Hypertensive diseases of pregnancy and risk of hypertension and stroke in later life: results from cohort study

Identification and tracing of women

From the databank we identified a cohort of women who were living in Aberdeen city at the time of delivery. The cohort comprised all women who had a diagnosis of pre-eclampsia or eclampsia during their first singleton pregnancy in the years 1951 to 1970; an individually matched sample of women with a diagnosis of gestational hypertension during their first singleton pregnancy and matched for age to the nearest year and year of delivery with the previous group; and an individually matched sample of women with no history of raised blood pressure during their first singleton pregnancy likewise matched with the first group. Gestational hypertension was defined as diastolic pressure ≥90 mm Hg on two occasions at least four hours apart or a single reading of ≥110 mm Hg from 20 weeks' gestation onwards in a previously normotensive woman. Pre-eclampsia was defined as gestational hypertension plus proteinuria of ≥0.3 g/24 hours. Eclampsia was defined clinically as convulsions occurring in the presence of pre-eclampsia. Women with chronic hypertension were separately categorised on the databank and excluded from this study. No other exclusion criteria were applied.

We carried out two tracing exercises for the cohort. The first, carried out in 1996, was at the local level using the Grampian community health index (a comprehensive regional primary care register²⁴) to identify the current name and address of all cohort members still alive and resident in the Grampian area. From this exercise we were able us to contact the current general practitioners of women in the original cohort who were still living in the area. The second tracing exercise was at national level and was undertaken in 1999. The information and statistics division (ISD) of the NHS in Scotland, Edinburgh, used probability matching to identify any women from the cohort who were admitted to hospital from 1980 to 1999. In collaboration with the NHS central registry in Southport, the division also ascertained the vital status of women in the cohort and obtained death certificates for any who had died, including those who died in the United Kingdom but outside Scotland. We retrieved the original databank records of women who could not be traced through either exercise to check whether additional information was available to maximise tracing rates.

Ascertainment of outcomes

Questionnaire and clinical examination—We contacted the general practitioner of each woman still living in Grampian and sought consent for us to contact her. We sent those for whom consent was given a self completion questionnaire, with questions covering general health, lifestyle and wellbeing, and obstetric history, including illnesses related to pregnancy. The questions were taken from previously validated questionnaires. The study hypothesis was not disclosed to prevent bias in reporting current health status. Non-respondents received two reminder questionnaires together with reply paid envelopes. The questionnaire included an invitation to contribute further by attending for a medical examination. Each medical examination was conducted by one of two trained research nurses in accordance with an examination protocol developed for the project. The nurses were blind to study group status of the women they were examining, and appointments were scheduled in a random manner so that no batching of groups occurred, which would undermine blinding. All women had blood pressure measured twice with a random zero sphygmomanometer with the appropriate cuff size in the seated position after a period of rest, and the mean of the two readings was taken. Height and weight were measured and an electrocardiogram produced.

Hospital discharges—In collaboration with the information and statistics division (ISD) of the NHS in Scotland we conducted a data linkage exercise²⁵ to obtain details of all hospital discharges in cohort members. The division holds data on hospital discharges and cancer registrations of individuals who have been resident in Scotland at any point. Probability matching was based on first name, surname, previous surname, and date of birth, and any obvious mismatches were checked manually by the division.

Deaths—We obtained details of the dates and causes of death from the UK NHS central register.

Sample size

The expected frequency of outcome measures varies for different cardiovascular conditions in women in Scotland, from 0.5% for electrocardiographic evidence of myocardial infarction to over 8% for angina defined by Rose questionnaire criteria.³ The frequency of hypertension varies depending on the definition used. One study of women aged 40-59 years in Scotland identified 7% with a diastolic pressure greater than 100 mm Hg.²⁶ For an outcome present in 5% of the control group, a sample size of 932 subjects per group would detect a doubling of risk, with 80% power at the 5% level of significance. On this basis, we included all women with pre-eclampsia or eclampsia (likely to be the group with fewest potential members) in the cohort and randomly sampled control and gestational hypertension groups (matched for age and year of delivery) of the same size as this group.

Statistical analysis

We used χ² and t tests as appropriate to test for baseline differences between the three groups, and, for all analyses, we used the control group as reference for comparisons with each of the two “exposed” groups (gestational hypertension and pre-eclampsia or eclampsia). For the questionnaire and clinical examination data, we computed odds ratios and used logistic regression methods to adjust for the influences of age, body mass index, and smoking behaviour at follow up. For the mortality and hospital discharge outcomes, we constructed diagnosis groups using international classification of diseases (ninth/tenth edition) codes: “hypertension” (401-4/I10-14); “ischaemic heart disease” (410-4, 428/I20-25, I50); “cerebrovascular disease” (430-8/I60-9); “kidney disease” (580-99/N00-39); “other circulatory disease” (390-8, 405, 415-27, 440-59/ I00-9, I15, I26-8, I30-49, I51-2, I70-99). We counted only first admissions in a diagnosis group, using the date of first admission as the incident date, but women could appear in more than one diagnosis group. We computed incident rate ratios for deaths and hospital discharges and used logistic regression to adjust for potential confounders.

Social class (UK registrar general) at time of follow up was used for analyses of questionnaire and clinical examination outcomes. This was allocated on the basis of the woman's own social class, unless this fell outside classes I-V, in which case the spouse or partner's social class was used if available. Social class at the time of the index pregnancy, based on the occupation of the spouse or head of household, was used for analyses of hospital admissions and deaths.

Tracing

Local tracing identified 1963 women (54.6%) who were still alive and resident in Grampian (figure). General practitioners gave consent for a questionnaire to be sent to 1876 (95.6%) of these women. The information and statistics division ascertained the current vital status of 2790 (77.7%) of the original cohort, of whom 265 (9.5%) were known to have died. Linkage with hospital discharge data identified 1829 (65.6%) women of the original cohort who had been admitted at least once to a Scottish hospital between 1980 and 1999. Compared with these untraced women, on average the traced women were a year older at delivery, the mean body mass index at the time of the index pregnancy was 0.6 kg/m² higher, and fewer were from non-manual social classes (31% (778/2481) v 39% (267/680)). The success of the tracing differed across the three comparison groups (figure).

Of the 1876 questionnaires, 1312 (71%) were returned with usable responses. Forty one were incorrectly identified as being eligible for the study by the Aberdeen neonatal and maternity databank. Table 2 summarises the self reported demographic and health characteristics of the questionnaire respondents at follow up. Respondents were more likely to be in manual social classes at baseline than the starting cohort. Respondents in gestational hypertension and pre-eclampsia or eclampsia groups had higher mean body mass index at follow up and were less likely to be former or current smokers, as they were at baseline. There were no clinically significant differences in alcohol intake, regular exercise, or use of exogenous hormones across the groups. A much higher proportion of the women were of non-manual social classes at follow up, using the woman's own profession, than at baseline using husband/partner's profession. There was no gradient in social class at follow up across the groups using this measure.

Questionnaire assessed outcomes

After adjustment for known risk factors for hypertension, women in both the gestational hypertension and pre-eclampsia or eclampsia groups had higher odds ratios for indicators of hypertension (table 3). This was not seen for the other conditions of interest, except for kidney disease in the pre-eclampsia or eclampsia group, which may have related to the original episode in pregnancy. Similar results were observed in the clinical examination data (table 4), with increased risk of hypertension defined by WHO criteria in both gestational hypertension and pre-eclampsia or eclampsia groups. In total, 41.1% of the former and 48.8% of the latter provisionally fulfilled the WHO criteria for hypertension compared with 26.7% in the control group.

Hospital admissions

Women in the gestational hypertension group were more likely to have been admitted to hospital for cerebrovascular disease, ischaemic heart disease, hypertension, and “other” circulatory disease, although this reached significance only for hypertension and other circulatory disease (table 5). Women in the pre-eclampsia or eclampsia group were also more likely to be admitted for hypertension, cerebrovascular disease, and other circulatory disease.

Mortality

Over 90% of the traced cohort were alive at the time of analysis, and there were 265 deaths available for analysis (table 6). By the time of follow up, there were 72 deaths in the control group, 89 in the gestational hypertension group, and 104 in the pre-eclampsia or eclampsia group. No deaths were coded to hypertension per se. After adjustment, mortality for all causes, stroke, and ischaemic heart disease was higher in both the gestational hypertension and pre-eclampsia or eclampsia groups. The pattern was consistent, though only the increased risk of stroke in the pre-eclampsia or eclampsia group was significant.

Influence of social class

Social class differed between the three study groups in the original cohort. This might be a result of chance or might indicate an association between social class and hypertensive diseases in pregnancy. Though such associations have been reported,²⁷,²⁸ there is no clear social class gradient. Pre-eclampsia or eclampsia is less common in smokers than in non-smokers²⁷,²⁸ and currently smoking is associated with social class, though in 1961 UK survey data showed little social class gradient in smoking in women.²⁹ We found a modest gradient in smoking across social class at the time of pregnancy but smoking information was recorded for only 28%. If there is a true inverse association between pre-eclampsia or eclampsia and smoking, and if recorded smoking habit during first pregnancy is an indicator of likely smoking habit in later life, then the pre-eclampsia or eclampsia group should in theory be at a lower than average risk of smoking related disease in later life. If so, this would operate against the hypothesis and might suggest that our results underestimate the association between pre-eclampsia or eclampsia and circulatory diseases.

Several studies have shown a risk of coronary heart disease associated with pre-eclampsia, though our findings were not significant. Retrospective studies of coronary heart disease that used patient recall of pre-eclampsia are subject to recall bias, while other cohort studies have combined coronary heart disease outcomes into a single cardiovascular disease group.³⁰

Tracing and investigation

The tracing rates varied across the three groups. Those who were normotensive during their first full term pregnancy were least likely to still be living in Grampian at follow up, suggesting selective migration from the area by the control groups. The national tracing exercise, however, was more successful and provided vital status and hospital discharge data for a larger proportion of women in each comparison group. Broadly similar patterns of cardiovascular sequelae were observed in the different groups, irrespective of whether the outcome measure used was obtained through the local or national tracing. Another important potential bias could have occurred if there were important differences in the response rates of those traced locally. There were in fact only small differences in the response rates to the survey by the three comparison groups (70%, 77%, and 77%) as shown in table 2, suggesting that response bias is unlikely to have materially affected our results. There were more women from higher social classes in the control group (table 1). Such women were more likely to emigrate from Grampian, resulting in a lower response rate in the control group.

Consistency of findings

There was consistency of the findings across the different methods of ascertaining outcomes. The association between gestational hypertensive disease and hypertension in later life was seen in the questionnaire and clinical examination data and also in the hospital diagnosis data and the mortality data. We suggest that our observations cannot be explained by chance or bias. The findings have implications for the aetiology and pathogenesis of circulatory disease, both in pregnancy and in later life. They also suggest that interventions that might minimise the risk of such conditions in later life should be identified and evaluated.