This study was conducted in Soweto, South Africa, from April 2018 to December 2018. Soweto is a large township within the city of Johannesburg with a total population of 1.3 million [32]. We used RDS, a well-established sampling method for populations for which there is no sampling frame, to calculate the proportion of people who have ever attempted abortion as well as the 1-year and 5-year cumulative incidence of abortion among women aged 15 to 49 years in Soweto.

With the assistance of well-known community-based organizations that provided a range of services, including, but not limited to, reproductive health services in Soweto, we recruited 11 women to serve as our initial seeds for RDS recruitment. These women were of various ages, income levels, and sexual and reproductive health experiences, including those with prior abortion. In accordance with the RDS methodology, seeds were members of the target population and purposively selected by our research team to initiate recruitment chains. We selected 2 study sites based on recommendations from our community-based organization partners about accessibility within the community and considerations of confidentiality—specifically, locations where people commonly gather or seek a range of services not specific to reproductive health. Seeds presented at one of the 2 study sites and completed an interviewer-administered questionnaire on their experiences with abortion. After completing the questionnaire, seeds were given 3 coupons to refer eligible peers to the study. Recruitment coupons contained information about the eligibility criteria, instructions on how to schedule an interview, and information about the study incentives. Potential participants contacted the study phone number via SMS text message or call and answered a short screening questionnaire to assess their eligibility. Participants were eligible if they identified as a woman; were aged between 15 and 49 years; spoke English, Tswana, isiZulu, Sotho, or Xhosa; and lived in Soweto. It is important to acknowledge that people of all genders have and need for abortions, and not all of them identify as women; some identify as men or another gender, and some people who identify as women do not have the capacity to carry a pregnancy. In the context of this study, we recruited people who identified as “women” and have referred to the study population accordingly throughout this paper.Eligible participants were invited to schedule an in-person interview at one of the 2 possible sites. The participants aged <18 years arrived at the interview with a signed parental consent form. Upon arrival at the study site, eligibility was confirmed, and consent was obtained. The consented participants completed an interviewer-administered questionnaire—administered by trained members of the study team who were South African women of reproductive age and spoke the above languages—and received 3 recruitment coupons. Participants received a participation incentive of 75 South African rand (approximately US $4) and a recruitment incentive of 50 South African rand (approximately US $2.50) for each eligible participant they successfully recruited. Participants returned to the study site to collect their recruitment incentive and completed an additional survey on their experiences participating in and recruiting for the study. Recruitment was tracked using coupon numbering.

Ethics approval for this study was obtained from the Human Sciences Research Council Research Ethics Committee in South Africa (REC 10/18/11/15: Experiences of women who self-induce abortion in Soweto, South Africa). The amount of compensation for participation in the study was arrived at in extensive consultation with the Human Sciences Research Council. No identifying information was collected from the participants. Study-related documents, including coupons, did not disclose abortion incidence as the primary aim of the study.

The main instrument in our study was a quantitative survey with questions on demographic characteristics, social network composition, health behaviors, sexual history, pregnancy history, and experience with abortion. The follow-up instrument contained questions on the recruitment process, including questions on refusals. Categories for out-of-facility providers and methods were informed by existing literature [28,30,33] as well as findings from formative research that comprised in-depth interviews conducted with 19 women from Soweto who had attempted to terminate a pregnancy outside the formal health setting [34]. In addition, once draft instruments were developed, we conducted cognitive interviews with 5 participants from the formative research phase (all of whom had consented to be recontacted) to ensure that the instruments were understandable and that the answer choices were appropriate. Minor refinements to the terminology and answer choices were made following the cognitive interview phase.

The primary outcome of interest for this study was the proportion of people who ever attempted abortion (facility-based or out-of-facility abortion), measured as the weighted proportion of women in the study who reported attempting at least 1 abortion in their lifetime. In addition, we calculated the 1-year and 5-year incidence of abortion attempts. We defined out-of-facility abortion as any abortion attempt, successful or unsuccessful, that did not take place under the supervision of (nor with a prescription from) a physician, nurse, or other advanced practice clinician at a government-run or privately operated health care facility. We defined facility-based abortion as any abortion attempt, successful or unsuccessful, that took place under the supervision of a physician, nurse, or other advanced practice clinician at a government-run or privately operated health care facility. The key sociodemographic variables used to compare the representativeness of our sample with the source population (women of reproductive age in Soweto) were age, educational attainment, employment status, and home language. Consistent with the RDS literature [21,35], we assessed network size using the following question: “How many women of reproductive age who live in Soweto do you know, who also know you, that you have seen in the past week?”

We additionally collected data on network characteristics such as recruiter-recruit relationships and recruitment experiences to assess whether several RDS recruitment assumptions were met in this study; full findings from this methodological assessment are published elsewhere [36].

Using the method proposed by Salganik [37] for calculating the desired sample size for a sample proportion, we arrived at a minimum sample size of 834 participants, which enabled us to detect a proportion of people who attempted abortion of 50% (417/834; maximally conservative estimate), with an SE <0.03 and assuming a design effect of 3. Data management was conducted using R (version 4.0.2; R Foundation for Statistical Computing) [38] and Stata (version 14; StataCorp LLC) [39]. We used RDS Analyst [40] to examine recruitment patterns, equilibrium (the point in recruitment at which the sample proportions of sociodemographic characteristics stabilize), homophily, waves of recruitment, and mean network size and compute weighted estimates of our primary outcomes. We used the RDS-II estimator to estimate the population proportions of demographic characteristics and our primary outcomes [41]. The RDS-II estimator reweights the sample population to account for homophily, the tendency of participants to recruit other participants who share similar characteristics [14]. Participants are weighted by the inverse of their degree (social network size); for example, participants with a degree of 10 would be given a weight of 1/10. We used imputed visibility for our measure of degree (effective network size), which incorporated self-reported social network size, the number of successful recruits, and the time to recruit to estimate each participant’s inclusion probability. Visibility was imputed using the impute.visibility_mle function in RDS Analyst [40]. We calculated 95% CIs using 1000 bootstrap replications.

Between April 4, 2018, and December 17, 2018, 849 eligible participants were recruited into the study. Recruitment occurred over 36 weeks, and the longest recruitment chain lasted 17 waves, with a mean of 6.6 recruitment waves for active seeds. A total of 2 seeds did not recruit any participants, and 56.5% (480/849) of the sample originated from 1 seed. Approximately one-third (n=837, 36.8%) of the 2277 distributed coupons were returned. Recruitment patterns based on lifetime experience of abortion are shown in Figure 1. A methodological assessment of RDS assumptions and recruitment dynamics has been previously published [36], and the key findings are summarized below. There was strong homophily (chi-square test for independence, P<.05) for age, educational attainment, employment status, and lifetime experiences with abortion, suggesting a strong tendency to recruit individuals with similar characteristics to theirs as compared with random recruitment. Sample proportions for age, home language, educational attainment, and socioeconomic indicators stabilized (reached equilibrium) by approximately 300 to 500 participants, well before our estimated sample size.

The unweighted sample proportions for the selected demographic characteristics, along with weighted population proportions, are reported in Table 1. Table 1 presents the population estimates of the selected demographic characteristics based on publicly available data. The unweighted median age was 27 (IQR 22-36) years. Approximately one-fifth (20.2%, 95% CI 17.4%-23.1%) of the target population were currently in school, and most had at least some secondary education (52.5%, 95% CI 49.1%-56.0%) or completed secondary education (39.6%, 95% CI 36.2%-43.1%). Most were unemployed (83.1%, 95% CI 80.5%-85.8%). Although a statistical comparison of the RDS-II sociodemographic estimates to the estimated source population estimates is not possible, the sample characteristics are largely similar to the source population for all variables, except for employment status.

The RDS-II estimates of the proportion of people who ever attempted abortion was 12.1% (95% CI 9.7%-14.4%; Table 2). A total of 7.1% (95% CI 5.4%-8.9%) reported a facility-based abortion, and 4.4% (95% CI 3.0%-5.8%) reported an out-of-facility abortion. The true design effect for the main outcome, any abortion attempt, was 1.14. Most participants (RDS-II estimate: 61.8%, 95% CI 58.4%-65.2%) reported that their best friend had an abortion (not displayed in the tables).

Because of likely underreporting, we present the unweighted proportions for various abortion experiences. A total of 106 out of 849 (12.5%) participants reported at least one abortion attempt at any point in their lifetime, and 9 (n=106, 8.5%) of them did not provide any subsequent information about their experiences. Among the remaining 97 (91.5%) participants who reported an abortion attempt and answered additional questions related to their abortion experience, 85 (88%) attempted an abortion once, 8 (8%) reported 2 abortion attempts, and 4 (4%) reported ≥3 abortion attempts. When asked about their most recent abortion attempt, 60 (62%) participants reported that they went to a health care facility. Among those who went to a health care facility, 9 (15%) attempted to do something on their own to end their pregnancy before seeking facility-based health care, most commonly taking a laxative. At the health care facility, 28 out of 60 (47%) participants reported taking medications for abortion, 10 (17%) participants reported a surgical procedure, and 2 (3%) participants did not know what method was used to end their pregnancy. Of the participants who received medications (n=28), 2 (7%) did not have a complete abortion and continued with their pregnancy. Of the participants who went to a health care facility (n=60), 17 (28%) did not ultimately end up having an abortion because they decided that they wanted to continue with their pregnancy, they were counseled to continue with the pregnancy, or their gestational age was beyond the legal limit.

Of the 97 participants, 37 (38%) of the participants who did not report going to a health care facility for their most recent abortion attempt reported using methods such as laxatives, aspirin, strong tea or coffee, pesticides, bleach, or combinations of the above. Of these 37 participants, 22 (59%) reported successfully terminating their pregnancy.

In our study, we explored the feasibility of applying the RDS methodology to estimate the proportion of women of reproductive age who have ever attempted abortion in Soweto, South Africa. The estimated proportion of ever attempting an abortion of 12% and 1-year incidence of 9.1 abortion attempts per 1000 women of reproductive age in our study likely represents a substantial underestimation of the actual abortion experiences in this study population [27]. Although no directly comparable measures exist, recently published, country-specific estimates of abortion incidence report an annual estimated 30 abortions occur per 1000 women of reproductive age in South Africa, representing a figure nearly 3 times the magnitude of the comparable estimate from our study [42]. We posit that this underestimation represents a failure of the RDS method to generate more reliable estimates of the abortion incidence in our study.

We previously published a methodological assessment of whether several RDS assumptions were met in these data [36]. In that paper, we found that although the approximation of sampling with replacement was met, the participants did not consistently report the same degree, nor did they randomly recruit from within their social network. It is likely that the failure to meet the assumptions yielded a sample with different employment characteristics than those of the target population, which was not resolved by standard RDS methods. However, without gold standard abortion estimates for the target population by sociodemographic characteristics, it is challenging to assess the impact of failing to meet these assumptions on inference for abortion. Although it is plausible that some of the underestimation of abortion may have been due to the overrepresentation of unemployed participants in the sample, it is more likely to have been due to underreporting.

This study highlights the limitations of RDS in measuring abortion. Although the social networking literature is lacking on the subject of abortion, public health evidence suggests that those who have abortions outside the formal health sector communicate with members of their social network to obtain information about self-managed or community-based abortions [43-45]. We hypothesized that RDS could offer a previously untested alternative method to more accurately measure abortion incidence by similarly relying on peer recruitment to help reduce the underreporting of abortion. However, it is possible that because we recruited from a general population of women of reproductive age, peer recruitment operated in the opposite direction in our study; if participants were recruited into the study by members of their social network who they would not want to know about their prior abortions, they might have been less likely to report their abortions to the researchers conducting the study. In this context, it is notable that most participants in our study reported that their best friend had had an abortion—potentially indicating a willingness, as seen in other studies, to discuss the abortion experiences of others but not themselves. In addition, as we lack representative sociodemographic data on reproductive-aged women localized to Soweto, we were unable to directly validate whether RDS sampling generated a sample with demographic characteristics similar to those of the overall population of women of reproductive age living in Soweto. However, based on the best available estimates (population-based data from Gauteng Province and the city of Johannesburg, where Soweto is located), our sample differs substantially from the source population, particularly with respect to employment, even after adjusting using RDS inference procedures.

However, the limitations of our study should be considered in the context of the strengths. We successfully recruited a large sample of women of reproductive age, demonstrating that RDS can be used to recruit a sample of participants who are willing to participate in a study about reproductive health, answer questions about abortion, and participate in peer recruitment. We hope that the lessons learned from our study will be instructive to future researchers exploring the use of novel sampling approaches for measuring abortion.

Despite the failure of RDS to generate more reliable estimates of abortion incidence, it may be a method best suited for sampling when selecting on stigmatizing characteristics (injection drug use, men having sex with men, sex work, and now, abortion), as it has most commonly been applied. For example, RDS could be used to sample a population with out-of-facility abortion experience and estimate the proportion of that population that has experienced one or more outcome of interest (eg, using medication abortion, seeking health care in the formal health sector, or experiencing complications). Other population size estimation methods could be deployed to arrive at estimates of prevalence in an RDS study that is specific to abortion experiences.

It is conceivable that RDS could reduce underreporting of abortion if it were deployed to estimate abortion incidence among a highly socially networked population (potentially in humanitarian settings, among sex workers, etc). However, it is also possible that asking questions about stigmatizing experiences in any general population–based survey will be subject to underreporting—especially when interviews are administered face-to-face. Using tools such as Audio Computer Assisted Self-Interviewing and other technologies has been shown to reduce underreporting in studies of some stigmatized behaviors and could prove useful in the context of abortion as well [46,47].

Accurate estimates of abortion incidence within and outside formal health settings are vital for developing targeted and effective programs, policies, and interventions to increase the access to safe abortions. In certain highly networked populations, RDS may prove to be a useful tool in the toolkit of abortion researchers, but to ensure that people seeking abortion have the information and support they need, regardless of where or how their abortion takes place, more work is needed to develop and validate tools that more accurately measure not only the incidence of abortion but also the experiences, quality, and outcomes of abortions.