The natural course of bone mineral density in transgender youth before medical treatment; a cross sectional study

Abstract

Introduction

In transgender persons, gender identity does not correlate with the sex assigned at birth. This incongruence can lead to gender dysphoria and the desire for medical treatment to obtain physical characteristics consistent with the gender identity. If gender dysphoria emerges before or during puberty, treatment can initially consist of a gonadotropin-releasing hormone agonist (GnRHa) to suppress pubertal development, which can be started from Tanner stage 2, and can be followed by gender-affirming hormone treatment (GAH).^1,2 As the sex hormone changes throughout puberty are pivotal for attaining peak bone mass,³ one of the biggest health concerns regarding GnRHa is insufficient accrual of bone mineral density (BMD) during treatment. Consequently, studies have focused on BMD development during GnRHa treatment. No controlled trials have been performed in transgender adolescents. Therefore, the course of BMD during treatment has been compared to data from the general population using Z-scores specific for age, sex, and ethnicity. The use of Z-scores is common in paediatrics to evaluate processes over time, such as growth, weight gain, and bone mineral accrual.

In general, a decrease in the BMD Z-score was noted during GnRHa treatment, followed by an increase during GAH.^4-9 However, a recent study reported that BMD Z-scores at the lumbar spine did not catch-up and remained low in individuals assigned male at birth (AMAB) at age 28 years after an average of 11.6 years of hormone treatment.¹⁰ Because of the lack of a control group, these data need to be interpreted with caution, as the observed change in BMD over time is not necessarily the result of the treatment. Interestingly, several studies on BMD in transgender adolescents made one common observation: BMD Z-scores were already relatively low prior to medical intervention in individuals AMAB.^4-9,11,12 Speculating on this finding, it has been suggested that besides medical treatment, lifestyle factors, such as less physical exercise, sunlight exposure, and dietary intake of calcium compared to peers, may affect accrual of BMD in transgender adolescents.¹³ This would provide an additional or even alternative explanation for the observed decrease in BMD Z-score during GnRHa treatment with incomplete catch-up during GAH. However, it has not been possible to determine the contribution of such factors versus treatment effects because the natural course of BMD development in transgender adolescents has not been studied. Therefore, the aim of this study was to assess the natural course of BMD development during puberty in transgender persons who had not (yet) received medical treatment and identify the determinants of BMD. Because withholding treatment from transgender adolescents who meet the criteria for treatment is considered unethical, it is not possible to collect longitudinal data over many years. Therefore, we used a cross-sectional design using BMD measurements from before the start of medical treatment from transgender youth across a wide age range from 12 to 25 years.

Methods

Study population and design

For this retrospective, cross-sectional study of the Amsterdam Cohort of Gender Dysphoria was used.¹⁴ This dataset contains clinical, biochemical, and imaging data on all people attending the gender identity clinic of Amsterdam UMC, with the location Vrije Universiteit Medical Center, at least once since its opening in 1972 until the 31st of December 2018. DXA scans were part of routine care. All people diagnosed with gender dysphoria according to the DSM-IV or DSM-5 who had a DXA scan prior to start of GnRHa or prior to start of GAH without preceding use of GnRHa between the ages of 12 and 25 years were included. Scans made up to 3 months after start of GnRHa or 3 months after start of GAH in people without prior use of GnRHa were allowed. If people had had multiple scans before start of medical treatment, the scan closest to the starting date of treatment was used. Per person, only one scan was included. Only people for whom DXA data on body composition were complete (ie, whole body DXA) were included. People with a history of conditions known to affect BMD were excluded.

Prior to treatment initiation, Tanner stage was assessed in adolescents by a paediatric endocrinologist. For individuals over the age of 18, assessment of Tanner stage was only done when indicated. If Tanner stage was not available in people over the age of 18, it was assumed they had reached Tanner stage G5/B5.

The local Medical Ethics Committee of Amsterdam UMC, Vrije Universiteit, Amsterdam, The Netherlands reviewed this study for legal and ethical aspects. The committee confirmed that the Medical Research Involving Human Subjects Act (WMO) did not apply because of a lack of interventions. The necessity for written informed consent was waived. The study was carried out according to the Declaration of Helsinki.

Dual-energy X-ray absorptiometry

Bone mineral density in g/cm² and body composition were measured on a Delphi DXA scanner manufactured by Hologic (Hologic Inc., Bedford, MA, USA) until February 2011 when a Hologic Discovery system was implemented. Updates to the software were done in 2012 and 2015. Direct comparison between devices and after updates was possible due to cross-calibration with a phantom. BMD Z-scores comparing BMD to the average BMD of age-, sex-, and ethnicity-matched persons were calculated for the sex assigned at birth at the lumbar spine, total hip, femoral neck, and total body less head. Data from the National Health and Nutrition Examination Surveys (NHANES) were used as reference.¹⁵ Reference data for total body less head were only available up to age 20. The reference values for age 20 were used for scans at ages 21 to 25 too, as BMD for the total body does not change after age 20.¹⁵

Whole body lean tissue and fat tissue were assessed by DXA. Lean mass was calculated as total mass minus fat mass in grams. Lean mass was adjusted for height (lean mass/height²) and converted to a Z-score based on NHANES reference values.¹⁵ Fat mass was measured in grams as well. As paediatric reference data for absolute fat mass are not available, percentage fat mass was used. Percentage fat mass was calculated as fat mass divided by total body mass multiplied by 100 and converted to a Z-score based on NHANES reference values.¹⁵ BMI Z-scores were calculated based on reference values from CDC growth data.¹⁶

Statistical analyses

Analyses were carried out using STATA Statistical Software version 15.1 (StataCorp LLC, College Station, Texas, USA). Continuous variables are reported as median with interquartile range. Dichotomous variables are shown as percentages.

Locally weighted regressions of BMD Z-score on age were carried out and plotted (locally weighted scatterplot smoothing, LOWESS) to examine their relationship. If a change in the linear association was observed in these plots, data were stratified by age based on this turning point. Data were analysed with a linear regression model with BMD Z-score as the main outcome and age at time of DXA as an independent variable. Additional adjustments were performed separately with height SDS, height-adjusted lean mass Z-score, and whole body percentage fat Z-score. These independent variables were collected at time of DXA. Individuals AMAB and assigned female at birth (AFAB) were analysed separately. For the results in Tables 1 and 2, age was rounded to integers.

DXA scans were part of the routine protocol for adolescents around start of medical treatment. As onset of puberty was a criterion for start of GnRHa, the timing of DXA could correlate with timing of puberty (ie, meaning that younger adolescents may be adolescents with early puberty and older adolescents with more delayed onset of puberty). As this might bias the results, we checked if pubertal development was comparable with the general Dutch population. Therefore, median Tanner genital stage and breast stage were calculated per age group for individuals AMAB and individuals AFAB, respectively. These medians were compared with nationwide reference values from The Netherlands.¹⁷

Standard deviation scores (SDS) were calculated for height using nationwide reference values.¹⁷ Reference values were available until age 21, when adult height has generally been reached. The reference values for age 21 were therefore used for ages 22 to 25 as well.

Results

People with a history of auto-immune disease (n = 32), syndromic disorders (n = 10), malignancy (n = 4), eating disorder (n = 11), neurological disorder (n = 8), kidney disease (n = 2), delayed or precocious puberty or hypogonadism (n = 5), and bone disease (n = 5) were excluded. In total, 889 individuals were included, of whom 333 were individuals AMAB and 556 individuals AFAB. Age at DXA ranged from 12 to 25 years. Characteristics by age group are listed in Table 1 for individuals AMAB and in Table 2 for individuals AFAB. Puberty development was not different from nationwide Dutch references.

AMAB

In individuals AMAB between age 12 to 22 years (n = 232), the BMD Z-score at all four regions of interest showed a negative association with age (Figure 1). Per 1 year increase in age, BMD Z-score was −0.13 (95% confidence interval, CI −0.17; −0.09, P < .001) lower at the lumbar spine, −0.04 (95% CI −0.08; −0.01, P = .01) at the total hip, −0.06 (95% CI −0.09; −0.03, P < .001) at the femoral neck, and −0.12 (95% CI −0.15; −0.09, P < .001) at the total body less head. BMD Z-scores were not associated with age between age 22 and 25 years (n = 101) (lumbar spine: 0.15 per year (95% CI −0.09; 0.39, P = .23), total hip: 0.09 per year (95% CI −0.07; 0.26 P = 0.), femoral neck: 0.06 per year (95% CI −0.13; 0.25, P = .52), and total body less head 0.17 per year (−0.01; 0.35, P = .06)). Adjusting for height SDS did not alter the results.

Figure 1.

LOWESS plot with individual observations of BMD Z-score at ages 12 to 25 years at four regions of interest in individuals AMAB and AFAB. Abbreviations: LOWESS: Locally weighted scatterplot smoothing, BMD: bone mineral density, AMAB: assigned male at birth, and AFAB: assigned female at birth. The dashed vertical line indicates a turning point in the linear association between BMD and age.

Open in new tabDownload slide

Height-adjusted lean mass Z-score showed a negative association with age, too (−0.10 per year, 95% CI −0.13; −0.07, P < .001) between age 12 and 22 years in individuals AMAB. When adjusted for height-adjusted lean mass Z-score, the association between BMD Z-score and age was attenuated at the lumbar spine to −0.07 per year (95% CI −0.10; −0.04, P < .001) and the total body less head to −0.06 per year (95% CI −0.09; −0.03, P < .001) and was no longer present at the total hip (0.02 per year, 95% CI −0.01; 0.04, P = .18) and at the femoral neck (−0.001 per year, 95% CI −0.03; 0.03, P = .93). Fat percentage Z-score was negatively associated with age (−0.04 per year, 95% CI −0.08; −0.01, P = .01) between age 12 and 22 years. Adjusting for fat percentage, Z-score attenuated the association of BMD Z-score with age at the total body less head (−0.06, 95% CI −0.08; −0.03, P < .001), but only slightly attenuated this association at the total hip (−0.03 per year, 95% CI −0.06; −0.002, P = .04) and at the femoral neck (−0.04 per year, 95% CI −0.07; −0.02, P = .002) and had no effect at the lumbar spine (−0.11, 95% CI −0.14; −0.08, P < .001).

AFAB

As shown in Figure 1, in individuals AFAB, no association between BMD Z-score and age was observed at the lumbar spine [−0.00 per year (95% CI −0.02; 0.02, P = .97)], the total hip [0.01 per year (95% CI −0.01; 0.03, P = .44)], and the femoral neck [−0.01 per year (95% CI −0.04; 0.01, P = .27)]. At the total body less head, an inverse association between BMD Z-score and age was observed [−0.08 per year (95% CI −0.12; −0.04, P < .001)] between age 12 and 20 years (n = 318). Between age 20 and 25 years (n = 238), BMD Z-scores of total body less head were not associated with age [0.05 per year (95% CI −0.03; 0.13, P = .21)]. Adjusting for height SDS did not change the results at the lumbar spine, total hip, and femoral neck, but slightly attenuated the association between BMD Z-score and age at the total body less head to −0.06 per year (95% CI −0.10; −0.03, P < .001). Adjusting for height-adjusted lean mass Z-score did not change the results. A positive association between fat percentage Z-score and age was found (0.03 per year, 95% CI 0.01; 0.05, P = .01), but adjustment for fat percentage Z-score did not change the association between BMD Z-score and age at all four regions.

Discussion

This study shows that in individuals AMAB diagnosed with gender dysphoria, who have not (yet) received medical treatment (ie, GnRHa and/or GAH), BMD Z-score at the lumbar spine, total hip, femoral neck, and total body less head is negatively associated with age. Moreover, this was in part mediated by height-adjusted lean mass. In individuals AFAB, an association between age and BMD Z-score was only found at the total body less head, but not in other regions of interest.

Our finding of a negative association between BMD Z-scores and age in a group of healthy individuals is extraordinary. In the general population, it is common for bone mass status to remain stable, meaning individuals keep their original position on the BMD chart as they grow older, a phenomenon referred to as “tracking”.¹⁸ Observational studies on the effect of puberty suppression and GAH on BMD have not included control groups of untreated transgender adolescents. Therefore, the impact of treatment has been assessed by studying changes in BMD Z-scores, with the implicit assumption that without hormonal interventions, BMD would track. Incomplete catch-up of BMD Z-scores to pre-treatment levels has been interpreted as a possible lasting negative effect of puberty suppression.⁶ However, the current study suggests that in individuals AMAB, BMD may not track. It is striking that a group of adolescents characterised by an aspect of their identity, without any medical conditions known to affect bone, might follow a different course of bone mineral accrual.

The findings could imply that the decrease in BMD Z-score previously observed in individuals AMAB during medical treatment should not simply be attributed to medical treatment. Lifestyle factors may be important determinants. The negative correlation between BMD Z-score and age was attenuated by height-adjusted lean mass Z-score at all four regions of interest in individuals AMAB. This suggests that reduced physical activity is an important factor leading to reduced bone mineral accrual, which was also proposed by others.^13,19 A previous study did indeed show that physical activity levels in transgender youth were lower compared with cisgender youth.²⁰ Additionally, individuals AMAB reported lower activity scores than individuals AFAB, making adolescents AMAB especially susceptible to inferior BMD.²¹ However, the association between age and BMD Z-score at the lumbar spine in individuals AMAB was not fully explained by differences in lean mass. This suggests that additional factors play a role in the decreased bone mineral accrual. One of these factors might be vitamin D deficiency.^21,22 Unfortunately, this could not be assessed in the current study because serum vitamin D concentrations were missing, particularly in people under the age of 18 years, but vitamin D deficiency is common in Dutch transgender adolescents.⁸

Although several studies have shown decreased BMD Z-scores before treatment initiation in transgender youth, there is a dearth of studies focusing on determinants of this finding. One case study on six individuals AMAB before start of pubertal suppression found a mean lumbar spine BMD Z-score of −1.07 (SD 2.0) and three participants (50%) with a Z-score ≤ −2.0.²³ Vitamin D, calcium intake, weekly exercise, pubertal stage, and EAT-26 score (measuring symptoms of an eating disorder) were not different between the participants with a BMD Z-score ≤ −2.0 and the ones with normal Z-scores, but the very small groups do not allow firm conclusions. In a larger cohort from the United States, 10 out of 33 individuals AMAB (30%) and four out of 30 individuals AFAB (13%) had a BMD Z-score < −2 before start of GnRHa.²¹ Female sex assigned at birth and higher vitamin D concentrations were associated with higher Z-scores at the total hip, but not at the lumbar spine or femoral neck in this study. Moreover, lower physical activity scores were found in the group with low BMD Z-scores (<−2.0) compared with the group with normal BMD Z-scores. Of interest, fitness and lean mass are both associated with positive deviation from tracking at the lumbar spine in cisgender males underlining the important influence of these factors on BMD.¹⁸

Engaging in weight-bearing physical activity should be encouraged in all adolescents, most especially in transgender individuals AMAB. Additionally, to promote optimal bone mass accrual, ensuring adequate vitamin D levels and calcium intake is recommended. These efforts should be continued throughout medical treatment.

Although BMD Z-scores remained stable in individuals AFAB at three regions of interest, an association between age and BMD Z-scores was observed at the total body less head. After adjustment for height SDS, the association attenuated slightly which indicates that the smaller height in this group plays a role in the lower BMD. An explanation for the remaining association could be that lifestyle factors, such as reduced physical activity among AFAB individuals²⁴ possibly due to persisting body dysphoria, have a more prominent effect on BMD of the total body less head compared to other regions. A larger decrease of BMD of the total body less head than at other sites has also been described in populations with other conditions that adversely affect BMD, such as avoidant/restrictive food intake disorder (ARFID).²⁵ The cause of this difference in change of BMD at different sites deserves further study.

In this study, reference values of the sex assigned at birth were used as people had not (yet) received medical gender-affirming treatment. Additionally, due to the difference in age at onset of puberty, cisgender girls are exposed to sex steroids at an earlier age than cisgender boys, which would complicate comparison to reference values of individuals with the same gender identity.

Limitations of this study include the retrospective and cross-sectional design. We cannot exclude that older presenting individuals were different from younger adolescents with respect to characteristics other than age. Another limitation is the lack of sufficient data on vitamin D levels. Nevertheless, this study adds valuable knowledge on BMD prior to treatment initiation in the largest cohort to date of transgender youth.

In conclusion, in transgender individuals AFAB aged 12–25 years pre-treatment, BMD Z-score was comparable to that in the general population with the exception of total body less head BMD Z-scores which were negatively associated with age between age 12 and 20 years. However, in transgender individuals AMAB, this inverse relationship between BMD Z-scores and age was observed in all four regions of interest; lumbar spine, total hip, femoral neck, and total body less head. This suggests that BMD may not track in this group characterised by gender dysphoria, but without any medical diagnoses that impair bone health. This must be taken into account when interpreting results from studies of BMD during puberty suppression and subsequent gender-affirming hormone treatment. Observed changes in BMD are not necessarily due to treatment alone, but may also represent the natural course of BMD development in transgender youth. The finding that the association between age and BMD Z-scores was attenuated by lean mass Z-score suggests that reduced muscle mass likely due to reduced physical exercise plays an important role. Lifestyle counselling, therefore, is an essential part of care for transgender adolescents, whether seeking hormonal interventions or not, to optimize bone health.

Funding

This research did not receive any funding.

Authors’ contributions

Maria van der Loos (Conceptualization [equal], Data curation [equal], Formal analysis [lead], Investigation [lead], Methodology [equal], Writing—original draft [equal]), Lidewij Boogers (Formal analysis [equal], Writing—review & editing [equal]), Daniel Klink (Supervision [equal], Writing—review & editing [equal]), Martin den Heijer (Supervision [equal], Writing—review & editing [equal]), Chantal M. Wiepjes (Conceptualization [equal], Methodology [equal], Supervision [equal], Validation [equal], Writing—review & editing [equal]), and Sabine E. Hannema (Conceptualization [lead], Methodology [equal], Supervision [equal], Writing—review & editing [equal]).

Data availability

Restrictions apply to the availability of some or all data generated or analysed during this study to preserve patient confidentiality or because they were used under license. The corresponding author will on request detail the restrictions and any conditions under which access to some data may be provided.

References

Author notes