Use of combined hormonal contraception and stroke: A case-control study of the impact of migraine type and estrogen dose on ischemic stroke risk

Abstract

Objective

To clarify how factors such as estrogen dose and migraine history (including migraine subtype) impact ischemic stroke risks associated with combined hormonal contraceptive (CHC) use.

Background

CHC use in those with migraine with aura has been restricted due to concerns about stroke risk.

Methods

We conducted a case-control analysis of stroke risk associated with estrogen dose and migraine history among CHC users in a large tertiary care center. All women aged 18–55 who used a CHC between January 1, 2010, and December 31, 2019, were identified. Those with a stroke diagnosis were identified using ICD codes and confirmed via chart and imaging review. Details of personal and family medical history, stroke evaluation, ethinyl estradiol dosing (EE; ≥30 vs. <30 μg), and demographics were collected. From a random sample of 20,000 CHC users without stroke, a control cohort (n = 635) was identified and matched based on patient characteristics, medical and family histories, as well as stroke risk factors, to assess association between migraine diagnosis, migraine subtype, estrogen dose, and stroke.

Results

Of the 203,853 CHC users in our cohort, 127 had confirmed stroke (0.06%; CI 0.05%, 0.07%). In unadjusted analyses, a higher number of patients in the case cohort had a diagnosis of migraine (34/127, 26.8%) compared to controls (109/635, 17.2%; p = 0.011). Stroke risk was higher with ≥30-μg EE doses compared to those using a <30-μg dose (OR, 1.52; CI 1.02, 2.26; p = 0.040). Compared to no migraine, personal history of migraine increased the odds of stroke (OR, 2.00; CI 1.27, 3.17; p = 0.003). Compared to no migraine, stroke risk was not significantly increased in those with migraine with aura, but migraine without aura increased the risk (OR, 2.35; CI 1.32, 4.2; p = 0.004).

Conclusions

Overall stroke risk in our cohort of CHC users was low. When CHCs are used in those with migraine, formulations containing ≤30 μg EE are preferred. Shared decision-making should include discussions about ischemic stroke risks in patients with migraine, even those without aura.

Abbreviations

CHC

combined hormonal contraceptive

CI

confidence interval

DVT

deep venous thrombosis

EE

ethinyl estradiol

EMR

electronic medical record

FH

family history

ICD

International Classification of Diseases

MWA

migraine with aura

OR

odds ratio

SMD

standardized mean difference

INTRODUCTION

There is a critical need to ensure that women of reproductive age have an effective form of contraception, including patients with migraines. Migraine is a common headache disorder affecting up to 20% of the general population, with women three times more likely than men to be affected.¹ Cumulative incidence in women across a lifetime is 43%, with highest prevalence from ages 30–39.^{2, 3} This peak of migraine burden affects people during their reproductive years and is a leading cause of disability, which may affect a woman's personal relationships, career, family planning, and overall health.^{4, 5} Reproductive age women with migraine commonly use oral contraceptives for both pregnancy prevention and medical indications. However, some guidelines discourage the use of combined hormonal contraceptives (CHCs; estrogen-containing contraceptives) in women who experience migraine with aura due to concerns about ischemic stroke.⁶ Given that one third of patients with migraine have accompanied aura (visual, sensory, aphasic, or motor), this limits women's choices of CHC use not only for contraception, but also for treatment of multiple other medical conditions such as acne, hirsutism, endometriosis, premenstrual dysphoric disorder, menstrual irregularity, prevention of ovarian cysts, or control of menstrual migraines.⁷ This may lead to issues of nonadherence to a contraceptive method if it does not align with patients' needs and interests. There are times when the clinical benefits of a CHC are felt to outweigh the risks for an individual patient, despite the presence of medical comorbidities that may impact these risks; thus, it is of utmost importance to know if there are ways to prescribe CHCs that may mitigate these risks.

How factors such as estrogen dose and migraine history (including migraine subtype) impact stroke risks associated with low-dose CHC use is not only based on limited observational data,⁸ but also has been contradictory.^{9, 10} The use of exogenous estrogen in contraceptives and migraine with aura (MWA) have been independently associated with increased risk of arterial thrombosis, though there is no clear evidence that these risk factors compound to further increase risk in an individual patient.¹¹ Migraine without aura has not been shown to consistently impact stroke risk; in contrast, women with MWA have higher odds of stroke, especially in those with a high migraine attack frequency (>12 attacks per year).^{8, 12} The risk of arterial thrombosis is also related to the dose of estrogen. Contraceptives with an ethinyl estradiol (EE) content of ≥50 μg are associated with the highest risk of ischemic stroke, whereas currently used regimens containing ≤30 μg are associated with lower risks. These findings are based on observational studies which have important limitations in that they have assessed higher dose formulations (not commonly used at the present time) or are in populations with high rates of tobacco use.^{1, 8} It is less clear whether currently used ultra-low dose formulations (containing ≤20 μg EE) confer a reduced stroke risk in those who suffer with migraines. Despite these gaps in the evidence, guidelines generally recommend avoiding CHCs in women who have MWA because, though stroke is rare in women of reproductive age, the consequences to an individual woman and her family may be severe.⁶

The primary aim of this study was to examine risk factors for ischemic stroke among CHC users, looking at a large population of reproductive age women who were using lower dose methods consistent with the currently accepted prescribing approach (<50 μg EE). We sought to understand whether migraine, especially MWA, further compounds ischemic stroke risks in CHC users, and whether there was a difference in stroke risks based on more commonly prescribed estrogen doses (30–35 vs. <30 μg EE). Our hypothesis was that the ischemic stroke risk in women with migraine using CHCs would be increased.

METHODS

This is a case-control study of all women between the ages of 18–55 who have used CHCs (including pills, patches, and vaginal rings) over the course of a decade in a large, urban, academic tertiary care center. We queried medication lists within the electronic medical record (EMR) to identify all women who were actively using a CHC between the dates of January 1, 2010, and December 31, 2019; menopausal dose estrogen–progestin replacement and progestin-only contraceptives were excluded. Among this cohort of CHC users, we identified individuals who had suffered an ischemic stroke based on the International Classification of Diseases, Ninth Revision (ICD-9) and Tenth Revision (ICD-10) codes. The records of the identified patients were reviewed by four neurologists and a women's health specialist with subspecialty training in headache to validate (1) stroke diagnosis, (2) migraine presence/subtype, and (3) active use of CHC and dose. Imaging findings were reviewed to confirm ischemic stroke diagnoses by the neurology team. Patients were excluded if the ischemic stroke diagnosis was not confirmed on chart review (determined by reviewing imaging and clinical notes). Patients diagnosed with transient ischemic attack, hemorrhage (subarachnoid, subdural, traumatic), neonatal or in utero strokes were excluded because most of the currently published evidence has shown an increase in ischemic stroke (as opposed to hemorrhagic) with CHC use (see Figure 1). Those not on CHCs at the time of stroke were also excluded. Duration of CHC use prior to the stroke date was not assessed. The intention was to exclude those who had been pregnant within the year prior to the stroke, but no patients fit this criterion. The cohort of CHC users was preferred because it is a much more clearly defined cohort, good for matching, and increased the chances of follow-up within our health system. In addition, this chosen cohort is more representative of patients typically encountered in a gynecologic or primary care practice where CHCs are prescribed for a variety of indications, despite the presence of medical comorbidities.

FIGURE 1

Case cohort record disposition. CHC, combined hormonal contraceptive; ICD, International Classification of Diseases.

Charts were reviewed to assess demographics, personal and family medical history, the stroke evaluation including cardiac records, and to confirm the use of CHCs prior to the stroke diagnosis within the consultant notes and medication lists. Migraine characteristics were diagnosed using ICD codes from the past medical history, problem list, or diagnosis fields. If migraine with and without aura ICD-10 codes were present for an individual patient, or a nonspecific ICD-9 diagnostic code was used, then chart review was done by the neurology headache team to confirm the final diagnosis (using the International Classification of Headache Disorders). The estrogen dose of the CHC in the case cohort was subdivided into categories (<20, 20–29, and ≥30 μg EE). Any reports requiring clarifications were addressed via discussion and repeat chart review. The EMR data include the following: inpatient, emergency room, and outpatient clinical notes (from the tertiary care institution as well as multiple community affiliates); past personal medical history and problem lists with ICD codes; medication lists (which include original clinician prescriptions, pharmacy reconciled medications, and patient reported medications); and family histories (in the form of predefined EMR subcategories and free text, as well as ICD codes). Historical medications reported by the patient in the case cohort were validated with a chart review, including review of medication lists in the neurology consultant notes (given that CHC use is consistently ascertained on history in the setting of a stroke in a young woman). If CHC formulations had been changed, the highest dose used within the year prior to the stroke date was recorded. Because personal and family histories may have developed for any individual patient over the course of the 10-year study period, relevant history was recorded as being present if noted at any time prior to or within 1 year after the index stroke date. We did not assess a temporal relationship between CHC use and migraine diagnosis, but we did ensure that the patient was on a CHC prior to the stroke diagnosis. Data were extracted by a systems analyst with expertise in data acquisition from the EMR. This EMR was the sole method used to capture data at our institution throughout the study period, and the population seen at the medical center is representative of the individuals residing within this geographic region, representing different races and varied socioeconomic status. The patient population includes those from a variety of private and government insurance plans, who may be in contact with the system for varied lengths of time. All data were entered in a Research Electronic Data Capture (REDCap) database.

From a random sample of 20,000 CHC users without stroke (from the same original cohort), a control cohort of 635 patients was identified by 1:5 greedy matching based on personal and family history variables as outlined in Table 1. The cohorts were compared to assess the variables of interest: estrogen dose (≥30 vs. <30 μg EE), presence of migraine (yes or no), and migraine type (aura vs. no aura). The records of the identified patients in the control cohort were reviewed to validate MWA (by reviewing clinician notes). Additional chart review was done if there was a discrepancy noted in the diagnoses or CHC dosing over time; that is, if a patient was on a variety of CHC formulations, the highest dose of CHC used was included. If migraine diagnosis or type was not clear after chart review then patients were not included in the control cohort, and replaced by patients where the migraine diagnosis and subtype were accurately and consistently identified. Several variables were used to match the cohorts, including age, body mass index, race, history of tobacco use, and personal medical history (presence of diabetes, deep venous thromboembolism, pulmonary embolism, hypertension, lupus, patent foramen ovale, hyperlipidemia, peripheral vascular disease, atrial fibrillation, and alcohol use). The cohorts were also matched for family history (stroke or cerebral embolism, cardiovascular disease or deep venous thrombosis [DVT], diabetes, hypertension, hyperlipidemia, and kidney disease) which was assessed by querying the family history field and based on ICD diagnosis codes. Cohorts were matched for family history only in a first-degree relative (mother, father, sister, brother, child). The family history risk factors were grouped into clinically meaningful categories. Table 1 lists all matching variables. Table S1 describes characteristics of the random sample of 20,000 CHC users (control pool), and the table can be an approximate depiction of the overall cohort of 203,853 patients.

TABLE 1. Characteristics of the stroke case cohort and the matched control.

CHC users	Total (N = 762)	No stroke (controls, N = 635)	With stroke (cases, N = 127)	p-value
Body mass index (in kg/m^2)	29.7 ± 8.0	29.8 ± 7.8	29.4 ± 8.9	0.647^a
Age at stroke for cases, CHC initiation for control (in years)	35.0 ± 9.3	35.1 ± 9.3	34.3 ± 9.0	0.361^b
Race				0.665^c
White	643 (84.4)	537 (84.6)	106 (83.5)
Black	103 (13.5)	86 (13.5)	17 (13.4)
Multiracial/multicultural	16 (2.1)	12 (1.9)	4 (3.1)
Smoking status				0.524^c
Current or passive smoker	60 (7.9)	50 (7.9)	10 (7.9)
Former smoker	141 (18.5)	113 (17.8)	28 (22.0)
Never-smoker	561 (73.6)	472 (74.3)	89 (70.1)
Diabetes	91 (11.9)	76 (12.0)	15 (11.8)	0.960^c
Deep venous thrombosis	25 (3.3)	20 (3.1)	5 (3.9)	0.590^d
Pulmonary embolism	2 (0.26)	0 (0.00)	2 (1.6)	0.028 ^d
Hypertension	255 (33.5)	211 (33.2)	44 (34.6)	0.757^c
Systemic lupus erythematosus	19 (2.5)	14 (2.2)	5 (3.9)	0.226^d
Patent foramen ovale	22 (2.9)	17 (2.7)	5 (3.9)	0.393^d
Hyperlipidemia	118 (15.5)	98 (15.4)	20 (15.7)	0.929^c
Peripheral vascular disease	4 (0.52)	1 (0.16)	3 (2.4)	0.016 ^d
Atrial fibrillation	0 (0.00)	0 (0.00)	0 (0.00)
Alcohol abuse	0 (0.00)	0 (0.00)	0 (0.00)
FH: Stroke or cerebral embolism	89 (11.7)	75 (11.8)	14 (11.0)	0.801^c
FH: Cardiovascular disease or deep venous thrombosis	183 (24.0)	164 (25.8)	19 (15.0)	0.009 ^c
FH: Diabetes	202 (26.5)	172 (27.1)	30 (23.6)	0.419^c
FH: Hypertension	353 (46.3)	305 (48.0)	48 (37.8)	0.035 ^c
FH: Hyperlipidemia	93 (12.2)	82 (12.9)	11 (8.7)	0.181^c
FH: Kidney disease	26 (3.4)	22 (3.5)	4 (3.1)	0.999^d

• Note: Statistics presented as mean ± SD, N (column %). Bold italic indicates p < 0.05 is considered statistically significant.
• Abbreviations: CHC, combined hormonal contraceptive; FH, family history.
• ^a Satterthwaite t-test.
• ^b t-test.
• ^c Pearson's chi-square test.
• ^d Fisher's exact test.

Statistical analyses

Approximately normally distributed continuous measures were summarized using means and standard deviations and compared using two-sample t-tests. Categorical factors were summarized using frequencies and percentages and were compared using Pearson's chi-square tests or Fisher's exact tests. Data on estrogen dose were summarized using means and standard deviations, medians and quartiles, and frequencies and percentages, and compared using Wilcoxon rank-sum tests. Standardized mean differences (SMDs) were calculated for demographic variables before and after matching.

To further examine the association between migraine and estrogen dose with stroke, mixed-effects logistic regression models were created to predict stroke, that is, the case cohort. Family history (FH) of cardiovascular disease or DVT, and FH of hypertension were controlled in these models; the two FHs were included because they were significantly different with an SMD of >0.2 after matching (thus are intended only for controlling unbalanced matching variables, and not interpretable for clinical meaning). Matching cluster effects were controlled by including random intercepts were used to control effects of matching clusters, and compound-symmetry covariance-structure was assumed Fixed effect odds ratios (ORs) and 95% confidence intervals (CIs) are presented in Tables 3 and 4. As a sensitivity analysis, variables with an SMD of >0.1 after matching were included in two additional models (including migraine, estrogen dose, smoking status, past medical history of systemic lupus erythematosus, FH of cardiovascular disease or DVT, FH of hypertension, and FH of hyperlipidemia). Although past medical history of pulmonary embolism and peripheral vascular disease were also significantly different, with SMDs of >0.2 after matching, they were too rare to be included in any of the logistic regressions.

This is the primary analysis of preplanned (i.e., a priori) data; no statistical power calculation was conducted prior to the study, as the sample size was based on the available data. All analyses were done using SAS (version 9.4, SAS Institute, Cary, NC) and a level of p < 0.05 in two-tailed testing was considered statistically significant. Greedy matching was performed via gmatch.sas (Erik Bergstralh & Jon Kosanke 08/2007). The study underwent review with the institutional review board (IRB) of Cleveland Clinic and was approved; written informed consent was waived.

RESULTS

We identified 203,853 distinctive patients aged 18 to 55 who used CHCs at some point between January 1, 2010, and December 31, 2019. Of this cohort, 313 patients had an ICD diagnosis code for stroke. Chart review showed only 127 had confirmed ischemic stroke on imaging after hospital admission and were on the CHC at the time of the diagnosis (0.06%; CI 0.05%, 0.07%; see Figure 1). Table 1 describes the characteristics of the case cohort and the matched controls. On average, patients in our cohort who suffered a stroke while on CHC were overweight (mean BMI 29.4 kg/m²), White (84%), and never-smokers (70%). Most of the women did not have obvious risk factors for stroke, but when a risk factor was present, the most common was hypertension (35%), followed by hyperlipidemia (16%), and diabetes (12%).

Table 2 shows the variables of interest in cases versus controls. In this unadjusted analysis, there were significantly different rates of migraine diagnoses in the stroke cohort compared to controls. A higher number of patients in the case cohort had a diagnosis of migraine: 26.8% reported migraine versus 17.3% in the controls (p = 0.011). Furthermore, 11.0% of the case cohort reported migraine with aura versus 8.5% in the controls, and 15.7% of the case cohort reported migraine without aura versus 8.7% in the controls (p = 0.025). Because of the small number of patients, the estrogen doses were classified into those that are <30 versus ≥30 μg EE. Use of a CHC with an EE dose of ≥30 μg was more frequent in the case cohort compared to the controls (62.2% vs. 51.7%; p = 0.030). Of note, there were four patients using a 50 μg pill among our cases and controls, and all were in the stroke case cohort. The numbers of CHC patch and ring users (as opposed to the use of CHC pills) were too small to run separate analyses to assess the impact of delivery method on stroke risk, but are categorized based on EE dose. Data on the type of progestin used were collected, but impact of the progestin formulation on stroke risk was not studied because the numbers were too small to draw meaningful conclusions.

TABLE 2. Unadjusted variables in cases versus controls.

Factor	Total (N = 762)	Control (N = 635)	Case (N = 127)	p-value
Estrogen dose in μg	26.4 ± 7.8	26.0 ± 7.7	28.0 ± 8.3	0.039 ^a
	30.0 [20.0, 35.0]	30.0 [20.0, 35.0]	30.0 [20.0, 35.0]
10	29 (3.8)	26 (4.1)	3 (2.4)
15	51 (6.7)	43 (6.8)	8 (6.3)
20	260 (34.1)	226 (35.6)	34 (26.8)
25	15 (2.0)	12 (1.9)	3 (2.4)
30	168 (22.0)	134 (21.1)	34 (26.8)
35	235 (30.8)	194 (30.6)	41 (32.3)
50	4 (0.52)	0 (0.00)	4 (3.1)
Estrogen dose in μg				0.094^b
<20	80 (10.5)	69 (10.9)	11 (8.7)
20–29	275 (36.1)	238 (37.5)	37 (29.1)
≥30	407 (53.4)	328 (51.7)	79 (62.2)
Estrogen dose in μg				0.030 ^b
<30	355 (46.6)	307 (48.3)	48 (37.8)
≥30	407 (53.4)	328 (51.7)	79 (62.2)
Migraine				0.011 ^b
No migraine	619 (81.2)	526 (82.8)	93 (73.2)
Migraine present	143 (18.8)	109 (17.2)	34 (26.8)
Migraine by aura				0.025 ^b
No migraine
Without aura	75 (9.8)	55 (8.7)	20 (15.7)
With aura	68 (8.9)	54 (8.5)	14 (11.0)

• Note: Statistics presented as mean ± SD, Median (P25, P75), N (column %). Bold italic indicates p < 0.05 is considered statistically significant.
• ^a Wilcoxon rank sum test.
• ^b Pearson's chi-square test (can only have one p-value and it is for overall distribution, so it does not differentiate cohorts when there are more than two cohorts). Estrogen dosing compared to each other in a relative measure within the case cohort.

Tables 3 and 4 show mixed-effects models to assess stroke risk based on estrogen dose, presence of migraine, as well as migraine type. Table 3 shows that, on average and controlling other effects compared to no migraine, having migraine was associated with increased odds of stroke (OR, 2.00; CI 1.27, 3.17; p = 0.003). Similarly, being on EE doses of 30 μg or higher was associated with increased odds of stroke compared to being on formulations containing <30 μg (OR, 1.52; CI 1.02, 2.26; p = 0.04). Table 4 shows that there was no significant difference in the odds of stroke between those with a history of MWA compared to those with no migraine (OR, 1.66; CI 0.87, 3.15; p = 0.126). In contrast, having migraine without aura was associated with increased odds of stroke compared to having no personal history of migraine (OR, 2.35; CI 1.32, 4.2; p = 0.004). The results from sensitivity analysis had minimal changes from the original: migraine versus no migraine (OR, 2.05; CI 1.29, 3.26; p = 0.003); EE doses of 30 μg or higher versus <30 μg (OR, 1.54; CI 1.03, 2.29; p = 0.035); MWA versus no migraine (OR, 1.69; CI 0.88, 3.25; p = 0.113); and migraine without aura versus no migraine (OR, 2.39; CI 1.34, 4.3; p = 0.003). Tables are not included to avoid redundancy.

TABLE 3. Mixed effect logistic model predicting cases versus controls (migraine vs. no migraine, N = 762).

Factor	Reference	OR (95% CI)	p-value
Migraine	No migraine	2.00 (1.27, 3.17)	0.003
Estrogen dose ≥30 μg	<30 μg	1.52 (1.02, 2.26)	0.040
^aFH: Cardiovascular disease or deep venous thrombosis	No FH	0.41 (0.24, 0.70)	0.001
^aFH: Hypertension	No FH	0.53 (0.36, 0.80)	0.003

• Note: Bold italic indicates p < 0.05 is considered statistically significant.
• Abbreviations: CI, confidence interval; FH, family history; OR, odds ratio.
• ^a Variables controlled in these models because they were significantly different after matching (thus are intended only for controlling unbalanced matching variables, and not interpretable for clinical meaning).

TABLE 4. Mixed effect logistic model predicting cases versus controls (migraine with or without aura vs. no migraine, N = 762).

Factor	Reference	OR (95% CI)	p-value
Migraine with aura	No migraine	1.66 (0.87, 3.15)	0.126
Migraine without aura	No migraine	2.35 (1.32, 4.17)	0.004
Estrogen dose ≥30 μg	<30 μg	1.52 (1.02, 2.26)	0.040
^aFH: Cardiovascular disease or deep venous thrombosis	No FH	0.41 (0.24, 0.70)	0.001
^aFH: Hypertension	No FH	0.53 (0.35, 0.80)	0.003

• Note: Bold italic indicates p < 0.05 is considered statistically significant.
• Abbreviations: CI, confidence interval; FH, family history; OR, odds ratio.
• ^a Variables controlled in these models because they were significantly different after matching (thus are intended only for controlling unbalanced matching variables, and not interpretable for clinical meaning).

DISCUSSION

The primary aim of this study was to understand the impact of migraine diagnosis and estrogen dose on ischemic stroke risks in CHC users, looking at a large population of reproductive age women who were using low dose estrogen-containing formulations. Although our findings did show an increased risk of stroke in individuals with migraine who were using a CHC, the study surprisingly indicated that the increased risk was limited to those who have migraine without aura. Because MWA has typically been considered a contraindication to CHC use, clinicians may have been hesitant to use a CHC in this patient population, leading to a smaller number of individuals with MWA on CHC, causing our results to be underpowered. Our findings suggest that all women with migraine who use CHC should be similarly counseled on ischemic stroke risks regardless of whether aura is present. This should include a shared decision-making process in which stroke risk associated with the use of CHCs is compared to stroke risks during pregnancy. The published evidence regarding the risk of ischemic stroke with the use of CHCs in the setting of migraine is sparse. Though prior studies have not consistently demonstrated an increased stroke risk in those who have migraine without aura, they have shown an approximately 2-fold increase in risk of ischemic stroke in the presence of MWA, a 6-fold increase with the use of CHC (only if aura was present), and up to a 9-fold increase when there were additional risk factors of smoking and the use of CHC.^{10, 13-15} The Centers for Disease Control have adapted guidelines from the World Health Organization for US use, stating that the risk of CHC in patients with MWA is considered “unacceptable” at any age.⁶ In contrast, the guidelines state that in those without aura, the benefits of CHC typically outweigh the risks, but encourage clinicians to consider other risk factors for stroke before finalizing decisions. Given that ischemic stroke is uncommon in this population, future studies are needed to investigate our study's findings. This includes studies that evaluate larger and more diverse populations of patients to best understand whether there is a significantly increased risk in CHC users who have MWA versus those without aura.

For women with migraine who prefer CHCs for contraception or a medical need, our findings suggest that formulations containing <30 μg EE may help minimize stroke risks, regardless of whether the individual has migraines with or without aura. This is consistent with systematic reviews of observational data suggesting that decreasing EE doses favorably impacts stroke, though data are sparse. A meta-analysis showed decreasing ORs for ischemic stroke with decreasing oral CHC doses: ≥50 μg EE, 30–40 μg EE, and 20 μg EE had ORs of 3.28 (95% CI, 2.49, 4.32), 1.75 (95% CI, 1.61, 1.89), and 1.56 (95% CI, 1.36, 1.79), respectively.^{16, 17} Approximately one third of our cases and controls were using 20–29 μg EE doses, while a minority (≤10%), were using formulations <20 μg EE (also referred to as ultra-low dose). It is not yet known whether the use of formulations containing <20 μg will further reduce ischemic stroke risk. More research is needed to understand the arterial thrombosis risks of 10–19 μg EE contraceptive formulations, though limited options and affordability concerns often limit access to these choices in clinical practice.

Our results show a very low risk of stroke (n = 127) in this cohort of 203,853 CHC users, yielding an overall stroke incidence of 0.06% in our patients. Previously published reports of ischemic stroke incidence in reproductive age women are limited, but vary between 3.4 and 64 cases in 100,000 females per year, though direct comparisons cannot be made to our data since the incidence varies depending on age, CHC formulation, and the presence of arterial thrombosis risk factors.^{10, 18} We controlled a comprehensive list of risk factors via matching, so the control cohort had elevated risk of stroke, it should be noted that our findings of increased ischemic stroke risk with migraine or higher EE doses are in relative measure. Thus, it is possible that our findings are not generalizable to a population without stroke elevated risk factors.

When pregnancy prevention is the primary goal, women at highest risk of arterial thrombotic events may benefit from highly effective, estrogen-free contraceptive options such as an intrauterine device (IUD) or progestin-only arm implant. However, women use estrogen-containing CHCs for a variety of reasons, including tolerability, desire for self-administration (contributing to the ease of initiating or ending the method), or to avoid procedures. Furthermore, many prefer a CHC to treat a medical condition that responds best to the use of an estrogen-containing method such as hirsutism, acne, or abnormal uterine bleeding. Lowering the EE dose to <30 μg can decrease stroke risk, but may also impact method efficacy and the side effect profile. Thus, the choice of using the lowest doses of CHC must be balanced with the high rates of unintended pregnancy in the US in all reproductive age groups, including adolescents and perimenopausal women. It is important to remember that the arterial and venous thrombosis risks of an unintended pregnancy are higher than the risk of any contraceptive. Because the use of CHC may be based on multiple clinical factors outside of pregnancy prevention, updated US and European Headache Society recommendations encourage individualized shared decision-making, even in individuals who have MWA.^{8, 11} Nonmigraine risk factors for stroke should be taken into consideration when treatment is offered.

There are several strengths to our study. We evaluated a large population of CHC users who came into contact with our system in the span of a decade, with a well-established EMR which ties together inpatient, outpatient, and emergency room data, as well as imaging, which helped to verify stroke diagnoses. Also, a large population of patients ensured available control cohorts where migraine had been diagnosed previously by using ICD codes with specific migraine subtypes, consistently identified throughout the chart. Given that making a clear migraine diagnosis is a known challenge in clinical care, additional chart review by the authors who are headache specialists was key to ensuring that migraine subtypes within both the case and control cohorts were properly categorized, using established criteria (based on clinician notes or questionnaires). If this was not clear based on chart review, then these patients were not included in the cohort, and replaced by patients where the migraine diagnosis and subtype were accurately identified.

Limitations of the study are related to the retrospective nature of the design. Use of EMR data for ICD codes, medication history, and patient characteristics could result in misclassification for both exposures and outcomes. We did try to control for this with a careful chart review of both the case and control cohorts, but for the controls only to confirm migraine diagnosis or estrogen dose. It is possible that our sample size was too small to examine the effects of migraine subtypes on stroke risk, or that these patients may have had other risk factors that were not accounted for within the analysis. This may differentially impact the MWA cohort who are less likely to be prescribed CHC, especially if other stroke risk factors are present. We were not able to verify whether patients were adherent to taking their medications, nor could we confirm the dosing of patient-reported historical medications. Though we ensured that patients were on a CHC at the time of their stroke diagnosis, we did not assess how long they had been on the treatment, given that thrombosis risks are greatest within the first year of starting treatment. Because the observation period spanned 10 years, the risk periods for the participants in the study could vary and individuals could cycle through different exposures during the observation period. There is concern that this may differentially impact individuals with migraine given they may be more likely to receive medical care and to be surveilled for stroke. However, our entire study population consisted of CHC users who are also more likely to receive consistent medical care and be surveilled for stroke risk factors prior to receiving a refill. We also did not control for the use of any antithrombotic or migraine preventive medications. We could identify whether a patient had ever smoked, but could not correlate the timing of tobacco use with contraceptive use. We excluded transient ischemic attacks and hemorrhagic strokes from our analysis given that most of the currently published evidence focuses on ischemic strokes; this may underestimate the impact of the studied variables on all cerebrovascular outcomes. Also, the absence of stroke in the control cohort (at a future time, outside of our system) could not be verified; any care received outside of our system may be missed, leading to an underestimate of stroke cases. Thus, duration of follow-up was also variable. However, this institution has over 200 outpatient locations and just under 20 hospitals within the system, and accepts a very wide array of private and government insurance plans, encompassing close to half of the care provided within the city. Thus, most of the patients in our sample have likely come into contact with healthcare systems and clinicians that also utilize the same EMR, allowing for the collection of data from patients who may have sought healthcare outside of our institution. Finally, our case cohort was primarily White and overweight, and we did not assess socioeconomic variables, which may limit the generalizability of the findings.

CONCLUSION

There has been an accepted counseling approach for young and otherwise healthy women diagnosed with migraine with aura that the use of CHCs is contraindicated, because of concern that the presence of aura significantly increases stroke risk in contrast to the absence of aura. However, this counseling approach has limited the contraceptive options available to this patient population, increasing the risk of unintended pregnancy. Our findings highlight the importance of appropriate patient education and shared decision-making for all women who start a CHC with a history of migraine, even those without aura. Women with migraines who prefer to be on estrogen-containing contraceptives for pregnancy prevention or a medical need should consider options that have <30 μg EE, and possibly <20 μg, if medically appropriate.

AUTHOR CONTRIBUTIONS

Study concept and design: Payal Soni, Maryann Mays, Julia Bucklan, Aarushi Suneja, Ruth Farrell, Meng Yao, Pelin Batur. Acquisition of data: Payal Soni, Julia Bucklan, Aarushi Suneja, Meng Yao, Pelin Batur. Analysis and interpretation of data: Payal Soni, Maryann Mays, Julia Bucklan, Aarushi Suneja, Ruth Farrell, Meng Yao, Pelin Batur. Drafting of the manuscript: Maryann Mays, Julia Bucklan, Ruth Farrell, Meng Yao, Pelin Batur. Revising it for intellectual content: Payal Soni, Maryann Mays, Julia Bucklan, Aarushi Suneja, Ruth Farrell, Meng Yao, Pelin Batur. Final approval of the completed manuscript: Payal Soni, Maryann Mays, Julia Bucklan, Aarushi Suneja, Ruth Farrell, Meng Yao, Pelin Batur.

CONFLICT OF INTEREST STATEMENT

Pelin Batur, Meng Yao, Julia Bucklan, Payal Soni, Aarushi Suneja, Ruth Farrell, and Maryann Mays declare no conflicts of interest.