did not change our estimates noticeably indicating no major confounding influence of these factors.

The novelty of this study is based upon the quality and quantity of our data utilized for our analysis. This dataset is unprecedented compared to other studies in this area and our focus has solely been on premenopausal women. It is reassuring that we do not find an increased risk of pancreatic cancer in premenopausal women using hormonal contraceptives due to the serious nature of the disease and high mortality rate.