The data described in this analysis were collected as part of a larger multi-method qualitative study exploring constructs of sexual and reproductive empowerment among female Kenyan adolescents and young adults (22). Data collection incorporated the present paper's focus area on contraceptive preferences, experiences, and decision-making, which were designed a priori to inform DCE attribute and level selection. We drew on Helter & Boehler's four stages of attribute development (data collection, data reduction, removal of inappropriate attributes, wording) (21) to provide a structural framework for our methods.

Our approach to the data collection and reduction stages was guided by Brofenbrenner's social ecological theory (23, 24), which is particularly apt in conceptualizing how interpersonal, community, and societal/policy contexts reinforce and influence, and are influenced by, individual-level contraceptive preferences and behaviors. We drew on the social ecological model to create a conceptual map of AGYW contraceptive preferences and their multifaceted influences (Figure 1), which was used to inform the interview guide and eventually to contextualize initial candidate attributes (Supplementary Table). We used qualitative in-depth interviews to gather rich narrative data, followed by structured cognitive interviews for DCE pre-testing and iterative revision.

This study was conducted in urban and peri-urban Kisumu, which is Kenya's third-largest city, bordering Lake Victoria. The predominant ethnic group in the region is the Luo people, and Dholuo and Kiswahili are widely spoken, as is English. We chose to situate the study in Kisumu county for two reasons: (1) our research collaboration's long-term commitment to improving SRH in the region, and (2) as a natural extension of our prior qualitative work in Kisumu and nearby counties indicating a critical need for person-centered contraceptive programs to better serve AGYW within their life contexts (7).

The study was approved by the Kenya Medical Research Institute (KEMRI) Scientific Ethics Review Unit (P00152/4,193), the Kenya National Commission for Science, Technology, and Innovation (NACOSTI/P/21/10,896), and the University of Washington Human Subjects Division (STUDY001172). Given the overall low risk of the cross-sectional survey procedures and potential invasion of privacy with parental involvement, a waiver of parental permission was approved. Written informed consent (age 18–23) or assent (age 15–17) was obtained for in-depth interview participants; verbal consent was obtained for cognitive interviews. All participants received 500 Kenyan Shillings (approximately USD$4) in appreciation of their time.

In-depth interviews: Female AGYW were purposively sampled for maximal variation with quotas for age (50% or more age 15–19 years) and prior contraceptive experience (50% or more with prior contraceptive use). Participants were eligible if they were 15–23 years old; spoke Dholuo, Kiswahili, or English; were sexually active in the last year; had the capacity for pregnancy (not currently pregnant or sterilized); and stated a desire to avoid pregnancy for the next 6 months. Eligible participants were recruited in-person in equal numbers from community venues (youth program meetings, markets, informal gathering spaces) and from an ongoing clinical trial cohort based at KEMRI research clinics (25).

Cognitive interviews: We used convenience sampling to recruit participants using the above eligibility criteria at KEMRI research clinics.

Sample size: The in-depth interview sample size was determined based on research team judgment that 30 interviews would be adequate to generate thematic saturation (26) based on the specific study population and research questions, in line with published guidance (27). Cognitive interview sample size was selected as a reasonable estimate of the number needed to iteratively revise the draft DCE instrument and materials. In-depth interview respondents were not eligible to participate in cognitive interviews.

The research team developed a semi-structured interview guide (Supplementary Material) to explore contraceptive perceptions, desires and preferences, experiences, and decision-making among AGYW, as well as the people and factors at various levels that influence their contraceptive priorities and choices. Interviewers attended a comprehensive training, which included material on research ethics, building rapport with adolescent research participants, research standard operating procedures, and one-on-one practice with the interview guide. We did not conduct pilot testing of the guide, but did iteratively revise the guide throughout the study. Interviewers did not have a prior relationship with participants, and the recruitment and consent/assent processes clearly delineated the researchers and institutions involved, and the goals of the research. Each participant was asked open-ended questions such as: “When you think about choosing a family planning method, what comes to mind for you? What are the most important factors for you when you think about choosing a family planning method?” Subsequently, participants were asked follow-up questions and specific questions relating to candidate attributes (Supplementary Table). All interviews (n = 30) were conducted in Dholuo, Kiswahili, and/or English by two Kenyan trilingual interview team members with extensive qualitative data collection experience (including AD). Each interview of 60–75 min' duration was digitally recorded, transcribed, and translated by the interview team.

Data collection and analysis were conducted concurrently. The analytic team (EKH, OC, SK) developed an initial codebook based on topics and concepts in the interview guide and an initial read of several raw transcripts. All transcripts were triple-coded, and the coding team met regularly to revise the codebook with data-specific concepts, review and resolve coding differences, and discuss emerging concepts. The data were analyzed with an inductive, constant-comparative approach (28) to continuously compare and contrast thematic elements within and between participants. Codes were distilled into analytic categories, and the analytic team summarized these categories, or themes, and the relationships between them in memos. Data were coded and managed in Dedoose software (29).

During and subsequent to data analysis, we iteratively revised the candidate attributes and levels in response to our findings. Based on general guidance in the DCE literature, we aimed to include 8 or fewer attributes with 2–4 levels each to avoid respondent fatigue and to follow recommendations for DCE sample size estimation (30). Attributes that were less relevant to participants' lived experiences and contraceptive decisions were removed. The research team then reviewed, discussed, and revised a final attribute and level table.

The research team drafted language explaining the structure of each DCE question, or choice task, and introducing each attribute and level in Dholuo, Kiswahili, and English. A graphic was chosen to represent each attribute level. We then recruited 15 female AGYW meeting eligibility criteria from KEMRI research clinics to participate in structured cognitive interviews. Interviews focused on comprehension and relevance of introductory material, instructions, and attribute and level wording. Participants completed 2–4 draft choice tasks and provided feedback on the structure and graphics. The interview team took detailed notes to record participant perceptions and feedback, which were summarized in debrief reports. The research team iteratively revised the wording and graphics of the final attributes and levels, as well as the introductory language throughout pre-testing (Supplementary Table).

Our binational (Kenya and the United States) team brought a variety of skillsets and perspectives to the current research. The two-person interview team (including AD) who collected all study data are cisgender Kenyan women, trilingual, and have 5–10 years of qualitative data collection and transcription/translation experience. The interview team had no pre-existing relationship with the participants prior to the study. The analytic team included two Kenyan researchers (OC, SK) with Kiswahili fluency and extensive familiarity with the social, economic, and health systems factors that impact Kenyan AGYW. The primary author (EKH) has collaborated on qualitative reproductive health research in western Kenya since 2008, and the second author (BH) is a health economist with extensive experiencing designing and conducting DCEs; both are based in the United States.

Participants named a variety of priorities and preferences related to choosing and accessing contraceptive methods, and described a variety of interpersonal and community-level influences on the factors that were most important to them. In the following results, we distill the themes into six categories: side effects; effectiveness; user control; privacy; source of services; and cost.

Concurrent data collection and analysis allowed the study team to iteratively assess the most relevant aspects of AGYW contraceptive decision-making and priorities. To minimize cognitive burden to DCE respondents and optimize the precision of preference estimates, our goal was to select fewer than 8 attributes (31). After completing qualitative data analysis, we selected 7 final attributes (Table 2) and removed the other, less appropriate, attributes using a process that combined findings from our qualitative work and multidisciplinary team consensus-building. First, we identified the attributes that were most relevant to AGYW's priorities and experiences, based on the key themes that emerged from the in-depth interviews. In tandem, we dropped 2 attributes (ability to stop method on your own, privacy during method access) that represented concepts or priorities that the majority of participants said were less important to them or were infrequently spontaneously introduced by participants; we also removed an attribute (timing of return to fertility) that overlapped conceptually with another (how long method will last). We chose to include cost in order to allow for calculation of willingness-to-pay. Attribute levels were also iteratively revised during the analytic process, guided by participant perspectives and team discussion.

After attribute selection, we drew on published guidance (11) to develop the structure of the choice tasks. Choice tasks include two full profiles (labeled “Option A” and “Option B”), meaning that all 7 attributes are represented in each profile, and an opt-out option. We defined the opt-out as “no method” in order to allow participants to indicate they would rather use no method and a higher risk of becoming pregnant than use either hypothetical contraceptive profile A or B. Our qualitative findings provided a rationale for including an opt-out: some participants, despite a strong desire to delay pregnancy, found the tradeoffs associated with contraceptive use unacceptable. Figure 2 presents a sample choice task.

Participants in cognitive interviews (n = 15) had a median age of 21 (IQR 19–23); no other demographic data were collected. Interviewers presented the participants with mock-up introductory material (5 in each language) and 4 example choice tasks, and assessed their understanding of the instructions, eliciting feedback on the clarity of language and translations, initial thoughts related to the images chosen to represent the attribute levels, and comprehension related to how to answer the questions. In response to participant comments and detailed interviewer notes, the team iteratively revised the introductory language, the wording of attributes and levels, and the images. For example, the lowest attribute level for cost was initially labeled as “free”, and the image used was a colorful red shape with the word “FREE” in it. AGYW explained that they chose the option with the “free” image in part because they liked how it looked; to avoid distracting survey participants and introducing bias, we removed the image and labeled that attribute level “0 KSH” (Kenyan shillings). Additionally, we received feedback that the image of a male partner used in the initial sample survey questions was distracting, as AGYW tended to interpret his facial expression as upset or suspicious. This image was replaced with one where the male partner's expression was perceived as more neutral. Finally, we used the cognitive interviews to assess AGYW's reaction to the various attribute levels for cost to establish the most appropriate upper limit of cost. AGYW were willing to choose the profile that cost 500 KSH when the other attribute levels in that profile were favorable, indicating that participants found it reasonable under specific conditions.