This study used data from the Taiwan Pubertal Longitudinal Study (TPLS). The TPLS follows a multicenter, population-based, prospective cohort. Since July 2018, we have conducted the TPLS by recruiting girls aged between 6 and 14 years and boys aged between 9 and 17 years at pediatric endocrinology outpatient clinics at Taiwan Medical University Hospital, Taipei Municipal Wanfang Hospital, Cathay General Hospital (CGH), and National Cheng Kung University Hospital (NCKUH). This study was approved by the Institutional Review Board of Taipei Medical University (N201802018), CGH (CGH-P108107), and NCKUH (B-BR-108-076) and complied with the principles outlined in the Helsinki Declaration (Supplemental Materials: Taiwan Puberty Longitudinal Study).

The main outcome variable was sexual maturation, which was evaluated every 3 months following enrollment. Tanner’s pubertal scale was used by pediatric endocrinologists at the outpatient clinic during each visit to determine the overall Tanner stage, the breast development for girls (staged as B1-B5) and the genital development for boys (staged as G1-G5) [8]. The development of the genitals for boys were also examined by measuring the testis volume (Tvol) [9]. We defined the timing of puberty onset as the month at which the participants achieved Tanner stage 2 for both sexes or achieved stage B2 (thelarche) for girls or testis volume ≥ 4 ml (the onset of testicular enlargement) for boys.

The primary exposure variables were anthropometric measures collected every 3 months since enrollment; these were body height, body weight, waist circumference, body composition analysis results, and the findings of questionnaires on physical activity.

Body mass index (BMI) was defined as the weight (in kilograms) divided by the square of height (in meters) and then transformed into age- and sex-specific Z scores (zBMI) according to the World Health Organization (WHO) Child Growth Standards for school-aged children and adolescents [10, 11]. Waist circumference was measured in centimeters by using a flexible tape at the natural waist (midpoint between the lower ribcage and iliac crest) and recorded to the nearest 1 mm. At the level of the maximal posterior extension of the buttocks, hip circumference was measured in centimeters. The waist-to-height ratio was calculated as waist circumference divided by height, and the waist-to-hip ratio was calculated as waist circumference divided by hip circumference [11]. Body composition, including BFP and fat-free mass (FFM), was measured in light clothing by using a portable bioimpedance analysis (BIA) electronic scale TT-BC418 (Tanita Corp., Japan). BIA is considered a reliable and practical tool to estimate BFP in children [12]. The children were asked to avoid fluid or food intake and vigorous exercise 2 h prior to BIA measurements. In addition, BIA measurements vary among different ethnic groups, and our BIA machine used predictive equations adjusted for the Asian population [13]. Standardized protocols were used in every participating hospital to prevent systematic bias.

The a priori confounding factors were prenatal factors (birth weight), postnatal factors (breastfeeding, diet composition and physical activity), and socioeconomic factors (parental education and family income) [14]. Breastfeeding was considered if children had been breastfed for at least 3 months after birth. Children were invited to record all instances of eating in written food records. Two trained dietitians independently disaggregated the foods into their constituent ingredients. The energy was averaged from the collected three-day data and calculated as kcal/day. Macronutrients include protein, fat, carbohydrates, and dietary fiber, and their basic units were calculated as grams [15]. The Chinese version of the International Physical Activity Questionnaire was used to determine the participants’ physical activity [16]. We calculated the participants’ metabolic equivalent values (METs) according to their physical activity in the past 7 days, and the physical activity level was classified into the following 3 categories: mild (< 3 kcal kg−1 h−1), moderate (3–6 kcal kg−1 h−1), and vigorous intensity (> 6 kcal kg−1 h−1) [16].

In the TPLS cohort, we included girls aged between 6 and 14 years and boys aged between 9 and 17 years who visited the pediatric endocrinology clinic for pubertal growth assessment. Among these children, we included those evaluated as having Tanner stage 2, breast stage 2, or testis volume ≥ 4 ml during at least one visit until the end of September 2020. [17] The reason for the inclusion is that, the first physical marker for the onset of puberty in girls is typically the transition from Tanner breast stage B1 to B2, which includes the early growth of the breast tissue. In boys, this marker is the change from Tanner genital stage G1 to stage G2, including the enlargement of the testis [17]. We reviewed several definitions of testicular volume cut-points representing the onset of puberty in boys and adopted the achievement of a testicular volume of ≥ 4 ml [18]. We excluded children with precocious puberty diagnosis and those without adiposity measures. According to a previous study [19], the proportion of R2 for both BFP and Tanner stage can be approximately 0.11. Hence, the sample size calculated at a 5% significant level and 80% power was 122 participants in both the case and control groups.

To investigate critical variables related to the timing of puberty onset, we calculated the regression coefficient (β) of age at the month of puberty onset (years, months) for each anthropometric measure (%, namely zBMI, BFP, FFM percent, waist-to-hip ratio, and waist-to-height ratio), and parental education, family income, breastfeeding, and physical activity were adjusted (Table 2).

To compare the importance of various anthropometric measures, determine the critical cutoff point of BFP for the timing of puberty, and understand the clinical applicability of anthropometric measures when combined with the radiographic examination, we used the classification and regression tree (CART) to investigate the association between the anthropometric measures (input) and age at puberty onset (target). The CART model derived the cutoff points of these predictors. In addition, the difference between bone age and chronological age (BA-CA) was included as an input factor (Fig. 2; details are in the Supplemental Materials). We used the data of the participants from NCKUH as a validation set to determine the model’s accuracy.

To investigate the critical time period of the BFP transition before and after puberty, longitudinal measures were evaluated at 3 months intervals. We used months before and after the onset of puberty instead of chronological age to determine the association of time to puberty onset with change in BFP. The month in which the participants achieved Tanner stage 2 was considered the time of puberty onset, and the month in which the participants achieved thelarche was used in the sensitivity test. Thelarche (the onset of breast development, Tanner breast scale 2, B2) often indicates the beginning of pubertal development and is related to the reactivation of the hypothalamic–pituitary–gonadal (HPG) axis [17]. We use the mean to impute the missing data of anthropometric measures individually. Statistical analyses were conducted using R software (Version 3.6.3; R Foundation for Statistical Computing, Vienna, Austria).