Approaches to Identify Factors Associated with Pubertal Timing in Self-Limited Delayed Puberty

Abstract

Introduction:

Children with self-limited delayed puberty (constitutional delay) enter puberty after variable waiting times, and the factors associated with their eventual pubertal timing are not well understood.

Methods:

We conducted a retrospective study of 99 girls and 228 boys with self-limited delayed puberty (DP) at an academic medical center between 2000 and 2015. To define features and potential subtypes of self-limited DP, we performed group-based trajectory modeling on childhood growth and latent-variable factor analysis on clinical characteristics. We then conducted time-to-event analyses to identify associations with pubertal timing.

Results:

We identified two distinct growth trajectories in individuals with self-limited DP: one with stable and the other with declining height percentiles. Latent-variable factor analysis identified five factors underlying clinical variation that appear to correspond to genetic height potential, body mass index, childhood growth, parental pubertal delay, and medical issues (attention-deficit/hyperactivity disorder and inhaled glucocorticoid use). We observed correlations between pubertal timing and bone age (p=0.01), childhood height (p=0.004), and midparental target height (p<0.001), but not with parental pubertal delay or with testosterone treatment in boys.

Conclusions:

By illustrating the heterogeneity within self-limited DP and identifying factors underlying this heterogeneity, our study suggests that there may be multiple causes of self-limited DP. However, our ability to determine when puberty will eventually occur remains limited. Dissecting self-limited DP into its component subtypes may inform future studies of the mechanisms contributing to pubertal delay as well as studies of the short- and long-term outcomes of self-limited DP.

INTRODUCTION

Delayed puberty is a common concern in pediatrics, affecting 2–3% of all adolescents [1–3]. The most common cause of delayed puberty is constitutional delay, a self-limited condition in which puberty eventually begins and progresses to achievement of adult reproductive endocrine function [4]. The clinical management of self-limited delayed puberty (DP) is hindered by uncertainty around when puberty will eventually start [4]. One management option is a “watch and wait” approach with reassurance and no medical intervention. Alternatively, sex-steroid treatment can safely induce development of secondary sex characteristics in both girls and boys, and it has been suggested that treatment may even accelerate the onset of puberty [1]. In some boys, testicular enlargement is observed during or just after testosterone therapy, indicating the action of endogenous gonadotropins and raising the possibility that sex-steroid treatment may “jump-start” puberty [5, 6]. Identifying the clinical features associated with pubertal timing in self-limited DP has the potential to inform our understanding of this condition and to guide clinical management and counseling [7].

One factor that may correlate with the timing of eventual pubertal onset is the pattern of childhood growth. Self-limited DP is often associated with a “constitutional delay of growth” pattern in childhood, before the pubertal growth spurt is a factor [8, 9], with childhood growth typically occurring along a percentile below that expected for midparental height [10] and in some but not all cases, with a decline in height percentiles over time [8, 9]. Skeletal maturation is often proportionally slowed, such that predicted adult height based on bone age is generally concordant with the midparental target height.

To understand the natural history of self-limited DP, we have leveraged longitudinal data from a large retrospective cohort at a single referral center. We sought to characterize the key factors underlying self-limited DP by using data-driven methods to analyze clinical features including anthropometric measures, bone-age X-rays to assess skeletal maturation, medical diagnoses potentially associated with delayed puberty (ADHD, use of inhaled glucocorticoids, refs. [11, 12], family history, and sex-steroid treatment). We then tested for associations between these features and pubertal timing.

MATERIALS AND METHODS

To characterize clinical features of individuals with self-limited DP and to associate these features with pubertal timing, we took three approaches: 1) group-based trajectory modeling to identify distinct patterns of growth in childhood, 2) latent-variable factor analysis to identify factors underlying clinical characteristics, family history, and conditions and treatments that have been associated with delayed puberty, and 3) time-to-event analyses to test whether testosterone treatment accelerates (“jump starts”) the onset of endogenous puberty in boys. The Boston Children’s Hospital (BCH) Institutional Review Board approved this study under protocol number IRB-P00017406 and waived informed consent.

Study participants

Individuals with a diagnosis of self-limited DP were identified by review of medical records as previously described [4]. Briefly, self-limited DP was defined as 1) absence of breast development in girls 12 years or older [13–16] and 2) testicular volume <4 mL in boys 13.5 years or older [17–19]. Individuals found to have idiopathic hypogonadotropic hypogonadism (IHH) were excluded [20]. Additional criteria for inclusion in specific analyses are described below.

Data collection and general statistical methods

Medical history, parents’ heights, parental pubertal history, medications (including sex-steroid use), height, weight, pubertal status, bone age, and treating provider were electronically imported or manually entered from the BCH clinical data repository into a Research Electronic Data Capture (REDCap) database [21]. Methods for quality control, data cleaning, and data processing and transformation are described in detail in Supplementary Materials. Briefly, height and BMI Z-score slopes were estimated using individual-level linear regression using available height measurements from age 4 to 10 years. To compare height Z-score across individuals, height and BMI Z-scores at age 7 years were interpolated/extrapolated.

We further collected data on attention-deficit/hyperactivity disorder (ADHD) and inhaled glucocorticoid use [22, 23]. Individuals with a diagnosis of ADHD and/or use of stimulant medications listed in the medical record were classified as having ADHD, and all others were classified as not having ADHD. A parental history of pubertal delay was defined by a reported history of delayed pubertal timing in either parent and/or maternal history of menarche ≥14 years.

Descriptive statistics were used to summarize patient characteristics, with mean and standard deviation for continuous variables and frequency and percentage for categorical variables. Paired t-tests were used to assess the difference between individuals’ height Z-scores and midparental height Z-scores. All statistical analysis were conducted using SAS software (version 9.4, Cary, NC). P-values <0.05 were considered significant.

Group-based trajectory modeling

To identify distinct trajectories of height in childhood, we applied group-based trajectory modeling using the PROC TRAJ procedure to those with at least three documented height measurements spanning at least one year from age 4 to 10 years. To account for variation in genetic height potential, we performed linear regression of height Z-scores at age 7 years on midparental height Z-scores and used the residuals from this regression in a second analysis using group-based trajectory modeling.

To compare demographic and clinical characteristics between subgroups identified by the analysis, Student’s independent-sample t-test was used for continuous outcomes and Fisher’s exact test for dichotomous outcomes. We used the PROC ICPHREG procedure to conduct interval-censored and informative right-censored time-to-event analysis on age at pubertal onset (see ref. 4 for the details of censoring). Proportional-hazard regression models were used to compare hazard ratios for pubertal onset between different childhood growth trajectories, with a higher hazard ratio indicating earlier pubertal onset. Variables with p <0.05 in univariate analysis were included in multivariate models.

Latent-variable factor analysis

To identify factors influencing clinical characteristics, family history, and conditions and treatments associated with delayed puberty, we applied latent-variable factor analysis using the procedure PROC FACTOR with the varimax orthogonal rotation procedure (for simpler structure and interpretability). This method aggregates variables based on the degree of correlation between these variables. The number of factors retained was determined by eigenvalues >1, inspection of Scree plots, and interpretability. In addition, Pearson correlation was used to assess correlations between the input variables used for latent-variable factor analysis.

Analyses of the effect of testosterone treatment on timing of endogenous puberty

To examine whether testosterone treatment accelerates pubertal onset, we analyzed time to pubertal onset as described above, defining the groups for comparison in three ways. First, we used the full cohort and compared all boys who had been treated with testosterone (treated boys) to all those who had not (untreated boys).

Second, we compared all treated boys to a subcohort of untreated boys matched for age at treatment, i.e., who were prepubertal at the age at which the treated boys started treatment. Details of the process of selecting the matching untreated subcohort are given in Supplementary Materials.

Third, we compared boys who were seen by providers grouped by their practices around testosterone treatment as a “fortuitous experiment” of earlier vs. later testosterone treatment. We reasoned that provider assignment was essentially random with respect to baseline confounders and could be used as an instrument for causal inference. Details for assigning treating providers to the “early-treating” and “late-treating” groups are described in Supplementary Materials. We then used nonparametric survival analysis (PROC ICLIFETEST) to compare the timing of pubertal onset between boys managed by “early-treating” vs. “late-treating” providers.

RESULTS

Patient Characteristics

Reviewing medical records at BCH identified 99 girls and 228 boys with self-limited DP (Table 1). This cohort was shorter and had slower growth and slower skeletal maturation than reference populations from the CDC and the Brush Foundation [24, 25]. Compared to these reference populations, this cohort had lower height Z-score at age 7 years (mean ± SD −1.20 ± 0.89) and bone age Z-score (−2.19 ± 1.10; p both <0.0001 for difference from zero; Table 1, Supplementary Fig. 1). There was also a decline in height Z-score across childhood (height Z-score slope −0.09± 0.12 per year, significantly different from zero, p <0.0001). The lower height Z-scores at 7 years are partially attributable to lower-than-average midparental height Z-scores for the cohort (mean ± SD −0.11 ± 0.74), but this does not entirely account for the difference between our cohort and the CDC reference population, as the individuals’ height Z-scores at 7 years were lower than their midparental height Z-scores (mean ± SD for difference: −1.07 ± 0.82; p <0.001).

Table 1.

Characteristics of the overall cohort and the subcohorts used in each analysis

	Overall (N=327)		Group-based trajectory modeling (N=150)		Factor Analysis (N=81)		Testosterone Treatment Analysis (N=108)
	Mean ± SD or N (% of total)	N	Mean ± SD or N (% of total)	N	Mean ± SD or N (% of total)	N	Mean ± SD or N (% of total)	N
Sex
Female	99 (30.3%)	327	50 (33.3%)	150	40 (53.2%)	81	108 (100%)	108
Male	228 (69.7%)	327	100 (66.7%)	150	41 (56.8%)	81	0 (0%)	108
ADHD	147 (45.0%)	327	66 (44.0%)	150	45 (55.6%)	81	56 (51.9%)	108
Use of inhaled glucocorticoids	39 (11.9%)	327	24 (16.0%)	150	12 (14.8%)	81	10 (9.3%)	108
Parental pubertal delay	112 (34.3%)	327	62 (41.3%)	150	40 (43.0%)	81	36 (33.3%)	108
Height Z-score at 7 years	−1.20 ± 0.89	129	−1.21 ± 0.85	111	−1.24 ± 0.76	81	−1.03 ± 0.95	41
Height Z-score slope (change per year)	−0.09 ± 0.12	129	−0.08 ± 0.12	111	−0.10 ± 0.11	81	−0.07 ± 0.11	41
BMI Z-score at 7 years	−0.15 ± 0.96	128	−0.19 ± 0.96	110	−0.32 ± 0.96	81	0.04±1.18	41
BMI Z-score slope (change per year)	−0.10 ± 0.21	128	−0.10 ± 0.20	110	−0.11 ± 0.19	81	−0.11 ± 0.25	41
Midparental height Z-score	−0.11 ± 0.74	200	−0.14 ± 0.73	149	−0.17 ± 0.76	81	−0.04 ± 0.68	61
Average bone-age Z-score	−2.19 ± 1.10	242	−2.09 ± 1.09	121	−2.25 ± 0.98	81	−1.97 ± 0.98	73

BMI, body mass index; ADHD, attention-deficit/hyperactivity disorder

Girls’ height Z-scores declined more rapidly than boys’ (slope −0.14 ± 0.12 vs. −0.06 ± 0.10 per year), and girls also had lower bone-age Z-scores (−2.7 ± 1.1 vs. −1.9 ± 1.0, p <0.001 for both). A high proportion of our cohort had ADHD (147 out of 327, 45%), a prevalence higher than that of the general US population (6.1 to 16.3%)[26], and 12% (39 out of 327) had a recorded history of inhaled glucocorticoid use. There were no significant differences in clinical characteristics between the overall cohort and the subcohorts used in each of our subsequent analyses (Table 1).

Identifying Distinct Growth Trajectories in Self-limited Delayed Puberty

To identify potential subgroups within our cohort based on prepubertal growth trajectories, we performed group-based trajectory modeling of height Z-scores from age 4 to 10 years and identified three trajectories that appeared to represent differences in midparental height (Supplementary Fig. 2). We repeated the analysis after adjustment for midparental height and found that a model with two trajectories was the best fit (Supplementary Table 1). In the first trajectory, termed “stable height percentiles” (N=121/150, 81%), height Z-scores remained stable over time, with an average change in the height Z-score from age 4 to 10 years of −0.04 ± 0.08 per year (Fig. 1, representative growth charts for each trajectory in Fig. 2). In contrast, the second “declining height percentiles” trajectory (N=29/150, 19%) was characterized by height Z-scores that declined with age at −0.23 ± 0.07 per year (p<0.001 for difference from “stable height percentiles” group; Supplementary Table 2). The declining height percentiles were unlikely to be due to growth hormone deficiency, as no individual in this cohort had IGF-1 below the reference range for prepubertal children (data not shown). There were disproportionately more girls in the “declining height percentiles” group than in the “stable height percentiles” group, and bone age Z-scores were lower (i.e., bone ages were more delayed) for the “declining height percentiles” group (p both <0.001; Supplementary Table 2).

Fig. 1. Patterns of childhood growth from group-based trajectory modeling of height.

Group-based trajectory modeling of height Z-score residuals after adjusting for midparental height showed two distinct trajectories, with participants in Group 1 exhibiting a minimal decline in height Z-score residuals and participants in Group 2 exhibiting a more substantial decline in height Z-score residuals over time. Dashed lines indicate standard errors.

Fig. 2. Representative growth charts by sex and growth pattern.

Growth charts of a girl (A) and a boy (B) with stable height percentiles across childhood and of a girl (C) and a boy (D) with declining height percentiles across childhood. Arrows indicate midparental height (MPH).

Clinical Features of Individuals with Self-limited Delayed Puberty

To further identify associations between clinical features within our cohort, we analyzed family history (midparental height, parental pubertal delay), medical history (ADHD, inhaled glucocorticoid use), anthropometric measurements (height and BMI Z-scores at age 7 y, height and BMI Z-score slopes from 4 to 10 y), and bone age Z-score. We assessed associations between these variables with pairwise correlation analyses and identified an expected correlation between midparental height and height Z-score at 7 years (r = 0.47, p <0.001; Supplementary Table 3). We also found correlations between height Z-score at 7 years and BMI Z-score at 7 years (r = 0.24, p = 0.006), between mean bone age Z-score and both height Z-score slope (r = 0.29, p = 0.003) and BMI Z-score at 7 years (r = 0.30, p = 0.002), and between BMI Z-score at 7 years and BMI Z-score slope (r = −0.21, p = 0.01).

We then conducted latent-variable factor analysis to identify factors underlying variation in these clinical variables and identified five factors (Table 2). Factor 1, which we term “genetic height potential,” had the largest contributions from height Z-score at 7 years and midparental height Z-score. Factor 2, termed “BMI,” had the largest contributions from BMI Z-score at 7 years and BMI Z-score slope from age 4 to 10 years. Factor 3, termed “childhood growth tempo,” had the largest contributions from height Z-score slope and bone-age Z-score. Factor 4, termed “parental pubertal delay,” had the largest contribution from presence/absence of a history of delayed puberty in at least one parent. Factor 5, termed “medical issues,” had the largest contributions from the presence/absence of ADHD and the use of inhaled glucocorticoids.

Table 2.

Factors and corresponding contributory variables identified by latent-variable factor analysis^*

	Factor 1 “Genetic height potential”	Factor 2 “Body mass index”	Factor 3 “Childhood growth”	Factor 4 “Parental pubertal delay”	Factor 5 “Medical issues”
Height Z-score at 7 years	83	−4	3	−14	−7
Height Z-score slope (change per year)	4	8	77	46	7
BMI Z-score at 7 years	20	−76	14	−24	−7
BMI Z-score slope (change per year)	17	80	1	−11	−9
Midparental height Z-score	86	4	−1	5	5
Bone Age Z-score mean	0	−20	82	9	−11
Parental pubertal delay	−9	5	0	84	0
ADHD	−28	22	15	−32	59
Use of inhaled glucocorticoids	12	−15	−13	15	83

BMI, body mass index; ADHD, attention-deficit/hyperactivity disorder

^*Correlation coefficients are multiplied by 100 and rounded to the nearest integer. Coefficients with absolute values greater than 50 are in boldface and italics.

Variables Associated with Timing of Pubertal Onset

Having identified different trajectories of childhood growth underlying clinical variation in our cohort, we sought to assess whether any of these trajectories and clinical variables were associated with the timing of pubertal onset. After adjusting for sex, the only trajectories and variables associated with later pubertal onset were lower bone-age Z-score (i.e., more delayed bone age), greater height Z-score at 7 years, and greater midparental height Z-score (p = 0.01, p = 0.004, and p <0.001, respectively; Table 3). In the multivariate model, only the association between higher midparental height Z-score and later pubertal onset remained significant (p = 0.03).

Table 3.

Associations of time to pubertal onset with growth trajectories and clinical characteristics

	Adjusted for Sex			Multivariate (N=93)
Variables	Hazard Ratio*	95% CI	P	Hazard Ratio*	95% CI	P
Sex (female vs. male)	7.20	4.28, 12.10	<0.001	11.28	5.16, 24.64	<0.001
Bone Age Z-score mean	1.28	1.05, 1.56	0.01	1.14	0.88,1.48	0.320
Use of inhaled glucocorticoids	0.85	0.50, 1.46	0.56
BMI Z-score at 7 years	1.18	0.97, 1.42	0.10
Height Z-score at 7 years	0.68	0.52, 0.88	0.004	0.88	0.61, 1.27	0.50
ADHD	1.05	0.72,1.55	0.79
Midparental height Z-score	0.52	0.38, 0.72	<0.001	0.59	0.37,0.95	0.03
BMI Z-score slope (change per year)	0.36	0.12, 1.02	0.054	1.09	0.28,4.23	0.90
Height Z-score slope (change per year)	2.29	0.28,18.77	0.44
Parental pubertal delay	0.97	0.66, 1.44	0.88
Growth trajectory (stable vs. declining height percentiles”)	1.03	0.52, 2.03	0.94	0.99	0.47, 2.09	0.99

BMI, body mass index; ADHD, attention-deficit/hyperactivity disorder

^*Higher hazard ratios indicate earlier pubertal onset

Inferring effects of testosterone treatment on timing of endogenous puberty

Given the paucity of data from randomized trials on whether exogenous sex-steroid therapy accelerates (“jump starts”) puberty in self-limited DP, we used three analytic strategies to test whether testosterone treatment was associated with earlier onset of testicular enlargement in boys. (Because estradiol directly induces breast development, the indication of girls’ pubertal onset used in this study, we were unable to conduct a parallel analysis in girls.)

We first simply compared the age at pubertal onset between boys treated and not treated with testosterone, and we found that pubertal onset occurred later in boys treated with testosterone (Fig. 3A). However, this analysis was likely confounded by indication because the decision to treat with testosterone was almost certainly influenced by age; that is, an older boy who has not yet entered puberty is more likely than a younger boy to opt for treatment.

Fig. 3. Effect of testosterone treatment on time to pubertal onset in boys with self-limited delayed puberty.

Survival analysis of time to pubertal onset comparing (A) all boys treated vs. not treated with testosterone, (B) testosterone-treated vs. a subset of untreated boys matched by age, and (C) boys managed by “early-treating” vs. “late-treating” providers. In the unmatched analysis (A), pubertal onset occurred later in boys treated with testosterone, but this difference was no longer observed after age-matching (B). Similarly, no difference in time to pubertal onset was seen between those managed by “early-treating” vs. “late-treating” providers (C). Gray shading indicates standard errors.

To control for this possible confounding by age, we next compared testosterone-treated boys to untreated boys matched based on prepubertal status at the index age at which treatment was initiated. In other words, at the age at which the treated boy started testosterone treatment, the matching untreated boy was prepubertal and therefore also had the potential option of starting testosterone yet was not treated. No difference in the timing of pubertal onset between treated and matched untreated boys was revealed by this analysis (Fig. 3B). Because the decision to treat may have been influenced by additional information that led to an expectation that puberty would start later, such as a family history of delayed puberty or a delay in bone age, we compared the presence of parental history of delayed puberty and mean bone age Z-scores between the two groups and found no significant differences between the groups (p both > 0.6); thus, confounding by severity was unlikely to have led the analysis to underestimate the effect of testosterone.

Our third approach addressed potential confounding bias by identifying a “fortuitous experiment” within our retrospective cohort. First, we noted that treatment practices vary among clinical providers, allowing us to classify providers as tending to treat with testosterone at earlier ages (“early-treating”) and tending to wait until later ages (“late-treating”; Supplementary Fig. 3). We then compared timing of pubertal onset between boys seen by “early-treating” vs. “late-treating” providers and observed no differences in the timing of pubertal onset (Fig. 3C). The lack of an effect of provider group on pubertal onset suggests that there is no average effect of testosterone therapy on pubertal onset.

DISCUSSION

In this study of girls and boys with self-limited DP (constitutional pubertal delay), we used data-driven approaches to identify factors that underlie the clinical heterogeneity within the cohort and sought to identify associations with pubertal timing. We found that most individuals maintained a consistent height percentile across childhood, but a subset had height Z-scores that declined across childhood. Latent-variable factor analysis identified five factors – “genetic height potential,” “BMI,” “childhood growth,” “parental pubertal delay,” and “medical issues” – as the primary contributors to variation in our cohort. We observed correlations between pubertal onset and bone age, childhood height, and midparental height, but not parental pubertal delay or, in boys, testosterone treatment. These findings highlight the wide variability within self-limited DP and identify key factors that contribute to this variability.

Self-limited DP is often associated with delayed growth and delayed bone age [27, 28]. In children presenting with delayed puberty, potential contributions to short stature (at least transiently) include: 1) genetics, reflected by lower midparental height, 2) a “constitutional growth delay” pattern, with lower height relative to midparental height, typically with a proportional delay in skeletal maturation (bone age), 3) declining growth percentiles during childhood, and 4) lack of a normally timed pubertal growth spurt [8, 9, 22]. Wehkalampi and colleagues previously showed that a decline in growth percentiles can start in early childhood, between the ages 3 and 8 years, before the expected time of the pubertal growth spurt [8, 9]. In our cohort, although childhood growth and bone age Z-scores were delayed overall, we observed considerable variability, such that not all individuals have heights shorter than expected for their genetic potential or delayed bone ages. These observations provide evidence that self-limited DP is a heterogeneous condition that may have multiple underlying mechanisms, and the application of quantitative methods to dissect delayed puberty into subtypes may disentangle these mechanisms.

In our latent-variable factor analysis, “medical issues,” which included the presence/absence of ADHD and use of inhaled glucocorticoids, emerged as a distinct factor. One potential and speculative explanation for the clustering of ADHD and inhaled glucocorticoid use in the same factor is that both may result in increased systemic glucocorticoid exposure. Use of ADHD medications has been shown to increase concentrations of circulating morning cortisol [29], and inhaled glucocorticoids, though intended to provide local effects without systemic absorption, may lead to systemic side effects, as potentially severe as adrenal insufficiency [30].

We also sought to identify factors associated with timing of pubertal onset and found that more delayed bone age and greater genetic height potential (greater childhood height and midparental height), were associated with later pubertal entry. While these associations raise the possibility of biological links between these factors and pubertal timing, we cannot exclude the possibility of ascertainment/referral bias [31]. For example, because patients with delayed puberty are often more concerned about height than about pubertal development per se, individuals who are taller due to greater genetic height potential may only seek specialist care if there is a greater degree of pubertal delay leading to a greater delay in growth. Furthermore, because each factor may be present to varying degrees within any one individual, this may have limited the statistical power to identify additional associations between individual factors and pubertal timing.

In addition, we tested the hypothesis that testosterone treatment accelerates (“jump starts”) the onset of puberty in boys. Sex-steroids have been shown to accelerate the onset of puberty in conditions of gonadotropin-independent sex-steroid production, such as congenital adrenal hyperplasia [32] and autonomous gonadal activity [33], and in boys with Klinefelter Syndrome who were treated with oxandrolone [34]. Prior randomized studies of testosterone treatment in boys with constitutional delay reported higher testicular volumes in boys after testosterone treatment compared to those not treated [35, 36], suggesting that testosterone treatment can accelerate the onset and/or progression of puberty. In a survey of pediatric endocrine providers, most providers believed that sex-steroid therapy can “jump start” pubertal onset in children with self-limited DP [7]. However, we found no evidence in our study that testosterone treatment results in earlier pubertal onset, although we cannot rule out the possibility that the decision to treat may have been influenced by additional information not captured by our study, such as subtle physical exam findings (as a speculative example: testicular volumes on the lower end of the prepubertal range). Furthermore, variation in the dose and duration of testosterone treatment could result in variability in responses, making it challenging to identify effects of testosterone treatment. To definitively investigate whether sex-steroid treatment “jump starts” pubertal entry in individuals with self-limited DP, a prospective randomized clinical trial of treatment versus observation is needed.

This cohort consists of patients seen in a specialty clinic at a referral center, which may limit generalizability of our findings to the general population. Specifically, individuals with milder delays in puberty and/or childhood growth may not be referred for specialty evaluation, such that individuals in our cohort may have more severe delays in pubertal timing and/or growth. In addition, our results raise the possibility of referral bias based on sex/gender. In our cohort, girls had more severe growth delay (as indicated by slower childhood growth and more delayed bone ages) compared to boys. Thus, there may be a higher threshold for girls to be referred for delayed puberty, as has been suggested by other studies [22, 37–39], such that only girls with more severely delayed growth are referred for specialist evaluation.

Additional limitations of our study include the retrospective design, which is susceptible to both selection bias, as noted above, and measurement bias. Only a subset of individuals within the cohort had all available clinical data for group-based trajectory modeling and latent-variable factor analysis; however, clinical characteristics of these and other sub-cohorts did not significantly differ from the overall cohort. Furthermore, data on ancestry, race, and/or ethnicity were not available, and we were therefore unable to analyze the effects of these variables.

By illustrating the heterogeneity within self-limited DP and identifying factors underlying this heterogeneity, our study suggests that there may be multiple causes of self-limited DP. This dissection of self-limited DP into its component subtypes can inform studies on the genetic and physiologic mechanisms contributing to self-limited DP and their adult consequences. This will ultimately allow us to predict when puberty will eventually occur, to provide timely sex-steroid treatment when indicated, and to mitigate any potential adverse consequences in this patient population.

Data Availability Statement:

The datasets generated during and/or analyzed during the current study are not publicly available due to their containing clinical information that could compromise the privacy of research participants but are available from the corresponding author [Yee-Ming Chan] upon reasonable request.