Intervenable Factors Associated With Suicide Risk in Transgender Persons: A Respondent Driven Sampling Study in Ontario, Canada

Greta R. Bauer; Ayden I. Scheim; Jake Pyne; Robb Travers; Rebecca Hammond


BMC Public Health. 2015;15(525) 

In This Article


Study Sample

Data were collected via respondent-driven sampling (RDS) in 2009–2010 as part of the Trans PULSE Project, a community-based research study in Ontario, Canada's most populous province. To be eligible, participants had to live, work or receive health care in Ontario, be age 16 or over, and identify as trans, broadly defined. Participants held a range of gender identities and were not required to have begun a social or medical sex/gender transition.

RDS is a method that combines systematic chain-referral sampling with statistical analysis strategies that account for differential levels of connectedness as well as non-independence within recruitment chains. This method was chosen as trans persons in Ontario were at least moderately networked, either in person or electronically, but constituted a hidden population that could not be randomly sampled. Moreover, given the lack of data on trans people, there was community motivation to participate, including to recruit others. Sampling began with 16 initial seeds, with 22 additional seeds enrolled once it was certain that sufficient recruitment chain length could be obtained. Each participant could recruit additional participants using a tracked coupon system, wherein three coupons were issued to each participant for distribution to eligible individuals. This allowed the research team to identify the network structure (i.e., who recruited whom) while allowing participants to remain anonymous if desired. Data collection continued for one year (n = 433), with a maximum chain length of ten waves. A network diagram of the sample is presented in Fig. 2. Our analysis is based on 380 participants (87.8 %) who completed items on past-year suicide ideation and attempts. Ethics approval for the project was provided by Research Ethics Boards at The University of Western Ontario and Wilfrid Laurier University.

Figure 2.

Network diagram of sample of trans people in Ontario, Canada (n = 433). Circles = male-to-female (MTF) spectrum, including genderqueer individuals. Triangles = female-to-male (FTM) spectrum, including genderqueer individuals. Grey = no past-year serious consideration of suicide. Blue = past-year serious consideration of suicide, but no attempt. Red = past-year suicide attempt(s)

Survey and Measures

Measures were derived from self-report data collected through a multi-mode survey (with visually identical online or paper versions), which could be completed anonymously and typically took 60 to 90 minutes. The survey was pre-tested by 16 members of the project's Community Engagement Team for content validity, clarity and length. A copy of the survey is available online (

Outcomes. Past-year suicidal ideation was coded from items that asked participants "Have you ever seriously considered committing suicide or taking your own life?" and "If yes, has this happened in the past 12 months?" Similar follow-up questions for those who indicated ideation asked "Have you ever attempted to commit suicide or tried taking your own life?" and "If yes, did this happen in the past 12 months?"

Background variables.Age, disability status, parental status, and chronic illness/pain were indicated by participant self-report. Gender spectrum was coded into two categories that each represent a spectrum of identities. Female-to-male spectrum individuals were those assigned female at birth, but who now identify as men or another non-female identity (e.g., genderqueer, bigender); similarly, male-to-female spectrum participants were those assigned male at birth who now identify as women, trans girls, two-spirit, or other non-male identities. Gender fluidity was coded as a separate dichotomous measure from a check-all-that-apply identity item into two categories: holding a more conventional gender identity (male or primarily masculine, female or primarily feminine), or only gender-fluid or third-gender identities. Ethno-racial group was comprised of three groups: participants were grouped as Aboriginal if they indicated they were First Nations, Métis, or Inuit, or another Indigenous ethnicity; non-Aboriginal participants were classified as white or racialized based on an ethno-racial background item, a write-in question describing background, and indication as to whether or not they were perceived by others as a person of colour. In Ontario, the term "racialized" is preferred over racial minority, visible minority, person of colour or non-White as it expresses race as a social construct rather than a description on perceived biological traits".[46] Immigration history was approximated by self-report of whether the participant was born in Canada. Sexual orientation was coded as sexual minority based on either identification as gay, lesbian or bisexual/pansexual, or having a past-year sex partner of the same gender. Region of province was classified based on first letter of postal code. Strong religious upbringing was defined as responding "quite" or "extremely" to the question "How religious or faith-based was your upbringing?" Childhood abuse prior to age 16 was indicated in two items describing physical and sexual abuse experiences. Major mental health disorders were coded using a self-reported checklist based on any prior diagnosis (e.g., bipolar disorder, schizophrenia, borderline personality disorder). Depression and anxiety disorders were excluded, as they are more likely to result from intervenable factors under study, and we hypothesized they would partially mediate their effects on suicide ideation or attempt; if we were to control for these mediators we would then remove a portion of the causal effect and would produce estimates for the effects of our intervenable factors as enacted only through pathways other than depression or anxiety.

Intervenable Variables – Social Inclusion. Social support was assessed using the 19-item Medical Outcomes Study Social Support Scale.[47] Items were averaged, with possible scale values from 1 to 5 (Cronbach α = 0.97). Strong support for gender was coded for participants' parents by combining self-report of either "very strong" support for gender identity or expression experienced from a parent or parents, or (for those who were not out to their parents) self-report of expectation of such support. Other gender support variables were similarly coded. "Peers" averaged completed items on friends, co-workers, or classmates, and "Leaders" on schools, teachers, supervisors or employers. Current religiosity or spirituality was measured on a 6-point Likert scale ranging from "not at all" to "extremely". Having one or more identity documents concordant with lived gender was coded for the participant sub-group who had socially transitioned and lived full-time in a male/masculine or female/feminine gender, based on having at least one document with a sex/gender designation ("M" or "F") matching one's lived gender. Identity documents included federal and provincial identification (e.g. drivers license, passport, Indian Status card, military ID).

Intervenable Variables –Transphobia. Transphobia was assessed using an 11-item scale;[21] it included items on enacted and internalized stigma as well as victimization, such as police harassment or feeling that being trans hurt or embarrassed one's family. Items were summed, with possible values ranging from 0 to 33 (Cronbach α = 0.81). Transphobic harassment and violence was defined as self-report of physical or sexual assault for being trans (analogous to assault as a hate crime); report of verbal harassment or threats related to being trans, but not of assault; and report of none of these.

Intervenable Variables –Transition. Medical transition status was self-reported as having completed a medical transition (self-defined), being in the process, planning to transition but not yet having begun, and either not planning, being unsure, or indicating that the concept of "transition" did not apply. Completing a medical transition involved varying hormone and/or surgical treatments.[16] Hormone use was self-reported. Social transition status was coded as living in one's felt gender full-time, part-time, or not at all. Being perceived as cisgender (non-trans) was coded for those living full-time in a non-fluid gender, based on a survey item that asked how often others knew you were trans without being told.

Statistical Analysis

Since eligible trans persons who knew fewer potential participants were less likely to be sampled than those with large network sizes, RDS II weights[48] were calculated based on the inverse of each participant's degree (personal network size), rescaled to sum to the total sample size of 433. This approach has been shown to produce frequency estimates that are asymptotically unbiased.[49] Here weighted statistics can be interpreted to apply to the population of networked (know at least one other eligible person) trans people age 16 and over in Ontario. All analyses were weighted, and adjusted for clustering by shared recruiter to account for non-independence within recruitment chains. Weighted frequencies or means, along with 95 % confidence intervals, were calculated for all background and potential intervention variables using SAS version 9.3.[50]

Prior to regression, simple imputation was used for background variables to reduce data loss in a complete case analysis. Of 380 participants with outcome data, 30 were missing data for one background covariate and 5 for more than one.

Multivariable logistic regression models were fitted for each intervenable factor variable separately, controlling for all background variables, using SAS-callable SUDAAN version 11.0.[51] Thus, 13 logistic regression models were fitted for the 13 intervenable factor variables for the suicidal ideation outcome, and an additional 13 models for the suicide attempt outcome; all models similarly controlled for background variables, but not for other intervenable variables, as mediated pathways between these variables are unknown. A domain analysis was used in order to limit the second analysis to the sample subgroup with suicidal ideation.

All categorical variables were dummy-coded to allow for independent estimation of effects for each group. Reference categories were chosen so that contrasts reflected effects in the direction of desired potential intervention effects (e.g., protecting from assault, increasing social support). Continuous variables were entered into the models as continuous. Since average marginal risks are estimated for specific points in the distribution (as if the entire trans population – standardized to the background variables – was at that level), the 10th and 90th percentiles were used as points for estimation since participant values did not necessarily cover the range of a scale and the extremes may not be achievable with any intervention (e.g., no transphobia would require not even having heard once that trans people were not normal).

Model-standardized risks and risk ratios for past-year suicidal ideation and attempts (standardized to the weighted sample or subsample on all background factors) were estimated for each of the thirteen intervenable variables under the counterfactual conditions where all participants were exposed or alternately all were unexposed to a dichotomous intervenable factor, or to an intervention target and reference level of a categorical factor. For continuous variables, these counterfactual risks were estimated for the 10th and 90th percentile levels. Standardized risks and risk ratios, and their 95 % confidence intervals, were calculated by Graubard and Korn's method[52] using the ADJRR option in SUDAAN PROC RLOGIST.[51] Where effects were statistically significant for a variable, counterfactual population attributable risks (cPARs) were then calculated by taking the weighted prevalence of the outcome in the sample (factual) and subtracting the model-standardized risk (counterfactual) based on having the entire population at intervention target levels of the exposure. These represent the potential proportion of the trans population affected (e.g., outcome averted) under the intervention target condition (e.g., all have parental support for gender). Counterfactual population attributable risk proportions (c%PARs) were similarly estimated by dividing cPAR by the weighted prevalence of the outcome in the sample; these represent the proportion of cases potentially averted within the trans population. C%PARs for different variables will sum to more than 100 %, given that a case of ideation or suicide attempt could be prevented through multiple means.