The Variability and Determinants of Testosterone Measurements in Children: A Critical Review

Abstract

Aims:

This critical review aimed to summarize: (1) the variability and determinants of testosterone (T) measurements; and (2) reference values for the variability and determinants of T measurements in children.

Background:

As T is a representative androgen, it has been widely used to explain male vulnerability to child health and developmental problems. T measurements in children, however, have been challenging because of low levels, diurnal and episodic secretion patterns, limited quantity and quality of the samples, and inconsistent study findings.

Methods:

The search strategy used PubMed, CINAHL, Cochrane Library, Embase, Scopus, and Google Scholar. Studies published between 2008 through 2020 that examined factors influencing T measurement were included. The final 30 studies were selected using two appraisal forms. We extracted five categories of data from the reports.

Findings:

Variability and determinants of T measurement included assay methods, the source of samples, and child demographic and environmental characteristics. T levels were higher 1–3 months after birth and in males up to 1 year; fewer sex differences were found up to 10–12 years. Serum T levels measured by using liquid chromatography-mass spectrometry were most reliable because immunoassays overestimated the levels, especially in neonates. T levels were stable at different temperatures and durations of storage, although sample collection remained an ongoing challenge for researchers.

Conclusion:

Depending on the study aims and feasibility, mass-spectrometry, multi-methods, and multi-materials are the recent trends in T measurement. Immunoassays may be an option if the study aims for relative rather than absolute comparisons.

The occurrence of preterm birth was 9.85%of live births in 2014–2016 (CDC, 2018; Griggs et al., 2020). Not surprisingly, child health and developmental problems are more common in preterm than full-term children, especially in males (Hintz et al., 2006). Male vulnerability for child health and development has been widely investigated since general theories of gender differences and extreme male brain theory of autism were introduced early in the 1980s and 2000s (Baron-Cohen, 2002; Geschwind, 1987). These theories hypothesized that exposure to elevated prenatal testosterone (T) levels is a biological risk factor for neurological and behavioral developmental problems. As T is a representative androgen and the levels are almost three times higher in preterm than full-term infants, especially 1–3 months after birth, T has been used to examine if this androgen is associated with health and the development of cognitive, motor, language, socioemotional, and behavioral outcomes in both preterm and full-term children (Cho et al., 2017, 2021; Whitehouse et al., 2012).

Many studies have shown male vulnerability in child health and development such that males are more likely to be born prematurely; have a higher incidence of health problems such as sepsis, intraventricular hemorrhage, and asthma; and have a higher prevalence of developmental problems such as autistic spectrum disorders and attention deficit hyperactivity disorder (Alexander & Saenz, 2011; Gissler et al., 1999; Hintz et al., 2006; Saenz & Alexander, 2013). Recently, the National Institute of Child Health and Human Development (NICHD) Neonatal Research Network team confirmed that neurodevelopmental impairment defined as Bayley Scales of Infant and Toddler Development III cognitive score less than 70, moderate/several cerebral palsy, a Gross Motor Function Classification System score of level 2 or more, vision impairment, or hearing impairment were associated with low gestational age, male sex, and medical complications at birth (Pappas et al., 2018).

However, T measurements in children have been challenging because of low levels, especially after 1–3 months until prepubescent ages of 10–12 years, diurnal and episodic secretion patterns, limited quantity and quality of the samples, natural sampling through passive drooling, and varied reference intervals related to the assay methods and the source of samples. Different T measurement methods including immunoassays (enzyme immunoassay [EIA], enzyme-linked immunosorbent assay [ELISA], and radioimmunoassay [RIA]), mass spectrometry (liquid chromatography-mass spectrometry [LC/MS], and gas chromatography-mass spectrometry [GC/MS]) have been used with different sources of samples including blood, saliva, urine, and hair. However, the reference values of T have been inconsistent not only due to assay methods and the source of samples but also related to child demographic characteristics such as age, sex, and environmental factors such as pre- and postnatal exposure to certain chemicals (Büttler et al., 2016; Huang et al., 2020). In addition to inconsistent reference values, study findings were sometimes conflicting, showing either positive or negative relationships between T levels and child health, such as autistic spectrum disorders (ASD; El-Baz et al., 2014; Kung et al., 2016), physical growth (Becker et al., 2015; Kiviranta et al., 2016) and language and motor development (Cho et al., 2017; Whitehouse et al., 2012).

Because boys are vulnerable to health and developmental problems (Pappas et al., 2018; Whitehouse et al., 2012), the exposure to elevated prenatal T levels as indicated by high T levels in amniotic fluid and cord blood, long anogenital distances, and low ratios of the lengths between the second and fourth digits are considered to be possible biological risk factors (Cho et al., 2012; Kallak et al., 2017; Knickmeyer et al., 2011; Saenz &Alexander, 2013). However, the variability and determinants of T measurements in children have not been critically reviewed, especially when considering birth outcomes such as birthweight and gestational age.

Several factors impact T variability in men and women, including acute and chronic disease, age, assay techniques, diurnal variation, ethnicity, genetics, geography, lifestyle factors, seasonal factors, circadian rhythms, intra-individual daily variability, transient stressors, and the phase of the menstrual cycle (Kanakis et al., 2019; Trost & Mulhall, 2016). To examine the variability of T measurements specifically in children, we focused on assay methods and materials as well as child sex and age because these factors are more relevant than other factors involved in adults. This critical review could assist readers and researchers in interpreting research outcomes related to T measurements in children. Therefore, the purpose of this review was to summarize: (1) the variability in T measurements including assay methods, the source of samples, and child sex and age and the determinants of T measurements including sensitivity, recovery, accuracy, and precision; and (2) reference values of T related to the variability and determinants of T measurements in children.

Method

Literature Search Strategy

Google Scholar and the database Academic Search Premier were used initially to identify criteria to narrow the literature review. Terms that were used in this preliminary search included precision, accuracy, T measurement, and very low birth weight infants. Then we searched the electronic databases, PubMed, CINAHL, Cochrane Library, Embase, and Scopus, to identify research articles between 2008 and 2020 for updated information about the variability and determinants of T measurement. The key terms used to conduct the search were T measurement, infant, child, serum, blood, urine, saliva, salivary, skin, and hair. These terms were put in the search builder as “(infant* OR child*) AND (T measurement) AND (serum OR blood).” The material for T measurement was replaced with saliva, salivary, urine, skin, and hair until all articles were retrieved. Inclusion criteria for publications were that they (1) be written in English within 2008−2020, (2) had a sample human population aged between 0 and 12 years old, and (3) included research on T measurement from blood, saliva, urine, skin, or hair sources with a method of immunoassays including EIA, ELISA, and RIA and a method of mass spectrometry including LC/MS and GC/MS. Publications were excluded from the review if they (1) were not written in English, (2) were case reports or review articles, (3) did not include the population of interest, or (4) were not relevant to T measurement.

Selection of Studies

The initial search led to the retrieval of 605 articles. After the removal of 125 duplicate articles, there were 480 articles. The titles and abstracts were reviewed by the authors. We excluded 162 articles because the participants were over the age limit of 12 years, another 196 articles due to their research topics not involving T measurement, six for being case studies, two for not having assay materials such as serum, plasma, saliva, urine, and hair, and the other 43 articles because they were case studies or had no full text. The full-text of the resulting 71 articles was obtained and read thoroughly by the authors. After reading the complete text of the articles, we removed 21 articles because the participants were over the age limit of 12 years, which had not been presented in the abstract. We removed another 22 articles because they did not relate to T collection. We added two articles with RIA and one article with LC/MS that were found through manual searches of the reference lists and PubMed so we would have the same number of articles using each analysis method. Using the same search terms, we retrieved dissertations and other sources on ProQuest and found 196 possible sources of literature. We did not add any articles from this second literature search because 135 articles did not meet the criteria for participants, 49 articles did not contain T measurements, and 12 articles were not able to be retrieved. As shown in Figure 1, the final count of 30 articles was included in this critical review.

Figure 1.

PRISMA flow diagram.

Data Extraction and Analysis

We used the critical review form for quantitative studies (Law et al., 1998) and the Critical Appraisal Skills Program (CASP, 2018) guidelines for a cohort study for appraising the studies. As shown in Table 1, study purpose, relevant background literature review, design, sample and setting, and results were reviewed. Major variables and measures including T measurement with assay methods and materials were extracted first. Then, the factors that affected variability and determinants of T measurements were identified and summarized followed by the study findings.

Table 1.

Summary Table for the Selected Studies.

Authors (Year)	Purpose	Design	Sample	Material(s)	Method(s)	Main Findings
Ahmed et al. (2010)	To determine a relationship between serum AMH and T	Longitudinal	26 boys only (on days 1, 4, & 22)	Serum	EIA	A significant relationship between AMH and T at day 4 and 22
Ankarberg-Lindgren et al. (2015)	To compare between RIA and MS	Longitudinal	80 girls and 77 boys (at different pubertal stages)	Serum	RIA & MS	T levels were similar between RIA and MS at Tanner stage B1 < 8 years: RIA = 0.09 & MS = 0.25.
Avidime et al. (2011)	To examine if T in cord blood is related to anogenital distance (AGD)	Cross-sectional	200 newborns	Umbilical cord blood	EIA	No correlation between T and AGD. No different T levels between sexes.
Becker et al. (2015)	To investigate the effects of the postnatal hormonal surgeon somatic and adipose tissue development	Longitudinal	35 healthy full-term infants	Blood at 4, 8, & 20 weeks	RIA	High T levels at 4 and 8 weeks, but not 5 months, were inversely associated with body weight and BMI up to 2–6 years.
Büttler et al. (2016)	To establish reference values for salivary T during puberty.	Longitudinal	123 boys and 131 girls (8–12 years old)	Saliva	ID-LC-S/MS	T levels were higher in girls between 8 and 10 years
Cho et al. (2017)	To determine associations between T and infant development	Longitudinal	62 very low birthweight infants (at 40 weeks PMA, and 3 and 6 months old)	Saliva	EIA	T levels were positively associated with the language development of boys.
Contreras et al. (2017)	To clarify the biological significance of minipuberty	Longitudinal	30 infant boys (1–4 months old) and 12 adolescent boys (11–17 years old).	Serum & saliva	LC-MS/MS	The low free T explains the absence of virilization. The significance of the minipuberty of infancy is still unclear
de Jong et al. (2012)	To describe the postnatal activation of the HPG axis	Longitudinal	21 VLBW male infants (day 1 to 6 months old)	Urine	ELISA	T levels decreased with increasing postnatal age
Dhayat et al. (2015)	To provide reference values for urinary steroid metabolome	Longitudinal	43 full-term infants	Urine	GC/MS	Urinary T levels were highest in the first 2 months with a peak around 3 weeks. Then, the levels decreased to 1/5 by week 25 in both sexes.
El-Baz et al. (2014)	To compare T levels and prevalence of autistic spectrum disorders (ASD) and its severity	Longitudinal	30 boys aged (6–15 years old)	Serum	ELISA	Hyperandrogenemia was prevalent in autistic boys and increased in autistic severity.
Fang et al. (2017)	To assess the possible hormonal effects of different feeding regimens	Longitudinal	166 full-term infants (birth to 12 months old)	urine, salivary, and serum	HPLC-MS/MS & RIC	Urine, saliva, and serum T levels showed strong correlations among the three milk groups.
Garagorri et al. (2008)	To obtain longitudinal reference plasma levels for 6 hormones	Longitudinal	138 healthy infants (birth to 6 months)	Umbilical cord blood	RIA	T levels were higher in boys and decreased from birth to 6 months of age
Hamer et al. (2018)	To determine if immunoassays are accurate to measure T levels	Longitudinal	78 infants (birth to 6 months old)	Plasma	Immunoassays & LC/MS	LC/MS was recommended to measure T levels in neonates because immunoassays overestimated levels
Huang et al. (2020)	To examine how maternal urinary PAH metabolites are related to hormone levels and birth outcomes	Cross-sectional	163 healthy pregnant women and 163 newborns (within 72 hours after birth).	Urine & cord blood	ELISA ELISA	T levels were lower in PAHs exposed newborns. Birth length was shorter and head circumference was lower in PAHs exposed newborns.
Jain et al. (2018)	To examine if T is related to AGD	Cross-sectional	117 neonates (1 day old)	Umbilical cord blood	LC/MS	No significant relationship between T and AGD for both sexes
Kareem et al. (2020)	To determine the relationship between total serum T and AGD	Cross-sectional	240 healthy full-term neonates	Serum	ELISA	T levels and AGD were two times higher and longer in male neonates. No sex difference in birth weight, length, and head circumference.
Kim et al. (2020)	To examine androgen levels during early childhood	Longitudinal	114 boys and 86 girls from 2, 4, and 6 years old	Serum	LC-MS/MS	No sex differences in T levels at 2, 4, and 6 years
Kiviranta et al. (2016)	To determine if T plays a role in the activation of linear growth in infants.	Longitudinal	84 healthy full-term infants (1–6 months old)	Urine	LC/MS	T levels were higher in boys. A positive correlation between T and growth velocity during the first 5 mo
Knickmeye et al. (2011)	To determine the relationship between 2D:4D and T levels.	Longitudinal	364 children (0–2 years old)	Saliva	EIA	The 2D:4D may not function well as a proxy measure of prenatal T exposure in infancy.
Kyriakopoulou et al. (2013)	To develop an accurate assay and establish the reference intervals for T	Longitudinal	Healthy 337 children (0–18 years old)	Serum	LC-MS/MS	T levels were stratified by child sex and age. No sex difference between 1 and 13 years.
Kung et al. (2016)	To examine relationships between T levels and autistic traits	Longitudinal	87 full-term infants (1–30 months of age)	Saliva	EIA	T levels were higher in boys. No correlations between T levels and autistic traits both in boys and girls.
Kushnir et al. (2010)	To develop an LCMS assay to simultaneously measure three hormones	Cross- sectional	2517 healthy children (6 months to 17 years of age)	Serum	LC-MS/MS & immunoassays	T levels were similar between 6 months and 7 years. Immunoassays overestimated T levels.
Minatoya et al. (2017)	To examine the association between BPA and T levels	Cross-sectional	285 full-term healthy newborns	Umbilical cord blood	LC/MS	No correlation between cord blood BPA and T in boys
Mouritsen et al. (2014)	To evaluate T levels during the pubertal transition	Longitudinal, every 6 mo for 5 yrs.	20 children (ages 9–11 years old)	Serum	LC-MS/MS &RIA	T levels increased in girls at 10.5 years and boys at 11.5 years. LC/MS was more quantifiable than RIA
Olisov et al. (2019)	To develop a method for the simultaneous measurement of five adrenal hormones	Cross-sectional	59 healthy children (ages 4–6 years old)	Leftover samples	HPLC-MS/MS & immunoassays	The results were comparable to immunoassays. T levels were lower in LC/MS than immunoassays.
Salameh et al. (2010)	To develop an online mass spectrometry method that will allow an accurate reading of low T levels	Cross-sectional	499 healthy volunteers (8–90 years old)	Serum	HTLC-MS/ MS & RIA	T levels in HTLC-MS/MS were storing correlated with RIA (r = 0.945). T levels were higher in girls between 1–5.9 yrs.
Shen et al. (2009)	To develop a method to measure five hormones in human hair	Cross-sectional	11 children (2–17 years old)	Head hair	GC/MS/MS	Determinants (sensitivity, accuracy, and precision) were within the required limits.
Smith et al. (2019)	To determine the relationship between T and DHEA across hair segments.	Cross-sectional	128 children (7–9 years old)	Head hair	ELISA	T was correlated with DHEA in the 0–3 cm (boys and girls) and 3–6 cm (girls) hair segments. T levels differed by sex in the 0–3 cm but not 4–6 cm.
Wen et al. (2017)	To determine if maternal urinary phthalate metabolite affects child T levels	Longitudinal	364 children (at 2–3, 5–6, 8–9, and 11–12 years of age)	Serum	RIA	Early exposure to maternal phthalate metabolite deceased child’s T levels over time
Whitehouse et al. (2012)	To determine if there is an association between prenatal testosterone and language delay	Longitudinal	767 full-term and preterm infants	Umbilical cord blood	LC/M	High prenatal Bio T levels were a risk factor for language delay in boys, but a protective factor in girls between 1–3 years.

Note. AMH = anti-müllerian hormone, HPG = hypothalamic-pituitary gonadal, PAH = polycyclic aromatic hydrocarbons, BPA = Bisphenol A.

Results

Characteristics of Studies

We reviewed 30 articles based on three categories: (1) the methods of T measurement, (2) the source of the samples, and (3) reference values of T measurement. All studies were multi-authored and were from 17 countries including the US (eight), Netherland (three), United Kingdom (two), Korea (two), China (two), Nigeria (two), Canada (one), Finland (one), Japan (one), Sweden (one), Egypt (one), Denmark (one), Australia (one), Taiwan (one), Switzerland (one), Germany (one), and Spain (one). Most of the studies were conducted in clinical settings with a sample size ranging from 11 (Shen et al., 2009) to 2,517 (Kushnir et al., 2010). The ages of the study participants ranged from 0 (birth) to 90 years old, but only data from participants who were 12 years or younger were included in this review. Nineteen studies included infants (0–2 months) who were either full-term (17) or very-low-birthweight (VLBW, birthweight < 1,500 gm) preterm (2). Four of 30 studies included only boys while 26 studies included both sexes. Of the 30 studies, 17 (56.7%) used a cross-sectional research design and 13 (43.3%) used a longitudinal research design with the study period ranging from short (22 days, Ahmed et al., 2010) to long (10 years, Wen et al., 2017). All studies had focused research methods and data analyses followed by the problem statement and aims of the study. All studies presented results and recommendations for future research and clinical practice implications.

Problem Statements and Aims of Selected Studies

The most common problem statement was low T levels in children until puberty. The possible factors affecting T levels were child age and sex, assay methods (e.g., immunoassays and LC/MS) and materials (e.g., blood and saliva), sample-type acceptability (e.g., serum and plasma), sampling time (e.g., morning and evening), sampling method (e.g., passive drooling and use of collecting devices), handling of samples (e.g., centrifugation and transfer supernatants), sample stability (e.g., sample storage at room temperature, in a refrigerator, or in a −20 °C freezer for 1, 3, 7, 14, 21, and 28 days), and storage stability (e.g., sample storage in 70 °C freezer for shorter or longer than 1 year (Kushnir et al., 2010).

We divided the aims of the studies into five categories. First, two studies examined the associations between T and child health including autistic spectrum disorder (El-Baz et al., 2014; Kung et al., 2016), whereas nine studies examined the associations of T levels with child development including physical growth (bodyweight, length, body mass index, skinfold triceps) (Becker et al., 2015; Kiviranta et al., 2016), motor and language development (Cho et al., 2017; Whitehouse et al., 2012), pubertal transition (Mouritsen et al., 2014), anogenital distance in neonates (Avidime et al., 2011; Jain et al., 2018; Kareem et al., 2020), and postnatal activation of HPG axis after insulin injection (de Jong et al., 2012). The second category included the five studies that aimed to identify T measurements with greater accuracy and precision by comparing methods and materials (Ankarberg-Lindgren & Norjavaara, 2015; Contreras et al., 2017; Fang et al., 2017; Hamer et al., 2018; Knickmeyer et al., 2011). In the third category, four studies developed a method that was capable of the simultaneous measurement of multiple hormones (Kushnir et al., 2010; Olisov et al., 2019; Salameh et al., 2010; Shen et al., 2009). The five studies in the fourth category provided reference values that were clustered based on child age and sex in relation to assay methods (EIA/ELISA, LC/MS, GC/MS, RIA, RIC) and sampled materials (serum, saliva, urine, hair) (Büttler et al., 2016; Dhayat et al., 2015; Garagorri et al., 2008; Kim et al., 2020; Kyriakopoulou et al., 2013). Finally, five studies examined if T was related to other steroid hormones and factors (Ahmed et al., 2010; Huang et al., 2020;Minatoya et al., 2017; Smith et al., 2019; Wen et al., 2017).

Study Findings Based on Study Aims

Associations between T and child health and development

T levels were associated with child health. For example, T levels and ASD and its severity were examined and showed different outcomes. El-Baz et al. (2014) reported that free T levels in serum were significantly associated with ASD, and hyperandrogenemia increased the severity of ASD in 30 boys between 6 and 15 years old, whereas Kung et al. (2016) reported that salivary T levels during the peak of mini-puberty (1–3 months age) were not correlated with autistic traits at 18–30 months of age in 87 children (40 boys and 47 girls). Mini-puberty is when postnatal T levels reach pubertal T levels. In addition, the associations between T levels and physical growth were examined, and the outcomes were conflicting. Becker et al. (2015) reported that serum T levels measured using RIA at 4 weeks and 8 weeks were negatively associated with body weight and body mass index (BMI) up to 6 years of age, whereas Kiviranta et al. (2016) reported that urinary T levels measured by using LC/MS were positively correlated with growth velocity between 1 to 6 months (r = 0.22–0.63), especially in girls (r = 0.41–0.69) at month 1 and in boys (r = 0.32–0.49) at month 6.

The associations between T and child motor and language development were examined and showed complicated findings. Cho et al. (2017) reported that high salivary T levels measured using EIA in 61 VLBW infants at 40 weeks postmenstrual age and 3 and 6 months corrected ages were associated with better language development in boys and worse motor development both in boys and girls up to 6 months corrected age, whereas Whitehouse et al. (2012) found that bioavailable T (BioT) levels calculated from cord blood in full-term and preterm children were associated language delay in boys at age 3 years but not in girls at age 1 and age 3.

The associations between T and pubertal transition, anogenital distance (AGD) in neonates, and postnatal activation of the HPG axis with insulin therapy showed inconsistent outcomes. Mouritsen et al. (2014) reported that pubertal transition-related serum T levels were quantifiable 1.5 years earlier by LC/MS rather than RIA, especially in boys. Avidime et al. (2011), Jain et al. (2018), and Kareem et al. (2020) examined whether AGD was related to T levels in cord and postnatal blood. Avidime et al. (2011) reported no gender differences in T levels measured by EIA between boys (2.78 ± 0.30 ng/ml) and girls (2.09 ± 0.22 ng/ml) as well as no correlation between AGD and T levels in both sexes. Jain et al. (2018) and Kareem et al. (2020) reported that T levels using LC/MS and ELISA were significantly higher for boys (13.0 ng/dl in cord blood and 357.4 ng/dl in postnatal blood) than girls (4.1 ng/dl in cord blood and 170.6 ng/dl in postnatal blood), but the AGD in both boys and girls were either not correlated with T levels after controlling for the potential confounders (Jain et al., 2018) or positively correlated with T levels (Kareem et al., 2020). De Jong et al. (2012) examined the association between urinary T levels measured by ELISA and postnatal activation of HPG with insulin therapy. They found that T levels decreased with age after 6 months, especially after 9 months, and the decrease was faster at five time points (at 1 and 4 weeks, and 3, 6, 12 months) in infants with continuous insulin therapy (44.3, 21.2, 20.9, 7.0, and 2.4 nmol/mmol) than in infants receiving standard neonatal care (33.3, 29.5, 31.8, 11.6, and 3.1 nmol/mmol).

Accuracy and precision of T measurements

Of 30 studies, the objective of five studies was to determine more accurate and precise T measurements by comparing methods and materials. Those studies compared methods including second-generation immunoassays and LC/MS (Hamer et al., 2018), RIA and LC/MS (Ankarberg-Lindgren & Norjavaara, 2015), EIA and 2D:4D ratio (Knickmeyer et al., 2011), and LC/MS and RIC (Fang et al., 2017). Those studies also compared saliva and serum (Contreras et al., 2017) and urine, saliva, and serum (Fang et al., 2017). Hamer et al. (2018) measured total T levels with two widely used second-generation immunoassays (the Architect and the Elecsys second-generation T assays) and LC/MS concurrently and reported that second-generation immunoassays overestimated T levels in neonates. Those discrepancies were particularly notable in the 1st week of life in both boys and girls. Ankarberg-Lindgren and Norjavaara (2015) compared T levels using RIA and measured by them in 1999 and 2004 and using LC/MS measured by another team in 2010. They reported that morning T levels at six pubertal stages were higher in RIA than LC/MS (0.09 vs 0.025 nmol/L) before 8 years of age, but similar (1.5 vs 1.4 nmol/L) after 8 years.

Knickmeyer et al. (2011) measured salivary T levels at 3 months using EIA and compared them to the relative lengths of the second and fourth digits (2D:4D) at 2 weeks, 12 months, and 24 months. Higher ratios of 2D:4D were expected to be associated with lower T levels but no relation was found in boys or girls at any age. Also, T levels in boys (39.42 pg/ml) and girls (36.45 pg/ml) were not significantly different. Fang et al. (2017) used LC/MS and RIC (recycling immunoaffinity chromatography) to measure T levels in urine, saliva, and serum in infants between 0 and 12 months of age, controlling for feeding regimens. They reported positive correlations between urine and saliva as 0.93 in boys and 0.91 in girls, urine and serum as 0.90 in boys and 0.92 in girls, and saliva and blood as 0.93 in boys and 0.95 in girls. They also validated RIC against ELISA for urine, saliva, and serum samples and reported correlations between RIC and ELISA as 0.85–.90. Contreras et al. (2017) collected saliva and blood samples from 30 infant boys ages between 1 and 4 months and 12 adolescent boys between 11 to 17 years old to determine if there was a correlation between salivary and serum T levels to clarify the biological significance of mini-puberty. All samples were analyzed using LC/MS and showed a strong correlation between salivary and serum T levels in infants (r = 0.637) and adolescents (r = 0.799). Salivary T levels in infant boys were significantly lower than those in adolescents, but not values from serum and had a narrower range in infants (1.02–18.6 pg/mL) than in adolescents (19.5–66.0 pg/mL). Salivary T levels were significantly higher at 1–2 months of age than at 3–4 months in infant boys. Serum T levels were also higher at 1–2 months of age than at 3–4 months, but these differences were not statistically significant.

Simultaneous measurement of multiple hormones

Four of the 30 studies aimed to develop a method that was capable of the simultaneous measurement of multiple hormones including T (Kushnir et al., 2010; Olisov et al., 2019); to develop an online mass spectrometry method such as a high-turbulence liquid chromatography-tandem mass spectrometry (HTLC-MS/MS) method that quantifies low serum T levels in male hypogonadism, congenital adrenal hyperplasia, and androgen excess in polycystic ovary syndrome (Salameh et al., 2010); and to develop a sensitive, specific, and reproducible method that measured T, epitestosterone, androsterone, etiocholanolone, and DHEA in human hair (Shen et al., 2009). Olisov et al. (2019) obtained leftover serum samples from children 4 to 6 years old and measured T levels with LC/MS and EIA. They reported that a simultaneous assay was comparable to EIA although the values obtained by LC/MS for T, 17-hydroxyprogesterone, and DHEA-S were lower than those obtained by EIA. Kushnir et al. (2010) analyzed androstenedione, DHEA, and T in blood samples from children aged 6 months through 6 years with their LC/MS method and reported acceptable agreements with other commercial LC/MS and immunoassays. They also reported that the results with immunoassays were overestimated and were less sensitive than the LC/MS method. T levels were higher in girls than boys at 2–3 and 4–5 years. Salameh et al. (2010) measured serum T with HTLC-MS/MS and compared it to the recommended RIA method (Endocrine Society). Analyses revealed strong positive correlations between HTLC-MS/MS and RIA across the range of 0–900 ng/dL (slope = 1.021, r² = 0.945). Shen et al. (2009) measured androgens in hair from different age groups using GC/MS and reported ranges for the limit of detection and quantification as 0.1–0.2 pg/mg, which was sufficient for the determination of T levels.

Reference values of T

The purpose of five of the 30 studies (Büttler et al., 2016; Dhayat et al., 2015; Garagorri et al., 2008; Kim et al., 2020; Kyriakopoulou et al., 2013) was to provide reference values longitudinally that were clustered based on child sex and age in relation to assay methods (immunoassays and mass spectrometry) and materials (blood, saliva, and urine). Büttler et al. (2016) reported that at 8, 9, 10, 11, and 12 years of age, salivary T levels with ID-LC-MS/MS were 1.6, 3.4, 5.6, 15.6, and 42.5 pmol/L for boys and 2.9, 5.0, 7.8, 10.5, and 13.3 pmol/L for girls. They also found that T levels were higher in girls in prepubescence at 11–12 years of age. Kim et al. (2020) examined changes differences in serum T levels measured by LC/MS between boys and girls from ages 2 (124 vs 119 pg/mL), 4 (146 vs 156 pg/mL), and 6 (123 vs 125 pg/mL) years. They reported no sex-specific differences between these time points but found age-specific changes from ages 2 to 4 years and ages 4 to 6 years. Kyriakopoulou et al. (2013) analyzed serum T by LC/MS in healthy children between 0 and 18 years old and reported upper/lower limit levels of T in boys and girls at 0–14 days, 15 days to 1 year, and 1–13 years. They compared T levels using LCMS and EIA and reported a strong positive correlation of r² = 0.894. Dhayat et al. (2015) measured urinary T by GC/MS in healthy full-term newborns at 13-time points from Week 1 to Week 49 (Weeks 1, 3, 5, 7, 8, 11, 13, 17, 21, 25, 33, 41, and 49) and reported that T levels were high during the first 2 months of life, especially at 3 weeks, then decreased to one-fifth by about 6 months in both boys and girls. Garagorri et al. (2008) measured plasma T levels using RIA. Cord blood was collected at birth and venous blood was collected on days 3, 15, 30, 60, 90, 120, 150, and 180. They reported that T levels were significantly higher in boys (2.39–0.19 ng/ml) than girls (1.95–0.13 ng/ml) and decreased from birth to 6 months of age regardless of infant gender. No gender differences were reported in weight, length, or gestational age.

Relationships between T and other factors

Finally, five studies were conducted to examine if T was related to other steroid hormones and factors (Ahmed et al., 2010; Huang et al., 2020; Minatoya et al., 2017; Smith et al., 2019; Wen et al., 2017). Ahmed et al. (2010) reported that T levels increased 13 fold following HCG stimulation from day 1 to day 4 and 24 fold from day 1 to day 22. They found that serum T levels measured by EIA were inversely associated with the child’s age on day 4. A significant correlation between T and anti-Mullerian hormone was found on day 4 and day 22 although the relationship was weak when T levels were low. Minatoya et al. (2017) collected cord blood to investigate whether Bisphenol A (BPA) levels in cord blood are associated with reproductive and thyroid hormonal levels in neonates. The levels of BPA and T were measured with LC/MS. The median and interquartile range of T levels were reported as 98.3 pg/ml (76.7–122.5) for boys and 69.0 pg/ml (51.8–94.9) for girls. They found no association between BPA. Wen et al. (2017) collected urine and blood samples from children to determine if prenatal exposure to phthalate metabolite (MEHP), which was measured by maternal phthalate metabolite levels in urine during pregnancy, was associated with T levels at 2–3, 5–6, 8–9, and 11–12 years of age. They analyzed children’s urine samples with LC/MS and serum samples with RIA. They reported that child MEHP concentration was significantly associated with decreased levels of free T for boys (β = −0.124). Total T levels were significantly higher in females at age 8 (3.23 vs 3.71 ng/ml, p = 0.026) and higher in males at age 11 (14.18 vs 6.10 ng/ml, p = 0.046). Similarly, free T levels were significantly higher in females at age 5 years (0.15 vs 0.20 pg/ml, p = 0.006) and in males at age 11 (0.42 vs 0.22 pg/ml, p = 0.006). Huang et al. (2020) examined how prenatal exposure to polycyclic aromatic hydrocarbons (PAHs) was related to birth outcomes of newborns. They measured PAHs levels in maternal urine and T levels in cord blood at birth using ELISA and reported an inverse relationship between prenatal exposure to PAHs and T levels, length, and head circumference but not length and head circumference of the newborns. Smith et al. (2019) collected hair samples from 7 to 9 years old children to compare T levels between hair segments (0–3 vs 3–6 cm) and to examine the relationships with anthropometric measurements. They analyzed T levels using ELISA and reported that the levels were positively correlated between hair segments 0–3 cm and 3–6 cm (r = 0.657) and between boys and girls only in the 0–3 cm segment (boys, −0.25 pg/mg and girls, −0.45 pg/mg). Hair T levels were not related to weight, height, waist size, or BMI.

Discussion

We expected that T levels might be inversely associated with health, physical growth, and development in children because elevated T levels have been considered a biological risk factor for health and developmental problems (Becker et al., 2015; Cho et al., 2017), based on theories of gender differences (Geschwind, 1987). The theory of sex differences and the extreme male brain theory of autism also emphasized that males are more vulnerable to ASD than females because of high T levels (Baron-Cohen, 2009) because males were four to five more likely to be identified with ASD than females in the US (Baio et al., 2018). However, the findings in the selected studies were conflicting. T levels were either positively or not associated with ASD depending on child sex and age as well as assay methods and materials. Serum T levels measured using ELISA in boys between 6 and 15 years old were positively associated with ASD and its severity (El-Baz et al., 2014), whereas salivary T levels measured by EIA in boys and girls between 1 and 3 months were not associated with autistic traits at 18–30 months. However, T levels and the autistic trait scores differed by child sex (Kung et al., 2016).

T levels were also expected to be inversely associated with physical growth as high T levels were reported to be a biological risk factor for lower birthweight and gestational age as well as smaller body weight, length, and head circumference in VLBW infants over 24 months after birth (Cho et al., 2012, 2021). However, T levels were also reported as a protective biological factor as the levels were positively associated with child physical growth. Serum T levels measured by RIA at 4 weeks and 8 weeks were negatively associated with body weight and BMI up to 6 years (Becker et al., 2015), but urinary T levels measured by using LC/MS were positively correlated with growth velocity between 1 and 6 months (Kiviranta et al., 2016).

The associations between T levels and child development varied in the selected studies. High salivary T levels in VLBW children measured by EIA at 40 weeks postmenstrual age and 3 and 6 months corrected ages were positively associated with language development in boys and negatively associated with motor development in both boys and girls at 6 months corrected age (Cho et al., 2017), whereas high Bio T levels in cord blood measured by LC/MS were negatively associated with language development in boys at 3 years, but positively associated in girls at 1 and 3 years (Whitehouse et al., 2012). Those conflicting associations between T levels and health, physical growth, and developmental outcomes should not be interpreted as the wrong outcomes against one another because many factors might impact T measurements. It would be better to understand why the outcomes are conflicting based on theories used, sample size, data analysis method, and variability and determinants of T measurements.

The variability in T measurement related to demographic factors such as sex and age as well as assay factors such as methods and materials needs to be investigated in different groups of children. For example, T levels varied with age: during the first months of life, the levels are highest followed by puberty and the second trimester and are almost three times higher in preterm boys than in full-term boys, although elevated T levels in early infancy may not yet to assess virilization (de Jong et al., 2012; Dhayat et al., 2015; Hamer et al., 2018; Kuiri-Hänninen et al., 2011). T levels were expected to be higher in males throughout life, however, this was not supported by the studies in this review. Researchers reported that T levels were significantly higher in boys during the first 2 months of life, especially at 3 weeks, then decreased to one-fifth by 6 months in both boys and girls (Dhayat et al., 2015; Garagorri et al., 2008). T levels were higher in boys between 0 and 1 year old, higher in girls between 1 and 8–10 years old, and higher again in boys between 11 and 12 years of age (Büttler et al., 2016; Kyriakopoulou et al., 2013).

Many assay methods have been used for T measurement in children. We found that the current trend for T measurement changed from RIA and GC/MS to EIA or ELISA and LC/MS over the past decade, especially because LC/MS has shown higher sensitivity, accuracy, and precision than other assay methods (Hamer et al., 2018). Due to its lower detection threshold, LC/MS measures a wider range of T levels than other methods (Kushnir et al., 2010). Compared to LC/MS, immunoassays are technically simple and feasible because of the affordable cost and availability of commercial reagents although with limited specificity and sensitivity (Fanelli et al., 2013). On the other hand, LC/MS is appropriate for the detection of very low T levels in children, straightforward to perform, and can detect multiple steroid hormones from a single sample. However, LC/MS method is expensive and special techniques are required (Fanelli et al., 2013; Turpeinen et al., 2008).

Thus, several of the selected studies compared the quality of T measurement, developed better methods, and evaluated reference values. A high correlation between LCMS and EIA suggests their values reflect actual T levels (Olisov et al., 2019). However, EIA and ELISA overestimated values compared to LC/MS (Ankarberg-Lindgren & Norjavaara, 2015; Hamer et al., 2018; Kushnir et al., 2010; Olisov et al., 2019). T levels measured by RIA and LC/MS were similar. However, study findings were not statistically significant and differed among studies (Ankarberg-Lindgren & Norjavaara, 2015; Hamer et al., 2018; Kyriakopoulou et al., 2013; Mouritsen et al., 2014; Salameh et al., 2010). That is, the study findings measured by LC/MS could be more reliable. Also, T levels measured to predict early peripubertal changes were more quantifiable by LC/MS than RIA in boys and girls (Mouritsen et al., 2014).

Many assay materials have been chosen for T measurements in children. All sources, including blood, saliva, urine, and hair, were consistently chosen although blood samples (serum, plasma, or cord blood) were more often selected, possibly because of the proliferation of LC/MS assays and reference intervals for serum T during the last decade (Fanelli et al., 2013). Saliva would be most popular because T in the saliva is largely free, free steroids (protein-unbound) in saliva are physiologically meaningful, and free T levels are strongly correlated with total (protein-bound) T levels in the blood (Fang et al., 2017; Gunnala et al., 2015). Saliva is also less invasive and simpler in assays than other samples such as blood and urine. Because saliva samples have been discouraged for measuring low T levels in children, especially using immunoassays, this finding of selecting blood as a more favorable assay material than saliva is understandable (Hamer et al., 2018).

Saliva sampling from small babies such as VLBW preterm neonates has been challenging because of the small volume of samples and the need for cautious handling of babies. Even though collecting saliva is non-invasive, it is difficult for the neonates because their mouths may be dry due to oxygen therapy and mouth breathing. Commercial collection devices interfere with T assays of saliva such that T levels were lower when using synthetic Salivettes and higher when using cotton Salivettes (Büttler et al., 2018). Also, saliva should be sampled at least 30 minutes before or after any oral intake to reduce interference.

Urine samples were chosen for serial measurements that might not be available using blood samples (Dhayat et al., 2015; Fang et al., 2017; Kiviranta et al., 2016; Wen et al., 2017). Head hair samples were chosen to identify long-term associations between T levels and child outcomes (Shen et al., 2009; Smith et al., 2019). The limits of detection and quantification, 0.1–0.2 pg/mg, were reported to be sufficient for the determination of T in human hair although repeated measurements may be imperative before making any clinical decisions (Shen et al., 2009). As hair sample has a unique advantage for providing retrospective information, further studies using different assay methods and hair segments stratified by sex and age would be necessary.

Other factors affect T measurement in children such as sampling acceptability (e.g., serum vs plasma), sampling time (e.g., morning vs evening), sampling method (e.g., passive drooling vs collecting devices), handling of samples (e.g., with or without centrifugation), sample stability (e.g., room temperature, 4 °C refrigerator, or −20 °C freezer for 1, 3, 7, 14, 21, and 28 days before storing into a −70 °C freezer), and storage stability (e.g., shorter storage vs longer than 1 year). As no statistically significant differences in the recovery of the liquid/liquid extraction were reported between serum and plasma samples, both might be acceptable for T measurement (Kushnir et al., 2010). However, serum samples could be more reliable than plasma because EDTA had a large positive bias with T measurements (Cao et al., 2018). Early morning sampling in fasting conditions has been recommended to reduce the diurnal effects of T and because of the low levels in the evening (Fanelli et al., 2013). T is stable at room temperature before storage at 70 °C freezer (Kushnir et al., 2010). No statistically significant differences were found in serum T levels before and after storage for more than a year at 70 °C freezer (Kushnir et al., 2010).

Finally, multiple studies provided reference values in children stratified by sex and age as the values are needed clinically (Büttler et al., 2016; Dhayat et al., 2015; Garagorri et al., 2008; Kim et al., 2020). As immunoassays have shown poor agreement, correlation, and calibration especially in children’s samples as indicated by poor correlations in girls compared to boys, the values from LC/MS should be used (Fanelli et al., 2011). The reference values with immunoassays for diagnostic purposes, such as for ambiguous genitalia and congenital adrenal hyperplasia, might not be useful (Fanelli et al., 2013; Kushnir et al., 2010).

Conclusion

The variability and determinants of T measurements in children remain challenging. In addition to the variability due to sex and age as well as assay methods and materials, multiple factors affecting the determinants of T measurements including sensitivity, accuracy, and precision need to be investigated continuously to provide the most reliable information to interpret the analytical performance. Although the poor performance of T measurements by immunoassays in children, especially in neonates, has been a concern, only a couple of studies were aimed to provide the reference values by using serum and LC/MS methods during the past decade. For researchers who are often faced with limited quantity and quality of samples, LC/MS would be a choice. However, even with LC/MS, sex- and age-specific differences commonly found within the 1st year may not be detected after the 1st year until puberty. Because of limited information about the determinants of T measurements in children, more work is warranted with extra factors related to sampling method (e.g., passive drooling vs commercially available collecting devices), multiple samplings (e.g., one vs three time samples), and sampling storage (e.g., higher than room temperature and longer than a year) that might affect T levels and the reference values. Future research needs to be focused on the aims of the study rather than merely following measurement trends as feasibility is another important issue in the analytical performance in children.