Main

Cardiovascular diseases (CVDs) are the leading cause of mortality among females1. Beyond CVD risk conferred by traditional risk factors, women may face an additional risk attributable to reproductive history2. However, current efforts did not adequately capture female-specific risk factors, and identification of more female-specific risk factors may help improve risk profiling and prediction of CVDs among females3.

Premenstrual disorders (PMDs), including premenstrual syndrome (PMS) and premenstrual dysphoric disorder (PMDD), constitute a constellation of emotional and somatic symptoms that cyclically emerge before the onset of menses. Around 20–30% of women of reproductive age are affected by PMS4,5, and 3–8% are affected by PMDD5,6,7,8. The mechanisms underlying PMDs are unknown, but an abnormal sensitivity to hormone fluctuation is a key mechanism9.

Although health outcomes of PMDs have been scarcely studied, studies have consistently found that women with PMDs are at a higher risk of vasomotor symptoms during the menopause transition10,11,12, which are important predictors of future cardiovascular risk13. Although causality has not been confirmed for PMDs and vasomotor symptoms, existing evidence has indicated an association between PMDs and CVDs. For example, PMDs have been associated with several risk factors for CVDs, including smoking and obesity14,15. In addition, studies have indicated potential common pathophysiology underlying PMDs and CVDs, including inflammation16,17 and dysregulation of the renin–angiotensin–aldosterone system (RAAS)18. However, only two studies have so far linked PMDs to CVD endpoints, with a focus on hypertension19,20. The long-term risk of CVDs, in addition to hypertension, is unclear in women with PMDs. Furthermore, the association between PMDs and CVDs could be confounded by factors shared within a family, for example, genetics21,22, and early-life exposures23,24, whereas no study has comprehensively addressed these confounders.

To this end, leveraging data from Swedish national registers, we aimed to examine the association between PMDs and subsequent risk of CVDs.

Results

We included 99,411 women with PMDs in the populaiton cohort and 36,061 women with PMDs in the sibling cohort (Fig. 1). The median age at index date was 35.4 years in the population analysis and 34.6 years in the sibling analysis. Individuals with PMDs were more likely to have a higher educational level, higher parity and a higher prevalence of disease history (for example, psychiatric disorders and menstrual disorders) at the index date than unexposed individuals, in both the population and sibling analyses (Tables 1 and 2). In addition, in the population analysis, individuals with PMDs were more likely to be born in Sweden and use hormonal contraceptives or hormone replacement therapy (HRT) at the index date than those without PMDs.

Fig. 1: Flow chart.
figure 1

Among women born between 1949 and 2006, we excluded those who emigrated, died, had bilateral oophorectomy or hysterectomy, had a diagnosis of PMDs or reached age 52 years, before immigration to Sweden, age 16 years or 1 January 2001, whichever came later, leaving 3,644,105 women eligible for inclusion. Then in the population cohort, each exposed woman was matched to 10 women without PMDs at the index date, on birth year and county of residence, using incidence density sampling. In the sibling cohort, exposed individuals were matched with their sisters who were free of PMDs at the index date. In both the population and sibling cohorts, women who had CVDs before the index date were excluded, leaving 99,411 women with PMDs and 947,263 matched women in the population cohort and 36,061 women with PMDs and 45,451 unaffected sisters in the sibling cohort.

Full size image
Table 1 Demographic and socioeconomic characteristics of women with and without PMDs, in the population and sibling analyses
Full size table
Table 2 Lifestyle and medical history of women with and without PMDs at the index date, in the population and sibling analyses
Full size table

In the population analysis, during a median follow-up of 6.2 years (maximum 22), we identified 8,944 (12.15 per 1,000 person-years) and 73,798 (10.67 per 1,000 person-years) women with CVDs among the exposed and unexposed groups, respectively (Table 3). Compared with women without PMDs, those with PMDs had a higher risk of any CVD (hazard ratio (HR) = 1.11 (95% confidence interval (CI): 1.08–1.13) in model 3). Similar results were found in the sibling analysis (HR = 1.10 (95% CI: 1.06–1.15) in model 3).

Table 3 PMDs with the risk of CVDs: a nationwide register-based study in Sweden, 2001–2022
Full size table

In the analyses of PMD subtypes, a seemingly stronger association was noted for PMDs diagnosed before age 25 years, compared with those diagnosed later, in both the population and sibling analyses (Table 4). Similar results were seen when the cutoff was set at ages 30 and 35 years (Extended Data Tables 1 and 2). In addition, a stronger association was suggested for PMDs with perinatal depression (PND) compared with those without PND, especially in the sibling analysis.

Table 4 Associations of subtypes of PMDs with the risk of CVDs: a nationwide register-based study in Sweden, 2001–2022
Full size table

When analyzing CVD subtypes, in both the population and sibling analyses, we found a positive association of PMDs with several subtypes of CVDs, including hypertensive diseases, essential hypertension, ischemic heart disease, cerebrovascular disease, ischemic stroke and arrhythmia (Table 5). Among all CVD subtypes, the association was seemingly most pronounced for arrhythmia in the population analysis (HR = 1.31, 95% CI: 1.25–1.37) and for ischemic stroke in the sibling analysis (HR = 1.27, 95% CI: 1.08–1.50).

Table 5 Associations of PMDs with the risk of subtypes of CVDs: a nationwide register-based study in Sweden, 2001–2022
Full size table

Educational level and history of psychiatric disorders modified the association between PMDs and CVDs in the population analysis (P for interaction < 0.05), but not in the sibling analysis (Extended Data Table 3). The association was only significant in women who did not initiate HRT in both the population and sibling analyses. In addition, the association was not clearly modified by initiation of hormonal contraceptives in both analyses (P for interaction > 0.05).

We observed a positive association between PMDs and CVDs in all time intervals in analyses stratified by time since the index date, although we lacked statistical power when analyzing HRs after the 10-year follow-up in the sibling analysis (Extended Data Table 4). In addition, comparable associations were found after additional adjustment for parity, or after additional adjustment for body mass index (BMI), and smoking among parous women (Extended Data Table 5); when additionally adjusting for diabetes, asthma, autoimmune diseases, menstrual disorders, polycystic ovary syndrome (PCOS), dyslipidemia and medication use (Extended Data Table 6); when restricting the analysis to women in counties with available primary care register data (Extended Data Table 7); or when restricting to women who received two consecutive PMD diagnoses that are at least 28 days apart (Extended Data Table 8).

Discussion

With prospectively collected data, comprehensive adjustment for important confounders (including familial factors) and robust findings in sensitivity analyses, our study illustrated that women who received a PMD diagnosis in specialist or primary care are at a higher risk of CVDs. The risk was particularly pronounced for PMDs diagnosed in earlier ages, as well as PMDs with comorbid PND. In addition, such positive association was significant for several types of CVDs in both the population and sibling analyses, including hypertensive diseases, essential hypertension, ischemic heart disease, cerebrovascular disease, ischemic stroke and arrhythmia.

Small cross-sectional studies have reported an adverse cardiovascular profile among women with PMDs. For example, a study in Nigeria found higher blood pressure in the late luteal phase than in the early luteal phase, only in women with PMDs but not in other women25. Other studies have found higher blood pressure26 and higher levels of triglycerides27 among women with PMDs. However, studies linking PMD to CVD endpoints are scarce and inconclusive. Chung et al. found that women with PMS did not have a higher risk of hypertension20, whereas Bertone-Johnson et al. found that women with PMDs had a 40% higher risk of hypertension19. In our study, we found that women with PMDs have a 5% higher risk of hypertensive disorders than others, supporting a mild association with hypertension.

We are not aware of studies assessing whether other cardiovascular events are associated with PMDs. In our study, both the population and sibling analyses showed a positive association for ischemic heart disease, cerebrovascular disease, ischemic stroke and arrhythmia. Considering that the median age at end of follow-up was 42 years in the population analysis, further studies with a longer follow-up to incorporate high-risk age groups of CVDs are warranted.

The mechanism underlying the association between PMDs and CVDs may be multifactorial. First, some studies indicated potential shared genetic risk factors between PMDs and CVDs, such as polymorphism in estrogen receptor alpha21,22. However, in our study, the association remained materially unchanged in the sibling analysis, suggesting that genetic confounders are unlikely to explain our results.

Alternatively, shared nongenetic risk factors between PMDs and CVDs may explain the association. For example, both PMDs and CVDs are associated with BMI15,28 and smoking14,29. However, we observed significant associations after adjusting for these factors. In addition, PMDs are highly comorbid with depression and anxiety30, which may underlie the association between PMDs and CVDs. However, we observed an association of PMDs with CVDs among individuals without any psychiatric disorder. In addition, studies have shown that some menstrual characteristics, for example, age at menarche, and menstrual disorders such as excessive menstruation, irregular menses and dysmenorrhea are associated with CVD risk31,32. Our results minimally changed after adjustment for menstrual disorders, suggesting that the association between PMDs and CVDs is independent of these menstrual variables. Although the confounding effect of age at menarche might have been partially controlled for in the sibling analyses33, such information was not available in registers and future studies could further explore such influence.

Furthermore, PMDs are plausibly biologically linked to CVDs. Premenstrual edema symptoms, including breast tenderness and swelling of extremities, are associated with dysregulation of the RAAS18, which plays a significant role in vasoconstriction and hypertension34. In addition, progesterone may have differential effects on metabolic profiles in women with and without PMDD35. Indeed, limited evidence has shown that women with PMDs have a higher level of total cholesterol than others, and premenstrual symptom burden is associated with metabolic syndrome27,36. Furthermore, emerging evidence has shown that women with PMDs have higher levels of inflammation16,17, which is fundamental in the process of atherogenesis37 and subclinical atherosclerosis38. Notably, we found a stronger association for women who have developed both PMDs and PND than for those having PMDs only. PMDs and PND are both characterized as abnormal sensitivity to hormone fluctuations39. PND has been linked to CVDs in previous studies, and both PMDs and PND entail dysregulated inflammation40 and cardiometabolic alterations41. However, future studies may further investigate the potential mechanisms.

Our study is a nationwide population-based study with a maximum of 22 years of follow-up, including comprehensive data on disease phenotypes from primary care, specialist care and death registers. The large sample size enabled us to conduct stratified analyses and estimate associations for PMDs with specific types of CVDs, whereas the sibling analyses allowed us to minimize confounding from familial confounders. However, our study has limitations. First, the diagnosis of PMDs has not been validated in the Swedish registers. However, analyses restricted to consecutive PMD diagnosis at least 28 days apart generated similar results. In addition, as women at the index date were not aware of the future status of their CVDs, such misclassification should be non-differential and would only have attenuated our results42. Second, there could be potential misclassifications in CVDs as well. For example, hypertensive disorders may be underestimated in regions where we did not have primary care data. However, analyses restricted to women who lived in counties where primary care register data were available yielded similar results. Third, women with PMDs may have more healthcare visits and thus be more likely to have CVDs detected. Nevertheless, we still observed positive associations after 10 years’ follow-up. Fourth, although we adjusted for a comprehensive set of confounders (for example, familial confounders in sibling analyses; smoking and BMI in both analyses), we cannot rule out residual confounding, for example, nutrition intake43,44 and alcohol use45,46. Finally, PMDs identified in this study may represent a patient group who experience more severe symptoms. Therefore, our findings may not generalize to individuals with milder PMDs.

In conclusion, our study shows that women with PMDs diagnosed in specialist or primary care have a higher risk of CVDs, suggesting that women with PMDs would face health challenges in the long run. Our results need to be replicated in other populations, and healthcare professionals can be aware of the potential CVD risk when managing women with PMDs, particularly those who developed both PMDs and PND. In addition, if the results are confirmed in other populations, PMDs may be considered as a factor for future CVD risk stratification and prediction among women in the real world.

Methods

Ethics approval

This study is approved by the Swedish Ethical Review Authority (2020-06540). Informed consent is waived for register-based studies in Sweden. We have complied with all relevant ethical regulations.

Study design

We conducted a nationwide population-based matched cohort study followed by a sibling comparison. We used multiple Swedish national registers, including the Total Population Register (TPR), National Patient Register (NPR), Causes of Death Register (CDR) and Prescribed Drug Register (PDR). We also used data from regional primary care registers in the regions of Stockholm, Skåne, Uppsala, Västra Götaland and Värmland, which cover 58–62% of women of reproductive age between 2001 and 2022 in Sweden. These registers are cross-linked through a unique personal identification number. Details of the registers are summarized in Extended Data Table 9.

The study base comprised women born between 1949 and 2006, namely, women aged 16 years (>96% of Swedish girls have menarche by 16 years of age47) to 52 years (average age at menopause in Sweden48) in 2001–2022. We excluded women who emigrated, died, had bilateral oophorectomy or hysterectomy according to the NPR, or had a diagnosis of PMDs, before immigration to Sweden, age 16 years or 1 January 2001, whichever came later, leaving 3,644,105 women eligible for inclusion (Fig. 1).

The NPR records diagnoses made by specialists in all inpatient care and >80% of outpatient care in Sweden, and the primary care register includes diagnoses made by general practitioners. In our study, we had access to data from the primary care registers of five counties in Sweden. We identified PMD diagnoses made by specialists from the NPR or by general practitioners from the primary care register, based on the International Classification of Diseases, Ninth Revision (ICD-9) and Tenth Revision (ICD-10) (Supplementary Table 1). To supplement PMD ascertainment in counties where data from the primary care register were not available, as described previously49, we searched the nationwide PDR for serotonin reuptake inhibitors or hormonal contraceptives with a written indication for PMDs. Although currently there is no validation study for the diagnosis of PMDs, gynecological and psychiatric diagnoses in the NPR have been shown with a good validity50,51,52,53. For example, the positive predictive value (PPV) is 98% for endometriosis50 and 88% for depression51 in the inpatient diagnoses of the NPR.

We then classified any individual who had a PMD diagnosis in 2001–2022 as exposed and used the date of diagnosis as the index date. For comparison, each exposed woman was individually matched to 10 women without PMDs at the index date, on birth year and county of residence, using incidence density sampling.

To address familial factors that may confound the association between PMDs and CVDs, we also designed a sibling matched cohort54. Briefly, based on the Multi-Generation Register, which included information on biological parents in Sweden55, we identified full sisters of the exposed women through a link to the common mother and father. The exposed individuals were matched with their sisters who were free of PMDs at the index date, that is, matched by sisterhood.

In both the population and sibling cohorts, women who had a history of CVDs recorded by ICD-9 and ICD-10 before the index date were excluded, leaving 99,411 women with PMDs and 947,263 matched women in the population cohort and 36,061 women with PMDs and 45,451 unaffected sisters in the sibling cohort.

The study was approved by the Swedish Ethics Review Authority (2021-02775). Informed consent is waived for register-based studies in Sweden.

Ascertainment of CVDs

Using ICD codes (Supplementary Table 1), we identified the first primary or secondary diagnosis of CVDs dated in the NPR, primary care register and CDR (that is, death attributable to CVDs). Moreover, we studied eight major subgroups of CVDs, including hypertensive diseases, ischemic heart disease, cerebrovascular disease, emboli and thrombosis, heart failure, arrhythmia, conduction disorder and other types of CVDs56,57. For this analysis, the first diagnosis for each subtype was used regardless of other subtype diagnoses. The validity of CVD diagnoses in these registers is high. For instance, the PPV was 98% for myocardial infarction in the NPR58 and 80% for cerebrovascular diseases and ischemic heart disease in the CDR59.

Covariates

Information on covariates was derived at the index date. We retrieved information on demographics, including birth year, country of birth and county of residence from the TPR. We collected information on educational level, civil status and income from the Longitudinal Integration Database for Health Insurance and Labor Market (LISA). A history of psychiatric disorders was identified using ICD codes from the NPR (Supplementary Table 1). Specifically, as PND may share etiologies with PMDs (for example, hormone sensitivity) and has been shown to be associated with CVD risk39,60, we also defined PND as having diagnoses of depressive disorders or having antidepressants prescribed from 1 year before pregnancy to 1 year after. We also collected information on use of hormonal contraceptives and HRT at the index date and during follow-up. Information on parity, smoking before pregnancy and BMI at early pregnancy was collected from the Medical Birth Register (MBR) for parous women from the latest pregnancy before the index date. Information on diabetes, asthma, autoimmune diseases, menstrual disorders (including excessive menstruation, irregular menses, dysmenorrhea), PCOS and dyslipidemia before the index date was collected from the NPR and primary care registers (ICD codes can be found in Supplementary Table 1). The number of any unique medications 1 year before the index date was accumulated by identifying prescriptions from the PDR and grouped as 0, 1–2 and ≥3.

Statistical analysis

All women were followed from the index date until the first diagnosis of CVDs, emigration, death or 31 December 2022, whichever occurred first (Fig. 1). Unexposed women were additionally censored at their first diagnosis of PMDs. We used stratified Cox regression to estimate the HR and 95% CI of CVDs after a PMD diagnosis, by conditioning on the matching set in the population analysis and conditioning on the sibling set in the sibling analysis. Time since the index date was used as the time scale. The proportional hazard assumption was found to hold according to visual inspection of the Schoenfeld residuals.

To illustrate the potential effects of different sets of covariates on the association between PMDs and CVDs, we built three models. We adjusted for birth year and county of residence in model 1, through conditioning on matching set in the population analysis and adjustment in the sibling analysis. This was considered a crude model in the population analysis and a model with adjustment for basic demographics in the sibling analysis. Then, to control for socioeconomic inequities in CVD incidence and potential confounders61, in model 2, we additionally adjusted for country of birth, and socioeconomic status, including educational level, civil status and income. In addition, hormone contraceptives or HRT may regulate menstrual cycles and inhibit or promote ovulation62 and subsequently influence premenstrual symptoms. Use of HRT and hormonal contraceptives might be associated with CVD risk63,64. Therefore, we additionally adjusted for use of hormone contraceptives and HRT, as potential confounders in model 2. In addition, psychiatric disorders are highly comorbid with PMDs30 and are associated with a higher risk of CVDs65; we additionally adjusted for history of psychiatric disorders in model 3, hypothesizing that the association between PMDs and CVDs may be attenuated to a larger extent. Model 3 was considered the main model because the maximum set of observed confounders were adjusted for.

To assess the potentially differential associations for PMD subtypes with CVDs, we estimated the association for (1) PMDs diagnosed before and after ages 25, 30 and 35 years and (2) PMDs with and without PND. In addition, to provide insights into CVD subtypes, we estimated HRs for CVD subtypes in relation to PMDs.

We performed stratified analyses. First, because individuals with lower socioeconomic status have a higher number of CVD risk factors61, we stratified on educational level to examine whether educational level and PMDs may have synergistic interaction in modulating the risk of CVDs. Furthermore, some common pathophysiological pathways have been indicated for both PMDs and CVDs, for example, inflammation16,66, which are also involved in many psychiatric disorders67; we therefore stratified on history of psychiatric disorders to examine the potential interaction between PMDs and psychiatric disorders. In addition, hormone contraceptive use can be used to treat PMDs, or other premorbid gynecological conditions, which may mitigate CVD risk63,68. We therefore also performed a stratified analysis by timing of use of hormone contraceptives, to examine whether hormone contraceptives confounded and mediated the association for PMDs and CVDs. Moreover, as previous studies have found that women with PMDs have a higher risk of early menopause and menopause symptoms10,12, and use of HRT may be associated with future CVD risk64, we also stratified the analyses by use of HRT, both before and after the index date.

We also performed other additional analyses. First, to examine whether the association between PMDs and CVDs remained robust after an extended period of follow-up, we analyzed by 1, 5 and 10 years since the index date. Second, to minimize residual confounding from parity, smoking and obesity, we additionally adjusted for parity, smoking and BMI—the analysis adjusted for smoking and BMI was performed only among parous women. In addition, diabetes69,70, asthma17,71, autoimmune diseases72, menstrual disorders31, PCOS73 and dyslipidemia74 have been associated with PMDs and/or CVDs. However, increased medication use may indicate worse health status and more healthcare visits, which may bias the observed association. Therefore, to further reduce residual confounding, we additionally adjusted for these variables based on the main model, including diabetes, asthma, autoimmune diseases, menstrual disorders (including excessive menstruation, irregular menses and dysmenorrhea), PCOS, dyslipidemia before the index date and number of unique medications used 1 year before the index date. Third, PMDs and CVDs (especially hypertensive diseases) are commonly attended by primary care; however, our study did not have primary care data for the entire country. We thus estimated the associations of PMDs with risk of hypertensive diseases by restricting to individuals who lived in counties where data from both primary and specialist cares are available at the index date. Finally, although prospective symptom charting is required in PMD diagnosis in Sweden75, such information was lacking in registers. Consistent with previous studies49,54, we restricted analysis to individuals with two consecutive PMD diagnoses that were at least 28 days apart, presumably leading to a higher validity of the PMD diagnosis.

Data were cleaned using SAS, version 9.4 (SAS Institute) and analyzed using Stata 18.0 (STATA). Two-tailed test with a significance level of 0.05 was used.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.