Timing of puberty in boys and girls: A population‐based study

Abstract Background A secular trend towards earlier puberty has been observed in girls, while a similar trend has been more uncertain in boys. We estimated current ages at pubertal development in both boys and girls. Methods In this population‐based cohort study, 14 759 of 22 439 invited boys and girls born from 2000 to 2003 in the Danish National Birth Cohort gave half‐yearly self‐reported information on puberty from the age of 11.5 years and throughout puberty. This late start of follow‐up limits the estimation of age at onset of puberty but not later pubertal milestones. We estimated mean age at attaining the following pubertal milestones in years with 95% confidence intervals (CI): age at menarche, voice break, first ejaculation of semen and Tanner stages for pubic hair development and breast development or genital development. Further, the difference in mean age at menarche between mothers and daughters was estimated. Results In boys, voice break occurred at 13.1 (95% CI 13.0, 13.1) years, first ejaculation of semen occurred at 13.4 (95% CI 13.3, 13.4) years, and Tanner Genital Stage 5 occurred at 15.6 (95% CI 15.5, 15.6) years. In girls, age at menarche occurred at 13.0 (95% CI 13.0, 13.1) years and Tanner Breast Stage 5 occurred at 15.8 (95% CI 15.7, 15.9) years. Daughters had menarche 3.6 (95% CI 3.1, 4.2) months earlier than their mothers had. Conclusion These data indicate that age at menarche has declined and to some extent support a decline in age at attaining other markers of pubertal development among boys.


| INTRODUC TI ON
Existing data show that the age at menarche, a marker of pubertal development in girls, has declined over the last century in Europe and the US, but it remains unsettled whether the decline is continuing or has stopped. 1,2 A decline in pubertal age in boys has also been suggested, but this is still not investigated in detail. 2 The reasons for a decline are unknown but is expected to be related to environmental and social factors as the genetic pool remains relatively stable over time. Childhood obesity, improved health, endocrine disrupting chemicals and prenatal exposures have been studied as potential causal factors. [3][4][5][6] Earlier timing of puberty is a potential concern 2 as it has been suggested as a risk indicator for some adult diseases, such as obesity, diabetes, heart diseases, breast cancer, and testicular cancer. [7][8][9][10] In this longitudinal, population-based cohort study, we estimate the current age at attaining various pubertal milestones in Danish boys and girls, using half-yearly self-reported information on pubertal development. Further, we investigate whether daughters had earlier age at menarche than their mothers as an indicator of a potential decline in age at menarche in girls.

| Study population
This study was based on the Puberty Cohort, nested within the Danish National Birth Cohort (DNBC). The DNBC includes approximately 92 000 pregnant women and their children and is described in detail elsewhere. 11 In short, women were recruited during their first trimester and gave information in

| Timing of puberty in boys and girls
We used a translated version of the growing and changing questionnaire from the British Avon Longitudinal Study of Parents and Children 14 and drawings of Tanner stages to collect information on puberty. Our questionnaire included Tanner stages for pubic hair (Tanner PH1-5) and breast (Tanner B1-5) or genital development (Tanner G1-5), axillary hair (yes, no), acne (yes, no), menarche (yes, no; age: year and month), first ejaculation of semen (yes, no; age: year and month), and voice break (yes-sometimes, yes-definitive changes, no, don't know). The questionnaire is available in Danish at www.dnbc.dk.
As the information on the pubertal milestones was collected half-yearly, the pubertal milestone occurred in an interval between an upper and a lower limit. The lower limit was the age at the last questionnaire where the child had not attained the pubertal milestone yet. Similarly, the upper limit was the first time the child had attained the pubertal milestone. The outcome was left-censored when a milestone was already attained by the first questionnaire, interval-censored when the pubertal milestone was attained between two questionnaires, and right-censored when the milestone was not attainted by the last completed questionnaire. When the child reported a specific age (years and months) for first ejaculation or menarche, this age was used as both upper and lower limits. Table   S1 gives an overview of the censoring of data.

| Maternal age at menarche
In the DNBC's first interview during pregnancy, the women, who were on average 30 years of age, recalled their age (in whole years) at their first menstrual period. Of the 7638 daughters that provided information on age at menarche, 7088 (93%) of the mothers had provided age at menarche.

| Background characteristics
We obtained self-reported information on maternal pre-pregnancy body mass index (BMI) and maternal smoking in first trimester from the first interview in the DNBC. We obtained information on mater-

| Statistical analysis
The following sampling strategy was employed in the Puberty Cohort. In total, 22 439 children were oversampled from 28 sampling frames: one sampling frame was the 56 641 eligible children from which we randomly sampled 8000 children, and the remaining 27 sampling frames were subgroups of 12 prenatal exposures believed to be important predictors of pubertal timing such as maternal smoking and alcohol consumption during pregnancy, pre-pregnancy BMI, birthweight, and diabetes mellitus. Hence, sampling weights were applied in all analyses. These sampling weights were the inverse of the probability of being sampled and were calculated based on the sampling fractions of the 28 sampling frames. All models were fitted with robust standard errors to account for the weighing approach for sampling and clustering of siblings.
Analyses were performed in STATA 13.1 MP software (StataCorp, College Station, TX) and R (×64 3.3.1). We estimated age in years at attaining the following pubertal milestones: Tanner Stage 2 to 5, acne and axillary hair in both boys and girls; menarche in girls; and first ejaculation and voice break in boys. To estimate the age at attaining the pubertal milestones in boys and girls, we used a parametric survival analysis for interval-censored data based on the normal distribution fitted by maximum-likelihood estimation. 15 The nonparametric distributions (plotted as the cumulative incidence function) of the interval-censored pubertal milestones based on the Turnbull Estimator were visually compared with the normal, Weibull, loglogistic, and gamma distributions. 16,17 The normal distribution had the best fit ( Figures S2 and S3). All pubertal milestones were analysed separately in a univariate model without explanatory variables.
Results were presented graphically as cumulative incidence functions and in tabular format as mean age with both 95% confidence interval (CI) and 95% prediction interval (PI). The latter estimates the interval where 95% of participants attained the given pubertal milestone.
Difference in age at menarche between mother and daughter was estimated as follows. In 7088 mother-daughter pairs with information on age at menarche in both mother and daughter, the mean age at menarche was estimated for mothers by univariate linear regression and then estimated for daughters by the interval-censored regression model. Then, we calculated the difference of the means between mothers and daughters and 95% CIs were computed using a nonparametric bootstrap procedure with 10 000 replications and normal-based standard error.
Two sub-analyses were performed for the analysis on age at attaining the pubertal milestones. In the first sub-analysis, we excluded all participants with inconsistent reporting of Tanner stages, that is reporting a lower Tanner stage in one questionnaire than in any previous questionnaires. In total, 6572 (86%) girls had consistent information on all Tanner Breast stages,

(81%) girls had consistent information on all Tanner Pubic
Hair stages, 5504 (77%) boys had consistent information on all Tanner Genital stages, and 6158 (87%) boys had consistent information on all Tanner Pubic Hair stages. In the second sub-analysis, we assessed the potential impact of selection bias due to nonparticipation (34%) using inverse probability weights. 18 This was performed by fitting a multivariable logistic regression model for participation using the following explanatory variables: the highest social status of parents, maternal pre-pregnancy BMI, maternal smoking in first trimester, maternal age at delivery, and parity with continuous covariates (maternal pre-pregnancy BMI and age at delivery) included as second-order polynomials. The inverse of the predicted values was used as selection weights. These selection weights were multiplied by the sampling weights from the design of the Puberty Cohort to compute the final weights used in this sub-analysis.
To assess the sensitivity of violations of the normality assumption, we conducted a simulation study by simulating data with varying degrees of skewness and censoring and assessed the performance of the interval-censored regression model used in this study (detailed description in Appendix S1, Tables S2 and S3, and Figures S2 through S7).

| Ethics
The Committee for Biomedical Research Ethics in Denmark approved the collection of data in the DNBC. A written informed consent was obtained from mothers upon recruitment covering both mother's and offspring's participation until they turn 18 years of age.
This study was approved by the Danish Data Protection Agency and the Steering Committee of the DNBC.  Figure 1 shows the cumulative incidence of the age at attaining Tanner stages, voice break and first ejaculation in boys, and Tanner stages and menarche in girls. Table 2 shows the average age at attaining the pubertal milestones with 95% CI and 95% PI. Results remained unchanged when excluding participants with inconsistent reporting of Tanner stages (Table 3) and when using selection weights to account for potential selection bias (Table 4).

| Simulation study
The simulation study showed that the estimate for the Tanner B2 with 90% left censoring is unbiased if the assumption of normality is correct but may be biased by up to +/−0.35 years under realistic violations of the assumption of normality, whereas estimates for all later milestones were biased with up to +/−0.05 years under the same realistic violations (Table S4).

| Difference in age at menarche
Age at menarche occurred 3.6 (−3.6 [95% CI: −4.2, −3.1]) months earlier in daughters than in their mothers. As the mean maternal age at delivery was 30 years, this potential decline in age at menarche occurred over approximately 30 years.

| Principal findings
This longitudinal cohort study provides current estimates for age at attaining various pubertal milestones in both boys and girls. The age at menarche in daughters was estimated to occur 3.6 months earlier than that in their mothers, which indicates that the age at menarche may have declined.

| Strengths of the study
The main strengths of this study include a large sample size with longitudinal data collection or short recall periods for age at menarche and first ejaculation, allowing us to estimate both age at attaining the pubertal milestones and difference in age at menarche between mothers and daughters with some precision. The large set of puberty markers allow comparability with both previous and future studies, and the high participation rate reduces the risk of selection bias. Further, our results were robust in sub-analyses in which we excluded children with inconsistent reporting of Tanner stages and took potential selection bias into account.

| Limitations of the data
A limitation of the analysis of age at attaining the pubertal milestone in Danish boys and girls is the uncertainty related to self-reported information on Tanner stages collected through questionnaires, which is expected to be more prone to misclassification than information collected in clinical examinations. 19 However, studies using clinical examinations  Years Girls Age at menarche used for data collection. For daughters, age at menarche was collected during puberty when their menarche occurred, thereby reducing the risk of misclassification, whereas age at menarche for the mothers was obtained during pregnancy when they were around 30 years, that is 16.4 (25 and 75 percentiles: 13.5, 19.8) years after their menarche. If mothers tend to systematically report an older age at menarche than their actual age, this misclassification could explain the observed difference. However, recalled age at menarche has been found to be remembered even in middle-aged women with some accuracy and without any systematic over-reporting or underreporting, indicating that our estimate need not be biased in one direction by recalled age at menarche. 20

| Interpretation
With regard to boys in this study, voice break occurred at an average age of 13.1 years, which is considerably earlier than former studies reporting between 15.5 and 14.0 years in the period 1968 to 2005. [21][22][23] However, the reporting of these ages may not be directly comparable as those studies used either an evaluator to assess whether the boys had voice break during speaking 21 or singing 23 or the boys recalled timing of voice break during adulthood. 22 Age at first ejaculation also occurred more than 1 year earlier in our study than an earlier study from 2005 using recalled age at first ejaculation (13.4 vs 14.7 year, respectively). 22 Tanner G2 and PH2 occurred around half a year earlier than most recent Danish data, whereas the age at Tanner G5 and PH5 were similar. 24,25 However, these results are not directly comparable as the former studies used clinical examination with low participation rates of 25%-35%, implying an increased risk of selection bias, but probably with less misclassified information on the pubertal milestones than in our study, especially for Tanner another study found a 4-month decline in age at voice break over a Girls in this study had menarche at the age of 13.0, which is similar to recent data from the Copenhagen Puberty Study. 27 These data are probably comparable with ours as both were self-reported as either current status or specific age with short recall, but there may be differences at the population level due to participation (35% and 66%, respectively). Our data indicated a decline in age at menarche of 3.6 months when comparing mothers with daughters. We speculate that this potential decline might be larger owing to the following reporting pattern. A mother may likely report an age at menarche of, The current data may be suitable to evaluate the cut-off ages for delayed puberty, which is usually defined as no sign of genital development in boys and breast development in girls at an age >2 or >2.5 standard deviations (SD) above the population mean, depending on the definition. 28 Traditionally, the cut-off ages for delayed puberty have been 14 years in boys and 13 years in girls, which is currently in use in Denmark. 28  A decline in timing of puberty may be of concern as early puberty may be causally related to later diseases in adulthood, such as obesity, diabetes mellitus, cardiovascular diseases, testicular cancer, and breast cancer. 7-10 This calls for continued surveillance for secular trends in pubertal timing. This study was aimed to estimate a wide range of pubertal markers as reliable as possible from a nationwide sample. These data were intended to serve as the basis for comparison with both former and future data on timing of puberty.
We only compared our data with former Danish data to increase the comparability as timing of puberty varies throughout the world. 34 However, a decline in pubertal timing in Denmark may well be generalizable to other countries with similar genetic background and distribution of social and environmental factors.

| CON CLUS IONS
In boys, most pubertal milestones were attained at younger ages than previously reported, but this apparent decline should be evaluated with caution. In girls, age at menarche occurred 3.6 months earlier in daughters than in their mothers.