Mental Health Outcomes in Transgender and Nonbinary Youths Receiving Gender-Affirming Care

This cohort study investigates whether gender-affirming care is associated with decreased depression, anxiety, and suicidality among transgender and nonbinary youths.

"Over the past 2 weeks, how often have you been bothered by thoughts that you would be better off dead or of hurting yourself in some way?" (Item-9 from the PHQ-9) Covariates

Mental Health Therapy
A readiness assessment is when the patient and their family meet with a mental health professional before starting any medical treatment. Other than having an assessment, are you receiving ongoing mental health therapy?

Tension with Caregivers
There is tension around my gender identity or gender expression… (check all that apply) • …between my parents or guardians • …between me and one or more of my parents or guardians • …between me and my extended family During the past 12 months, did you: 1. Drink any alcohol (more than a few sips)? (Do not count sips of alcohol taken during family or religious events) 2. Smoke any marijuana or hashish? 3. Use anything else to get high? ("Anything else" includes illegal drugs, over the counter and prescription drugs, and things that you sniff or "huff") With response options: yes, no, and I don't know Connor-Davidson 10-item Resilience Scale (CD-RISC 10) 1. I am able to adapt when changes occur. 2. I can deal with whatever comes my way. 3. I can see the humorous side of things when I am faced with problems. 4. Having to cope with stress can make me stronger. 5. I tend to bounce back after illness, injury, or other hardships. 6. I believe I can achieve my goals, even if there are obstacles 7. Under pressure, I can focused and think clearly 8. I am not easily discouraged by failure 9. I think of myself as a strong person when dealing with life's challenges and difficulties. 10. I am able to handle unpleasant or painful feelings like sadness, fear and anger. With response options: not true at all, rarely true, sometimes true, often true, true nearly all the time, and I don't know.

II. Generalized Estimating Equation (GEE) Model Specification
GEE is a marginal model and models population averages (compared to mixed-effect models which are conditional and can model subject-specific effects). We specified the following GEE models to estimate the average change in the outcome variable ( ) at each time point ( ) relative to baseline ( 0 ) (Model 1) and the association between the exposure ( ) and outcome (Model 2) adjusted for -many baseline covariates ( 0 ).

Model 1:
( We allow the exposure (receipt of PB/GAH) to vary over time, where indicates the month, and thus use an independent working correlation structure. This model assumes there are no time-varying covariates associated with the exposures and that the exposure is exogenous. A visual schematic of this model is included below in eFigure 1, the counts and percentages of participants in the exposure group at each timepoint is included in eTable 2, and the prevalence of the outcome variables over time stratified by exposure group is included in eTable 3.

B. Interpretation
We can interpret these findings to suggest that (1) the observed OR of 0.40 could be explained away by an unmeasured confounder that was associated with both the PB/GAH and the moderate to severe depression by a risk ratio of 2.56-fold each, above and beyond the measured confounders, but weaker confounding could not do so, and (2) the observed OR of 0.27 could be explained away by an unmeasured confounder that was associated with both the PB/GAH and the moderate to severe depression by a risk ratio of 3.25-fold each, above and beyond the measured confounders, but weaker confounding could not do so. This is evidence that our findings are robust to a moderate to high degree of unmeasured confounding, since "In the context of biomedical and social sciences research, effect sizes ≥2 or 3-fold occasionally occur but are not particularly common; a variable that affects both treatment and outcome each by 2-or 3-fold would likely be even less common." 3 In observational studies, unmeasured confounding and lack of exchangeability pose the greatest barrier to drawing causal inferences from observational cohort studies. In addition, there are notable pitfalls in overly relying on pvalues for the interpreting the significance of results. For instance, studies with a large sample size often have the statistical power to precisely estimate associations and obtain very small p-values; the p-value may be made arbitrarily small by increasing the sample size, even for small effect sizes. In contrast, the E-value depends on the magnitude of the association; it cannot be made arbitrarily large simply by increasing the sample size. Thus, bias adjustments, such as calculating the E-value, assess robustness of study findings to unmeasured confounding, thereby offering an important supplement to p-values.

A. Disaggregated Exposure Variable
We separately examined the association of PB and GAH with the outcomes of interest, although we a priori did not anticipate being powered to detect statistically significant associations due to our small sample size and the relatively low proportion of youth who accessed PB (n=19).

B. Restricting Analysis to Youth Age 13-17 Years Old
We restricted our analysis to minor youth age 13-17 (n=90), since they were subject to different laws related to consent and pre-requisite mental health assessments.

C. Dichotomous Outcome for Depression Based on the PHQ-8
We conducted sensitivity analyses using the PHQ-8 score, 4 which is equivalent to the PHQ-9 with item-9 regarding self-harm/suicidal thoughts removed. We conducted these analyses in order to determine whether item-9 was driving any associations between moderate to severe depression since we analyzed self-harm/suicidal thoughts as a separate outcome. For these analyses we define moderate or severe depression as a PHQ-8 score ≥10.