Achieving Conceptual Clarity in LGBTQ+ Health Research: Measurement, Methodology, and an Application in Reproductive Epidemiology

Abstract

In 2016, the NIH designated sexual and gender minority people (SGM; e.g., lesbian, gay, bisexual, transgender, queer, etc.) as a health disparities population. In recent years, there has been substantial growth in the number of studies on SGM reproductive and perinatal health. The growing interest in this area of research merits revisiting the methodological approaches researchers employ. In Chapter 1, I discuss both best practices and the real-world challenges of measuring and analyzing sexual orientation data using the Nurses’ Health Study 2, a longitudinal cohort study of female registered nurses enrolled in 1989, as a case study. After measuring sexual orientation identity in 1995 and 2009, the cohort added a new identity measure with additional response options and new dimensions of same-sex attraction and same-sex sexual contact to the 2017 questionnaire. The proportion of participants classified as sexual minorities increased substantially with the addition of new measures in 2017 (11%) compared to 1995 (1%) and 2009 (1%). Most of this increase is attributable to the ability to identify heterosexuals with same-sex attractions, same-sex contact, prior sexual minority identity, and/or those who identified as “mostly heterosexual” (a new response option that was previously unavailable). I discuss the considerations researchers must weigh when deciding which measures to use and how, as well as make recommendations for future data collection. In Chapter 2, I outline common errors in covariate adjustment in SGM health research and argue that adopting directed acyclic graphs (DAGs) can help researchers clarify their research question of interest and identify appropriate adjustment sets. To illustrate these points, I simulated 1,000 datasets with a sample size of 10,000 individuals and demonstrated how different covariate adjustment sets for a variety of research questions can produce widely varying estimates. I motivate why covariates that are commonly used in SGM health disparities research (e.g., use of medically assisted reproduction) are mediators, not confounders, and how adjusting for these variables in causal research can induce bias by blocking part of the indirect effect of exposure on the outcome. Next, I illustrate the complexity of mediation analyses with social exposures due to mediator-outcome confounding induced by exposure and compare potential approaches. Then I demonstrate how collider stratification bias can arise from our sample recruitment and selection. Finally, I demonstrate how incorporating heterosexism (i.e., stigma and discrimination) as an unobserved node in our DAG can guide decision-making on appropriate adjustment sets. Finally, in Chapter 3, I apply these methodological considerations to assess sexual orientation-related disparities in pregnancy loss using data from three longitudinal cohorts: the Nurses’ Health Study 2 and 3 and Growing Up Today Study (N=235,214 pregnancies from 85,289 participants). I used log-binomial generalized estimate equation models to compare the risk of pregnancy loss among pregnancies to completely heterosexual participants (reference) to those among heterosexual participants with same-sex attractions/partnerships, mostly heterosexual, bisexual, and lesbian participants. To address confounding, selection, and multiple pregnancies per participant, models were weighted by the product of inverse probability of treatment, inverse probability of censoring, and inverse cluster size weights. Cohort-specific results were combined using fixed-effects meta-analysis. Pregnancies among heterosexual participants with same-sex attractions/partnerships (RR:1.07; 95%CI:1.02–1.13) as well as those among mostly heterosexual (1.26; 1.19–1.34), bisexual (1.46; 1.25–1.70), and lesbian (1.59; 1.36-1.86) participants were more likely to end in a loss than those to completely heterosexuals. Notably, risk of stillbirth was elevated among pregnancies to lesbian (2.58; 1.52–4.37) participants compared to those among completely heterosexuals; stillbirth risk was not elevated among other sexual minority subgroups.