Study area and study design
This study analyzed the baseline data of the evaluation of the “In Their Hands” (ITH) program. Three partners—Marie Stopes Kenya (MSK), Triggerise, and Well Told Story (WTS)—implemented the ITH program in 18 counties across Kenya to increase adolescents’ use of high-quality sexual and reproductive health services. The 18 counties were prioritized because they have high levels of teenage pregnancy, the highest unmet need for contraception among adolescents, and high rates of new sexually transmitted infections (STIs) and human immunodeficiency virus (HIV) infections [27]. As part of the project's evaluation, a baseline survey and qualitative study were undertaken in Homa Bay and Narok, out of the 18 counties involved in the main program because they have the highest rates of unintended pregnancy in Kenya. The study adopted a convergent mixed-methods cross-sectional design involving the concurrent collection of both qualitative and quantitative data between September and October 2018.
Study population and sampling approach
The study involved adolescent girls aged 15–19 years who were residents and members of the sampled households in the selected counties for at least 6 months preceding the survey. A sample size of 1885 was calculated based on Cochran's formula for categorical data, but we successfully interviewed 1840 adolescent girls for the quantitative study. For this analysis, we included adolescent girls that have begun sex at the time of the survey—a total of 1110 respondents from both Counties. The sampling procedure involved a purposive selection of two counties with the highest prevalence of teenage pregnancy from the 18 counties where the ITH program is implemented. We then selected three sub-counties in each of the two counties. In each of the three sub-counties, we selected three wards based on the ITH affiliated health facilities' distribution. For each of the health facilities sampled for each ward, we identified seven and six catchment villages served by the health facility. Accordingly, we sampled 22 villages in Narok county and 24 villages in Homa Bay county. We conducted household listing in each of the villages and identified households with adolescent girls. We randomly selected only one adolescent girl for the interviews from each household with adolescent girls.
For the qualitative component, we purposively identified and interviewed 45 adolescent girls, 25 from Narok county and 20 from Homa Bay (varying by age, occupation, marital status, location, education level). Of the 45 adolescents interviewed, 19 (10 and nine from Narok and Homa Bay, respectively) met the criteria for inclusion in this paper—being pregnant or having had a baby at the time of the study. Community mobilizers identified adolescent girls in the study community, and we screened them to ensure that they met the inclusion criteria for the study. These included being a teenage girl aged 15–19 years, usual residence in any sampled communities two study counties. We defined “usual residence” as living in the household continuously for 6 months before the study.
Instruments and procedures
Quantitative data were collected from a representative sample of adolescent girls living in urban and rural ITH areas to understand adolescents’ access to information, use of sexual and reproductive health (SRH) services, and SRH-related decision-making autonomy before implementing the intervention. We used an interviewer-administered structured questionnaire to collect the quantitative data using the ODK-based SurveyCTO platform, whereby interviewers used tablets in face-to-face interviews. We trained research assistants for five days on the study protocol, the study tools, and the informed consent process. Following training, we pre-tested the tools among adolescent girls in the Korogocho informal settlement in Nairobi county, which was not part of the study. We used the pilot test to check for consistency, sensitiveness, appropriateness, readability, and ease of understanding of the question; the feedback helped improve the quality of the tools. Piloting also helped in testing the quality of the instrument as programmed in SurveyCTO for the quantitative component. A team of 19 experienced research assistants (17 for quantitative and two for the qualitative) conducted the interviews.
The in-depth interviews with adolescent girls were conducted mostly in the Swahili language and partly in English, Dholuo, and Maasai languages, depending on the respondent's preference. We audio-recorded all interviews using digital recorders after obtaining permission from the interviewees. On average, each interview lasted an hour. We conducted all interviews in private spaces to guarantee confidentiality and allow respondents to freely voice their views and perspectives on contraceptives, sexual activities, and unplanned pregnancy, which are sensitive issues. We developed an interview guide for the study covering the mentioned topics.
Ethical considerations
Amref Health Africa Ethics and Scientific Review Committee (AMREF-ESRC) approved our study protocol and materials (AMREF-ESRC P499/2018). Also, Kenya’s National Commission for Science, Technology and Innovation (NACOSTI) granted the research permit. Additional approvals were sought from local administrators and Ministries of Health and Education in the respective counties where the study was conducted. In the case of participants younger than 18, both parents/guardians and adolescents signed informed consent and assent forms, respectively, before interviews took place. However, married adolescents (considered emancipated) and those aged 18 years or older signed informed consent forms.
Variables and measures
The primary outcome measure in this study was unintended pregnancy, which we define as not wanting to have or ambivalent about having a child or becoming pregnant at the time of index pregnancy. Responses were categorized into wanted, wanted to wait, wanted no more, and did not mind either way, but later re-categorized into the binary category of intended and unintended pregnancy for binary logistic regression analysis. The ambivalent cases were combined with the unintended category.
The covariates included in the study are individual-level factors (such as age, marital status, employment status, educational attainment, school attendance status, having ever used any contraceptives, and ownership of mobile phones), household /family-level factors (parents living status), and community-level factors (county of residence, and place of residence). These variables were included in the analysis based on existing literature on factors associated with unintended pregnancy [3, 28,29,30,31]. Age was measured as a continuous variable by asking respondents to state their age at their last birthday. We asked all respondents if they were currently married or not and if they own a mobile phone. Also, we measured parent living status by asking participants to indicate if any or both of their parents were alive or dead. Responses were classified as both parents alive, both parents dead, and only one parent alive. Place of residence was classified as rural or urban areas based on the population and amenities in the residence community. Employment status was measured by asking respondents if they have engaged in income-generating activity in the last six months preceding the interview. Binary choices of “yes” and “no” were provided. We also asked whether participants had dropped out or still in school. Lastly, we asked them to indicate the highest level of education attended and their highest grade completed. We classified education level into primary or less, and secondary or higher for the bivariate and multivariable analysis.
Analytical approach
Based on evidence from previous studies that showed that estimating the prevalence of unintended pregnancy among adolescents, in general, may lead to biased estimates because a large proportion of them are not sexually active [32], we limited our analysis to 1110 adolescent girls who have initiated sex. Descriptive statistics-frequency counts and percentages, and inferential statistics-Pearson Chi-Square and Logistic Regression were performed. To examine the factors (individual, household, and community) associated with unintended pregnancy among adolescent girls, we fitted both adjusted and unadjusted regression models. The unadjusted model was the baseline model used to examine the relationship between each variable and unintended pregnancy. The adjusted model was used to investigate each individual level factor's overall effect after controlling for other relevant covariates (family and environmental level factors). The analysis was performed at a 95% confidence interval limit, and a P-value less than 0.05 was deemed to be statistically significant. We used Stata’s survey design data analysis feature to account for the complex sampling employed while also adding sampling weights.
All in-depth interviews were transcribed and translated into English. To ensure the translation's accuracy, we used two translators who understood the communities’ local languages and the English language. The transcripts were then verified against the related audios by two members of the study team, who are fluent in the Swahili language and English Language. The qualitative data coding was performed, using NVIVO, by two of the experienced qualitative researchers in qualitative methods. Emerging codes were compared and discussed among the research team, and the consensus was reached on the codes to include in the study. We used the thematic approach in developing the codebook and coding scheme and in the analysis.