点击显示 收起
1 From the London School of Hygiene and Tropical Medicine, London, United Kingdom
2 The Aberdeen Children of the 1950s Cohort Study was funded as a component project (G9828205) of a Medical Research Council Co-operative Group: "Life-course and Trans-generational Influences on Disease Risk" (G9819083). 3 Reprints not available. Address correspondence to DA Leon, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1 7HT, United Kingdom. E-mail: david.leon{at}lshtm.ac.uk.
ABSTRACT
Background:Few studies have examined whether the inverse association of age at menarche with adult body mass index (BMI) is due to the tendency of BMI to track between childhood and adult life, with childhood BMI playing a causal role in determining age at menarche.
Objective:The objective was to investigate whether the association of younger age at menarche with a high BMI and increased likelihood of obesity in middle age is due to confounding with early childhood BMI.
Design:In a historical cohort of 3743 Scottish females born between 1950 and 1955, height and weight were measured in early childhood, and age at menarche and height and weight in middle age were obtained by questionnaire.
Results:The age-adjusted change in mean adult BMI per additional year of age at menarche was –0.64 (95% CI: –0.78, –0.50). Adjustment for childhood BMI measured between 4 and 6 y reduced this value to –0.57 (–0.71, –0.43). Adjustment for childhood and adult social class, parity, smoking, and alcohol intake had little effect. The odds ratio for being obese compared with not being obese in adulthood was 0.82 (0.76, 0.86) per 1-y increase in age at menarche and was unchanged by adjustment for childhood BMI and other covariates.
Conclusions:The inverse association of age at menarche with BMI and obesity in middle age is not explained by confounding by early childhood BMI. Instead, age at menarche may simply be a proxy marker for the pace of sexual maturation, which itself leads to differences in adiposity (and BMI) in the peripubertal period that track into adult life.
Key Words: Growth • sexual maturation • menarche • body mass index • life course analysis • epidemiology • Aberdeen Children of the 1950s Study
INTRODUCTION
Obesity is a major public health problem. It is associated with adverse physical health outcomes [including diabetes, coronary heart disease, cancer, and respiratory problems (1)] and psychosocial difficulties (2). The prevalence of obesity is increasing in adults in the United States (1), the United Kingdom (3), and worldwide (1).
Early age at menarche, together with other indicators of early biological maturity, has been shown to be associated with increased adult body mass index (BMI) (4). Between 1972 and 2003, 10 longitudinal studies found a negative relation between age at menarche and adult weight-for-height (5–14), assessed by BMI in all but one study (5), but at least one longitudinal study showed no relation between age at menarche and adult BMI (15). If there is a true causal link between age at menarche and adult BMI, it could be argued that this association may play a role in explaining the temporal trends in obesity. Age at menarche has been declining (16–20) at the same time as adult BMI has been increasing, although the rate of decline has slowed or stopped in some countries in recent times (21).
Whether early age at menarche is causally associated with increased adult obesity is unclear because many factors are related to both age at menarche and adult obesity. In particular, girls who have an early age of menarche have the highest childhood BMIs (9, 11, 22, 23). In turn, BMI tracks between childhood and adult life (12, 24), with high childhood BMI being predictive of high BMI in adulthood. Thus, the association between early menarche and adult obesity may be largely (or wholly) due to the association between earlier and later obesity.
Other factors across the life course (23) may also explain or mediate the relation between age of menarche and adult BMI. Socioeconomic position at birth (14) and in adult life (11), parity (25), current smoking status (26), and alcohol intake (27) are all associated with BMI in adult life. Social class is also negatively associated with age at menarche (13). No studies of the association of age at menarche and adult BMI have taken all of these factors into account simultaneously.
The aim of this study was to assess the association of age at menarche with adult BMI and adult obesity, taking into account childhood BMI and other potential confounders and mediators.
SUBJECTS AND METHODS
The Aberdeen Children of the 1950s Cohort Study is a cohort study of 12 150 children who were born in Aberdeen, Scotland, between 1950 and 1955 and took part in a child development study in the early 1960s (28). This study collected comprehensive medical and social information on these children. As described elsewhere (29), we revitalized this cohort by attempting to trace them in adult life. We were successful in identifying 99% of the original cohort and in ascertaining their vital status and area of residence. Between 2000 and 2002, a questionnaire was sent to the traced survivors believed to be resident in Scotland, England, or Wales. The study population considered in this paper consists of the female responders to the postal questionnaire.
The primary outcome of interest was BMI calculated from self reported height and weight on the questionnaire (at age 45–52 y). We asked the subjects to weigh themselves on scales while wearing light clothing and no shoes and to say whether they had done so (93% had done so). The primary exposure of interest was self-reported age at menarche, which was also taken from the questionnaire responses.
The factors considered as potential confounders or mediators were father's occupational social class at birth, childhood BMI, and, in middle age, occupational social class, parity, alcohol intake, and smoking status. Father's occupation was derived from birth records. Childhood height and weight were measured at a routine medical exam on school entry. Height, weight, and age at the time of the medical exam were obtained from school medical records. The children's age range was 48 to 149 mo (22% <5 y, 75% >5 to <6 y, 0.5% >6 to <7 y, and 2.5% > 7 y). We excluded from the multiple regressions any girls who had had their first medical examination after the age of 7 y, because they were unusual in that for various reasons their school entry had been delayed, and the older girls might have been prepubertal. Occupational social class, age, parity (number of live-born children), smoking status, and alcohol intake were derived from responses to the postal questionnaire.
BMI in childhood and adult life was calculated as weight in kg/height2 in m. Because of the variation in age at which childhood weight and height were measured, we computed childhood BMI z scores based on the study mean and SD of BMI within consecutive 3-mo bands from 48 to 83 mo of age. We estimated the amount of alcohol consumed in units per week from the responses to a question about how many of what type of alcoholic drink had been taken in the past week [1 glass of wine, 0.5 pints (236 mL) of beer, or one shot of spirits equals one unit].
We categorized childhood BMI z scores for age at examination into quarters; age at menarche into <12, 12, 13, 14, and 15 y; adult BMI into underweight, normal weight, overweight, and obese on the basis of World Health Organization definitions (30); alcohol consumption into drinkers (0–5, 6–14, or 15 units/wk), never drinks, and exdrinkers; and parity into 0, 1 or 2, or 3. Father's occupation at the time of birth of the child was classified as I (professional), II (managerial), IIIn (nonmanual clerical), IIIm (skilled manual), IV (semiskilled manual), V (unskilled manual), and unemployed or divorced. The subject's current occupation was classified according to the Registrar General's Social Class 1991 based on Standard Occupational Classification and categorized into the same groups as the father's occupation.
To assess bias, we compared the characteristics of those who did or did not respond to the postal questionnaire and the distribution of the covariates in all responders and in those for whom a complete data set was available for multiple linear regression.
The relation between each of the covariates and the age of menarche and adult BMI, both for those who responded to the questionnaire and for those for whom a complete data set was available for multiple linear regressions, was assessed by analysis of variance and linear regression. Multiple linear regression was used to explore the effects of the covariates (separately and in combination) on the relation between age at menarche and adult BMI. We conducted a similar analysis using logistic regression to investigate the association of age at menarche with the binary outcome of obesity in adult life. All analyses were done by using STATA 7 (Stata Corporation, College Station, TX). The Scottish Multi-Centre Research Ethics Committee gave ethical approval for the Aberdeen Children of the 1950s Cohort Study.
RESULTS
Of 5874 women (sex defined by birth records) in the Aberdeen Children of the 1950s Cohort Study, 334 women were not sent questionnaires: 165 had died, 139 had emigrated, 18 were in institutions or armed forces, and 12 were not mailed questionnaires for other reasons, eg, refusal by subject's doctor for permission to forward the questionnaire (31). Of the 5540 women sent a questionnaire, 3752 (68%) responded and returned the questionnaires by November 2003. Of the 3752 "female" (defined on the basis of sex on birth records) responders, 9 reported their sex as "male" on the questionnaire or did not give either their sex or their age at menarche. These 9 subjects were omitted from the study, which left 3743 responders who reported their sex as "female" on the questionnaire or reported an age at menarche.
In Table 1 we compare mean childhood BMI-for-age z scores and social class at birth between questionnaire responders (n = 3752) and nonresponders (n = 1788). Responders were of higher social class at birth but the responders did not differ from nonresponders with respect to childhood BMI. The mean (±SD) childhood BMI in responders was 16.20 ± 1.54 and in nonresponders was 16.19 ± 1.45 (t = 0.25, P = 0.804)
View this table:
TABLE 1. Characteristics of female questionnaire responders and nonresponders
In Table 2 we compare responders who had complete data with those who had incomplete data on adult BMI (outcome), age at menarche (exposure), or one or more of the other covariates and were thus omitted from the multiple linear regression. Responders with complete data (n = 2968) were of slightly higher social class (at birth and at the time of the questionnaire), were less likely to be current smokers, and were less obese in middle age than were those who were omitted (n = 755).
View this table:
TABLE 2. Characteristics of female questionnaire responders included or excluded from the multivariate analysis
In Table 3 we examine the association of covariates with mean adult BMI and with mean age at menarche for the subgroup of subjects included in the multiple linear regression analyses. Adult BMI was inversely associated with mean age at menarche, social class in childhood and in adulthood, and smoking. Childhood BMI and parity were positively associated with adult BMI. On the basis of current US Centers for Disease Control and Prevention standards, the mean BMI z score in childhood for the subjects included in our main analysis was 0.5.
View this table:
TABLE 3. Association of selected covariates with mean adult BMI and mean age at menarche for the questionnaire responders with complete data1
Childhood BMI-for-age z score (derived internally in the study as described in Methods) and adult obesity were significantly inversely associated with age at menarche. Smoking status was significantly associated with age at menarche. Adult but not childhood social class was associated with age at menarche. These univariate associations were similar for all study participants, regardless of whether they were or were not included in the multiple linear regressions (data not shown).
The correlation coefficient between childhood BMI and adult BMI was 0.193 (P = 0.008) and between childhood BMI and age at menarche was –0.132 (P < 0.001). The increase in adult BMI per unit increase in childhood BMI was 0.722 (95% CI: 0.60, 0.76) and was reduced to 0.64 (95% CI: 0.57, 0.76) after adjustment for age at menarche.
In Table 4 we show the mean change in adult BMI by age at menarche after adjustment for age and several potential confounders or mediators. The negative association of age at menarche with adult BMI was largely unaffected by adjustment, except in the case of childhood BMI. Adjustment for childhood BMI had only a small effect on the strength of the association of age at menarche on adult BMI, reducing it by 11% [1– (0.57/0.64)]. Adjustment for the other covariates had almost no effect on the strength of association (data not shown).
View this table:
TABLE 4. Differences in mean adult BMI by age at menarche (and 95% CIs), adjusted for selected covariates1
The age-adjusted odds ratio for being obese compared with not being obese in adulthood was 0.78 (95% CI: 0.73, 0.84) for a 1-y increase in age at menarche. Adjustment for childhood BMI z score attenuated this effect very slightly to 0.80 (0.75, 0.86). The corresponding odds ratios for being obese or overweight compared with being normal weight or underweight were 0.82 (0.77, 0.86) with and 0.82 (0.76, 0.86) without adjustment for childhood BMI.
DISCUSSION
Our results confirm the findings of other studies, ie, there was an inverse association of age at menarche with BMI in later life (5–7, 9–14). However, this is the first time that the association has been reported in a population-based cohort of middle-aged women. Most of the other studies in the literature investigated adult BMI at an appreciably younger age, ie, late teens to early 30s.
We found that the inverse association between age of menarche and adult BMI was independent of social class at birth, and there was no evidence that it was either confounded or mediated by social class in middle age, parity, smoking status, or alcohol intake. The findings with regard to social class, smoking, and alcohol intake were in keeping with those of Okasha et al (13).
We found that the association between age at menarche and mean adult BMI was only slightly confounded by early childhood BMI and that the relation between age at menarche and adult obesity was not confounded at all by childhood BMI. Only 2 other studies, the Bogalusa Heart Study (32) and the 1958 British Birth Cohort (11), have addressed the question of whether the associations of age at menarche with adult BMI and obesity are attenuated after adjustment for childhood BMI. Unlike our study, they found a substantial confounding effect of childhood BMI on the relation between age at menarche and adult BMI. The Bogalusa Heart Study (32) reported that white girls who underwent puberty before age 12 y had a mean BMI at age 26 y that was 3.6 higher than that of girls who reached menarche at an older age. However, this effect was reduced to 1.4 after adjustment for BMI and triceps-skinfold thickness at 9 y of age. The 1958 British Born Study (11) reported a mean BMI difference of 4.1 at the age of 33 y between women who had menarche at <12 y compared with those who had menarche at 15 y. Adjustment for childhood BMI at 11 y was described as having "substantially reduced" this association.
Strengths and limitations of the study
Particular strengths included the fact that the study was relatively large, and social class at birth was based on records rather than recall in adult life. Moreover, compared with most other studies it had the advantage that childhood BMI in childhood was measured at a younger age and BMI in adulthood was measured at an older age.
However, in our study adult BMI was based on self-reported weight and height. Although there are moderate-to-strong correlations of self-reported height and weight to measured height and weight (33–35), those who are really at the upper end of the BMI distribution tend to have underestimated BMIs based on self-reports of height and weight (36–40). If this bias is independent of self-reported age at menarche, it will result in an underestimation of the true strength of effect. Age at menarche was also self-reported, but recalled age at menarche has been shown to be generally accurate both for teenagers and for women up to 30 y of age and older (35–41).
Although we traced almost all the eligible subjects, 32% of the women did not respond to the questionnaire. Those women included in the multiple linear regressions were of higher social class, were less obese, and were less likely to be current smokers than were those who were omitted. Selection bias due to nonresponse and missing data would have biased our findings only if the association in questionnaire nonresponders or those with incomplete data were different. Although we had no data on which to assess this, we did not regard this as likely.
Age at which childhood BMI is measured
Why is it that we found little effect of adjustment for childhood BMI on the association of age at menarche with adult BMI, whereas 2 other similar studies (11, 32) found that childhood BMI appears to be an important confounder? We suggest that the effect of adjustment for childhood BMI may differ depending on the age at which it is assessed. This could be because the determinants of variation in BMI in early childhood are different from those in later childhood. If assessed between the ages of 4 and 6 y (as in our study), childhood BMI is largely unaffected by pubertal changes. However, if childhood BMI is measured later in childhood [ie, at 9 or 11 y as in the Bogalusa Heart Study (32) or in the 1958 British Born Cohort Study (11)], it is going to reflect to a greater degree the changes in growth and body composition associated with the process of prepubertal development and sexual maturation.
Possible explanations
It has been suggested that the underlying reason for the association of age at menarche with adult BMI is that childhood BMI drives (or is at least permissive of) the age at onset of sexual maturation and hence the age at menarche (11, 32). This view is represented in Figure 1. However, the fact that BMI also increases as a consequence of, or is entrained by, the process of sexual maturation means that BMI in later childhood may also be a marker of the stage of sexual maturation and hence of the age at menarche. This more complex scenario is represented in Figure 2.
FIGURE 1.. Previously proposed model for the relation between childhood BMI, age at menarche, and adult BMI.
FIGURE 2.. Our proposed model for the relation between childhood BMI, age at menarche, and adult BMI.
What evidence is there that Figure 2 is a more realistic depiction of what is occurring? High rates of postnatal growth and high early childhood BMIs are associated with an increased rate of sexual maturation (42, 43). In addition, childhood BMI tracks from early to later childhood. Finally, the rate of sexual maturation, in turn, appears to have a direct effect on BMI in later childhood and adolescence, as supported by recent results from the Fels Longitudinal study (44). All 3 pathways are shown in Figure 2. If this more complex model is correct, it would mean that differences that arise at any particular age in childhood, due to differences in the stage of sexual maturation, become effectively fixed and track into adulthood. This could be regarded as an amplification of earlier childhood differences in BMI. Indirect evidence of this comes from the 1958 British Born cohort (11). This shows that BMIs at age 11 and 16 y are associated with age at menarche to almost the same extent, but both are more strongly related to age at menarche than is BMI at age 7 y. In addition, the data from the Fels Longitudinal Study show that the difference in BMI between early and later maturing girls increases throughout adolescence (44)
Our more complex model would explain why adjustment for BMI in early childhood has a smaller effect on the association of age at menarche with adult BMI than does adjustment for BMI in later childhood. Early childhood BMI is unlikely to be the sole determinant of age at onset of puberty and hence will only be a weak confounder. In contrast, adjustment for later childhood BMI will attenuate the association to a greater degree because it will not only have tracked from an earlier age but will be a proxy for the changes in growth and body composition that have occurred as a consequence of the onset of puberty.
The hypothesis we set out to test was that the relation between age at menarche and adult BMI was largely explained by confounding with childhood BMI. However, our results are not consistent with this, because this hypothesis does not take into account that BMI in later childhood is itself in part determined by the onset of puberty. Instead, we suggest that early age at onset of puberty in girls may itself influence BMI in later life, directly through its effect on BMI and adiposity in childhood, and lead to an earlier age at menarche. Unfortunately we were not in a position to test this hypothesis because we did not have data on age at onset of puberty and later childhood BMI.
What are the implications of our hypothesis? Over the past century there has been a steady decline in age at menarche in all industrialized countries that have examined this (16–18). This appears to have slowed or ceased in some European countries (Britain and Holland) (17, 45), but not in others (Italy) (46), and in the US white population but not in African Americans (21). The effect of better nutrition and circumstances in childhood results in healthier children who mature and undergo puberty earlier. If we are right, this in itself will result in higher BMIs in later life. This long-term process may thus have contributed to the increase in BMI in the women.
ACKNOWLEDGMENTS
We are very grateful to Raymond Illsley for providing us with the data from the Aberdeen Child Development Survey and for his advice about the study. Graeme Ford played a crucial role in identifying individual cohort members and in helping us initiate the process of revitalization. Sally Macintyre, Doris Campbell, George Davey Smith, Marion Hall, Bianca De Stavola, Susan Morton, David Batty, David Godden, Diana Kuh, Debbie Lawlor, Glyn Lewis, and Viveca Östberg collaborated with David Leon to revitalize the cohort. Heather Clark managed the study at the Dugald Baird Centre, Aberdeen, with the assistance of Margaret Beveridge. We also thank the staff at the Information and Statistics Division (Edinburgh), General Register Office (Scotland), and National Health Service Central Register (Southport) for their substantial contributions; Debbie Lawlor and Anne-Marie Nybo for their helpful comments on an early draft of the manuscript; and the study participants who responded to a mailed questionnaire 40 years after the original survey was completed.
MBP developed the idea for the paper, conducted the analyses, and drafted the initial text. MBP and DAL were responsible for the analytic strategy and for the interpretation and drafting of the final text. Neither author had any conflict of interest.
REFERENCES