Cohort

All Danish residents are assigned a unique personal identification number via which vital statistics and contacts with the health care system are registered. Using the Civil Registration System [20], the National Patient Register [21], and the Medical Birth Register [22], we identified all pregnancies in Denmark ending in live birth or stillbirth between 1978 and 2012. We excluded pregnancies of <20 weeks’ duration and all pregnancies in women registered in the National Patient Register with any cardiovascular disease (International Classification of Diseases, 8^th revision [ICD-8] codes 390.99–458.99, 10^th revision [ICD-10] codes I00.0–99.9) or diabetes (ICD-8: 249.00–250.09; ICD-10: E10.0–14.9) up to 30 days before their first delivery in the study period. Women could contribute more than one pregnancy to the study, but pregnancies occurring after a registration of cardiomyopathy (see Outcome, below), heart failure (see Outcome, below) or ischemic heart disease (ICD-8: 410.09–414.99; ICD-10: I20.0–25.9) in the National Patient Register were not considered.

Hypertensive disorders of pregnancy (exposure)

A woman was considered to have an HDP in a given pregnancy if she had a diagnosis of gestational hypertension, preeclampsia, eclampsia, or hemolysis, elevated liver enzymes and low platelets (HELLP) syndrome registered in the National Patient Register within a given timeframe. Since by definition an HDP involves incident hypertension in a pregnant woman with onset after 20 weeks’ gestation, we considered only diagnoses registered after this point. To maximize the validity of registered HDP diagnoses, we also required that at least one diagnosis be registered in the period from 30 days before to 7 days after delivery. As registered in the National Patient Register, gestational hypertension is defined as hypertension (systolic blood pressure ≥140 mmHg or diastolic blood pressure ≥90 mmHg) without accompanying proteinuria (ICD-8 code 637.00, ICD-10 codes O13.9 or O16.9), while in moderate preeclampsia, mild/moderate hypertension is accompanied by proteinuria (ICD-8 codes 637.03, 637.09 or 637.99, ICD-10 codes O14.0 or O14.9). Severe preeclampsia fulfills the criteria for moderate preeclampsia, with the addition of one or more of the following: severe hypertension (systolic blood pressure ≥160 mmHg or diastolic blood pressure ≥110 mmHg), severe proteinuria, signs of organ failure (including the HELLP syndrome), generalized seizures (ICD-8 codes 637.04, 637.19, 762.19, 762.29, or 762.39, ICD-10 codes O14.1, O14.2, or O15.0–15.9). (See S1 Definitions in the online Supporting Information for additional details.) If more than one HDP diagnosis was registered, the pregnancy was classified according to the most severe diagnosis. For women contributing more than one pregnancy to the study, exposure status was reset for each pregnancy, such that HDP in a previous pregnancy did not contribute to HDP status in the next pregnancy; resetting exposure status between pregnancies allowed us specifically to investigate the association between a current HDP and the immediate risk of PPCM in that pregnancy.

We also classified preeclampsia based on timing of delivery, with preeclampsia requiring delivery at <34 weeks’ gestation defined as early-onset and preeclampsia with delivery ≥34 weeks defined as later-onset, regardless of registered severity.

Peripartum cardiomyopathy and peripartum heart failure (outcomes)

PPCM was defined as cardiomyopathy registered in the National Patient Register or Causes of Death Register (ICD-8: 425.99; ICD-10: I42.0–43.8, O90.3) in the peripartum period, defined as the 6-month period beginning 1 month before delivery and ending 5 months postpartum [23,24], without registration of cardiomyopathy prior to the pregnancy. In Denmark, cardiomyopathies are diagnosed according to the European Society of Cardiology’s position paper on cardiomyopathy [25] and guidelines on heart failure [26]; since the mid-1990s, echocardiography or magnetic resonance imaging have been required to confirm a cardiomyopathy diagnosis. A PPCM diagnosis requires a left ventricular ejection fraction <45% and an absence of pre-existing (pre-pregnancy) dilated cardiomyopathy or other conditions that could produce similar structural changes in the heart (e.g. thyrotoxicosis), as per the European Society of Cardiology position paper on PPCM [1].

Since PPCM might also have been coded as heart failure, we also identified women with heart failure in the peripartum period (PPHF) (ICD-8: 427.09–427.19, 427.99, 428.99, 782.49; ICD-10: I50.0–50.9) for a separate, parallel analysis. As with the cardiomyopathies, heart failure in Denmark is diagnosed according to European Society of Cardiology guidelines [26] and requires a left ventricular ejection fraction <45%, as measured by echocardiography or magnetic resonance imaging.

Covariates

Parity (1, 2, ≥3 live births or stillbirths), maternal age at delivery (<25, 25–29, 30–34, ≥35 years), year of delivery (5-year intervals), multiple gestation (yes/no), first-trimester smoking (yes/no) and pre-pregnancy body mass index (BMI, <25, 25–29, 30–34, ≥35), all available from the Medical Birth Register, were considered as potential confounders of the association between HDP and PPCM. We also considered the impact on our results of incident diabetes (ICD-8 codes 249.00–250.09 and ICD-10 codes E10.0–14.9 registered in the National Patient Register) and hypertension (≥2 filled prescriptions for anti-hypertensive medication [Anatomic Therapeutic Chemical codes C02, 03, 07–09] registered in the National Prescription Register [27] within 6 months in a previous non-user) that developed between pregnancies. Information on parity, maternal age, delivery year, multiple gestation, and diabetes was available for the entire study period; information on smoking, BMI, and anti-hypertensive medication use was available from 1991, 2004 and 1994, respectively.

Statistical analyses

We used log-linear binomial regression with generalized estimating equations to estimate risk ratios (RRs) for the association between HDP and PPCM in the same pregnancy while accounting for potential within-woman correlations in women contributing more than one pregnancy. All RRs were adjusted for parity, maternal age, delivery year, and multiple gestation. We additionally adjusted for smoking and BMI in sub-cohorts of women with pregnancies in or after 1991 and 2004, respectively. Furthermore, we examined the effect of excluding pregnancies occurring after diabetes and hypertension registered after the first pregnancy in the study period, in the full cohort and a sub-cohort of women with pregnancies in or after 1995, respectively. Analyses were performed using SAS statistical software, version 9.4 (SAS Institute, Inc., Cary, N.C.).

Ethics

Studies based solely on data from the Danish national registers do not require approval from the Danish research bioethics committees, as study participants are never contacted, and consent is not required for the use of register information. The study’s use of register data was covered by the approval extended by the Danish Data Protection Agency to all register-based studies conducted by Statens Serum Institut (approval no. 2015-57-0102).

Main findings and interpretation

Preeclampsia has long been considered a risk factor for PPCM, and previously published data suggested that PPCM risk might depend on HDP severity [8–11]. In a nationwide study of >2 million pregnancies, we showed that PPCM risk in women with HDP was dramatically increased (5–21 times) compared with the risk in women with normotensive pregnancies, and demonstrated that the risk associated with severe preeclampsia was greater than that associated with moderate preeclampsia and gestational hypertension.

A mechanistic link between HDP and PPCM is biologically plausible. An angiogenic imbalance, with increased levels of anti-angiogenic factors (e.g. sFlt-1), is believed to play a role in the pathophysiology of HDP [28], and sFlt-1 and anti-angiogenic prolactin cleavage products may also be involved in the pathophysiology of PPCM [17–19]. The altered hemodynamics of HDP also increase cardiac stress during pregnancy. Up to 80% of preeclamptic women exhibit adaptive cardiac changes such as myocardial remodeling, and up to 50% have left ventricular global diastolic dysfunction; furthermore, 20% of women with severe forms of preeclampsia experience even greater cardiac dysfunction (left ventricular hypertrophy and radial systolic dysfunction) [29]. Why only some women with HDP develop PPCM can perhaps be explained by a “two-hit” hypothesis, whereby the anti-angiogenic state and/or cardiac changes common in HDP push women already susceptible to cardiac insult (due to e.g. reduced cardiac proangiogenic defenses, myocarditis, genetic variants, viral infection [14,17,19]) over some homeostatic threshold. Our finding that severe preeclampsia was most strongly associated with PPCM fits well with this hypothesis, since women with severe preeclampsia often have high levels of anti-angiogenic factors [30] and significant cardiac maladaptation [31].

Despite the magnitude of the associations between HDP and PPCM, most PPCM (69%) occurred in apparently healthy women with normotensive pregnancies. The hemodynamic load of a normal pregnancy may also exceed the heart’s adaptive capacity in a certain proportion of women, a suggestion supported by the recent finding that 18–28% of apparently healthy nulliparous women had left ventricular diastolic chamber dysfunction and/or impaired myocardial relaxation at term [32]. Alternatively, pregnancy may simply trigger cardiomyopathy in women already predisposed to cardiac pathology. The distribution of truncating variants in genes previously associated with dilated cardiomyopathy (e.g. the gene for the sarcomere protein titin) is similar for women with PPCM and persons with idiopathic dilated cardiomyopathy [14], suggesting that in women already predisposed to dilated cardiomyopathy, the stress of pregnancy may simply hasten the process, resulting in PPCM.

As observed in other studies, higher maternal age [8,11] and multiple pregnancy [8,11] were independently associated with the risk of PPCM. Smoking was also strongly associated with PPCM, as seen in other studies [12,13]. We also found a strong association between high BMI and PPCM risk, which is at odds with the prevalent opinion about the importance of obesity in the etiology of PPCM [8,13,19]. Interestingly, adjustment for BMI did not meaningfully affect the associations between HDP and PPCM, suggesting that the observed associations cannot be explained by obesity. Other studies have observed associations between both chronic hypertension and diabetes, and PPCM [8,10,11], but our results were robust to the exclusion of pregnancies following the development of diabetes and hypertension. There were too few adverse outcomes among women with PPCM (e.g. only one death among women with both an HDP and PPCM, and 12 deaths in women with PPCM and no HDP) to allow us to assess whether there were HDP-related differences in PPCM outcomes.

Strengths and potential limitations

By basing our study on all women in Denmark giving birth in a given period, we minimized the possibility of selection bias. Using information from the Civil Registration System minimized loss to follow-up, and the use of prospectively collected diagnosis information from Denmark’s health registers ensured freedom from recall bias.

The validity of any register-based study depends on the correctness and completeness with which diagnoses of interest are registered. Although the sensitivity of the National Patient Register for HDP and heart failure ranges from low to moderate (10% for gestational hypertension, 69% for preeclampsia overall and 29% for heart failure), the specificity for both conditions is very high (≥99%) [33,34]. Consequently, non-differential bias due to misclassification of HDP or heart failure status is likely to have been negligible.

Preeclampsia definitions have evolved considerably in the 40 years since the beginning of our study period. Of particular relevance to our study, over the past few years, the American College of Obstetrics and Gynecology, the Royal Society of Obstetrics and Gynecology, and the International Society for the Study of Hypertension in Pregnancy [ISSHP], the professional societies whose definitions and guidelines are accepted internationally, have moved away from making clinical distinctions between mild or moderate preeclampsia and severe preeclampsia. The ISSHP’s 2018 guidelines no longer include the term “mild” or “moderate preeclampsia” [35]. However, for the majority of the study period, these terms were routinely used. Both the ICD-8 (used in Denmark until 1996) and ICD-10 (in use since 1996) systems code preeclampsia diagnoses as “moderate” or “severe”, and physicians continue to code preeclampsia using ICD-10 codes, as the ICD system has not yet caught up with the guidelines. Consequently, although we use outdated terms (defined carefully in the Methods and online Supporting Information), our findings remain clinically relevant and show a clear pattern with increasing disease severity. For comparison, we also provided results based on timing of preeclampsia onset (any preeclampsia, categorized by timing of delivery), as is now preferred.

Cardiomyopathies registered in the National Patient Register have also been validated against medical records; in a large validation study, 90% of all registered cardiomyopathy diagnoses, and 75% of registered dilated cardiomyopathy diagnoses (the form of cardiomyopathy typically seen in PPCM), could be confirmed [36]. Danish cardiologists follow European Society for Cardiology guidelines when diagnosing PPCM and the other cardiomyopathies, and assignment of a cardiomyopathy diagnosis is preceded by a thorough clinical work-up that since the mid-1990s has included echocardiography. To further ensure that our results were based on valid PPCM diagnoses, we excluded from the study women with any form of cardiovascular disease up to one month before their first delivery in the study period. Consequently, one would expect the specificity of PPCM diagnoses included in the study to be excellent. In fact, the prevalence of PPCM in our study (ca. 1:16,500 deliveries) was lower than prevalences reported in many other studies [37], suggesting that our definition of PPCM was conservative. Our study also stretched over a 35-year period and PPCM was likely underdiagnosed before the advent of echocardiography, which would tend to lower the PPCM prevalence averaged over the whole study period. (Indeed, the prevalence in the period 2008–2012 was ca. 1:6,500 deliveries.) In addition, we set the upper limit of the peripartum period at five months postpartum, whereas other groups studied a longer period; however, our previous work has shown that very few additional cases of cardiomyopathy were identified between 5 months and 5 years postpartum [38]. Since the timing of PPCM diagnoses was similar for women with and without HDP, surveillance bias (due to increased medical surveillance of women with HDP) is also unlikely to explain the observed association between HDP and PPCM. Finally, since differences in access to health care among immigrant and non-immigrant women in Denmark are modest [39,40], and PPCM is such a severe condition that all cases will come to medical attention, surveillance bias due to differential access to health care for immigrants is unlikely.

Since PPCM manifests as idiopathic heart failure in the peripartum period, we expected that a certain proportion of PPCM might be coded as PPHF, particularly early in the study period (before the advent of echocardiography). We did indeed observe strong time trends, with higher incidences of PPCM later in the study period and of PPHF in earlier years. For most affected women, PPCM and PPHF were diagnosed in either the last month of pregnancy or the first month after delivery, suggesting a great deal of overlap in the conditions, and not unexpectedly, HDP, severe preeclampsia in particular, was also associated with large increases in PPHF risk.