Risk of Stroke in Women Using Levonorgestrel-Releasing Intrauterine Device for Contraception

Requests for data from this study are available from the corresponding author on reasonable request.

In this historical cohort study of all Danish women aged 18 to 49 years who lived in Denmark during some or all of the period between January 1, 2004 and December 31, 2021, we studied the association between incident IS and ICH and use of the LG-IUD by linking data from Danish registries with national coverage.^5–7 The cohort was linked to the registries by a unique personal identification number given to all residents in Denmark.⁶ All women hospitalized with an incident IS or ICH (study outcomes) in the study period were identified through the Danish Stroke Registry.⁵ Linking to the Danish National Prescription Registry,⁷ we obtained information on current use of LG-IUD and other HCs. Data on contraceptives, pregnancies, and variables used for adjustments in the statistical analyses (age, ethnicity, education, calendar year, and medication use for stroke risk factors) were obtained from Statistics Denmark⁶ and the Danish National Prescription Registry.⁷

If a woman was not present in Denmark at the start of her follow-up period, that is, her turning 18 or on January 1, 2004, whichever comes last, she was excluded from the study. Women were censored during pregnancy (in the 9 months before delivery) and 3 months after delivery and at the first prescription of ovarian stimulation drugs (Anatomical Therapeutic Chemical classification [ATC] code G03G). Women were censored at death, first emigration from Denmark, on their 50th birthday, or at the end of follow-up period. Additionally, women were censored during their use of non-IUD HCs.

The study protocol was approved by the Danish Stroke Registry and the Danish Data Protection Agency (FSEID-00006480). This study is reported as per the guidelines for the STROBE checklist—observational studies (Supplemental Material).

Following exclusion and censoring of data (Figure S2), a total of 1 681 611 nonpregnant women contributing 11 971 745 person-years (py) of observation were included in the study (Table 4). Mean age at inclusion was 30.0 years (interquartile range [IQ]: 25–75; 19.1–39.2 years); mean length of follow-up was 7.1 years (IQ, 25–75; 2.4–11.4 years). Denmark was the country of origin for 89.5% of the women (Table 4).

As for pregnancy, mean age at first pregnancy was 28.9 years. Of the women, 462 244 became pregnant (1×, 192 554; 2×, 173 795; 3 or more times, 95 895). Censored time due to pregnancy was 752 898 years.

During the study period 2916 women (24.4 per 100 000 py) had a first-ever IS while 367 (3.1 per 100 000 py) had a first-ever ICH (Table 4).

Incidence rates and adjusted IRR of IS and ICH related to age, and calendar year are shown in Table 1. Educational level, country of origin, and medication use for risk factors are shown in Table 2. Incidence rate of IS and ICH increased almost exponentially with age. The incidence rate of IS was rather stable and for the most part did not change significantly, although there is a slight pattern of decrease in risk in the later study years. The incidence rate of ICH did not change significantly in the study period.

Women originating from non-Western countries had significantly lower incidence rates of IS when compared with native Danes while incidence rates of ICH were not affected by country of origin.

Adjusted IRR of IS and ICH was increased for women with medication use for risk factors. The only exception was women using triptanes who had an increased IRR of IS but not of ICH.

In total, 364 784 users of LG-IUDs contributed 1 720 311 py to the investigation: mean age at start of LG-IUD use was 34.6 years; mean length of usage was 4.7 years. In total, 1 653 959 women contributed 10 251 434 py of nonuse of HC to the investigation; the mean age at inclusion was 30.0 years; the mean length of follow-up was 6.2 years (Table 4).

Women using LG-IUDs containing 52 mg levonorgestrel (Mirena or Levosert) contributed 1 561 571 py, users of LG-IUDs containing 13.5 mg (Jaydess) and 19.5 mg (Kyleena) contributed 52 515 and 106 225 py, respectively.

Figure S1 shows use of LG-IUD and non-IUD HC for different age groups across the study period. Use of LG-IUD increases both with age and throughout the study period, with barely any use in the early 2000s compared with almost 25% 50-year-old women using LG-IUD in the last year of the study period. Use of non-IUD HC remains relatively stable across study period for younger population, whereas the percentage of older women using non-IUD HC rises steadily throughout the study period.

Compared with nonusers of HC (adjusting for education, country of origin, medication use for risk factors, age, calendar year, and use of contraception) users of LG-IUD had a 22% lower incidence rate of IS (IRR, 0.78 [CI, 0.70–0.88]) while incidence rate of ICH was not affected (IRR, 0.94 [CI, 0.69–1.28]; Table 3); not different from unadjusted incidence rate (IS: IRR, 0.76 [CI, 0.68–0.86]; ICH: IRR, 0.96 [CI, 0.72–1.29]).

A conservative sensitivity analysis where usage time was set to 2 years for every LG-IUD implant and follow-up time was censored all the way until a new HC prescription was cashed in reduced the py of use to 841 202 still showed significant relative incidence rate reduction (IRR, 0.83 [CI, 0.70–0.98]).

Incidence rates of IS did not differ significantly between age groups 18 to 29 years, 30 to 39 years, and 40 to 49 years. Incidence rates of ICH did not differ significantly between age groups 30 to 39 years and 40 to 49 years; risk was however increased for the age group 18 to 29 years when compared with the age group 40 to 49; IRR, 4.49 (CI, 1.65–12.19; Table 3).

Due to the differing exposure patterns over the study period, we performed a sensitivity analysis excluding all the data before 2013 (half of study period) focusing only on the most recent data. The results remain unchanged (Table S2).

For further clarification, we did a comparison between LG-IUD users and users of non-IUD HCs (Table S3). Compared with users of non-IUD HCs (adjusting for education, country of origin, medication use for risk factors, age, and calendar year), users of LG-IUD had 36% lower incidence rate of IS (IRR, 0.64 [CI, 0.56–0.72]) while incidence rate of ICH was not affected IRR, 0.87 [CI, 0.64–1.20]). For IS, incidence rate was reduced independent of age. For ICH, incidence rate was not affected for women >30 years of age but increased (IRR, 4.76 [CI, 1.79–12.65]) for women 18 to 29 years of age (as seen when LG-IUD users were compared with nonusers of HC).

We predicted stroke incidence rate differences (IRDs) between LG-IUD users and nonusers of HCs (Table S4). The differences were very small. The overall unadjusted IRD for IS was −5.98 (CI, −19.04 to 7.09) per 100 000 py, or equivalently, a reduction of about 6 stroke cases per 100.000 women in 1 year if these women were using LG-IUD instead of no HC. In the adjusted model, IRD for IS per 100 000 varied from −0.35 in 18- to 24-year-old women to −6.48 in women aged 45 to 49 years. For ICH, IRDs are not measurable.

As users of LG-IUDs containing 52 mg levonorgestrel (Mirena or Levosert) contributed to >90% of py, we refrained from comparisons between preparations in relation to levonorgestrel content.

Over a 5-year follow-up period, which included 1.7 million user-years, we found that the incidence rate IS was reduced by 22% in users of LG-IUDs compared with those not using HC.

LG-IUDs are commonly used as an effective contraceptive method and are not associated with an increased risk of stroke according to guidelines.¹³ The risk of stroke associated with this method has, however, only been investigated in 1 study by Lidegaard et al.³ Based on 184 875 py, they found that risk of stroke was reduced by RR, 0.73 (CI, 0.54–0.98). Our study, which contributes to an almost 10-fold increase in py of LG-IUD use, is in line with the results of that study therefore providing further support to the safety of LG-IUDs regarding the risk of IS.

The identification of women using LG-IUDs is consistent between our study and that of Lidegaard et al³ as we followed their rules for allocation to HC usage groups. Both studies are from Denmark and designed to be nationwide, but they differ in terms of the time periods covered (1995 to 2009 versus 2004 to 2021) and registries used for recruitment of stroke patients. Lidegaard et al utilized the Danish National Registry of Patients, which contains discharge diagnoses from all Danish hospitals. In contrast, we used the Danish Stroke Registry, to which all Danish hospitals by law are obligated to report information about patients admitted and treated for stroke (see Methods). An evaluation of the validity of the registries revealed that the diagnostic sensitivity/positive predictive value of the Danish Stroke Registry was 97%/90%, while in the Danish National Registry of Patients 79%/79%.¹² The Danish Stroke Registry includes over 90% of all strokes in Denmark, the Danish National Registry of Patients includes about 75%. The Danish Stroke Registry distinguishes between IS and ICH in over 90% of cases while in Danish National Registry of Patients, half of the included stroke are in the unspecified category.

We did not find an increased/decreased incidence rate of ICH associated with LG-IUD use. The risk of ICH related to LG-IUD use has not been previously investigated, so there are no other studies for comparison. We did not observe a significant correlation between age and the lower incidence rate of IS. The incidence rate of ICH was also not age-dependent for women over 30 years. However, we identified a significantly higher incidence rate of ICH associated with LG-IUD use in women under 30 years, although this finding should be interpreted cautiously due to low number of events resulting in wide CIs.

The use of LG-IUDs is also associated with reduced risk of venous thromboembolism. Based on women with 239 841 py of observation, Lidegaard et al⁴ found the risk of venous thromboembolism to be reduced by RR, 0.57 (CI, 0.42–0.81) concluding that “LG-IUDs did not confer any increased risk, but perhaps even protection.” In 2 minor studies, risk was reduced by RR, 0.3¹⁴ and 0.6¹⁵ although significance was not obtained.

We currently have no explanation for the findings of reduced risk of stroke and venous thromboembolism with the use of LG-IUDs. Progestins, including levonorgestrel, are known to counteract the prothrombotic effect of ethinyl estradiol¹⁶ and they influence hemostasis by interacting with protein S,¹⁷ factor VII activity, fibrinogen, and plasminogen activator inhibitor.¹⁸ However, we are not aware of studies on the thrombosis risk associated with other forms of levonorgestrel administration beyond the use of LG-IUDs (oral or parenteral in other ways). The absolute risk of stroke in women of childbearing age is very low, so a 22% reduction in stroke risk among women using LG-IUDs would likely not be measurable in absolute terms as shown in our analysis of IRDs. Nevertheless, the observed risk reduction is in line with what is seen with the use of platelet inhibitors such as aspirin.¹⁹ Our study contributes evidence that could serve as a basis for further investigations into whether levonorgestrel in some form of administration might be a potential stroke protective agent in populations at higher risk.

Our study’s strength lies in the large dataset containing information about the age, education, and country of origin of Danish women aged 18 to 49 years over an 18-year period. Information about stroke and stroke types is derived from the Danish Stroke Registry, which has high validity with a diagnostic sensitivity of 97%, a high degree of certainty in differentiation between IS and ICH, and covers over 90% of all stroke hospitalizations in Denmark.¹² HCs, including LG-IUDs, are prescription-only in Denmark. The purchase and insertion of LG-IUDs in Denmark can only occur after electronic registration in The Danish National Prescription Registry ensuring complete data on the redemption and type of LG-IUDs for Denmark during the study period.

It is a limitation that the identification of women with stroke risk factors (cardiac arrhythmia, diabetes, hyperlipidemia, hypertension, and migraine) is only indirect through the registration of prescribed and redeemed medication for treating these risk factors. However, registration of these data is complete. It is a limitation that we lack information on smoking, alcohol consumption, and BMI, which is not available in public Danish registries. Therefore, bias due to these stroke-relevant risk factors is a possibility that cannot be excluded. However, according to studies by Lidegaard et al, neither BMI²⁰ nor smoking^3,20 had a confounding influence on the relationship between hormonal contraception and stroke risk. We adjusted for educational level, which in Denmark is strongly related to smoking and BMI,²¹ thus contributing to weakening bias by these confounders.

Although we have data on the redemption of LG-IUD prescriptions, we do not have secure information on their duration of use. As many women have their LG-IUDs removed before the expiration date, we set the duration of use to be 1 year less than the maximum approved usage duration. The result of our study, however, did not change significantly in our sensitivity analysis where user time was set to 2 years.

We did not exclude women with cancer as in the Danish cancer population cancer itself was not found to be associated with an increased risk of stroke.^22,23 The relationship between hysterectomy and risk of stroke is uncertain.²⁴ Risk was increased in a study from China.²⁵ However, in the Nurses’ Health Study II and the UK biobank studies^26,27 while risk was increased for cardiovascular disease risk was not increased for stroke. Hence, we did not exclude women with hysterectomy from the study. In the study by Lidegaard et al,³ cancer as well as hysterectomized women were excluded; yet, their results were fully consistent with ours. However, we cannot exclude that a cancer diagnosis or any other comorbidity may influence contraceptive preferences and choices, potentially introducing bias.

Our study supports the view that the use of LG-IUDs is not associated with the risk of stroke. However, regarding a possible potential for stroke prevention, our study cannot stand alone and requires confirmation in further research. Although the incidence rate for IS and ICH did not significantly change after adjustment for various factors, bias through residual confounding and confounding by indication cannot be fully ruled out.

Our study contributes to reinforcing that the use of LG-IUDs for contraceptive purposes is not associated with the risk of stroke. The use of LG-IUDs was associated with a 22% lower incidence of IS. Incidence of ICH was not increased in women over 30 years old. Whether the reduced stroke risk is due to a stroke preventive effect of levonorgestrel on coagulation remains to be studied.