Children with Early Sexual Maturation Exhibit Changes in Dairy and Soybean Consumption: A 6-Year Nationwide Study

Early sexual maturation (ESM) is associated with behavioral disorders in adolescence and hormone-related cancers in adulthood. Dietary pattern (DP) has been noticed in association with ESM. However, to our knowledge, no study has shown that association between DP derived from reduced-rank regression (RRR) and ESM, and whether ESM inuences tracking dietary habit in children. This study was therefore to examine the prospective association between childhood dietary pattern (DP) and ESM, and whether ESM inuenced children’s dietary habit during 6 years of pubertal growth.

disease, adiposity, nutrition intake, and pubertal growth. The present study involved schoolchildren aged 11 years in 2011 who had already received nutrition education regarding the appropriate types and proportions of foods to consume. Thus, the children were able to answer a food frequency questionnaire (FFQ) in 2011 and 2017 (Fig. 1).
Body composition including body weight, fat mass, and body fat percentage were measured using a body composition machine (IOI 353, Jawon Medical, Korea; accuracy: 0.1 kg). Body height was measured to the nearest 0.1 cm using a wall-mounted stadiometer (NAGATA, BW-120, Taiwan) at each school. Body mass index (BMI) was calculated as weight (kg) divided by (height) 2 . We divided family income into 3 groups: below US$1,666 per month, US$1,666-6,666 per month, and over US$6,666 per month (1 US$ = 30 New Taiwan Dollar). Parental level of education was determined by either mother's highest level of education or father's highest level of education.

Measurement of ESM
Puberty outcomes were assessed in 2012, one year after the FFQ assessment. Children self-reported their Tanner-derived composite stage (TDCS) [17], revealing their pubertal stage. The TDCS contains drawings of the 5 stages of secondary sexual characteristics (pubic hair and breast development for girls and pubic hair and genital development for boys). The two sex characteristics were combined and averaged to create a single Tanner score [18]. Furthermore, ESM was de ned as a child reaching a certain pubertal stage earlier than the median age for that stage [19], referencing 2 large-scale population-based Chinese studies [20,21]. We validated the Chinese version of the TDCS, and the consistency between Tanner stages from self-reports and physical inspection by a pediatric endocrinologist was high [22].

DP measurement and analysis
We collected dietary intake information in this study by using an FFQ-a modi ed version of the Nutrition and Health Survey in Taiwan. The survey included 35 questions on 71 food items, which were categorized into the following 9 food groups: vegetables, fruits, staple foods, eggs, milk, yogurt, cheese, meat and sh (meat, sh, shell sh, and organ meat), and soybean products (soybean milk and tofu). The consumption frequency for each food item was evaluated using a scale from 0 to 7 days per week. If a food item was left blank, intake was set to zero [23]. Each food item in the FFQ was assigned a portion size using appropriate local tableware, such as bowl (200 g) and cup (240 mL), for Taiwanese foods, unless the item was from the fast food group. Daily portion sizes were calculated by multiplying the weekly intake frequency by the number of portions and dividing it by 7. Well-trained staff, including one professional dietitian, provided instructions to the participants, who were asked to complete the FFQ by themselves. Participants could ask our staff questions at any time throughout the sessions.
We employed 2 emperical methods, FA and RRR, to derive DPs from the 9 de ned food groups in 2011. FA with principal component estimation, an explorative multivariate statistical technique, was used to extract noncorrelated factors from the 9 food groups (predictors). To minimize the number of indicators that have a high loading on one factor, we used varimax orthogonal rotation. Consequently, 3 factor solutions of eigenvalue > 1 were extracted and examined using break point in the scree plot, and food groups with a factor loading greater than 0.4 were used to interpret the factor. Furthermore, RRR considers both predicting variables and the outcome, which, in this study, were food items and ESM (responses with 0 = non-ESM and 1 = ESM), as previously described [9]. In both techniques, the DP score for each individual was calculated as the sum of portion size of each food item, weighted according to the corresponding factor loading greater than 0.4. Each participant's DP score was then categorized into quartiles in which quartile 4 represented those whose diet adhered most to that particular DP.
Six-year tracking was performed using con rmatory analysis in which all factor loadings were the same ones evaluated at baseline, derived from either FA or RRR. Changes in DP score represent actual differences in food consumption frequencies and portion sizes [24].

Physical activity
Participants were asked to report their physical activity during the preceding 7 days by using the validated Chinese version of the International Physical Activity Questionnaire [25]. To express the intensity of physical activities, the daily metabolic equivalent of task (DMET) was calculated as the sum of total energy expenditure for physical activities per body weight per hour (kcal/kg/hour). Screen time per day was determined by average screen time during the whole week (hours/day).

Statistical analysis
Data are expressed as frequency (%) and mean ± standard deviation for categorical and continuous variables, respectively. The chi-square test and Student's t test were used to compare differences among groups. The association between the DP score at age 11 and ESM at age 12 in 2593 children was assessed using logistic regression analysis and expressed as odds ratio (OR) and 95% con dence intervals (CIs). The adjusted model was established using multivariate logistic regression that adjusted for parental education, family income, body fat percentage, total energy intake, and screen time.
Paired Student's t tests were used to determine the mean difference of DP score in subgroups between the initial evaluation and after achieving sexual maturation (participants who completed both evaluations = 1018). Statistical power was analyzed considering the effect size measured using Cohen's d parameters: < 0.2 = small effect, 0.2 to 0.8 = medium, and > 0.8 = large [26].
Statistical signi cance level was set at P < 0.05. FA and factor loadings derived from RRR were extracted by using PROC FACTOR and PROC PLS DATA = DIET METHO = RRR, respectively, in SAS version 9.4 (SAS Institute Inc.; Cary, NC, USA). SPSS Statistics for Windows, version 25 (IBM Corp., Armonk, N.Y., USA) was used for descriptive and analytical statistics.

Characteristics of the population
In total, 2593 children aged 11.1 ± 0.3 years were recruited in 2011 and were followed-up to assess the pubertal outcomes in 2012 (aged 12.1 ± 0.3). After 6 years of follow-up, only 1018 children (aged 17.3 ± 0.5) remained in the cohort and completed the questionnaire, Figure 1. Only minor differences were noted between the basic characteristics of those who completed the follow-up in 2018 and those who did not [Additional le 1]. Table 1 presents the baseline characteristics of study participants. Females had a higher total body fat percentage at baseline than did males. BMI increased signi cantly between the 2 time points in both sexes, and energy intake decreased in females but not in males. We observed signi cantly decreased DMET and increased sedentary time during growth development in males. In females, although DMET was not signi cantly different, we noted increased sedentary time over the 6-year interval. We found meaningful changes in dietary habits in terms of intake proportion for carbohydrate, protein, and fat in females. Similarly, in males, we observed that the intake proportion of carbohydrates decreased, whereas that of fats increased; however, we observed no change in protein proportion, Table 1.
Baseline DPs derived from FA and RRR FA revealed 3 DPs with the highest eigenvalues, which explained 54.9% of the total variance of food intake from the 9 food groups calculated as portion size. The "dairy-soybean diet" included high factor loadings of milk, yogurt, cheese, and soybean products. The "traditional diet" was highly loaded with staple foods, eggs, and meat and sh. The "vegetarian diet" involved a high intake of vegetables and fruits. RRR-derived DP was characterized by high factor loadings of yogurt, cheese, and soybean milk, Table 2.
Association between DP score changes and ESM Table 4 lists the mean differences in the score changes of DPs derived from FA and RRR over a 6-year interval. The FA-derived "dairy-soybean diet" DP scores dropped signi cantly more in children with ESM than in children without ESM in both sexes. For the RRR-derived DP, only males with ESM exhibited a signi cantly decreased DP score, Table 4.
We observed no differences in the "vegetarian diet" and "traditional diet" DP scores between children with and without ESM, Table 4. However, in general, females and males changed their dietary habits, as evidenced by signi cantly decreased "vegetarian diet" DP scores and an increasing tendency in "traditional diet" DP scores after achieving sexual maturation [Additional le 2].

Discussion
We investigated for the rst time the prospective DPs derived from both FA and RRR in association with ESM. Healthy children who adhered to "dairy-soybean diet" during the peripubertal period were more likely to experience ESM. RRR-derived DP yielded a similar association with ESM. The DP score for the "dairy-soybean diet" in children with ESM decreased signi cantly more than it did in children without ESM over the 6-year period. In general, children were prone to resume the "traditional diet" and follow the "dairy-soybean diet" and "vegetarian diet" to a lesser extent by the end of puberty. These ndings have illustrated the need for appropriate dietary modi cations to promote healthy pubertal growth.
The 3 major DPs, the "dairy-soybean diet", the "vegetarian diet", and the "traditional diet", contributed 54.9% of the total variation of the 9 food groups identi ed using the FFQ. Children with ESM adhered to "dairy-soybean diet", derived from FA and RRR. This diet, which includes milk, yogurt, and dairy products has been associated with accelerated pubertal development in longitudinal studies [27][28][29]. Milk and dairy products contain measurable quantities of complex steroid hormones and growth factors, such as estrone and 17beta-estradiol [30], insulin-like growth factor 1 (IGF-1), and IGF-binding protein 3 [31] that act on the hypothalamic-pituitary-gonadal axis to modulate reproductive function [4]. Despite the fact that steroid hormones and IGF-1 levels in cow milk are small relative to endogenous production rates in humans [30,31], their structures are identical [32]. This might affect breast development [33] and pubertal timing [27] in children with high consumption of milk and dairy products. Notably, on average, 35% of total calcium intake among girls 9-12 years is derived from dairy products [28]. 12-month calcium supplementation in children aged 8-12 years advanced the Age at Peak Height Velocity [34]. Also, a higher consumption of total calcium and milk was associated with a high risk of early menarche [28,35]. In addition, soy products contain considerable quantities of phytoestrogen, which is hormonally similar to mammalian estrogens and their active metabolites [36]. We found that children with ESM consumed more soy milk, which highly contains the iso avone genistein [37]. Dietary genistein intake at 1-2 years before the Age at Take-Off had a weak positive association with Age at Take-Off, Age at Peak Height Velocity, and age at pubertal stage 2 for breast development [8]. This weak association might be due to the aforementioned study's smaller sample size compared with ours (227 vs. 1018). However, other cohort studies have found no association between soy-based products and pubertal timing. This disparity may be due to dissimilar exposure timing (aged less than 3 and 6-8 years) [29,38]. Meanwhile, the adherence to soymilk intake at age 11 may strongly exert the effects of soymilk-derived iso avone genistein on ESM in the present study. Thus, the time window of phytoestrogen exposure on sexual maturation merits further investigation.
The "traditional diet" was not associated with ESM in children. Similarly, Chen et al. [11] found no association between precocious puberty and "traditional diet" included white meat, seafood, vegetables, fruits, and dairy products. Our "traditional diet" pattern included a high intake of animal protein which has been found not associated with early menarche [39,40]. Meat consumption was not associated with IGF-1 and IGF-binding protein 3 levels that regulate growth and pubertal development [41,42].
Dietary tracking is increasingly used to determine the stability of dietary habits in children from childhood to adolescence [24,[43][44][45][46]. However, the rationale for understanding of factors that in uence children's eating habit is limited [14]. We found that children with ESM exhibited a signi cantly larger decrease in DP scores for the "dairy-soybean diet" than did children without ESM over a 6-year period with the moderate effect size. This is consistent with that of the GINIplus cohort study [45]. However, Harris et al. did not evaluate the puberty onset as an in uence of tracking dietary habits [45]. In young adulthood, those who experienced ESM were likely to resume the "traditional diet", which was characterized by meat, sh, eggs, and staple foods. The transition from childhood to adolescence seemingly led to children increasingly to adopt the "traditional diet", which is the principal DP in adulthood, regardless of ESM [Additional le 2]. Harris et al. [45] found that males increased their meat intake during puberty; whereas, females shifted to a vegetarian diet that protect them from the risk of breast cancer [47]. However, both sexes exhibited signi cantly lower adherence to the "vegetarian diet" from childhood to adolescence [Additional le 2], with a signi cant decrease in carbohydrates proportion in their servings, Table 1. Consistently, the low to moderate tracking of fruits and vegetables was observed from childhood to adulthood [45]. These ndings emphasize the importance of encouraging healthy eating habits, which include fruit and vegetable consumption, in young people.
Concerning sex differences, the association between the quartiles of the "dairy-soybean diet" and ESM revealed a signi cant dose-responsive relationship in females but not in males. This suggests that females are seemingly more vulnerable to the effects of the "dairy-soybean diet" than males. By contrast, males who adhered to "vegetarian diet" at age 11 were more likely to experience ESM. Inconsistent with our ndings, DPs characterized by fruit and vegetable consumption during early childhood (at age 3) were associated with a delay in pubertal onset [10]. The fact that children consume more fruits and vegetables high on glycemic index, which have been associated with an increased prevalence of obesity, a critical risk factor of ESM [48,49]. Hence, further research is required on how the consumption of fruits and vegetables with high glycemic index affect the ESM.

Strengths And Limitations
The strengths of our study include its longitudinal design and nationwide representation of children across Taiwan. This is the rst study to elucidate the prospective association between DPs deriving from RRR method [9] and ESM. Our DPs, derived from FA with varimax orthogonal rotation were characterized by non-overlapping food groups in each DP that facilitates the interpretation of the DP score's changes over a 6-year period in association with ESM, which has not been done before. In particular, those associations were controlled by parent education, family income, body fat percentage, physical activity, and total energy intake. These potential factors have been considered to modulate the onset of puberty and dietary habits in children [22,[50][51][52]. Our analysis of DP score changes between one year before the sexual maturation assessment (aged 11) and early adulthood (aged 17) incorporated the critical period of sexual maturation.
The present study also had some limitations. Recall bias in FFQ dietary assessment is a possibility, particularly in children. Children's dietary habits may be partially affected by their family's food choices. We did not assess the role of micronutrient intake in pubertal development [7]. The follow-up rate was about 39%, limiting the external validity of our results. However, we observed no signi cant differences in the demographics between those who completed follow-up and those who did not [Additional le 1].

Conclusions
FA-and RRR-derived dairy-soybean DPs, characterized by high factor loadings of milk, yogurt, cheese, and soybean products, were consistently associated with increased risk of ESM in both sexes. A "vegetarian diet" in childhood signi cantly accelerated ESM in males but not in females, meriting further investigation. The onset of sexual maturation modi ed the children's dietary habits: we noted a downward trend for the "dairy-soybean diet" pattern and the "vegetarian diet" pattern and an upward trend for the "traditional diet" pattern, implying that healthy dietary habits tend to decrease from childhood to adolescence. Therefore, policies should encourage healthy dietary habits, which include consumption of fruits, vegetables, and milk alternatives, during sexual maturation and until adulthood. Abbreviations BMI, body mass index; CI, con dence interval; DMET, daily metabolic equivalent of task; DP, dietary pattern; ESM, early sexual maturation; FA, factor analysis; FFQ, food frequency questionnaire; IGF-1, insulin-like growth factor 1; OR, odds ratio; RRR, reduced-rank regression; TCHS, Taiwan Children Health Study; TDCS, Tanner-derived composite stage Declarations Availability of data and materials The dataset used and/or analyzed in this study are available from the corresponding author upon reasonable request.
Authors' contribution YCC, designed research; JWH, and CY conducted research; NTKN, HYF, and SYH analyzed data; NTKN, and YCC wrote the paper; YCC had primary responsibility for nal content. All authors read and approved the nal manuscript.
Ethics approval and consent to participate Informed consent was obtained from both the parents and the children prior to the study. The present study followed the principles of the Declaration of Helsinki and was approved by the Institutional Review Board of National Taiwan University Hospital.

Consent for publication
Not applicable. The number of participants did not add up to the total number because of missing data.
*P < 0.05, indicates differences in characteristics between children at age 11 and those followed-up at age 17.
# Indicates signi cant difference in body fat percentage between females and males at baseline.
: DMET, daily metabolic equivalent of task; ESM, early sexual maturation; TCHS, Taiwan Children Health Study; USD, United States Dollar.  1 Greater consumption of dairy and soybean products. 2 Greater consumption of staple foods, eggs, and meat and sh. 3 Greater consumption of vegetables, fruits, and fruit juices.
Food groups with a factor loading greater than 0.40 (in boldface) were included in the DP.
De nition of abbreviations: DP, dietary pattern; FA, factor analysis; RRR, reduced-rank regression; TCHS, Taiwan Children Health Study. 0.02* Adjusted models were adjusted for parent education, family income, body fat percentage, total energy intake, and screen time.
De nition of abbreviations: CI, con dence interval; DP, dietary pattern; ESM, early sexual maturation; FA, factor analysis; OR, odds ratio; RRR, reduced-rank regression; TCHS, Taiwan Children Health Study. Data are expressed as mean ± standard deviation.
DP score change was calculated as follows: DP score at age 17 − DP score at age 11 * P < 0.05.