This analysis uses data from a cross-sectional study designed to develop a new measure of psychosocial burden of obtaining abortion care in the US [8]. From January to June 2019, we recruited participants from 4 abortion facilities located in 3 states (California, Illinois, and New Mexico) with minimal abortion restrictions, but due to their geographic location serve people traveling from more restrictive settings. At each site, clinic staff or a research assistant presented patients in the waiting room with a study flyer and asked if they were interested in participating in a study on “the challenges people face trying to access care to end a pregnancy.” We restricted eligibility to those ages 15 years or older, able to speak and read English or Spanish, seeking an abortion that day, and not pre-medicated with narcotics for a planned procedure.

After being introduced to the study by a research assistant and having patients’ eligibility confirmed, interested participants provided electronic consent, completed a 20-min self-administered, one-time iPad survey in English or Spanish, and received a $30 gift card for compensation. The University of California San Francisco, Institutional Review Board approved this study.

We explored consideration of SMA by creating a dichotomous variable to participants’ responses to the question: “Would you consider ending this pregnancy on your own if you are unable to obtain care at a health care facility?” Prior research using this question language demonstrated good comprehension of the phrasing “on your own” and “ending [this] pregnancy” [9]. We considered those who responded, “definitely yes” or “probably yes” to the question as considering SMA and those who responded “probably no”, “definitely no”, or “I don’t know” as not considering SMA.

To identify covariates, we reviewed the literature to identify factors associated with considering self-managed abortion [3,4,5]. We included questions on sociodemographic characteristics, including age, race/ethnicity, health insurance status, and state of residence, specifically whether the participant had traveled from a different state to access abortion care.

We assessed logistical and practical obstacles to accessing facility-based abortion by asking participants whether any of the following delayed them from obtaining care: finding a place that provided abortions (overall and at their gestation), figuring out how to get to the clinic, finding the money for the cost of care, finding money for the cost of travel, needing multiple visits, parental notification or consent requirements, and travel times. We collapsed responses to contrast those selecting “Yes” vs. “No” or “Don’t know” and then summed the number of delays to create a composite score ranging from 0 (no obstacles) to 7 (all obstacles). To further characterize financial circumstances, we asked participants how difficult it was to “find the money to pay to end the pregnancy.” Those who responded “Very” or “Somewhat” (vs. “Not at all” or “A little bit”) were categorized as having difficulty paying for the abortion. If a participant selected “Very much” (vs. “Somewhat,” “A little bit,” or “Not at all”) to a question that asked how worried they were about other people finding out that they were ending the pregnancy, we considered them very concerned about privacy.

To characterize pregnancy circumstances, we asked participants to select the reasons they were seeking an abortion, which included, but were not limited to, the response options, “I'm concerned about my mental health,” “I'm concerned about my physical health,” and “I'm concerned about the health of the fetus.” We collapsed concerns about maternal mental or physical health, resulting in two binary variables summarizing health concerns for the pregnant person and fetus. Certainty about their pregnancy decision was assessed using a 5-item version of the Decisional Conflict Scale (DCS), which includes statements such as “I feel sure about what to choose,” “I am clear about the benefits and risks of each option,” and “This decision is easy for me to make.” Participants respond to each statement on a Likert Scale ranging from “Strongly agree” to “Strongly disagree.” Consistent with guidance from scale developers [10], DCS scores were summed and then scaled to range from 0 to 100, with lower scores reflecting higher certainty. We assessed pregnancy intention using the question, “Thinking back to just before you got pregnant, how did you feel about becoming pregnant?” Those who “didn’t want to be pregnant then or at any time in the future” were classified as having an unintended pregnancy, while those who wanted to be pregnant “sooner,” “then,” or “later” as intended or mistimed. Finally, we asked participants the date or number of weeks since their last menstrual period started and used this to estimate pregnancy duration in weeks.

We present descriptive characteristics on the study sample, overall and by whether they would consider trying to end the pregnancy on their own if unable to get care at a healthcare facility. We test for differences in the distribution of sociodemographic, pregnancy and care-seeking characteristics by consideration of SMA using Poisson regression models that include a fixed effect for recruitment site. Then, we examine the multivariable association between a reduced set of individual sociodemographic, pregnancy and care-seeking characteristics and likelihood of considering SMA, again using Poisson regression models. We chose Poisson models recognizing that when an outcome is common, prevalence ratios represent a more interpretable and conservative measure of association than odds ratios (ORs) [11,12,13]. Given missingness on covariates, we re-ran multivariable analyses after multiple imputation of missing values using chained equations. Multivariable results were similar with and without imputation; we present only imputed results. To further characterize the role of logistical obstacles and consideration of SMA, we also describe the proportion that reported each type of obstacle by whether or not they would consider SMA, and assess whether this difference is statistically significant using a Poisson regression model that includes fixed effects for site. All analyses were conducted in Stata 15.0.