From January to June 2019, we recruited people presenting for abortion while waiting for their appointment at four abortion facilities located in three states (California, Illinois, and New Mexico). A University of California San Francisco (UCSF)-trained research assistant approached patients in the waiting rooms to briefly introduce the study and invite patients to screen for eligibility on tablets. Patients were eligible if they were 15 years or older, could speak and read English or Spanish, and were seeking but had not completed their abortion. Patients were ineligible for the study if they were known to be pre-medicated with narcotics for a planned aspiration procedure. Eligible patients who provided informed consent completed a self-administered anonymous survey on paper or an iPad. The survey took an average of 20 min to complete. Participants received a $30 gift card to thank them for their time. UCSF’s Institutional Review Board reviewed and approved the study. Additional details including site and patient recruitment details have been described elsewhere [22]. This sample was powered to develop a scale to measure psychosocial burden accessing abortion care.
After drafting the survey instrument based on a review of the literature [23,24,25], the research team sought feedback from 11 experts, including four clinicians who provide abortion care, two abortion counselors, one clinic director, three abortion researchers, and one lawyer who supports youth in navigating the judicial bypass process. Based on their feedback, we added, removed, and modified survey items.
Next, we held cognitive interviews with 11 patients from three San Francisco Bay Area abortion facilities, to ensure that the survey language was clear and addressed relevant barriers to accessing abortion care. We also asked patients’ comfort level answering the survey items and how answering the items affected their mood or stress levels to ensure that study participation was not burdensome. We added the survey item on abortion terminology preference after cognitive testing because some patients indicated discomfort with the word “abortion” and a preference for other terms. Also based on this feedback, we included language at the beginning of the survey acknowledging that people use different terms but that we would use the terms “end this pregnancy” or “abortion” to include medication abortion, aspiration or surgical abortion, D&E (dilation and evacuation), D&C (dilation and curettage), or an induced miscarriage or termination.
After we analyzed the study data, we presented findings from the study to the clinic staff at the four abortion facilities that participated in recruitment for the study. We did not collect identifying data from our patient participants so were unable to share the findings directly with them.
Measures
We assessed our primary outcome, preference on abortion terminology, using the question: “We used the terms ‘end this pregnancy’ and ‘abortion’ to describe what you are seeking today. Given the choice, what word(s) do you prefer to describe it?” Response options included “abortion”, “pregnancy termination”, “ending a pregnancy”, “dilation and curettage (D&C)”, “dilation and evacuation (D&E)”, “no preference”, and an “other” category which allowed participants to write in their response. We coded responses into a 4-part categorical variable which included: (1) abortion, (2) ending a pregnancy, and (3) pregnancy termination for people who selected a single term, and (4) no preference (participants who selected two or more terms or selected “no preference”), which served as our reference category.
To understand whether participants’ circumstances around the pregnancy and abortion may be associated with a preferred term, we included the following independent variables: gestational age (calculated as the difference between the self-reported date of or weeks since last menstrual period and survey date), whether the reason for seeking abortion was because the fetus had a medical condition, pregnancy intention/feelings about the timing of becoming pregnant just before being pregnant, and, as a proxy for anticipated abortion stigma, how worried they were that other people might find out that they ended the pregnancy, which was based on a four-point Likert scale (not at all, a little bit, somewhat, and very much). We chose these independent variables given their documented association with abortion stigma [26]. We also adjusted for the following demographic characteristics: age; race/ethnicity; belonging to a church or religious community; receipt of any governmental financial assistance in the past year (receiving any of the following: Temporary Assistance for Needy Families, WIC, food stamps, social security or disability, another form of government aid); ability to come up with $2,000 in a month if needed; and whether born in the U.S.
In our regression models, we selected reference groups following the guidance of Johfre and Freese [27], using the negation of the variable for binary variables and the lowest quantity group when the variable categorized a quantity. We chose “did not want to become pregnant” as the reference group for pregnancy intention, as other levels of intention “unfold” from this group. Since the meanings of the race/ethnicity categories did not provide a rationale for choosing a reference group, we selected non-Hispanic white because this group had the most frequent responses for not having a preference for a single term.
Analysis
We ran frequencies on participant characteristics and preference for abortion terminology. We conducted bivariate and fully adjusted multinomial logistic regression analyses estimating relative risk ratios (RRRs). Bivariate analyses, using multinomial regression models, assessed relationships between circumstances around the pregnancy and participant demographic characteristics with abortion terminology preference, including clinic site as a covariate to adjust for clustering of observations by site. The fully adjusted model, using multinomial regression, examined whether circumstances around the pregnancy were associated with preference for a single term, adjusting for participant demographics, pregnancy circumstances, and clinic site. We also conducted a sensitivity analysis to test the robustness of our findings with clinic site included in the fully adjusted model as a random effect rather than a covariate. To handle missing covariate data in our bivariate and adjusted multinomial regression models, we used multiple imputation then deletion using chained equation [28]. We ran ten imputations based on the largest fraction of missing values, including for the outcome variable and covariates [29]. We removed observations with missing outcome data in all regression analyses. We conducted all analyses using Stata 15 (StataCorp, College Station, TX).