We conducted a retrospective cohort study using data from a 5 % random sample of Medicare beneficiaries. Medicare is a government health insurance program for US adults aged 65 years and older, and adults younger than 65 years with end-stage renal disease or who are disabled. Medicare data were obtained from the Centers for Medicare and Medicaid Services (CMS) Chronic Condition Data Warehouse. The institutional review board at the University of Alabama at Birmingham approved the study and waived the requirement to obtain informed consent.

We identified Medicare beneficiaries with a history of migraine between January 1, 2008 and December 31, 2017, using inpatient and outpatient healthcare utilization claims for migraine and pharmacy fills for acute migraine medications as defined in Supplemental Table 1. To provide adequate time to identify patient characteristics including the presence of comorbidities, we restricted the study population to Medicare beneficiaries who were living in the US and had continuous inpatient, outpatient, and pharmacy coverage for the 365 days prior to meeting the definition of history of migraine. As we were interested in studying history of migraine in older adults, we further restricted the study population to Medicare beneficiaries ≥66 years of age on the date they met the definition of history of migraine which means they were ≥ 65 years of age 365 days prior to meeting the definition of history of migraine (i.e., at the start of the look-back period). We required each beneficiary to be alive on the date they met the criteria for having a history of migraine. For each beneficiary, the earliest date they met all of the inclusion criteria described above was defined as their index date.

To serve as a control group, we identified beneficiaries who did not have a history of migraine, as defined above, using all available Medicare claims on or before December 31, 2017. For each beneficiary without a history of migraine, we randomly selected a day between January 1, 2008 and December 31, 2017 to serve as their index date. We applied identical inclusion/exclusion criteria for beneficiaries without a history of migraine as for those with a history of migraine. Specifically, we restricted the analysis to patients without a history of migraine who lived in the US and had continuous inpatient, outpatient, and pharmacy coverage for the 365 days before their index date. We further restricted the analysis to beneficiaries ≥66 years of age who were alive on their index date. For patients with and without a history of CVD, separately, beneficiaries without a history of migraine were frequency matched to those with a history of migraine four to one based on the calendar year of their index date, age in years on their index date, and sex.

We used Medicare beneficiary and claims data between January 1, 2007 and each patient’s index date to define characteristics (Supplemental Table 2). In addition to age on their index date, sex, and history of CVD, patient characteristics analyzed included race/ethnicity, receiving a low-income subsidy to pay for their health insurance, smoking status, history of diabetes, hypertension, chronic kidney disease (CKD), heart failure, dementia, depression, an anxiety disorder, insomnia, cancer, epilepsy, and hospitalization within the past year. Among beneficiaries with migraine, we further identified those with a history of migraine with aura. Beneficiaries who did not meet the definition for comorbidities listed above were assumed not to have those conditions. Area-level income was defined by the median income level within the patient’s zip code of residence according to data from the 2017 American Community Survey [16]. We also used Medicare pharmacy claims within 90 days prior to each patient’s index date to identify the use of antihypertensive medication, glucose-lowering medication, statins, non-statin lipid-lowering medications, medications for insomnia, and, among women, hormone replacement therapy. A list of these medications is provided in Supplemental Table 3.

For patients with a history of migraine, we used Medicare pharmacy claims within 90 days prior to each patient’s index date to identify use of acute migraine medications (i.e., triptans, ergotamine, nonsteroidal anti-inflammatory drugs [NSAID], and opioids) and preventive migraine medications (i.e., select antiepileptic agents, select antihypertensive agents, select antidepressants, botulinum toxin, and other agents). A list of the medications in each of these classes is provided in Supplemental Table 4.

Patients were followed from their index date through December 31, 2017 for ischemic stroke events and CHD events (i.e., myocardial infarction hospitalization or coronary revascularization), separately, with censoring occurring if they lost Medicare fee-for-service inpatient or outpatient coverage or died. The definitions of ischemic stroke and CHD events are provided in Supplemental Table 5. As a secondary outcome, we analyzed the risk for CVD events, defined as the composite of an ischemic stroke or CHD event.

All statistical analyses described below were conducted among Medicare beneficiaries with and without a history of CVD, separately. We calculated summary statistics of patient characteristics for those with and without a history of migraine. We calculated the cumulative incidence and incidence rates of ischemic stroke, CHD and CVD events for patients with and without a history of migraine accounting for the competing risk of all-cause mortality as described by Fine and Gray [17]. We used Cox proportional hazard models with the Breslow method for tie handling to estimate hazard ratios (HR) and 95% confidence intervals (CI) for the risk of ischemic stroke, CHD events, and composite CVD events among patients with versus without a history of migraine. Four models with progressive adjustment for covariates were used. Model 1 was unadjusted. Model 2 included adjustment for age, sex, and race/ethnicity. Model 3 included adjustment for age, sex, race/ethnicity, low-income subsidy, area-level income, smoking status, diabetes, hypertension, CKD, heart failure, dementia, depression, insomnia, cancer, epilepsy, and hospitalization within the past year. Model 4 included adjustment for the variables in the third model and use of antihypertensive medication, glucose-lowering medication, statins, non-statin lipid-lowering therapy, medication for insomnia, and among women, hormone replacement therapy. As prior studies have shown an association between migraine with aura and increased risk of ischemic stroke [18, 19], we also estimated incidence rates and HRs for ischemic stroke associated with a history of migraine with and without aura, separately.

The calculation of rates and HRs for ischemic stroke, CHD events, and CVD events described above was repeated comparing patients with migraine but not taking migraine medications, those with migraine taking each migraine medication class and taking two or more migraine medication classes, each versus their counterparts without migraine. Few patients with migraine with and without a history of CVD were using ergotamines or botulinum toxin and the number of outcome events during follow-up in these subgroups was low. Therefore, we did not analyze the risk for outcome events among patients with migraine taking ergotamines or botulinum toxin. All analyses were conducted using SAS 9.4 (SAS Institute) with a two-sided level of statistical significance of 0.05.