Puberty Timing and Adiposity Change Across Childhood and Adolescence

Disentangling Cause and Consequence

Linda M. O'Keeffe; Monika Frysz; Joshua A. Bell; Laura D. Howe; Abigail Fraser

Disclosures

Hum Reprod. 2020;35(12):2784-2792. 

In This Article

Materials and Methods

Study Participants

Data were from the Avon Longitudinal Study of Parents and Children (ALSPAC), a prospective birth cohort study in southwest England (Boyd et al., 2013; Fraser et al., 2013). Pregnant women resident in one of the three Bristol-based health districts with an expected delivery date between 1 April 1991 and 31 December 1992 were invited to participate. The study is described elsewhere in detail (Boyd et al., 2013; Fraser et al., 2013). ALSPAC initially enrolled a cohort of 14 451 pregnancies, from which 13 867 live births occurred in 13 761 women. Follow-up has included parent- and child-completed questionnaires, clinic attendance and links to routine data. Research clinics were held when the participants were ~7, 9, 10, 11, 13, 15 and 18 years.

Ethical Approval

Ethical approval for the study was obtained from the ALSPAC Ethics and Law Committee and the Local Research Ethics Committees. The study website contains details of all the data that are available through a fully searchable data dictionary https://www.bristol.ac.uk/alspac/researchers/our-data/ (University of Bristol, 2020).

Data

Assessment of Puberty Timing. Puberty is a period of intense hormonal activity and rapid growth, of which the most striking feature is the spurt in height (Cole et al., 2014). aPHV is a validated measure of pubertal timing (Cole et al., 2014) captured using Superimposition by Translation and Rotation (SITAR), a non-linear multilevel model with natural cubic splines which estimates the population average growth curve and departures from it as random effects (Cole et al., 2010; Simpkin et al., 2017). Using SITAR, PHV was identified in ALSPAC participants using numerical differentiation of the individually predicted growth curves, with aPHV being the age at which the maximum velocity is observed (Cole et al., 2010; Simpkin et al., 2017; Frysz et al., 2018). Repeated height data included measurements from research clinics. Individuals with at least one measurement of height in each of the age ranges 5 to <10, 10 to <15 and 15–20 years are included here. Data were analysed for females and males separately. The model was fitted using the SITAR package in R version 3.4.1, as described elsewhere (Frysz et al., 2018). Further details on how aPHV was derived are described elsewhere (Frysz et al., 2018) and in Supplementary Methods, Table SI and Figure S1 of Supplementary Material.

Assessment of Adiposity. Adiposity was assessed via total body fat mass (in kilogram, less head) as derived from whole-body DXA scans performed five times at ages 9, 11, 13, 15 and 18 years using a GE Lunar Prodigy (Madison, WI, USA) narrow fan-beam densitometer.

Confounders. We considered the following as potential confounders in our analysis: birth weight, gestational age, maternal education, parity, maternal smoking during pregnancy, maternal age, maternal pre-pregnancy BMI, household social class, marital status, partner education and ever breastfeeding all measured by mother-or mother's partner-completed questionnaires; details in Supplementary Methods). The distribution of confounders included in our analyses, including the proportion of missing data for each confounder by fourths of aPHV, is also shown in Supplementary Table SII; note the table demonstrates minor differences in the proportion of missing confounder data by fourths of aPHV.

Statistical Analysis

Multilevel models were used to examine change in fat mass during childhood and adolescence (Laird and Ware, 1982; Goldstein, 1995). Using terms such as polynomials and splines to account for non-linearity in the trajectory, such models can estimate mean trajectories of the outcome while accounting for the non-independence or clustering of repeated measurements within individuals, change in scale and variance of measures over time, and differences in the number and timing of measurements between individuals (using all available data from all eligible participants under a missing at-random assumption) (Howe et al., 2013; Tilling et al., 2014). Participants that reported being pregnant at the 18-year clinic were excluded from the multilevel models at that time point only (N = 6). Participants that had a measure of aPHV, at least one measure of fat mass from 9 to 18 years and complete data on all confounders were included in analyses, leading to a total sample of 4176 (2186 females and 1990 males).

All analyses were performed separately for females and males. aPHV was normally distributed in both sexes. Linearity of associations of aPHV with fat mass was examined by comparing the model fit of regressions of fat mass on aPHV, with continuous aPHV and fourths of aPHV examined as continuous exposures. Model fit was then formally tested using a likelihood ratio test. Prior to analysis, aPHV was centred on the sex-specific mean of aPHV for females and males. Fat mass was also log transformed due to its skewed distribution. All models were adjusted for height using the time- and sex-varying power of height that best resulted in a height-invariant measure, described in detail elsewhere (O'Keeffe et al., 2019a,b). We performed unadjusted and confounder-adjusted analyses on participants with complete data (N = 4176) for all models.

From all models, we back-transformed the difference in fat mass trajectories per year of aPHV and the average trajectory for the 10th, median and 90th sex-specific percentile of aPHV. The back-transformed difference in fat mass per year of aPHV is a ratio of geometric means, expressed here as a percentage difference per year of aPHV. The average trajectories back-transformed from the log scale are in original units (kg) and are presented in figures.

Models for Fat Mass Trajectories. A common approach to modelling change over time using multilevel models involves examining change by chronological age (O'Keeffe et al., 2018a,b, 2020). However, when change before or after a specified event is of interest (e.g. onset of puberty or menopause), it is also possible to model change according to other time metrics such as time before and/or after the event. Thus, to gain a greater understanding of the association of aPHV with change in fat mass during childhood and adolescence, we modelled trajectories of fat mass in two ways: by chronological age, and separately by time before and after puberty.

Model 1: Chronological Age-based Models. Fat mass was previously modelled according to chronological age using linear spine multilevel models, with three periods of linear change (9 to <13, 13 to <15 and 15–18 years) (O'Keeffe et al., 2018a,b, 2019a,b). Thus, for this analysis, we examined whether this model was appropriate for modelling change over time within quartiles of pubertal age to ensure that model fit was adequate across the entire distribution of pubertal age. Subsequently, the association between aPHV and chronological age-based trajectories was then examined for females and males by including an interaction between centred sex-specific aPHV and the intercept (age 9 years) and each spline period, providing an estimate of the difference in the average trajectory of fat mass from age 9 to 18 years, per year later aPHV. Confounders were included as interactions with both the intercept and linear slopes; note, inclusion of interaction terms between variables (exposures/confounders) and the outcome trajectory is the standard approach to examining associations of an exposure with an outcome trajectory and adjusting this outcome trajectory for confounders in multilevel models (Howe et al., 2013; Tilling et al., 2014). A main effect for the exposure/confounder plus interaction terms between exposures/confounders and linear spline terms ensures that the effect of the exposure/confounder on the intercept (here, the value of fat mass at 9 years) and each linear spline term is modelled. For exposures (here aPHV), this allows us to estimate differences in trajectories of fat mass from 9 to 18 years by different values of aPHV. For confounders, this ensures that the full trajectory of fat mass from 9 to 18 years is adjusted for the confounders of interest.

Model 2: Pubertal Age-based Models. The purpose of the pubertal age-based model was to examine whether changes in fat mass before or after puberty onset differ by aPHV. In order to select an appropriate model, we examined observed data for fat mass in females and males by sex-specific quartiles of aPHV. Based on the observed data, a selection of suitable models was examined, each with different numbers of pre- and post-pubertal change periods. We compared observed and predicted values of fat mass for these models by sex-specific quartiles of pubertal age to examine model fit. In females, the final model selected for fat mass included two periods of change (pre-puberty and post-puberty). The final model for fat mass in males had three periods of change (from age 9 to 3 years before puberty, from 3 years before puberty to puberty (i.e. aPHV) and from puberty to the end of follow-up at 18 years). Differences in the rate of change in fat mass before and after puberty by aPHV were then modelled by including an interaction between centred sex-specific aPHV and the intercept (fat mass at puberty) and each linear spline period (one pre- and post-pubertal spline period for females and two pre- and one post-pubertal spline period for males). This model provided insight into whether different ages at PHV were accompanied by different rates of change in fat mass before and after puberty. Confounders were included as interactions with the intercept and linear slopes, as described above for chronological age-based models. All trajectories were modelled in MLwiN version 3.04 (University of Bristol Centre for Multilevel Modelling, 2019), called from Stata version 16 (StataCorp, 2019) using the runmlwin command (University of Bristol Centre for Multilevel Modelling, 2016).

Further details on model selection are included in Supplementary Methods and details of model fit for both models are included in Supplementary Tables SIII and SIV.

Additional and Sensitivity Analyses. We examined the characteristics of mothers of participants included in our analysis compared with mothers of participants excluded from our analysis due to missing exposure, outcome or confounder data to better understand generalisability and the potential for selection bias. We performed unadjusted analyses on the sample of participants that had data on aPHV and at least one measure of fat mass from 9 to 18 years; this analysis included an additional 1517 participants excluded from our main analysis due to missing confounder data (total N = 5693). We regressed observed fat mass at 9 years (first available measure) and 18 years (last occasion of measurement) on aPHV in females and males and compared results to those obtained from the multilevel models at these ages. We performed sensitivity analyses restricting the sample to participants with at least one fat mass measure before and one after aPHV to examine whether results from the main analysis were driven by participants with only a single pre- or post-puberty fat mass measure. We examined whether the association of self-reported age at menarche with fat mass during childhood and adolescence was similar to findings for the association of aPHV and fat mass among females. In addition, we examined whether our findings were similar in analyses when non-White ALSPAC participants were excluded from analyses (N = 178 excluded).

processing....