Progression of Gender Dysphoria in Children and Adolescents: A Longitudinal Study

This study uses data from the Study of Transition, Outcomes, and Gender (STRONG) cohort. The STRONG cohort includes participants from Kaiser Permanente (KP) integrated health care systems in Georgia, Northern California, and Southern California. The 3 KP sites collectively provide comprehensive health care to ∼9 million individuals.^34,35  The current transgender health care protocols at KP follow the Endocrine Society and WPATH guidelines, which include mental health support, GnRH agonists, feminizing and masculinizing hormones, and surgery.³⁶  The Emory University Rollins School of Public Health served as the coordinating center. The study protocol received approvals from the institutional review boards of all 4 institutions (3 KP sites and the coordinating center) with exemption of informed consent. The details of STRONG cohort ascertainment and data collection were described in previous publications.^17,32,37  Participants were identified in the electronic health record from January 1, 2006, to December 31, 2014, by searching for keywords in free-text clinical notes reflecting GD behaviors. Each participant’s index date was defined on the basis of the first evidence of GD behaviors mentioned in the notes, as evidenced in the presence of keywords such as “transgender,” and “gender identity.” The full list of keywords can be found in our previous publication.³²  All notes were reviewed to confirm eligibility. The analytic data set was limited to participants who were aged <18 years at index date, had at least 1 follow-up appointment, and had evidence of GD behavior, as reflected in the keywords, but did not have a diagnosis related to gender dysphoria and had not received any GAHT. Participants whose sex recorded at birth could not be determined (n = 14) were excluded from the analyses because sex recorded at birth was considered a key variable in the analyses. Two types of events of interest were ascertained during follow-up: an assignment of the first gender dysphoria–related diagnosis (GDRD) and receipt of GAHT. The GDRDs were based on the International Classification of Diseases, Ninth Revision and included codes for conditions such as transsexualism (302.5) and gender identity disorder in children (302.6). The Diagnostic and Statistical Manual of Mental Disorders uses the International Classification of Diseases coding scheme with direct match between the 2 systems. For this reason, any International Classification of Diseases codes used in this analysis would be equivalent to the same diagnosis in the contemporaneous Diagnostic and Statistical Manual of Mental Disorders. Cohort ascertainment and follow-up were undertaken before the health plans switched to International Classification of Diseases, 10th Revision codes. GAHT receipt was determined from pharmacy records and date of therapy initiation was based on the first prescription for a puberty suppression medication or cross-sex hormonal therapy.

The follow-up for each participant extended from the index date until the event of interest (diagnostic code or first ordered GAHT prescription, depending on the analysis), disenrollment from KP, or end of follow-up (December 31, 2014). The data were examined by using time-to-event analyses to take into account censoring and variable duration of follow-up.

Kaplan Meier curves were constructed to compare timing and occurrence of GDRD and GAHT initiation across subgroups of participants. The independent variables in these analyses included age category at index date (categorized as 3–9, 10–14 and ≥15–17 years), recorded sex at birth, and race and ethnicity (non-Hispanic white individuals versus individuals of other races and ethnicities). These age groups represent 3 distinct populations: children who have not started puberty, based on national averages (age 3–9 years); children after the onset of puberty who may be considered as candidates for puberty suppression (age 10–14 years); and teenagers who might be offered feminizing or masculinizing hormones on the basis of current treatment guidelines (age ≥15 years).³⁶ 

Multivariable Cox proportional hazards models were used to evaluate the associations of all 3 independent variables (site, sex recorded at birth, age, and race and ethnicity) considered individually and simultaneously with each event of interest. The results of the Cox models were expressed as crude and adjusted hazard ratios (HRs) and corresponding 95% confidence intervals (CIs). All models were evaluated for validity of proportional hazard assumptions by inspecting log-log curves. If proportional hazard assumptions were violated, stratified Cox models were used. The data analyses were performed by using SAS software version 9.4 (SAS Institute, Inc, Cary, NC).

After applying eligibility criteria (Fig 1), 958 children were included in the final analysis data set (Table 1). Of those, 431 individuals were assigned male sex at birth (AMAB) and 527 individuals were assigned female sex at birth (AFAB). Children aged <10 years at index date represented 21% of the total cohort, 30% of the AMAB group and 14% of the AFAB group. In both the AMAB and AFAB groups, non-Hispanic white individuals made up >45% of the children. A majority of individuals (63%) received care at KP Northern California. Of the total analytic cohort, 29% of participants received a GDRD and 25% were prescribed GAHT during an average follow-up of 3 years (maximum 9 years). Among 677 cohort members without a diagnosis, 74 (11%) received GAHT treatment, whereas, among 281 children and adolescent with a diagnosis, 162 (58%) initiated GAHT.

Compared with the AFAB group, a lower proportion of AMAB children and adolescents received a GDRD (24% vs 33%) or initiated GAHT (14% vs 33%) during follow-up (Table 1). When time to diagnosis was compared across age groups among AMAB and AFAB children (Fig 2A and C, respectively), the youngest age group (3–9 years of age) was most likely to remain diagnosis-free to the end of follow-up for both (85% in AMAB and 72% in AFAB). Among AMAB participants, the proportion of those who received a diagnosis before the end of follow-up was lower in the middle age group (10–14 years) than in the oldest age group (≥15 years) (20% vs 35%, respectively) (Fig 2A) but this was not the case among AFAB participants (Fig 2C). Whereas few of the youngest children received GAHT by the end of the follow-up (5% in the AFAB group and 4% in the AMAB group), treatment was generally more delayed in older AMAB than older AFAB children (Fig 2B and D).

In the multivariable model (Table 2), the difference in diagnosis rates between AFAB and AMAB participants was attenuated (adjusted HR = 1.3; 95% CI: 1.0–1.7), but the difference in GAHT receipt remained evident (adjusted HR = 2.5; 95% CI: 1.8–3.3). By contrast, the differences in diagnosis rates remained pronounced across the 3 age groups; using the youngest age group (3–9 years) as reference, the adjusted HRs (95% CI) were 2.0 (1.3–3.0) for age 10 to 14 years and 2.7 (1.8–3.9) for age ≥15 years. The proportional hazard assumption for the age variables was violated in the analyses that used GAHT initiation as the end point of interest, and for this reason the corresponding adjusted HR estimates across the age groups were not generated by the model (Table 2).

Relative to non-Hispanic white children, children of minority racial and ethnic groups were less likely to receive a GDRD (26% vs 33%) or be prescribed GAHT (21% vs 29%) during follow-up, and the time to diagnosis and GAHT were also different in the 2 groups (Fig 3). Controlling for other variables (Table 2), the difference in diagnosis rates was less evident (adjusted HR = 0.8; 95% CI: 0.6–1.0) whereas the difference in GAHT receipt was greater (adjusted HR = 0.6; 95% CI: 0.5–0.8).

When the data were examined by site, rates of diagnosis appeared to be lower among children residing in Georgia compared with their California counterparts (Fig 4). The HR estimates, however, were imprecise because of the small size of the KP Georgia cohort. The proportional hazard assumption for site variable was violated for GAHT initiation, and for this reason we controlled for site in the stratified Cox model, but the adjusted HR estimates for this variable are not presented (Table 2).

This electronic health record-based cohort study nested in 3 large integrated health care systems revealed that gender-expansive behaviors in most children do not lead to a GDRD and initiation of GAHT. We observed that the rates of diagnosis and GAHT initiation differed across demographic categories of participants and tended to be higher in (1) older adolescents, (2) AFAB cohort members, and (3) non-Hispanic white people. Because of the limited sample size, differences across study sites could not be definitively tested and no conclusion could be reached. In our cohort, data have overrepresentation from subjects who were AFAB, and, although this is also the case in more recent publications,^38,39  the skew toward subjects who were AFAB was not as pronounced as in those studies. This is likely due to the fact that our cohort dates back to 2006, and thus the difference is not as pronounced as in studies including subjects from more recent years.

Our results need to be viewed in the context of similar findings reported previously in European research.^21,28  Two similarly designed, but nonoverlapping, studies performed a follow-up assessment of children treated for gender dysphoria at a specialized clinic in the Netherlands.^21,28  The first study included 77 children who had been referred to a gender-specific clinic between 1989 and 2005 for gender dysphoria at age <12 years at initial presentation.²¹  After an average follow-up of 10 years, 27% of the initial cohort continued experiencing gender dysphoria; however, this result may have been affected by the relatively high (30%) proportion of participants who did not respond to the survey. The authors also reported that individuals with gender dysphoria had more extreme gender dysphoria observed during childhood and were more likely to meet criteria for a gender dysphoria diagnosis during childhood.²¹ 

The second Dutch study originated from the same clinic but sampled a different group of adolescents (n = 127) between 2000 and 2008.²⁸  As in the earlier study, participants received a GDRD at <12 years old and were followed-up at 15 years old or older. Approximately 37% of adolescents in the overall cohort still experienced gender dysphoria at follow-up, although a high percentage of nonrespondents (22%) were counted in the study as no longer experiencing gender dysphoria. It is noteworthy that this study did not record pubertal Tanner staging at enrollment; thus, although participants were all <12 years old, some could have been pubertal and some others prepubertal. Factors associated with unalleviated gender dysphoria included more pronounced dysphoria symptoms and older age at presentation. In addition, continuous gender dysphoria was more common among individuals who were AFAB.²⁸ 

Both Dutch studies found that most participants did not experience gender dysphoria beyond puberty. This result is consistent with our observation that less than one-third of children presenting with GD behaviors received a GDRD and only approximately one-quarter initiated hormone therapy during follow-up. The Dutch researchers also reported a greater likelihood of unalleviated gender dysphoria in children who presented at an older age and among AFAB participants, both results in agreement with our findings.

Perhaps the most important methodologic feature of our study compared with previous research is the use of system-wide cohort ascertainment that was not limited to a particular clinical center. Although many of the transgender members of KP receive specialized coordinated care through Multispecialty Transitions Clinics, this was not an inclusion criterion in the study. Rather the present data set included any member with evidence of transgender status or gender diversity documented in the medical records. Thus, the resulting patient population represents an unselected group with diverse pathways to care. This may explain the relatively low percentage of GD children who received a GDRD, and the relatively high proportion of children who received care without a diagnosis. The deidentified data permitted inclusion of all eligible persons in the analyses because participation did not require subject opt-in. In addition, the keyword-based approach to identify eligible study participants offered a rare opportunity to evaluate the course of events in children at earlier stages of gender-variant behavior, which is rarely possible in specialized clinic-based studies.

The second methodologic feature that distinguishes our study from previously published studies conducted in the Netherlands is the more recent time period (1985–2008 vs 2006–2014). As the size and the composition of transgender populations are rapidly changing, especially in the youngest age groups,⁴⁰  it appears likely that the results may also be affected by sociopolitical and medical advances, increased access to medical care, less pronounced cultural stigma and evolving social norms with differential impact across generations.^41–44  On the other hand, greater awareness of the available gender-affirming care options, without adequate access to this care, may also exacerbate gender dysphoria.

Another notable aspect of the current study is the diverse sample and ability to examine racial and/or ethnic disparities in the care of GD children. We observed that non-Hispanic white participants had an earlier progression to diagnosis and a more rapid initiation of GAHT compared with other racial and ethnic groups. Although relative sparsity of data for specific racial and/or ethnic minority groups precluded a more granular examination of differences by race and/or ethnicity, our findings are in broad agreement with other studies that have shown cultural biases adversely affecting accessing care for gender dysphoria.^45–47  There was also a suggestion that geographical location may play a role; however, the analyses lacked power to evaluate modification by geographic region.

The observed disparities in receiving gender-affirming care may be the result of differences in attitudes toward gender diversity among children, their peers, and their families.^48–50  More broadly, a number of social factors, including political environment, culture, and religious taboos, may influence the degree to which GD children are accepted.^51–55  It is also possible that geographic region and/or ethnicity play a role in social acceptance and ability to access gender-affirming care.^56–58 

The difference in rates of diagnosis and especially treatment initiation between AMAB and AFAB children requires further investigation. Although some reports indicate that transgender boys and transgender girls experience differences in parental acceptance,⁵⁹  a more likely reason is that endogenous puberty occurs later in natal boys, and, because gender dysphoria typically worsens with pubertal progression, children who were AMAB might seek and receive gender-affirming care later than their AFAB counterparts.⁶⁰  Moreover, current Endocrine Society and WPATH guidelines recommend GAHT initiation once Tanner 2 stage is reached; this occurs on average later in AMAB children compared with their AFAB peers.^11,12  One could also argue that the observed disparity in GAHT initiation between AMAB and AFAB participants is attributable to the differences in the cost of therapy. Whereas the use of GnRH agonists is routinely preferred in AMAB youth, in AFAB children and adolescents use of contraception offers a cheaper alternative. Recently, use of GnRH agonists has expanded as it is increasingly covered by insurance plans⁶¹ ; however, the cohort in the current study was followed to the end of 2014, which might have been just before this change in medical practice.

Although our findings indicate that most children with GD behaviors did not receive a GDRD and did not start GAHT during follow-up (3 years on average), these results cannot be interpreted as evidence of long-term outcomes. Thus, it is likely that the proportion of individuals who eventually start gender-affirming care will increase with extended follow-up; however, we expect that some GD study participants will remain dysphoria-free and not require intervention.

It is worth noting that the methodologic features of our study can be viewed as both its strengths and its weaknesses. Because the analyses were based exclusively on the information obtained from medical records, this design precluded collection of patient- and family-reported measures. As such, the degree of documentation by medical professionals can influence the index date and therefore the timing of follow-up initiation. Other limitations of our analyses include the lack of data on social environment or psychological support, pubertal status, and the inability to distinguish children who identify as transgender from those who present with nonbinary or other gender-nonconforming identities. In addition, GD children enrolled in integrated health care systems come primarily from families with health insurance and may not be representative of all GD youth in the United States. On the other hand, this cohort does include patients enrolled in Medicaid plans, allowing that at least some of the study participants come from populations with lower socioeconomic status. Although GAHT protocols do not differ substantially across study sites, the relatively long interval of data collection and follow-up (2006–2016) means that the therapeutic approaches changed over time, and thus children identified earlier in the study may have different pathways to care compared with their counterparts in more recent years.

Our analyses reveal that GD adolescents are more likely to receive a GDRD or hormone therapy compared with younger children. We also found that both diagnosis receipt and treatment initiation were more common among non-Hispanic white children and AFAB children relative to their respective counterparts. The observed differences by geographic locations require confirmation because of the limited sample size.

Taken together, these results indicate that, even in the presence of similar access to care, use and timing of services may differ across groups of GD children and adolescents. With respect to their clinical interpretation, our findings are more likely to inform primary health care providers who first encounter GD children rather than practitioners specializing in gender-affirming care. In future studies, researchers should explore the possible reasons for the observed differences by recruiting a cohort with a wider range of sociodemographic characteristics, especially with regards to race and ethnicity, and include data on parental perceptions of transgender care and pubertal staging. It would be interesting to know how many of the youth presenting to care for GAHT have already socially transitioned, as well as examine deeper the effect of therapy on psychological functioning. Perhaps the most important next step in this area of research is to compare health outcomes and quality of life among GD children and adolescents who began receiving care at different ages. These types of data are needed to inform clinical practice and facilitate development of evidence-based guidelines.

AFAB

assigned female sex at birth

AMAB

assigned male sex at birth

CI

confidence interval

GAHT

gender-affirming hormonal treatment

GD

gender-diverse

GDRD

gender dysphoria–related diagnosis

GnRH

gonadotropin-releasing hormone

HR

hazard ratio

KP

Kaiser Permanente

STRONG

Study of Transition, Outcomes, and Gender

WPATH

World Professional Association for Transgender Health