Number of children and coronary heart disease risk factors in men and women from a British birth cohort

Transcription

1 DOI: /j x Epidemiology Number of children and coronary heart disease risk factors in men and women from a British birth cohort R Hardy, a DA Lawlor, b S Black, a MEJ Wadsworth, a D Kuh a a Medical Research Council National Survey of Health and Development, Department of Epidemiology and Public Health, Royal Free and University College Medical School, London, UK b Department of Social Medicine, University of Bristol, Bristol, UK Correspondence: Prof DA Lawlor, Department of Social Medicine, University of Bristol, Canynge Hall, Whiteladies Road, Bristol BS8 2PR, UK. d.a.lawlor@bristol.ac.uk Accepted 31 January Objective To examine the association between number of children and coronary heart disease (CHD) risk factors in women and men. Design Prospective cohort study. Setting Britain. Sample A total of 2977 individuals (51% women) from the Medical Research Council National Survey of Health and Development, a birth cohort study of individuals born in Britain in 1946 and followed up regularly throughout life. Main outcome measures Blood pressure, body mass index (BMI), waist to hip ratio (WHR), total, high-density lipoprotein and low-density lipoprotein cholesterol and triglyceride levels, and glycated haemoglobin (Hb A1C )measured at age of 53 years. Results Number of children showed no consistent relationship with CHD risk factors at age 53 years in either men or women, and no obvious and consistent sex differences were observed. Mean BMI (95% CI) increased with increasing numbers of children (P = 0.01) in women from 27.4 kg/m 2 ( ) in those with one child to 28.6 kg/m 2 ( ) in those with four or more children. WHR and type II diabetes in women and Hb A1C in men were the only other risk factors exhibiting a linearly increasing trend with increasing number of children. These associations were largely explained by adjustment for behavioural and lifestyle variables. Conclusion Our findings suggest that any association between number of children and CHD risk factors is a result of lifestyle and behaviours associated with family life rather than being as result of the biological impact of pregnancy in women. Keywords CHD risk factors, epidemiology, parity. Please cite this paper as: Hardy R, Lawlor D, Black S, Wadsworth M, Kuh D. Number of children and coronary heart disease risk factors in men and women from a British birth cohort. BJOG 2007;114: Introduction Of the several studies that have considered the association between parity and coronary heart disease (CHD) risk among women, most have found increasing disease and mortality 1 4 with increasing number of children. Other studies have also found that women with more children have poorer CHD risk factors and increased carotid intima-media thickness. 4 6 Biological mechanisms that have been proposed, which might explain the association, include pregnancy lowering lifetime estrogen exposure and pregnancy resulting in permanent, not just temporary, detrimental changes to lipid and glucose metabolism. The alternative that social and behavioural factors associated with child rearing and family life might underlie any association has received less attention. Comparing the associations of number of children with CHD risk for men and women is one way to assess the relative contribution of social and biological factors. 4 Only two studies have performed this. 3,4 The larger of these studies found a J -shaped association between number of children and CHD in both women and men aged years, with lowest risk in those with two children. 4 However, the effect was stronger and more robust to adjustment for a range of potential confounding factors in women than men. The authors concluded that greater number of children, beyond two, was associated with increased CHD and risk factors in both sexes, but while this was explained by socio-economic factors in men, other (possibly ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology 721

2 Hardy et al. hormonal mechanisms) factors appeared to result in some increased effect in women. In that study, the women and men were from two separate cohorts, and any differences between women and men could be explained by differences in the sampling frames and protocols. Further, the participants provided information on their number of children by retrospective report when they were aged years, and there may be differences in the accuracy of retrospective report between women and men. Using data from the Medical Research Council (MRC) National Survey of Health and Development (NSHD), a birth cohort study originally of 5362 men and women born in Britain in 1946, we assessed the association between number of children (obtained prospectively throughout adult life by repeated follow up) and CHD risk factors (blood pressure, body mass index [BMI], waist to hip ratio [WHR], total cholesterol, high-density lipoprotein cholesterol [HDLc] and low-density lipoprotein cholesterol [LDLc] and triglyceride levels, and glycated haemoglobin [Hb A1C ]) measured at age of 53 years. Further, we investigated the relationship between number of children and change in blood pressure and BMI between 36 and 53 years using repeated measures of these two risk factors (other risk factors had not been assessed at the 36- or 43-year follow ups). Methods The MRC NSHD is a birth cohort study consisting of a socially stratified sample of 2547 women and 2815 men born in 1 week in March There have been 21 follow ups of the whole cohort, with the most recent being at the age 53 years when 3035 cohort members (1563 women, 1472 men) provided information. The majority (n = 2989) were interviewed and examined in their homes by research nurses, with others completing a postal questionnaire (n = 46). Contact was not attempted for the 1979 individuals who had previously refused to take part, were living abroad, were untraced since last contact at 43 years or had already died. The responding sample at the age of 53 years is in most respects representative of the national population of a similar age who were born in Britain. 7 The data collection received multicentre research ethics committee approval, and informed consent was given by respondents to each set of questions and measures. Height (standing and seated) and weight were measured using standard procedures at 53 years, and BMI, defined as weight/height 2, was calculated. Waist and hip circumferences were measured using a standard protocol at 53 years, and WHR was calculated. Blood pressure at 53 years was measured while the survey member was seated and after 5 minutes of rest using the validated Omron HEM-705 (Omron Corp., Tokyo, Japan) automated digital oscillometric sphygmomanometer. Nonfasting venous blood samples were taken. Total cholesterol, HDLc and triglycerides were measured as previously reported. 8 LDLc was calculated using the Friedewald formula (LDLc = total cholesterol minus [HDLc + triglycerides 0.45]). 9 Samples were analysed for Hb A1C with the Tosoh A1C 2.2 Plus Analyzer (Tosoh Corp., Tokyo, Japan) using high-performance liquid chromatography. Height, weight (from which BMI can be calculated) and blood pressure were measured at similar home visits when the cohort members were 36 and 43 years of age. The interview nurses recorded any information on participants current medication. Information collected on antidiabetic medication, lipid-lowering medication and medicines for high blood pressure were used in these analyses. Physician-diagnosed diabetes was self-reported at the three home visits at ages 36, 43 and 53 years and on a postal questionnaire at 31 years. Age at onset of 31 years or more was classified as type II diabetes. Records of all live births, including date of birth, have been collected throughout the adult life of the cohort, and number of children at 36, 43 and 53 years was derived by adding all new reported births to the previous number of reported births. Where there was no information at the previous age, survey members were asked to provide details of all births. For the purposes of analyses, the number of children was classified as 0, 1, 2, 3, 4+ at age 53 years because of the small number of families with more than four children. Socio-economic position was measured by social class (categorised into nonmanual and manual) in childhood, based on father s occupation when the cohort member was aged 4 years, and adulthood, based on cohort member s own occupation at age 53 years (or if this was not available, at the nearest available age). Physical activity was assessed at 53 years by participants being asked the number of occasions in the previous 4 weeks in which they had participated in sports, exercises or other physical activities in their leisure time. Those who had performed no activities were classified as inactive. Cigarette smoking status at 53 years (current smoker and current nonsmoker) was defined. For women, menopausal status at 53 years was classified into one of four of the following mutually exclusive groups: premenopause/ perimenopause, postmenopause, hysterectomy and hormone replacement therapy (HRT). Perimenopause and postmenopause statuses were defined based on the criteria used in the Massachusetts Women s Health Study 10 using data on menstrual characteristics collected annually in postal questionnaires to the women in the NSHD. 11 Self-reports of hysterectomy operations and use of HRT were also obtained from these questionnaires. Statistical methods For each continuous CHD risk factor at age 53 years, the difference in means between those with no children and those with at least one child at that age was tested. This was followed by a test to assess the differences in mean by number of 722 ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology

3 Association between number of children and CHD risk factors children in those who had at least one child. Where there was no significant deviation from a linear trend, a P value for trend was obtained. As a result of the skewed distribution of triglycerides and Hb A1C, a logarithmic transformation was carried out. To account for the fact that some individuals were taking medications for diabetes (n = 63), hypertension (n = 438) or high cholesterol (n = 85) at the time the CHD risk factor measurements were taken at age 53 years, additional censored regression models 12 were fitted. These models censored the outcome measure of those on medication at the observed value so that the true measurement is assumed to be that observed or higher. The analyses were carried out separately for men and women. Tests for interaction between sex and number of children were also performed using multiple regression models to assess formally whether a difference in effect existed between the sexes. Chi-square tests and logistic regression models were used to carry out the equivalent analysis for the categorical outcomes at 53 years. Multiple regression models (normal or logistic) were used to assess the potential confounding of any associations between number of children and each CHD risk factor by childhood and adult social classes, cigarette smoking and physical activity. For outcomes other than BMI and WHR, the models were additionally adjusted for BMI at 53 years. Finally, among women, menopausal status was added. Analyses were carried out using SPSS 13 (SPSS Inc., Chicago, IL, USA) and Stata 14 (Stata Corp., College Station, TX, USA). To assess the relationship between number of children and change in blood pressure and BMI between ages 36 and 53 years, multilevel models were fitted using the statistical package MLwiN (MLwiN, Bristol, UK). 15 These models take into account the correlation between repeated measures (taken at ages 36, 43 and 53 years) on the same individual and allow for incomplete outcome data as long as it can be assumed that a missing-at-random process is operating. First, the change in each outcome with age was modelled. The intercept represented mean level at 36 years, and linear and quadratic age terms were used to model the nonlinear change in outcome with increasing age. The intercept was then allowed to vary according to number of children modelled as a time-varying covariate. The change in outcome with age was also allowed to vary by number of children by adding a number of children by age interaction term. In all models, the variance of the outcome was modelled to allow for increases with age, and both intercept and slope were allowed to vary between survey members (random effects). Separate models were fitted to compare the effects of not having children versus at least one and then number of children in those with at least one child. Results There were 2977 cohort members with at least one outcome measure and a valid record for number of live births. Men excluded from analyses had higher blood pressure at ages 36 and 43 years than those included, but there was no difference for women. There was no difference in BMI at these earlier ages for either sex. A greater proportion of excluded women had only one child (21.1 versus 13.9%) or four or more children (15.2 versus 9.8%) at age 43 years. Descriptive statistics for the CHD risk factors and the distribution for numbers of children in men and women are presented in Table 1. Table 1. Descriptive statistics for CHD risk factors and number of children by sex Variable Women Men n (%) Mean (SD) n (%) Mean (SD) BMI at 53 years (kg/m 2 ) (5.5) (4.0) WHR at 53 years (%) (6.7) (6.2) Systolic blood pressure at 53 years (mmhg) (20) (20) Diastolic blood pressure at 53 years (mmhg) (12) (12) Total cholesterol at 53 years (mmol/l) (1.09) (1.08) LDLc at 53 years (mmol/l) (0.99) (0.96) HDLc at 53 years (mmol/l) (0.49) (0.42) Triglycerides at 53 years* (mmol/l) Hb A1C at 53 years* (%) Number of children at 53 years 1516 (100) 1461 (100) (12.4) 243 (16.6) (12.1) 163 (11.2) (44.1) 679 (46.5) (22.8) 265 (18.1) (8.7) 111 (7.5) *Geometric mean. ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology 723

4 Hardy et al. The censored regression models for blood pressure, lipids and Hb A1C (censoring observations for individuals on medications for hypertension, high cholesterol or diabetes, respectively) gave very similar results to the standard regression models, hence only the results from the standard models are presented. In women, there were no significant differences in risk factor means or prevalence of diabetes, hypertension or antihypertensive medications when comparing those with no children and those with at least one child (Table 2). Men without children had a better risk profile in terms of BMI, LDLc and triglyceride levels than men with at least one child, but they had higher mean Hb A1C. In women, mean BMI and WHR at 53 years increased with increasing number of children (Table 2). There was also a nonlinear association between number of children and HDLc where women with four or more children had lower levels than those with fewer children. In men, Hb A1C increased with increasing number of children. Tests for sex interaction were not significant for any of these outcomes. In women, the prevalence of self-reported type II diabetes increased with increasing number of children, with 5.3% of those with four or more children having the condition compared with 2.2% of those with one child. For men, there was no such association (test for sex interaction P = 0.07).Women with four or more children had a higher prevalence of using antihypertensive medication, whereas this was not the case in men (test for sex interaction P = 0.02). However, because of the low prevalence of use of antihypertensive medication, these estimates were imprecise, and even in women, there was no strong statistical evidence of an association between number of children and antihypertensive medication (P = 0.08). Women without children were less likely to be from a manual social class in both childhood and adulthood than women with children. In contrast, men without children were more likely to be from a manual social class in adulthood (Table 3). The proportion from an adult manual social class increased with increasing number of children for both sexes, with the trend being stronger among women. Women with no children were more likely to be postmenopausal at 53 years and less likely to have had a hysterectomy. For women and men, the greatest proportions of smokers were observed in those with four or more children (28% for women and 33% for men). Similarly, the greatest proportions of being inactive were observed in those with four or more children (59% for women and 62% for men). The association of number of four or more children with inactivity in men represented a linear trend, with the proportion increasing with each additional child, whereas in women, the association was a threshold effect with levels of activity similar in those with zero to three children (test for sex interaction P = 0.06). In the reduced sample with full confounding variable information, unadjusted relationships with number of children (Table 4) were almost the same as those in the maximum samples presented in Table 2. Adjustment of the relationship between number of children and the biological risk factors at age 53 years for childhood and adult social classes, smoking and physical activity resulted in a weakening of unadjusted associations. In women, the mean difference in WHR between those with four or more children and those with one child reduced from 1.5 to 0.9% (Table 4). The mean difference between these two extreme groups was also reduced for BMI (from 1.1 to 0.8 kg/m 2 ) and for HDLc (from 0.10 to 0.07 mmol/l). Additional adjustment for BMI and menopausal status in women had little further impact (models not shown). For type II diabetes, the linear association was reduced after adjustment for the lifestyle factors (Table 4) and further reduced after additional adjustment for BMI (P = 0.2). The odds ratio for four or more children compared with that for one child reduced from 2.1 (95% CI: ) in the unadjusted model to 1.4 (95% CI: ) in the fully adjusted model including BMI. There was evidence, particularly strong for men (P < for BMI increase), that the difference in BMI between those without children and those with at least one child changed with increasing age. Those with children had a faster increase in BMI with age compared with those without children. For men, the estimates provided by the model (Table 5) corresponded to a predicted mean difference in BMI between these two groups of 0.1 kg/m 2 at age 36 years compared with one group of 0.6 kg/m 2 at 53 years. For women, these differences were 0.2 and 0.1 kg/m 2, respectively. In women, but not men (test for sex interaction P = 0.02), mean BMI at age 36 years increased with increasing number of children, and this relationship changed little with age (P = 0.2) (Table 5). In both sexes, mean systolic blood pressure (SBP) and diastolic blood pressure (DBP) was lower in those with at least one child compared with those with none at age 36 years, but blood pressure increase was greater in those with children. There was variation in SBP by number of children for women (test for sex interaction P = 0.07); those with one child had lower mean SBP than those with two, three or four or more. Increase in SBP was similar for all groups (P = 0.7). There was no significant effect of number of children on DBP for either sex. Discussion Number of children showed no consistent relationship with CHD risk factors at age 53 years in either men or women, and no obvious and consistent sex differences were observed. BMI, WHR and type II diabetes in women and Hb A1C in men were the only risk factors exhibiting a linearly increasing trend with increasing number of children. Of note in both sexes, those with four or more children were more likely to be smokers, inactive, from manual social classes in childhood and still in manual social classes in adulthood. The associations between number of children and CHD risk factors at age 724 ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology

5 ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology 725 Table 2. Means (95% CI) for continuous and percentages (95% CI) for categorical CHD risk factors at 53 years by number of children at 53 years of age Outcome n Number of children at 53 years P value, children versus no children P value (trend across children) P value, sex by number of children interaction BMI (mean kg/m 2 ) 0.4 Women ( ) 27.4 ( ) 27.1 ( ) 27.8 ( ) 28.6 ( ) Men ( ) 27.9 ( ) 27.4 ( ) 27.4 ( ) 28.3 ( ) * WHR (mean %) 0.2 Women ( ) 80.6 ( ) 80.4 ( ) 80.7 ( ) 82.1 ( ) Men ( ) 94.1 ( ) 93.5 ( ) 93.7 ( ) 94.2 ( ) SBP (mean mmhg) 0.3 Women ( ) ( ) ( ) ( ) ( ) Men ( ) ( ) ( ) ( ) ( ) DBP (mean mmhg) 0.8 Women ( ) 82.1 ( ) 81.7 ( ) 81.6 ( ) 81.7 ( ) Men ( ) 86.7 ( ) 87.3 ( ) 87.0 ( ) 87.2 ( ) Total cholesterol (mean mmol/l) 0.5 Women ( ) 6.20 ( ) 6.14 ( ) 6.06 ( ) 6.13 ( ) Men ( ) 5.94 ( ) 6.09 ( ) 6.08 ( ) 5.94 ( ) LDLc (mean mmol/l) 0.5 Women ( ) 3.52 ( ) 3.51 ( ) 3.45 ( ) 3.59 ( ) Men ( ) 3.48 ( ) 3.56 ( ) 3.63 ( ) 3.55 ( ) HDLc (mean mmol/l) 0.6 Women ( ) 1.81 ( ) 1.85 ( ) 1.86 ( ) 1.71 ( ) * Men ( ) 1.46 ( ) 1.48 ( ) 1.51 ( ) 1.43 ( ) Triglycerides (mean mmol/l)** 0.8 Women ( ) 1.55 ( ) 1.54 ( ) 1.53 ( ) 1.59 ( ) Men ( ) 2.07 ( ) 2.10 ( ) 2.08 ( ) 2.08 ( ) Hb A1C (mean %)** 1.0 Women ( ) 5.65 ( ) 5.58 ( ) 5.66 ( ) 5.69 ( ) Men ( ) 5.59 ( ) 5.60 ( ) 5.66 ( ) 5.71 ( ) Type II diabetes (%) 0.07 Women ( ) 2.2 ( ) 1.2 ( ) 2.6 ( ) 5.3 ( ) Men ( ) 3.0 ( ) 2.5 ( ) 3.0 ( ) 1.8 (20.7 to 4.3) Antihypertensive medication (%) 0.02 Women ( ) 18.3 ( ) 13.7 ( ) 15.9 ( ) 22.1 ( ) * Men ( ) 15.2 ( ) 15.3 ( ) 10.9 ( ) 10.8 ( ) Hypertension (%) 1.0 Women ( ) 44.7 ( ) 42.4 ( ) 44.6 ( ) 43.8 ( ) Men ( ) 55.6 ( ) 57.4 ( ) 56.7 ( ) 57.3 ( ) *Test for heterogeneity across groups when test for deviation from linearity gives P **Geometric mean. Association between number of children and CHD risk factors

6 726 ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology Table 3. Percentages (95% CI) of CHD lifestyle risk factors by number of children at 53 years of age Outcome n Number of children at 53 years P value, children versus no children P value (trend across children) P value, sex by number of children interaction Cigarette smoker at 53 years 0.2 Women * ( ) ( ) ( ) ( ) ( ) Men ( ) 29.9 ( ) 19.8 ( ) 25.3 ( ) 33.3 ( ) * Inactive at 53 years 0.06 Women ( ) 52.7 ( ) 49.9 ( ) 49.0 ( ) 58.8 ( ) * Men ( ) 46.3 ( ) 43.5 ( ) 51.1 ( ) 62.2 ( ) Father s manual social class 0.3 Women ( ) ( ) ( ) ( ) ( ) Men ( ) 62.9 ( ) 53.9 ( ) 63.4 ( ) 63.0 ( ) * Own manual social class 0.07 Women ,0.001 ( ) ( ) ( ) ( ) ( ) Men ( ) ( ) ( ) ( ) ( ) Postmenopause at 53 years , ( ) ( ) ( ) ( ) ( ) Hysterectomy by 53 years ( ) ( ) ( ) ( ) ( ) HRT user at 53 years ( ) 19.5 ( ) 20.3 ( ) 18.6 ( ) 14.2 ( ) *Test for heterogeneity across groups. Hardy et al.

7 ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology 727 Table 4. Regression coefficients from unadjusted and adjusted models for number of children among those who have at least one child (adjusted models include cigarette smoking and physical inactivity at 53 years and father s and own social class) Outcome n Unadjusted Adjusted for social class and behaviour factors at 53 years Regression coefficients (95% CI) compared with baseline group of one child P value, trend Regression coefficients (95% CI) compared with baseline group of one child BMI (kg/m 2 ) Women (21.2 to 0.6) 0.3 (20.7 to 1.3) 1.1 (20.1 to 0.6) (21.2 to 0.5) 0.3 (20.7 to 1.3) 0.8 (20.4 to 2.0) 0.06 Men (21.2 to 0.2) 20.6 (21.3 to 0.2) 0.3 (20.7 to 1.3) 0.1* 20.6 (21.3 to 0.1) 20.7 (21.4 to 0.1) 0.3 (20.7 to 1.2) 0.07* WHR (%) Women (21.3 to 0.8) 0.1 (21.1 to 1.1) 1.5 ( ) (21.2 to 0.9) 0.0 (21.1 to 1.1) 0.9 (20.5 to 2.4) 0.2 Men (21.7 to 0.4) 20.5 (21.7 to 0.7) 0.0 (21.5 to 1.5) (21.6 to 0.5) 20.6 (21.8 to 0.6) 20.3 (21.8 to 1.2) 0.7 SBP (mmhg) Women (23.8 to 2.8) 0.6 (23.0 to 4.3) 21.0 (25.5 to 3.6) (24.0 to 2.7) 0.5 (23.2 to 4.1) 21.6 (26.2 to 3.0) 0.9 Men (25.0 to 1.8) 22.1 (25.9 to 1.8) 24.5 (29.2 to 0.3) (25.0 to 1.7) 22.1 (25.9 to 1.8) 24.5 (29.2 to 0.3) 0.2 DBP (mmhg) Women (22.2 to 1.6) 20.3 (22.4 to 1.8) 20.2 (22.8 to 2.5) (22.2 to 1.6) 20.4 (22.5 to 1.7) 20.5 (23.2 to 2.2) 0.7 Men (21.5 to 2.7) 0.2 (22.2 to 3.8) 0.8 (22.2 to 3.8) (21.6 to 2.7) 0.1 (22.4 to 2.5) 0.6 (22.4 to 3.6) 0.98 Total cholesterol (mmol/l) Women (20.24 to 0.15) (20.34 to 0.09) (20.32 to 0.21) (20.07 to 0.2) (20.33 to 0.09) (20.37 to 0.20) 0.5 Men (20.05 to 0.34) 0.11 (20.11 to 0.34) (20.29 to 0.26) (20.04 to 0.35) 0.12 (20.11 to 0.34) (20.29 to 0.27) 0.2 LDLc (mmol/l) Women (20.20 to 0.16) (20.19 to 0.13) 0.05 (20.19 to 0.30) (20.20 to 0.16) (20.25 to 0.14) 0.03 (20.22 to 0.28) 0.6 Men (20.10 to 0.26) 0.13 (20.08 to 0.33) 0.07 (20.19 to 0.33) (20.09 to 0.27) 0.14 (20.07 to 0.35) 0.08 (20.18 to 0.34) 0.6 HDLc (mmol/l) Women (20.05 to 0.12) 0.04 (20.06 to 0.13) (20.21 to 0.02) 0.04* 0.06 (20.05 to 0.12) 0.04 (20.06 to 0.13) (20.19 to 0.05) 0.2* Men (20.06 to 0.10) 0.05 (20.04 to 0.14) (21.41 to 1.56) (20.05 to 0.10) 0.06 (20.03 to 0.15) (20.13 to 0.10) 0.8 Triglycerides (mmol/l)** Women (29.6 to 9.0) 20.1 (210.2 to 10.1) 3.3 (29.4 to 16.0) (29.5 to 8.9) 20.1 (210.1 to 9.9) 21.7 (214.3 to 11.0) 0.9 Men (210.3 to 11.1) 20.7 (212.9 to 11.5) 20.7 (215.9 to 14.4) (29.7 to 11.7) 21.2 (213.4 to 11.0) 22.2 (217.4 to 12.9) 0.6 Hb A1C (%)** Women (23.1 to 0.6) 0.1 (21.9 to 2.1) 0.3 (22.2 to 2.8) 0.2* 21.3 (23.1 to 0.5) 20.1 (22.0 to 1.9) 20.7 (23.2 to 1.8) 0.5* Men (21.8 to 1.8) 1.1 (21.0 to 3.1) 2.1 (20.4 to 4.6) (21.5 to 2.1) 0.9 (21.1 to 2.9) 1.6 (20.9 to 4.1) 0.1 *Test for heterogeneity. **Logarithms. P value, trend Association between number of children and CHD risk factors

8 728 ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology Table 5. Results of multilevel models of the relationships between number of children and BMI and blood pressure at ages 36, 43 and 53 years Outcome n Model a* Model b** Regression coefficient (95% CI), reference group 5 no children P value Regression coefficient (95% CI), reference group 5 one child P value (trend across children) One or more child P value, sex by children interaction BMI at 36 years (kg/m 2 ) 0.02 Women (20.7 to 0.3) (20.5 to 0.5) 0.3 (20.3 to 0.8) 0.9 ( ) 0.01 Men (20.4 to 0.3) (20.5 to 0.1) 20.1 (20.5 to 0.3) 20.1 (20.7 to 0.4) 0.7 BMI increase (kg/m 2 /year) 0.4 Women 0.02 (20.01 to 0.06) (20.03 to 0.04) (20.05 to 0.02) (20.07 to 0.03) 0.2 Men 0.04 ( ), (20.02 to 0.04) 0.00 (20.03 to 0.03) 0.00 (20.04 to 0.4) 0.98 SBP at 36 years (mmhg) 0.07*** Women (25.0 to 20.9) (24.8 to 20.7) 21.8 (24.1 to 0.5) 22.7 (25.7 to 0.3) 0.06**** Men (24.3 to 21.1) (20.9 to 2.9) 0.9 (21.4 to 3.2) 20.4 (23.6 to 2.9) 0.9 SBP increase (mmhg/year) 0.5*** Women 0.28 ( ) (20.11 to 0.29) 0.12 (20.10 to 0.34) 0.04 (20.24 to 0.31) 0.7**** Men 0.08 (20.08 to 0.24) (20.32 to 0.08) (20.17 to 0.08) (20.59 to 0.37) 0.7 DBP at 36 years (mmhg) 0.5 Women (23.1 to 0.1) (23.0 to 0.2) 20.5 (22.3 to 1.3) 21.3 (23.7 to 1.0) 0.3**** Men (23.4 to 20.7) (21.6 to 1.6) 0.5 (21.4 to 2.4) 0.3 (22.2 to 2.9) 0.6 DBP increase (mmhg/year) 0.8 Women 0.12 (20.01 to 0.25) (20.07 to 0.19) 0.03 (20.12 to 0.18) 0.06 (20.13 to 0.25) 0.8**** Men 0.11 ( ) (20.11 to 0.17) 0.04 (20.12 to 0.21) 0.08 (20.14 to 0.30) 0.5 *Model a: none versus at least one child. **Model b: number of children among those who have at least one (n for model b will be smaller than n in Table as those with no children are excluded. ***Using categorical version of child variables. ****Test for heterogeneity. Hardy et al.

9 Association between number of children and CHD risk factors 53 years could be explained largely by these socio-economic and behavioural characteristics. We also found evidence that the differences in BMI and blood pressure comparing those with children and those without varied with age, with those who had children showing a greater increase in BMI and blood pressure with increasing age. Our assays were undertaken on nonfasting samples. Hb A1C is an accurate measure of mean blood glucose concentration over the half-life of a red blood cell (usually 60 days) that does not require fasting to be accurate and reliable. Total cholesterol and HDLc assessed on nonfasting samples have also been shown to be reliable, 8 but triglycerides and therefore LDL cholesterol are less reliable if assessed using nonfasting samples. Ascertainment of type II diabetes was by self-report, and the relatively small numbers reporting the disease also limits interpretation. 16 It may be that the biological effects of pregnancy are seen in women with a large number of pregnancies, and we had to group those with four or more pregnancies into a single group because of lack of women with five or more children. However, this is a nationally representative sample reflecting the fact that few parents have more than four children in contemporary western societies. Thus, if having a very large number of children were detrimental to cardiovascular health, this would not have a major population impact. Further, in a recent very large prospective study from Finland that was able to examine parity categories up to ten or more pregnancies (accounting for 4% of the sample), the association of parity with cardiovascular mortality attenuated markedly towards the null with adjustment for socio-economic position, pre-pregnancy smoking, pre-pregnancy BMI and age at birth. 17 These findings were consistent with our overall conclusions that any association of parity with CHD risk factors is likely to be primarily explained by confounding caused by socio-economic position and lifestyle than caused by a specific pathophysiological response to pregnancy. We were restricted to looking at the associations with CHD risk factors rather than with CHD morbidity or mortality because of a lack of numbers with these outcomes, and it is not known whether the relationship between number of children and the CHD risk factors studied here is the same as that between number of children and CHD mortality. Our findings may have been influenced by survivor bias. Those who have died of CHD are very likely to have had increased levels of the CHD risk factors that we are examining here. The findings would be biased if these individuals also had a different distribution of number of children to those included in the analyses. If these individuals also tended to have many children, the observed associations would be weakened. Alternatively, those who died prematurely of CHD may have been less likely to have children because of ill health. However, it is unlikely that the small number of deaths up to the age of 53 years in this cohort (45 deaths from cardiovascular disease, including just 21 from CHD) will have importantly biased the results presented here. It is possible that any association of greater number of children or parity with CHD events and risk factors in women may be mediated by complications of pregnancy, specifically pregnancy-induced hypertension, gestational diabetes, pregnancy loss, preterm birth and low birthweight, all of which are associated with CHD risk. 18 However, as the associations of greater number of children with CHD risk factors that we found in this study were not specific to women and were largely explained by socio-economic position and behavioural risk factors, it was not necessary to try and explore femalespecific mechanisms for the effect of greater parity on CHD risk factors in this study. Our study has the advantage that the number of children was reported in the same way for men and women and was collected prospectively over the reproductive years rather than being recalled in later life. We also had measures of some of the CHD risk factors at an earlier age (36 or 43 years) and were thus able to assess whether the relationships with parity changed with time since childbearing. Our findings in relation to BMI and WHR at 53 years in women were consistent with previous studies that have shown increasing obesity with increasing numbers of children. 4,6 In addition, we showed that these differences in BMI were already evident by the age of 36 years. The higher prevalence of type II diabetes in women of high parity in our study was also in agreement with some, 4,6,19 but not all, 5,20 previous studies. We showed, however, that all these gradients were attenuated by addition of social and behavioural confounding variables. We were able to adjust for childhood and adult social classes, something that only few other studies have been able to do. Our findings therefore support the conclusions of Lawlor et al. that lifestyle risk factors associated with child rearing lead to obesity in both sexes. Our findings in relation to lipid levels were mixed, as has been the case in previous studies. 4 6 The differences between our study and others may be explained by the younger age of the NSHD cohort compared with that of the other studies in women past the age of childbearing. The associations may be influenced by menopause status, and the NSHD women were at various stages of the transition at the age of 53 years. However, adjusting for menopausal status had little impact on the results. There was little evidence to suggest that those without children remained childless because of ill health. We did find that men and women without children had higher blood pressure than those with children at the age of 36 years, but the relationship weakened, and for women reversed, with age. The only differences in risk factor levels at 53 years between those with at least one child and those without were among men rather than women, and the men without children had a healthier risk factor profile. Women with no children at age 53 years in this study were the most likely to be from ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology 729

10 Hardy et al. a nonmanual social class possibly suggesting that some of these women may have chosen not to have children to pursue a career. One study found that CHD risk was higher in those with one child and lowest in those with two children. 4 Women suffering from pregnancy complications such as gestational diabetes, pre-eclampsia, low birthweight and preterm births are at higher CHD risk in later life 18 and may limit the number of further pregnancies. We found that women with one child had higher mean SBP than women with two or more when including measures from 36 and 43 as well as 53 years. Otherwise we found no evidence to support such a J-shaped relationship. Our study suggests that at age 53 years, any association between number of children and CHD risk factors is because of lifestyle and behaviours associated with family life rather than being caused by the biological impact of pregnancy in women. Thus, obstetricians can reassure women who have up to four children that this does not appear to directly result in them having a greater risk of obesity and CHD risk factors. On the other hand, women and men with four or more children have adverse lifestyle risk factors, including a greater prevalence of inactivity and smoking, and our results lend some support to recommendations that exploration of the family environment and family-based interventions might be effective in reducing obesity and CHD risk in all family members. Acknowledgements The authors are grateful to the National Centre for Social Research, whose research nurses carried out the data collection when survey members were 53 years of age, and to the cohort members for their continuing participation in the study. D.A.L. is supported by a UK Department of Health Career Scientist Award. The NSHD is funded by the UK MRC. The views expressed in this article are those of the authors and not necessarily of any funding body. The funding bodies had no influence of the analyses or their interpretations. j References 1 Green A, Beral V, Moser K. Mortality in women in relation to their childbearing history. BMJ 1988;297: Ness RB, Harris T, Cobb J, Flegal KM, Kelsey JL, Balanger A, et al. Number of pregnancies and the subsequent risk of cardiovascular disease. N Engl J Med 1993;328: Dekker JM, Schouten EG. Number of pregnancies and risk of cardiovascular disease. N Engl J Med 1993;329: Lawlor DA, Emberson JR, Ebrahim S, Whincup PH, Wannamethee SG, Walker M, et al. Is the association between parity and coronary heart disease due to biological effects of pregnancy or adverse lifestyle risk factors associated with child-rearing? Findings from the British Women s Heart and Health Study and the British Regional Heart Study. Circulation 2003;107: Humphries KH, Westendorp IC, Bots ML, Spinelli JJ, Carere RG, Hofman A, et al. Parity and carotid artery atherosclerosis in elderly women: The Rotterdam Study. Stroke 2001;32: Wolff B, Volzke H, Robinson D, Schwahn C, Ludemann J, Kessler C, et al. Relation of parity with common carotid intima-media thickness among women of the Study of Health in Pomerania. Stroke 2005; 36: Wadsworth ME, Butterworth SL, Hardy RJ, Kuh DJ, Richards M, Langenberg C, et al. The life course prospective design: an example of benefits and problems associated with study longevity. Soc Sci Med 2003;57: Skidmore PM, Hardy RJ, Kuh DJ, Langenberg C, Wadsworth ME. Birth weight and lipids in a national birth cohort study. Arterioscler Thromb Vasc Biol 2004;24: Warnick GR, Knopp RH, Fitzpatrick V, Branson L. Estimating lowdensity lipoprotein cholesterol by the Friedewald equation is adequate for classifying patients on the basis of nationally recommended cutpoints. Clin Chem 1990;36: Brambilla DJ, McKinlay SM, Johannes CB. Defining the perimenopause for application in epidemiologic investigations. Am J Epidemiol 1994; 140: Kuh D, Hardy R. Women s health in midlife: findings from a British birth cohort study. J Br Menopause Soc 2003;9: Tobin J. Estimation of relationships for limited dependent variables. Econometrica 1958;26: SPSS Inc. SPSS: Version Chicago, IL: SPSS Inc., StataCorp. Stata Statistical Software: Release 8.0. College Station, TX: Stata Corporation, Goldstein H, Rasbash J, Plewis I. A User s Guide to MLwiN. London: Institute of Education, Wadsworth M, Butterworth S, Marmot M, Ecob R, Hardy R. Early growth and type 2 diabetes: evidence from the 1946 British birth cohort. Diabetologia 2005;48: Koski-Rahikkala H, Pouta A, Pietilainen K, Hartikainen AL. Does parity affect mortality among parous women? J Epidemiol Community Health 2006;60: Sattar N, Greer IA. Pregnancy complications and maternal cardiovascular risk: opportunities for intervention and screening? BMJ 2002; 325: Kritz-Silverstein D, Barrett-Connor E, Wingard DL. The effect of parity on the later development of non-insulin-dependent diabetes mellitus or impaired glucose tolerance. N Engl J Med 1989;321: Collins VR, Dowse GK, Zimmet PZ. Evidence against association between parity and NIDDM from five population groups. Diabetes Care 1991;14: ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology