Pregnancy Loss and Cardiovascular Disease: A Nationwide Cohort Study

Data sources

Healthcare in Denmark is universal and free of charge. Data from a wide variety of healthcare services are collected in registries for administrative and research purposes. Using these nationwide registries, we extracted information on pregnancies, diseases and demographics for all Danish women between 1973 and 2017 with at least one registered pregnancy. The main data sources used were: (i) the Danish Central Persons Registry, containing date of birth and death, since 1968;[18] (ii) the Danish Medical Births Registry, containing data on live and stillbirths since 1973;[19] (iii) the Danish National Health Registry containing data on hospital discharge diagnosis codes since 1977;[20] (iv) the Danish National Prescription registry containing data on dispensed prescription medication since 1995,[21] (v) the Danish Cancer Registry, containing diagnoses of cancer, available since 1942.[22] A summary of registers used is listed in eTable 1. Search strings used to extract data from registries can be found in eTable 2. For diagnoses extracted from the National Health Registry, the admission date of primary and secondary diagnoses was used, and emergency room contacts were not used.

Identification of unique pregnancies

Live births and stillbirths are registered in the Medical Births Registry. From 1973 to 1996 these events were filled manually for each birth. From 1997 data on births were extracted from the National Health Registry. In Denmark, the gestational threshold for a stillbirth to be reported in the Medical Births Registry, was 28 weeks until 2004, and 22 weeks hereafter.[19] Early pregnancy complications such as miscarriage, blighted ovum, ectopic pregnancy, and molar pregnancy were extracted from the National Health Registry. As these pregnancies can lead to multiple hospital visits during clinical care, a set of restriction periods were made to ensure that each pregnancy was only counted once: (i) at least 90 days between two early pregnancy complications, (ii) at least 154 days (22 weeks) between an early pregnancy complication and a succeeding delivery date of a live or stillbirth. (iii) At least 60 days between the delivery date of a live or stillborn and a succeeding early pregnancy complication. (iv) At least 154 days between a live or stillbirth and a succeeding live or stillbirth. In the rare event two pregnancies were overlapping, the first was kept.

Study design

The study was conducted as a historical prospective cohort study.

Pregnancy loss was defined as a diagnosis code of miscarriage, spontaneous abortion, missed abortion, blighted ovum, or stillbirth. Three different pregnancy loss exposures were considered: (i) The number of pregnancy losses: 0, 1, 2, 3, or ≥4. (ii) Recurrent pregnancy loss, defined as either three consecutive registered pregnancy losses, or a diagnosis of recurrent pregnancy loss. Primary recurrent pregnancy loss was defined as women with no deliveries before a series of three or more consecutive pregnancy losses, and secondary recurrent pregnancy loss was defined as women with one or more births before the consecutive series of pregnancy losses. (iii) Stillbirth was also studied as an independent exposure.

Outcomes were defined as the first hospital contact, or the underlying cause of death of the following:(i) venous thromboembolism, including deep vein thrombosis, and pulmonary embolism (superficial thrombosis was not considered a valid outcome); (ii) myocardial infarction; (iii) ischemic stroke (transitory ischemic attack was not considered a valid outcome). Immigrants to Denmark were eligible for inclusion if they immigrated to Denmark before 20 years of age.

The study consisted of two separate designs in order to calculate both absolute and relative estimates: The first part evaluated absolute risk with a landmark approach at age 40 using Cause-specific Cox regression and G-formula;[16] the second part evaluated hazard ratios using time-dependent Cox regression.[17]

Study population, study part one

In order to calculate 20-year absolute risks of outcomes, all women born from 1957 through 1977 with a registered pregnancy, were eligible for inclusion. Inclusion was based on a landmark at age 40.

Women were stratified in exposure groups according to pregnancy losses occurring before age 40. Women with a history of venous thromboembolism, myocardial infarction, ischemic stroke, thrombophilia, cancer, severe cardiovascular disease, emigration over six months, emigration with no return date, or death before inclusion were excluded.

Statistical analyses, study part one

Risk time started at included women’ s 40^th birthday and ended at first occurrence of (i) an outcome of interest, (ii) death from other causes, resulting in a competing risk event. (iii) Emigration over six months, or end of follow-up (31^st of December 2017) resulting in censoring. Two Cox proportional hazards regression models were fit, one for the cause-specific outcome, and one for the competing outcome of death from other causes. Estimates were combined using G-formula to estimate the absolute risks.[23–25] The horizon for risk predictions was set to 20 years, corresponding to the probability for a 40 year old women to experience an event within the next 20 years, reported in percent. Absolute risks and risk differences were shown in a forest plot alongside 95% confidence intervals, calculated using 1,000 bootstrap samples.

Study population, study part two

In this design women born from 1957 and onwards, with a registered pregnancy were eligible for inclusion. Women with a history of venous thromboembolism, myocardial infarction, ischemic stroke, thrombophilia, cancer, severe cardiovascular disease, or emigration over six months before first pregnancy were excluded. Women were included at first pregnancy after 1^st of January 1977 and followed until the earliest occurrence of the following: (i) venous thromboembolism, myocardial infarction, or ischemic stroke, resulting in an event for the respective outcome, or (ii) thrombophilia, malignant cancer, severe cardiovascular disease, emigration over six months, emigration with no return date, or end of follow up 31^st of December 2017, resulting in censoring.

Statistical analyses, study part two

Included women were followed in a time-dependant manner, with maternal age as the underlying time axis. Each woman’ s risk time was split during follow-up, to account for covariates changing. The following occurrences caused the risk time to be split: (i) pregnancy loss, (ii) recurrent pregnancy loss, (iii) stillbirth, (iv) live birth, (v) diabetes, (vi) obtainment of a bachelor’ s degree, (vii) diagnosis of rheumatoid arthritis, (viii) diagnosis of systemic lupus erythematosus, and (ix) total risk time surpassed a 5 year interval. Women could increase exposure level to pregnancy loss during follow up, but not revert. Estimates were obtained using the Cox proportion hazards model, and the Cox proportional hazards assumption was tested by plotting the scaled Schoenfeld residuals against time, with acceptable horizontal linearity for all predicter variables, and no clear pattern with time. Estimates and 95% confidence intervals were displayed in forest plots.

Sensitivity analysis, study part two

A sensitivity analysis was carried out for exposure to stillbirth and outcome of myocardial infarction. In this analysis women were censored as above or at age 40, whichever came first.

Covariates

Covariates to adjust for in both study parts, were chosen after considering a causal graph (eFigure 1) of known and suspected confounders.[26] Analyses were adjusted for (i) calendar year using a spline with 3 degrees of freedom (ii) the number of live births using a spline with 3 degrees of freedom, (iii) diabetes, (iv) rheumatoid arthritis, (v) systemic lupus erythematosus, and (vi) educational level above or below bachelor degree, (vii) maternal age, as maternal age in years was used as the underlying time axis.

Hypertension was defined as at least two hospital admissions with a diagnosis of hypertension or fulfilling prescriptions of at least two classes of anti-hypertensive drugs (Classes: i. Adrenergic antagonist, ii. Diuretic, iii. Vasodilator, iv. Beta blocker, v. Calcium Antagonist, vi. Renin–Angiotensin– Aldosterone System Inhibitors). Diabetes included diagnoses for gestational diabetes and was defined as at least two hospital contacts with a diagnosis or fulfilling at least two prescription of an anti-diabetic drug. Thrombophilia, rheumatoid arthritis and systemic lupus erythematosus, and polycystic ovarian syndrome was defined as a relevant hospital diagnosis.

All data management and statistical analysis was performed using R Version 3.6.3.[27] Cox proportional hazards models were calculated using the survival package version 3.1-11.[28] Absolute risks were calculated using the riskRegression package version 2020.02.05.[29] Plots were produced using the ggplot2 package version 3.3.2.[30] Causal graphs and minimal sufficient adjustment sets were obtained using the dagitty package version 0.3-0.[31]

Patient and public involvement

Patients contributed data to the current study during hospital contacts and when purchasing prescription drugs, but were otherwise not involved in the research process, in forming the research question, in the design of the study, in the choice of outcome measures, and conduct of the study.

Strengths and weaknesses of the study

The major strength of this study is the size of the study population, length of follow-up, and the clinically confirmed endpoints.

The study was limited by data registered. Variables such as smoking status and body mass index commenced during follow-up and was only available for women who ever experienced a live or stillbirth. As body mass index and smoking were not part of the suggested minimal sufficient adjustment set of the causal graph (eFigure 1), this was not seen as major limitation.

A diagnosis of pregnancy loss was defined by diagnoses in the National Health Registry. These diagnoses have been shown to have a positive predictive value of 97% (92.7-99.5).[32] The health registry only contains hospital-treated pregnancy losses, and we have therefore likely underestimated the true number, as those managed at home, by general practitioners, and gynaecologists in private practice, are not registered. Thrombophilia was likely underdiagnosed, but not under-reported, as during much of follow-up, thrombophilia evaluation was not carried out until after a thromboembolic event or evaluation for recurrent pregnancy loss. A hospital diagnosis of venous thromboembolism has been shown to have a positive predictive value of 91.2%,[33] while validation of a hospital diagnosis of myocardial infarction has shown a positive predictive value of 93.6% and sensitivity of 77.6%.[34] For stroke the corresponding positive predictive value was 97.0%.[35] No studies have to our knowledge validated the equivalent diagnoses in the cause of death register, however very few of outcome diagnoses were only identified in the cause of death register (eTable 5).

Strengths and weaknesses in relation to other studies, discussing important differences in results

Results from studies, examining the association between pregnancy loss, and later myocardial infarction and ischemic stroke have shown ambiguous results,[10,11,14] and meta-analyses have as a result concluded that there is not enough evidence to draw meaningful conclusions about the possible link, although prior stillbirth is increasingly accepted to be a risk factor for later cardiovascular disease.[9,36]

To our knowledge, no studies have evaluated the outcome of venous thromboembolism, and further no studies have presented clinically meaningful absolute risk estimates. Adjustment of confounders varied among studies examining the risk of myocardial infarction and stroke after pregnancy loss. In the largest cohort study to date, risk estimates were adjusted for current age, calendar period, and number of live births.[14]. Our study supports the main conclusions from the study, although there are some discrepancies in our findings, e.g. the study found a decreased incidence rate ratio for myocardial infarction after three miscarriages compared to none.

The rare autoimmune disease antiphospholipid syndrome is linked to both pregnancy loss and thrombotic events, and is present in 5-15% of patients with recurrent pregnancy loss.[37,38] Possibly other immunological disturbances influence both pregnancy loss and later risk of thrombosis. Our findings suggest that women with secondary recurrent pregnancy loss are at an increased risk of venous thrombosis, and myocardial infarction, while women with primary recurrent pregnancy loss have an increased risk of later ischemic stroke, however, the increase in risk was not substantially different than for women with multiple non-recurrent losses or stillbirth.

Meaning of the study: possible explanations and implications for clinicians and policymakers

A woman’ s pregnancy profile could provide clinicians with a valuable and easily assessable tool containing information about future risk of venous thromboembolism, myocardial infarction, and ischemic stroke. Our study provides easily interpretable 20-year absolute risks of outcomes from age 40, based on history of pregnancy loss. For implementation in current risk stratification schemes, further investigations into the added effect of including pregnancy loss history, and the effect of lifestyle modification and medical intervention strategies.

Unanswered questions and future research

The mechanisms behind the associations cannot be directly deduced from our results. The pregnancy losses could themselves have increased the risk of outcomes, possibly through specific physiological changes caused by the pregnancy loss. Alternatively, a genetic, metabolic, vascular, or immune precursor exists, which is associated with both pregnancy loss and cardiovascular disease, possibly through endothelial dysfunction. Further, the large increase in risk of myocardial infarction among women under 40 warrants further investigation.