The study is being implemented in four counties, Nairobi and Mombasa as the intervention counties and Uasin Gishu and Nakuru counties are serving as the comparison counties. The intervention counties are the first and third highest urban cities with HIV incidence among youth aged 15–24 years in Kenya ⁷ . The comparison counties were selected for their comparability to the intervention counties based on the following characteristics: comparable high population level ²⁰ and urban settings with informal settlements that have vulnerable AGYW.

This study employs a concurrent mixed-methods design to facilitate triangulation of findings between the quantitative and qualitative methods. The study was implemented in three phases—formative phase, implementation phase, and endline phase—conducted at different study points. The study objectives are: i) to assess changes of the level of awareness and attitudes of the community members and health workers toward HIV and pregnancy prevention services for AGYW; ii) to assess levels of comprehensive knowledge and utilization patterns for PrEP and pregnancy prevention services among AGYW; iii) to examine the barriers and facilitators of uptake of HIV and pregnancy prevention services among AGYW; and iv) to evaluate changes in uptake of integrated SRH and PrEP services among AGYW.

Quantitative data consists of two household surveys, one involving AGYW (utilization survey) and another the general population (awareness survey), and collection of routine data on utilization of HIV prevention and pregnancy prevention services abstracted from registers in 11 health facilities. Qualitative data consists of key informant interviews (KIIs) with relevant county health managers, focus group discussions (FGDs) with health service providers, and FGDs and in-depth interviews (IDIs) with AGYW. Quantitative and qualitative data analysis were conducted simultaneously comparing findings as they emerge. Comparison of formative and endline indicators will document changes during the study period. During the intervention phase, the proposed package of interventions was executed in Nairobi and Mombasa counties while the comparison counties continued with the current standard of care delivering HIV and pregnancy prevention services for AGYW. The study lasted for 12 months.

Study participants include AGYW, the general population (male and female), health service providers, and county health managers. Details of each of these study populations are outlined below:

Participant eligibility: The inclusion and exclusion criteria are detailed in Table 1.

The overall expected outcomes of the package of interventions include increasing demand for services, supporting delivery of AGYW-responsive services, and strengthening use of data. The specific activities and their expected outcomes are summarized in Figure 2.

a) Utilization survey: AGYW were recruited at the household level within the catchment areas of the study health facilities. Trained research assistants established eligibility using a recruitment script and, thereafter, administer consenting procedures in a private location within the precincts of the household. Those who agreed to participate in the study provided written consent and only one AGYW per household will be invited to participate in the survey.

b) Awareness survey: General population participants were recruited at the household level within the catchment area of study health facilities. Community health workers guided the trained research assistants to the selected households with an adult male or female. The research assistants used a recruitment script to invite the potential participants. Those who met the eligibility criteria and expressed interest provided a written consent to participate. Only one participant per household participated and the research team alternated between male and female participants to ensure a balanced sex representation.

a) FGDs and IDIs for AGYW: Participants were invited by facility staff who approached AGYW in their community using a recruitment script. The AGYW were purposely selected to include those who had used PrEP, those who use contraceptives, and those who had neither used PrEP nor contraceptives. Consenting for FGDs was conducted as a group by a trained qualitative researcher on the day of the FGD, and those who agreed to participate provided written consent. Consent for the IDIs was administered individually in private.

b) FGDs for health providers: Participants were recruited through facility in-charges guided by a recruitment script. Providers who expressed interest were linked to the qualitative researchers who confirmed eligibility and invited participants for the FGDs. Consent was administered as a group on the day of the FGD, followed by individual written consent.

c) KIIs for county health managers: County health managers were approached by a qualitative researcher through a phone call or a face-to-face meeting and were invited to participate in a KII. If the manager accepted the invitation, the qualitative researcher scheduled a convenient time and agreed on the venue of the interview. Consenting procedures were conducted in a private location.

Quantitative: The sample size for the utilization survey was powered to provide estimates of the difference of PrEP and pregnancy prevention services’ utilization among AGYW aged 15–24 years. The sample size for the general population was powered to provide estimates of the difference in PrEP awareness between the intervention and comparison counties. The sample sizes for both surveys was powered to detect a 10 percentage point difference in the difference between the endline and formative assessment in the intervention relative to the comparison counties at 80% power and a 95% confidence interval. In both surveys participants were recruited from clusters consisting of several households in a community unit affiliated to the selected facilities. We assumed an average cluster yield of 30 participants meeting the study criteria and an intra-class correlation of 0.03, which resulted into a design effect (DE) of 1.87. After accounting for a 10% non-response, the sample size was adjusted to about 264~270 for each of the intervention and comparison counties. The cumulative sample size was 540 for each population type and also for each of the two data collection periods (formative and endline assessments) resulting into a total sample size of 1,080 AGYW and 1,080 general population participants. The formula for sample size calculation is:

m = ( Z 1 − α 2 + Z 1 − β ) 2 ∗ P 1 ( 1 − P 2 ) + P 2 ( 1 − P 2 ) ∗ [ ( 1 + ( n − 1 ) ρ ) ] Δ 2 ^{22, 23}

where P ₁ is the probability of an event in the intervention group, and P ₂ the probability of an event in the comparison group; Δ represents the difference in proportions, P ₁− P ₂; and design effect = 1+ (n−1) ρ where n is the number of individuals per cluster and ρ the intra-class correlation.

A summary of the parameters and sample size estimates for the utilization and awareness surveys are summarized on Table 2

Qualitative study population: FGDs, IDIs, and KIIs were conducted during the formative and endline phases where the participants were purposively selected at community, health facility and county level. The study included eight FGDs among AGYW and four with health providers during the formative phase. Each FGD constituted an average of 10 participants. IDIs were conducted among 20 AGYW allocated equally to the study counties. Four health managers in each of the four counties were interviewed during the formative assessment. A similar number of FGDs, KIIs, and IDIs were conducted during the endline assessment to exhaustively saturate the themes related to each group of participants. A summary of the estimated sample sizes for the quantitative and qualitative data is presented in Table 3.

We employed multi-stage sampling for the two surveys and purposive sampling for the FGDs, KIIs and IDIs.

This sampling approach included selection of clusters, households, and participants at the household level. The sampling procedure at formative and endline assessments were similar.

Step 1. Selection of the clusters

During formative, community units (CUs) affiliated with the selected health facilities were used as clusters. All CUs were listed to form a sampling frame and simple random sampling method used to select CUs/clusters to be involved in the study. Nine clusters were selected for the intervention and a similar number for the comparison counties.

Step 2. Selection of the households

Households for each selected cluster was line-listed by the study team with support from the CHVs in each CU as guided by the community health strategy team in each county. The updated list of households (with unique serial identifiers) served as the sampling frame for the selection of households in the second stage. Random sampling was used to select 100 households in every cluster as described below:

Determining the number of households

Based on the 2019 census distribution ¹⁹ , we will compute the expected distribution of the population in the respective age categories, as shown in Table 4.

Based on these computations, and assuming a household size of four, the number of households needed per population type is described below:

General population: It was estimated that we would be able to find at least one respondent in each household. Therefore, to reach the estimated sample size, we needed a ratio of 1:1 for the number of households to sample size. Considering that every cluster was assigned an average of 30 participants, it was possible to reach 30 participants in 30 households in a cluster. We included 10 additional households per cluster to account for household refusals and non-traceable eligible household members. Therefore, we required 40 households for each cluster. Using the sample size estimate, we needed 360 households to reach the required sample size for general population at both formative and endline phases.

AGYW: It was estimated that, at least one AGYW would be found in every three households. Therefore, in order to obtain the required sample size for the AGYW, we needed a ratio of 3:1 for the number of households to sample size. Every cluster was assigned an average of 30 AGYW, hence it was to be possible to reach the 30 AGYW participants in 90 households per cluster based on the population distribution in Table 4. We included a reserve of 10 households to each cluster to account for household refusals and non-traceable eligible household members and thus we required 100 households for each cluster. Using the sample size estimate, we needed to cover 900 households for the AGYW to reach the required sample size in each arm at both formative and endline assessments.

Step 3. Selection of the survey participants

The survey participants (AGYW and general population) were selected at the household level. For the AGYW, we visited all the households selected until we reached the sample size. For the general population, participants were drawn specifically from households with an odd number on the line list. Only one AGYW and one general population respondent participated in the survey per household. In case two or more eligible participants were found within a household, the study team used to paper-pick method where the participant who picks a “Yes” was interviewed. For the general population survey, the research team alternated between male and female participants to ensure a balanced sex representation.

This approach was used to sample participants for FGDs, KIIs, and IDIs:

Diverse data collection procedures were employed to generate quantitative and qualitative data. The same procedures conducted at formative, were conducted at endline assessments. Surveys were conducted by trained research assistants at private locations within the precincts of participants’ households. Data was collected using an interviewer-administered survey questionnaire in English or Kiswahili. Data was collected using Research Electronic Data Capture (REDCap) software ²⁴ hosted on password-protected tablets and the data submitted to the server daily.

FGDs, IDIs, and KIIs were moderated by trained qualitative researchers using guides and the appropriate language (Swahili or English) for each target audience. Sessions were conducted in private rooms within public health facilities, community halls, or offices of the county health managers. The researchers took handwritten notes and audio recorded all the sessions. The handwritten notes and the downloaded audio data files were properly labelled, encrypted, and securely stored in lockable cabinets and password secured computers. A brief demographic profile consisting of age, gender, education, marital status, and current living standards of the FGD, KII and IDI participants was also obtained.

During the intervention phase, data on utilization of PrEP and family planning (FP) services was collected on a monthly basis from the health facilities in both the intervention and comparison counties through a data abstraction form.

Different data collection tools were used for each sub-study described above. Table 5 outlines the variables that were collected using each of the study tools.

To ensure data quality, tools and the methods for data collection were pre-tested before the actual study and necessary changes made. The data collectors were trained on the process to ensure quality data is captured. All the completed questionnaires were checked for completeness, legibility, and consistency and any errors noted corrected before uploading of the data to the server. Continuous on-the-job training for service providers and focal persons in charge of data in facilities, monthly data verification, data review meetings, and quarterly data assessments was done during the intervention phase, to ensure data quality was maintained.

We did not anticipate that this study would cause any harm. However, there was minimal risk related to breach of confidentiality in case information discussed in the FGDs was shared by participants outside the research setting. Further, it was possible for participants to experience emotional distress, embarrassment, or discomfort discussing their sensitive information with the study team. We also identified that participants might have felt uncomfortable sharing their personal sexual and behavioral experiences. The study team carefully tracked the occurrence of any of such events.

Given that the study was conducted during the COVID-19 pandemic era, the study implemented Kenya medical Research institute guideline ²⁵ and MOH guidelines to prevent and address any potential threat of transmission of the COVID-19 virus to study participants and staff. These includes: 1) provision of alcohol-based sanitizers, 2) enforcement of donning of three-ply masks by participants and staff, and 3) maintaining physical distancing precautions including conducting data collection in private outdoor spaces.

The outcomes of interest for this study include service utilization of PrEP and pregnancy prevention services among AGYW and knowledge and attitudes toward PrEP and contraceptives among the general population. For the service utilization, we will compare difference in the proportion of AGYW reporting PrEP and contraceptive method use between the intervention and comparison counties, expecting a 10% difference between formative and endline assessment. We also expect a 10% difference in knowledge and attitudes in the general population in the intervention and comparison counties comparing formative and endline assessment.

Data reduction techniques will be used to summarize the observed variables for data collected through Likert scales for the knowledge and attitude questions into a few dimensions through latent variable modelling using the ltm R package ²⁶ . Component internal consistency and reliability used for computing the composite scores will be evaluated by calculating Cronbach’s α. Pairwise associations corresponding to two-by-two contingency tables for all possible pairs will be computed and variables with a low association with other items will be dropped leaving variables with Cronbach’s α of 0.78. Factor scores will then be generated by fitting the Rasch model ²⁷ or multiple item response theory analysis for outcomes, which have more than two levels. Depending on the distribution of factor scores, a new outcome variable will be generated, which will be either dichotomous or ordinal and used in new analysis.

Descriptive analysis of quantitative variables will be conducted using measures of central tendency (mean, median) and measures of dispersion (range, standard deviation) as appropriate. We will then conduct an exploratory analysis to compare outcomes by sociodemographic, cultural, and economic characteristics (age, education, religion, marital status, wealth quintile, etc.) using the Rao-Scott chi-square test to determine if there is any association. We will further check the balance between the intervention and comparison counties in terms of potential confounding factors. To determine the relative importance of these factors on the outcomes, generalized estimating equations, univariate and multivariate logistic regressions will be used to adjust for clustering within the clusters. Results from the regression analysis will be reported as odds ratio (OR) and 95% confidence intervals. All statistical analysis will be considered statistically significant if the p-value is less than 0.05.

Qualitative analysis will commence by verbatim transcription of the audio-recordings from the FGDs, KIIs, and IDIs. Data collected in Swahili will be translated to English and back translated to confirm accuracy of the meanings during translation. The transcripts will be compared with the typed handwritten notes for concurrence. The data will be imported to NVivo 11.0 for analysis. First, open coding will be conducted to unearth all the emergent codes from the data. This will be augmented by a priori codebook derived from the qualitative guides. A master codebook will then be developed to make sure the codes are applied consistently but also allow the qualitative analysts to propose new codes as they emerge. Next, all the data will be coded using the master codebook. Thereafter, axial coding will be conducted using the constant comparison method to identify linkages and relationships in the data, and begin structuring the data according to emerging concepts. Finally, we will conduct selective coding using constant comparison techniques to organize the findings in an organized framework, drawing on the research questions, existing literature, and seeing what new relationships of phenomena emerging from the data.