Setting and study design

The Danish tax-funded healthcare system provides the entire population with access to both primary and secondary healthcare services free of charge [8]. As part of the primary care system, general practitioners (GPs) diagnose, monitor, and treat a variety of diseases. Aside from emergencies, GPs are the first point of contact in the healthcare system and are responsible for referring patients to specialists in private practice, hospital-based outpatient specialty clinics, and inpatient hospital care when necessary. The costs of prescription drugs are partially reimbursed through the national healthcare system [8].

A unique registration number, the Civil Personal Register (CPR) number, is assigned to all Danish citizens and legal residents at birth or immigration [9]. Contacts with the Danish healthcare system are recorded together with patients’ CPR number in national registries. The CPR number thus allows researchers to link data from administrative and medical registries at the individual level [8]. We conducted a nationwide population-based cohort study using Danish registries during the period 1 January 1996 to 31 December 2018.

Study cohorts

In Denmark, the majority of individuals with migraine are treated by GPs. Individuals may be referred to a hospital in cases of severe migraine or diagnostic uncertainty. While hospital diagnoses are available in the Danish registries, diagnoses given solely by GPs are not. We therefore used records of redeemed migraine medications (including triptans and ergotamine) identified from the Danish National Prescription Registry (NPR) [10] to identify a cohort of males with migraine and a cohort of females with migraine. All redeemed prescriptions issued by GPs, specialists in private practice, and hospitals (treatments administered during admissions not included) have been recorded in the NPR since 1995. Thus, we assumed that this approach would allow us to identify a large and representative sample of individuals with migraine. At redemption, the patient’s CPR number, the identifier for the prescriber and dispending pharmacy, and dispensing information are recorded. This information includes the anatomical therapeutic chemical (ATC) classification system code, dispensing date, product name, drug strength and formulation, and number of defined daily doses per package [10].

All men and women aged 18 to 60 years with at least 2 redeemed prescriptions for migraine medication recorded in the NPR were included in one of the 2 sex-specific migraine cohorts (ATC codes in Table A in S4 Text). At least 2 prescriptions were required to increase accuracy of the migraine diagnosis. The index date was defined as redemption date of the second prescription. Individuals were excluded if they had 2 or more prescriptions for migraine medication prior to the study period or a diagnosis of MI or stroke prior to or on the index date. Given the importance of the selection in the study cohort, we also did a sensitivity analysis using migraine diagnoses codes for enrollment in the study; see “Sensitivity analysis” below.

Female and male general population comparison cohorts were constructed using the Danish Civil Registration System [9]. For each individual with migraine, we sampled 5 individuals from the general population matched on sex, birth year, and calendar year. These individuals were assigned the index date corresponding to that of their matched migraine individual. Individuals were eligible for these cohorts if they were alive on the index date, had no previous diagnosis of MI or stroke, and had not redeemed 2 or more prescriptions for migraine medications. If a member of one of the general population cohorts redeemed a second prescription for a migraine drug during follow-up, the individual joined the migraine cohort. However, to avoid informative censoring, the person was not censored from the original cohort.

Premature MI, ischemic stroke, and hemorrhagic stroke

Cardiovascular events examined were MI, ischemic stroke, and hemorrhagic stroke occurring at age 60 or younger and recorded as primary or secondary discharge diagnoses in the Danish National Patient Registry (DNPR) [11]. We defined these events as premature. Admissions to Danish hospitals have been registered in the DNPR since 1977; emergency room and hospital-based outpatient specialist contacts were added in 1995. Records of each admission or contact include the patient’s CPR number, dates of admission and discharge or contact, one primary diagnosis, and up to 19 secondary diagnoses. Diagnoses were recorded according to the World Health Organization’s International Classification of Diseases, Eighth Revision (ICD-8) until 1994 and then according to the Tenth Revision (ICD-10) [11]. To ensure high validity of diagnoses, only inpatient codes were used (ICD codes in Table A in S4 Text). Due to their different pathogenesis, we chose to examine ischemic and hemorrhagic stroke as separate outcomes.

Covariates

The Integrated Database for Labor Market Research (IDA) [12], established in 1980, contains data on labor market participation for all Danish citizens. This is recorded on a yearly basis together with individuals’ CPR number [12]. IDA data on highest level of education achieved and income were used to describe socioeconomic position. For each individual, income was calculated as average income during the 5 years prior to the index year. Income was then categorized for men and women separately as low, medium, high, or very high, based on income quartiles during the year prior to the index year.

We used inpatient, hospital-based outpatient specialty clinic, and emergency room diagnoses recorded in the DNPR to identify comorbidities for each individual in the study prior to or on the index date. The following comorbidities were included: hypertension, atrial fibrillation or flutter, thyroid disease, venous thromboembolism (VTE), valvular heart disease, obesity, alcohol-related disease, mood disorders, and chronic obstructive pulmonary disease (COPD). Furthermore, we identified diseases included in the Charlson Comorbidity Index (CCI) (excluding cerebrovascular disease and MI) [13] and assessed the burden of comorbidity using the CCI score to categorize all individuals into one of 3 CCI categories: low = CCI score of 0 (no comorbidities), medium = CCI score of 1 to 2, or high = CCI score of 3 or more.

For each individual in the study, the NPR was used to identify ever-redeemed prescriptions for beta-blockers, angiotensin-converting enzyme inhibitors or angiotensin II antagonists, diuretics, vitamin K antagonists, direct oral anticoagulants, platelet inhibitors (including clopidogrel and low-dose aspirin), combined oral contraceptives, and systemic progesterone-only contraceptives prior to or on the index date. Additionally, we identified prescriptions for nonsteroidal anti-inflammatory drugs (NSAIDs), which may be used as a migraine abortive treatment. Although these drugs can be purchased at a reduced price if prescribed, they are also available over the counter. No individual-level records exist of over-the-counter purchases. Finally, we identified hyperlipidemia either as an existing diagnosis of hyperlipidemia in the DNPR or as a redeemed prescription for a lipid-lowering drug (ICD and ATC codes in Tables B and C in S4 Text).

Statistical analyses

We described the study cohorts according to age on the index date, calendar period, socioeconomic position, CCI category, comorbidities, and redeemed prescription drugs. Cohort members were followed from their index dates (date of second redeemed prescription for individuals with migraine) until date of cardiovascular outcome, emigration, 61st birthday, death, or end of follow-up (31 December 2018), whichever came first. To assess the entire impact of migraine on each outcome, we conducted 3 separate analysis—one for each cardiovascular outcome (MI, ischemic stroke, or hemorrhagic stroke). Thus, in the analysis examining impact of migraine on ischemic stroke, we ignored whether or not an MI or hemorrhagic stroke occurred during follow-up and vice versa. S1 Fig illustrates windows for exposure, exclusion, and covariate assessment and follow-up based on template by Schneeweiss and colleagues [14].

The original purpose of this study, as stated in the prespecified protocol (S1 Text), was to examine the interaction between sex and migraine on the risk of premature MI and stroke. However, an interaction analysis demands 2 exposures that are or are not present (with 1 group having none, 2 groups having either, and 1 group having both). It was not clear to us whether men or women would be the “exposed” sex. We therefore changed the aim of the study to examine the impact of migraine on the risk of MI and stroke among women and men separately. The reason for this change was that earlier studies revealed clear sex differences for the risk of ischemic stroke but did not examine sex differences in the risk of MI. We therefore focused on whether the sex differences associated with ischemic stroke also applied to MI and hemorrhagic stroke. The impact of migraine on each outcome for men and women was assessed on the absolute scale, using risk differences (RDs), and on the relative scale, using hazard ratios (HRs). As a measure of absolute risk (AR), cumulative incidences of the outcomes for the follow-up period of 1 to 20 years were calculated for all cohorts using the Aalen–Johansen product-limit estimator, treating death as a competing risk. From these estimates, the RD comparing individuals with migraine to individuals without migraine was calculated for men and women separately. We used Gray’s test to compute the p-value for comparison of the cumulative incidence curves for men and women separately.

We used Cox regression to calculate crude and adjusted HRs for the comparison of each cohort to women without migraine. We chose to use a Cox model rather than a Fine and Gray model, because the Cox model has been suggested to be more appropriate for etiologic questions [15]. Moreover, we calculated HRs comparing men with migraine to men without migraine. We conducted 2 different sets of adjustments. The first set of adjustments were based on the directed acyclic graphs provided in S2 Text. In this analysis with MI as the outcome, adjustments were made for age (as a continuous variable), calendar period (1996 to 2000, 2001 to 2005, 2006 to 2010, 2011 to 2014, and 2015 to 2018), hypertension, thyroid disease, hyperlipidemia, VTE, obesity, alcohol-related disease, and COPD. Where ischemic stroke was outcome, we adjusted for atrial fibrillation or flutter in addition to the factors adjusted for in the MI analysis. Where hemorrhagic stroke was outcome, we adjusted for age, calendar period, hypertension, alcohol-related disease, COPD, and anticoagulant treatment (vitamin K antagonist, direct oral anticoagulants, or platelet inhibitors). Additional adjustments were made in the second version of the adjusted HRs, and the adjustments were the same for all outcomes. Here, we adjusted for age, calendar period, education, income, CCI score, diabetes, hypertension, atrial fibrillation or flutter, VTE, Raynaud’s disease, obesity, thyroid disease, alcohol-related disease, COPD, hyperlipidemia, and ever-redeemed prescriptions of angiotensin-converting enzyme inhibitors/angiotensin II antagonists, diuretics, and anticoagulant treatments. There were no missing data for these analyses. The likelihood ratio test was used to compute p-values for the HRs.

Proportionality of hazards was checked using log–log plots (Figs A-C in S3 Text). As the log–log plots showed lack of proportionality for some hazards during the first year, we computed HRs for the follow-up period of years 1 to 20.

Sensitivity analysis

To investigate the robustness of the main analyses to potential misclassification, we conducted 2 sensitivity analyses. First, we repeated the computation of ARs and HRs of ischemic stroke, including the ICD-10 code for unspecified stroke (I64) in the definition of ischemic stroke (log–log plots in Fig D in S3 Text). An earlier validation study of stroke codes in the DNPR has shown that the majority (57%) of unspecified strokes are ischemic [16].

Second, we repeated all analyses using first-time ICD codes recorded in the DNPR to identify migraine (rather than prescription data). These refer to inpatient or outpatient diagnoses in hospital (rather than the prescription data from the NPR that refer to all redeemed prescriptions in community pharmacies and, therefore, also include prescriptions initiated by GPs; the NPR was the basis of migraine identification in our main analysis). The hazard rate functions were not completely parallel even looking beyond the first year of follow-up (log–log plots in Figs E and F in S3 Text). Our estimated HRs thus ignore some time-varying effects and show the average impact of migraine [17].

Use of migraine medication changed during the study period. To investigate the importance of this change, we conducted a third sensitivity analysis. In this analysis, we computed HRs of each outcome stratified by index year (1996 to 2006 or 2007 to 2018). To allow for the same potential follow-up, HRs were computed for follow-up of 1 to 10 years.

Ethics statement

The study was reported to the Danish Data Protection Agency by Aarhus University (record number 2016-051-000001, serial number 603). According to Danish law, no further approvals were required. The funding source had no role in the study design; data collection, analysis, or interpretation; writing the report; or decision to submit the manuscript for publication. Study cohorts were constructed using version SAS version 9.4 (SAS Institute, Cary, NC, USA), and analyses were performed using R software, version 4.2.1.

Descriptive data

We identified 179,680 women and 40,757 men with migraine (Table 1). From the general population, we sampled 898,365 women and 203,775 men without migraine. Median age on the index date was 41.5 years for women and 40.3 years for men. A somewhat larger proportion of individuals were identified in the study’s earliest calendar period.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Baseline characteristics [n, (%)] of women and men with and without migraine (identified by prescription data), 1996–2018.

https://doi.org/10.1371/journal.pmed.1004238.t001

Overall, individuals with migraine appeared to have slightly lower socioeconomic position than migraine-free individuals of the same sex (Table 1). Women appeared to have a higher level of education than men. Approximately 86% to 87% of men and women with migraine had none of the comorbidities included in the CCI, whereas approximately 90% to 91% of men and women without migraine were free of these comorbidities on their index date. Among persons of the same sex, prevalence of hypertension, VTE, obesity, thyroid disease, mood disorders, COPD, and hyperlipidemia appeared higher for individuals with migraine than for those without. Prevalence of alcohol-related disease was higher among men with migraine than among men without migraine. Compared to migraine-free individuals of the same sex, a larger proportion of individuals with migraine had redeemed prescriptions for beta-blockers, angiotensin-converting enzyme inhibitors or angiotensin II antagonists, diuretics, platelet inhibitors, and NSAIDs. A larger proportion of women with migraine had redeemed prescriptions for combined oral contraceptives and systemic progesterone-only contraceptives compared to women without migraine. See Table 2 for frequencies of comorbidities and redeemed prescription drugs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Individual baseline comorbidities and use of prescription drugs [n, (%)] among women and men with and without migraine (identified by prescription data), 1996–2018.

https://doi.org/10.1371/journal.pmed.1004238.t002

Risk of premature myocardial infarction, ischemic stroke, and hemorrhagic stroke

For all outcomes, total follow-up was approximately 12.4 million person years and the cohorts were followed for a median of 8.8 years (interquartile range: 4.3 to 4.4, 14.0 years). See Table 3 for number of events and AR estimates for 1 to 20 years of follow-up. The baseline risk of premature MI among individuals without migraine was higher for men than for women: AR 3.1% (95% confidence interval (CI) [2.9%, 3.2%]) versus 1.0% (95% CI [1.0%, 1.1%]). This was also the case for risk of premature ischemic stroke: AR 1.9% (95% CI [1.8%, 2.0%]) versus 1.2% (95% CI [1.1%, 1.2%]). The baseline risk of premature hemorrhagic stroke was nearly the same in men and women without migraine: AR 0.7% (95% CI [0.6%, 0.8%]) versus 0.6% (95% CI [0.5%, 0.6%]) (Fig 1 and Table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. AR of premature MI, ischemic stroke, and hemorrhagic stroke among women with and without migraine and among men with and without migraine (migraine identified by prescriptions).

The 95% CIs illustrated by the shadowed areas. Difference in impact of migraine was evaluated for men and women separately using Gray’s test for the entire follow-up period. For MI and ischemic stroke, the p-value was <0.001 for both women and men. For hemorrhagic stroke, the p-value was 0.001 for women and 0.517 for men. AR, absolute risk; CI, confidence interval; MI, myocardial infarction.

https://doi.org/10.1371/journal.pmed.1004238.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Number of events, ARs, and RDs for 1–20 years of follow-up for premature MI, ischemic stroke, and hemorrhagic stroke for women and men with and without migraine (identified by prescription data).

P values reflect Gray’s test for RDs.

https://doi.org/10.1371/journal.pmed.1004238.t003

The impact of migraine was estimated on the absolute scale as RDs and on the relative scale as HRs. The RD for premature MI among persons with migraine versus no migraine was 0.3% (95% CI [−0.1%, 0.6%]; p = 0.061) among men and 0.3% (95% CI [0.2%, 0.4%]; p < 0.001) among women. For premature ischemic stroke, the RD was 0.5% (95% CI [0.1%, 0.8%]; p < 0.001) for men and 0.3% (95% CI [0.2%, 0.4%]; p < 0.001) for women. For premature hemorrhagic stroke as the outcome, the RD was −0.1% (95% CI [−0.3%, 0.0%]; p = 0.176) for men and 0.1% (95% CI [0.0%, 0.2%]; p = 0.011) for women.

The following reported HRs are those with adjustments based on the directed acyclic graphs (Table 4). The adjusted HR of premature MI among persons with migraine versus no migraine was 1.07 (95% CI [0.97, 1.17]; p = 0.164) among men and 1.22 (95% CI [1.14, 1.31]; p < 0.001) among women. For premature ischemic stroke, the adjusted HR was 1.23 (95% CI [1.10, 1.38]; p < 0.001) for men and 1.21 (95% CI [1.13, 1.30]; p < 0.001) for women. For premature hemorrhagic stroke, the adjusted HR was 0.85 (95% CI [0.69, 1.05]; p = 0.131) for men and 1.13 (95% CI [1.02, 1.24]; p = 0.014) for women. The HR with additional adjustments showed similar results (Table 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Crude and adjusted HRs for 1–20 years of follow-up for premature MI, ischemic stroke, and hemorrhagic stroke (migraine identified by prescription data).

P values reflect likelihood ratio tests for HRs.

https://doi.org/10.1371/journal.pmed.1004238.t004

Sensitivity analyses

Risk estimates of ischemic stroke (including the code for unspecified stroke) are available in S1 Table. The RDs and HRs comparing individuals with migraine to individuals without migraine were similar to the main analysis.

Defining migraine by ICD codes rather than by redeemed medications, we identified 25,274 women with migraine and 7,397 men with migraine (Table A in S5 Text). The analyses where MI or ischemic stroke was outcome yielded similar results to the main analysis, albeit that the impact of migraine appeared greater for both women and men. For premature hemorrhagic stroke, the RD comparing migraine versus no migraine was −0.1% (95% CI [−0.6%, 0.4%]; p = 0.403) for men and 0.3% (95% CI [0.0%, 0.7%]; p = 0.011) for women. The adjusted HR was 1.16 (95% CI [0.73, 1.83]; p = 0.534) among men and 1.40 (95% CI: 1.07, 1.84]; p = 0.015) among women. Risk estimates are presented in Table B and Fig A in S5 Text.

When results were stratified by index year (1996 to 2006 or 2007 to 2018), we found a similar impact of migraine on risk of MI for women, although slightly attenuated, for both time periods. For men, we only saw an association with MI in the more recent time period (2007 to 2018). The HRs of ischemic stroke were similar to those of the main analyses for both men and women for both time periods. For premature hemorrhagic stroke, it may be there was an impact of migraine for women identified in 1996 to 2006. However, estimates for this outcome were imprecise. The HRs are available in S2 Table.