This is a retrospective population-based cohort study using Danish national administrative health registers and databases. The registers are nationwide and cover all residents. Danish residents have a unique civil registration number recorded in the Danish Civil Registration System [13]. Therefore, it is possible to cross-link registers and databases at the individual level and obtain complete follow-up.

Study population

The study population consists of individuals living in Denmark between 1996 and 2017 identified as T1D individuals based on data from the Danish health care registers and databases containing diabetes-defining information. A detailed description of the identification of individuals with T1D can be found in Carstensen et al. [2]. In short, the cohort was identified from Registry of Medicinal Products Statistics (RMPS) [14], Danish Adult Diabetes Database (DADD) [15], National Patient Register (NPR) [16], Danish National Health Service Register (NHSR) [17], and Danish Clinical Quality Assurance Database for Screening of Diabetic Retinopathy and Maculopathy (DiaBase) [18]. Individuals were classified as having diabetes with a proxy for diabetes diagnosis date being the earliest date of any of the following: (1) first occurring diagnosis of diabetes (International Classification of Diseases (ICD)-8 codes: 249 and 250; ICD-10 codes: E10 and E11; with the exclusion of gestational diabetes) in NPR (valid from 1977), (2) first occurring use of diabetes podiatry in NHSR (valid from 1990), (3) first date of purchase of any anti-diabetic medication (Anatomical Therapeutic Chemical Code (ATC) A10xxx) in RMPS (valid from 1995), (4) earliest mentioned date of diagnosis in DADD, (5) earliest date of eye examination recorded in DiaBase (valid from 2009).

Individuals were classified as having T1D if the criteria for diabetes and any of the following criteria were met; (1) purchase of insulin before age 30, (2) classified as T1D in the majority of the individual’s DADD records, (3) Not classified in DADD but with a majority of the records from NPR being classified as T1D. An individual could not be classified as having T1D if they had no recorded date of insulin purchase.

Cardiovascular risk factors

HbA1c, estimated glomerular filtration rate (eGFR), urine albumin creatinine ratio (UACR), and lipid levels were sourced from the Danish National Laboratory Database (NLD) [19]. The measurements were identified by the Nomenclature for Properties and Units (NPU) codes listed in Additional file 3: Table S1. Blood pressure, BMI and smoking habits were obtained from the DADD. All cardiovascular risk factors were obtained for the period 2010–2017.

Cardioprotective medications

Data from the RMPS was used to map cardioprotective medication usage patterns in the population for the period 1996–2017. RMPS includes all filled prescriptions since 1996 with information on ATC code and amount at the individual level. The lipid lowering drugs (LLDs) with the following ATC codes were extracted; Statins: C10AA01-07, Fibrates: C10AB01, C10AB02 and C10AB04, Bile acids: C10AC01, C10AC02 and C10AC04, Nicotinic acids: C10AD06 and C10AD52, Ezetimibe: C10AX09, PCSK9: C10AX13-14, Statin combinations: C10BA02 and C10BA05. The antihypertensive drugs (AHDs) with the following ATC codes were extracted: C02*, C03*, C07*, C08* and C09*.

Individuals were followed from 1st January 1996 or date of diabetes diagnosis until date of emigration, death or 1st January 2017. LLD-exposure was computed for the entire follow-up period using the gen.exp-function in the ‘Epi’-package in R(21). From records of drug purchase which include dates of purchase, amount purchased in number of pills and dose per time, the gen.exp-function generated LLD-exposure covariates for a particular LLD for the entire follow-up of each person. LLD exposure was assessed each 1/10 of a year time interval from start of follow-up. A grace period of 1 month was used in the definition of exposure to an LLD type, meaning that an individual was considered exposed 1 month after the end of the formally computed exposure interval. Concomitant use of different LLDs was allowed. As such, individuals could discontinue and resume therapy with LLDs several times during follow-up. The annual frequency of use for each type of LLD among the T1D population was then visualized in a bar plot.

Cardiovascular complications

First events of diabetes-related micro- and macrovascular complications were ascertained from the National Patient Register [16]. The following complications were considered for analysis: ischaemic and haemorrhagic stroke, ischaemic heart disease, heart failure, atherosclerotic macrovascular disease, albuminuria, end-stage renal disease, lower limb amputation and moderate to severe retinopathy. A CVD composite of ischaemic and haemorrhagic stroke, ischaemic heart disease, heart failure, atherosclerotic macrovascular disease and atrial fibrillation was also defined. Events of end-stage renal disease were defined by initiation of dialysis or kidney transplantation. Measurements of albuminuria were sourced from the National Laboratory Database. NPU codes for measurements of UACR were used for ascertainment. The following thresholds were applied for classification of albuminuria; UACR < 30 mg/g as normal, UACR ≥ 30 mg/g for microalbuminuria and UACR > 300 mg/g for macroalbuminuria. Additional file 4: Table S2 supplemental data lists the specific ICD codes and Danish procedure codes used to define the disease entities.

Emigration and death

For follow-up, death and emigration was obtained by linkage to the Central Person Register [13].

Statistical analysisCardiovascular risk factors levels

To describe cardiovascular risk factor levels over time, we calculated the mean cardiovascular risk factor concentration for each individual per calendar year and CV risk factor type. Trends in CV risk factor levels were estimated using an additive mixed effects model containing sex, age, date of CV risk factor measurement, and duration of T1D as fixed effects and the within-individual variation as random effect. The effect of age, date of CV risk factor measurement and duration of T1D was assumed linear. Using this model, 95% prediction intervals of estimated CV risk factor values were calculated. Triglyceride and UACR values were log transformed before modelling due to skewness.

Proportions of individuals within risk factor thresholds were calculated for each year. The LDL-C thresholds were defined according to (ESC/EAS) guidelines for the management of dyslipidaemias [20]. Attainment of treatment targets are referred and compared according to the 2011/2016 version of the guidelines which were in effect at time of the study period and the most recent 2019 version. HbA1c and blood pressure thresholds were defined according to the Danish Society of Endocrinology’s guidelines [21]. eGFR and UACR thresholds are according to Improving Global Outcomes (KDIGO) Guideline for Diabetes Management in Chronic Kidney Disease [22] and BMI thresholds are defined according to World Health Organization’s classification of BMI [23].

Cardioprotective medications in relation to incidence of cardiovascular complications

During follow-up of LLD and AHD exposure, individuals could move from “never use” (the period before the first use of the drug) to “continued use” (the period(s) a person is using the drug) and at discontinuation to “discontinued use" (period(s) where the person is not using the drug, but previously has) and back to “continued use” when treatment is resumed. As such, a person can contribute with risk time in several medication exposure groups during follow-up.

Incidence rates of diabetes-related micro- and macrovascular complications by medication-exposure status (never use/continued use/ discontinued use), sex, current age, T1D duration and calendar time were estimated using Poisson regression for time-split data.

Analyses were carried out on a remote, secure server for researchers provided by Statistics Denmark with no access to the unique personal registration number. Data management was done in SAS version 9.4 and analyses were performed in R version 3.5.1.