We used the unique personal identity number assigned to each Swedish resident (about 9 million in total) to link together information from seven population-based registers (Fig. 1) [3].
The Prescribed Drug Register, the Cancer Register, and the Causes of Death Register, all maintained by the National Board of Health and Welfare were used to obtain information on targeted person-time and outcome. We retrieved variables reflecting potential confounding factors [5] from the Swedish National Diabetes Register (maintained by the local health authorities), the Prescribed Drug Register, the National Patient Register [6], and the Medical Birth Register [7] (all maintained by the National Board of Health and Welfare). Information on educational level was extracted from the National Education Register [8], which is maintained by Statistics Sweden. By law, local health authorities must report all new cases of cancer to the Cancer Register, all in-patient information to the National Patient Register, all births to the Medical Birth Register, and all deaths to the Causes of Death Register [9].
The Swedish Prescribed Drug Register contains details of all the prescriptions dispensed in Sweden [10]. Updated monthly, there are presently around 100 million prescriptions dispensed each year. Before 1 July 2005, the personal identity number was not recorded in the register. We therefore had to start recruiting subjects for observation from 1 July 2005.
The Swedish Cancer Register was set up in 1958, and since that time every clinician, pathologist and cytologist in Sweden must notify the National Board of Health and Welfare of each person who has been diagnosed with a new primary malignancy. The Cancer Register includes primary malignancies and certain benign tumours and precancerous lesions [11]. A comparison with death certificates revealed the rate of non-reporting to the National Cancer Registry to be less than 2% during the late 1970s [12], and in a comparison with the National Patient Register, the rate of non-reporting to the National Cancer Registry was estimated to be 3.7% in 1998 [13].
Launched as a quality assurance register in 1996, the National Diabetes Register includes clinically relevant information [14]. Trained physicians and nurses report the data collected during visits to hospital outpatient clinics and primary healthcare centres via the internet or via clinical record databases. The National Education Register is updated annually with information on the highest formal education achieved.
Ethical considerations
The Swedish National Board of Health and Welfare is a government agency and may, in accordance with Swedish law, use population-based registers to follow and analyse health and social conditions among the general population. Data were made available to us in such a way that individuals could not be identified.
Targeted person-time
We have studied all 114,841 individuals who were aged 35–84 years old at the end of 2005, had at least one prescription dispensed for insulin (Anatomical Therapeutic Chemical [ATC] code A10A) between 1 July and 31 December 2005, and who were alive at the start of follow-up (1 January 2006). We studied first diagnosis of a primary malignancy as an outcome measure, excluding individuals who received this diagnosis at any time between 1 January 1958 and 31 December 2005. That is, a subject with a record of having been diagnosed with any type of malignancy was excluded from the analyses of outcomes including ‘all malignancies’, and men who had a record of having been diagnosed with prostate cancer before 2006 were excluded from the analyses of the outcome ‘prostate cancer’. We followed the subjects from 1 January 2006 to 31 December 2007. The number of person-years of follow-up for each individual was from 1 January 2006 to death or loss to follow-up (censorship) or the outcome being analysed or study end. Consequently, the number of observed person-years varies according to the particular type of malignancy studied.
Categories of insulin use
Information on exposure to insulin and analogues was obtained from prescriptions dispensed between 1 July and 31 December 2005. Individuals registered as having had at least one prescription dispensed for insulin glargine (ATC code A10AE04), but no prescriptions dispensed for other types of insulin (ATC code A10A) were classified as using insulin glargine alone (no other types of insulin). Having a prescription dispensed for both insulin glargine and another type of insulin classified the individual as a user of insulin glargine and other types of insulin. Having a prescription dispensed for insulin but not for insulin glargine, classified the individual as a user of types of insulin other than insulin glargine.
Outcomes
We studied five malignancy outcomes; death from any cause and acute myocardial infarction were also used as endpoints. Following the routines for the Cancer Register, the outcome ‘all malignancies’ included a carcinoid tumour, a granulosa cell tumour, a thymoma, an adamantinoma, a chordoma, a transitional cell papilloma of the urinary tract, a hormonally active tumour from at least one endocrine gland (except the thyroid), an enterochromaffin or a neuroendocrine tumour. We also included precancerous lesions, including gastrointestinal polyps with suspected malignancy, bronchial adenomas, carcinoma in situ of the breast, fibro-adenoma with suspected malignancy, adenoma phyllodes, precancerous endometrial lesions, hydatidiform moles of placental tissue, ovarian cystadenomas of borderline type, histologically benign tumours of the central nervous system and meningomas. In situ cases were included with malignant tumours as a second outcome. We defined outcomes for three different anatomical areas: ‘breast cancer’ (International Classification of Diseases, 10th revision [ICD-10] code C50), ‘prostate cancer’ (ICD-10 code C61) and ‘gastrointestinal cancer’ (ICD-10: codes C16–C20). For these three areas we only included tumours that were histopathologically classified as adenocarcinoma (WHO/HS/CANC/24.1 histology code 096).
Individuals registered with any type of malignancy between 1 January 1958 and 31 December 2005 were excluded from the analysis when ‘any type of maligancy’ was the outcome. When studying breast cancer among women we excluded those who had previously been diagnosed with breast cancer; similar exclusions were made when studying prostate cancer and gastrointestinal cancer. We retrieved the date of death from the Causes of Death register. Subjects who were not registered as dead and who did not have a prescription dispensed for any drug in 2008 were classified as having been lost to follow-up. The date for loss to follow-up was set at 90 days after the last date for a dispensed prescription.
Variables reflecting potential confounding factors
Sex and age were retrieved from the Prescribed Drug Register. We obtained data on age at onset of diabetes from the National Diabetes Register or estimated it from the time for first admission to hospital care with diabetes as the main diagnosis (ICD-8 code 250; ICD-9 code 250; ICD-10 codes E10–E14) from data in the Patient Register for 1969 to 2005. An age at onset of diabetes of less than 30 years, as recorded in the National Diabetes Register (primary choice) or by data from the Patient Register (secondary choice), defined an individual as having type 1 diabetes; an age at onset above 30 years defined an individual as having type 2 diabetes; the absence of information on age at onset defined the individual as having missing information on type of diabetes. The highest BMI reported to the National Diabetes Register from 2003 to 2005 for each individual was used as the value for BMI. We retrieved information on smoking habits from 2003 to 2005 from the National Diabetes Register. Anyone who reported smoking during 2005 was classified as a current smoker. Anyone who reported not smoking in 2005 but reported smoking in 2003 or 2004 was classified as a former smoker. Anyone who reported not smoking in 2003, 2004 and 2005 was classified as a non-smoker. A record of a prescription dispensed for an oestrogen or for metformin in the Prescribed Drug Register from 1 July to 31 December 2005 defined oestrogen and metformin use, respectively.
A record of at least one hospital admission with a main diagnosis of any cardiovascular disease (ICD-10 codes I00–I99) in the National Patient Register, during the period 1 July 2004 to 30 June 2005 (i.e. 1 year prior to definition of exposure), classified the individual as having cardiovascular disease.
Educational level refers to the highest attained educational level at the end of 2005. Educational level was classified into the following three categories, representing distinct levels in the Swedish educational system: (1) 9 years or fewer of schooling, equivalent to elementary school or less; (2) 10–12 years of schooling, equivalent to secondary school; and (3) more than 12 years, equivalent to university. Age at birth of first child (women only) was categorised into no children, <30 years, ≥30 years, and missing information. A large group of women (49%), mainly the older women in the study population, had information missing on childbearing.
Statistical methods
As a measure of the relative occurrence of malignancies, we used the incidence rate ratio. For example, we calculated the incidence rate of having been diagnosed with any type of malignancy among users of insulin glargine alone and compared this with the incidence rate among users of other types of insulin. We cite this measure of relative occurrence, the incidence rate ratio, as a relative risk. Poisson regression analyses were used to evaluate the association between the three groups of insulin users and malignancy outcome. These models were fitted with the logarithm of observed person-years as the offset and they also provided 95% CIs of the incidence rate ratio. When adjusting for potential confounding factors, we categorised the numeric variables as presented in Tables 1 and 4, and in order to avoid a substantial reduction of the number of subjects, we accepted ‘missing value’ as a single category in our main analyses. The Genmod procedure in the SAS statistical software package (SAS Institute, Cary, NC, USA) was used for the calculations.
Table 1 Baseline characteristics of the subjects