Study design

The full study protocol is available in S1 Appendix. We conducted a prospective pre-post assessment of static, mobile and mixed models each implemented in a purposively selected cluster of wards in Siaya and Migori counties of Kenya for a total of six clusters. Wards are Kenya’s smallest electoral units within each county with a typical population of 10,000–20,000 people. Since program inception, all the three models evaluated in this study had been implemented seamlessly across the country at the discretion of implementing partners but without formal assessment focused on sustainability. Under the evaluation protocol, the target age band for all three models was boys aged 10–14 years. For safety, 10 years was the lower age limit for PEPFAR-supported VMMCs outside infancy at the time. The 10–14 year age band was chosen for this evaluation because of: a) its rapid contribution to maintaining coverage among older, sexually active age groups as the boys age; b) its intrinsically high demand for VMMC, e.g., 60% of all Kenyan VMMC clients were aged 10–14 years in 2017 [10]; c) logistical efficiency of accessing them through schools, given that Kenya has 72% net male primary school attendance [18]; and d) the necessity of circumcising adolescents for at least the next decade regardless of strategy, because even with adoption of an infant circumcision strategy, adolescent MC would still remain available to cover infants past two months who would not safely get MC until they reach the lower age cutoff. In each study cluster, the VMMC services were reorganized to prioritize boys 10–14 years and to be provided solely through one of the three models. Across all study clusters, clients above 14 years who presented for VMMC were offered the service but were neither actively recruited nor counted towards the evaluation targets. Mobilisers focused on activities reaching younger adolescents such as direct meetings with schoolteachers, parents, and village leaders, and leveraging health promotion events planned in the wards by ministry of health team. They were refreshed monthly and their outputs reviewed weekly to ensure fidelity. It was envisioned that exclusive focus on 10-14-year-olds in clearly defined clusters would make it possible to measure any efficiency or program quality gains associated with each model, which are critical for sustainability.

Each study cluster consisted of 2–6 contiguous wards where only one service delivery model was implemented. The clusters were purposively selected through consensus with the county Ministries of Health and the implementing partners operating locally. Factors considered in selection and assignment of models to clusters were: a) to leave ‘buffer’ areas (not implementing any study model) between different clusters; b) to avoid including in the study cluster wards that border areas with large traditionally circumcising populations, from where many adolescents would potentially cross over to seek VMMC as an alternative to cultural circumcision; and c) to include one sparsely populated rural cluster for mobile model, one peri urban cluster with moderately dense population for mixed model and one densely populated urban cluster for static model.

1. A. In the clusters assigned a ‘static’ model, general clinicians stationed in a static health facility offered VMMC to clients who come requesting it. Outreach and mobile services were suspended. The VMMC teams determined the number and distribution of static sites sufficient to meet all VMMC needs of the cluster. Mobilizers were trained to refer 10–14 year old potential VMMC clients to these sites, as well as any older men who requested the service.
2. B. In the clusters assigned a ‘mobile’ model, a dedicated mobile VMMC team was responsible for maintaining VMMC coverage over a large catchment area via year-round outreach visits to multiple venues. Teams had a ‘base’ site but followed a set monthly schedule of service delivery in outlying health facilities and community sites within the cluster. Clients aged 15+ years were served at the ‘base’ sites. Clients aged 10–14 years were referred by mobilisers to the mobile/outreach services only.
3. C. In clusters assigned a ‘mixed’ model, providers offered both static and mobile service models. Teams chose their strategy mix between the two models freely. Adolescent-targeted campaign-style demand creation and services were conducted at opportune times in the school year.

The national routine VMMC data collection tool used in the study clusters was expanded to include evaluation-specific variables such as the ward of residence for each 10-14-year-old client circumcised. The evaluation was based on numerical target achievement and expenditure data plus key informant data in each model area, as well as consensus lessons learned from the investigators and implementers. The broad mix of outcomes was needed to cover the wide spectrum of elements that are potentially important to program sustainability and to explore for additional considerations needed to make VMMC sustainable.

VMMC services were provided by the implementing partners already responsible for these services in the study areas: Center for Health Solutions (CHS) in Siaya, and University of Maryland Baltimore (UMB) in Migori through subgrantee Impact Research and Development Organization (IRDO). These partners were responsible for reporting service delivery and expenditure data from each study cluster. Jhpiego, an international academic partner providing technical support and coordination for VMMC services in Kenya, conducted data collection from key informant interviews and consensus lessons learned from the investigators and implementers as well as analysis. Jhpiego also tracked any administrative markers of the progress of VMMC service transition from implementing partners to the county ministries of health, such as staff employment and inclusion of VMMC in government strategic documents and budgets.

In the analysis described here, primary outcomes in the baseline period (November 2016-October 2017) and Year 1 (actually the 10-month period December 2017-September 2018, truncated to align with the PEPFAR fiscal year) were compared. November 2017 was a planning period falling between the baseline and year one implementation periods.

Baseline conditions, interventions and data collection

At baseline, all selected study clusters had ongoing VMMC services, provided through delivery approaches selected by partners but typically more similar to the ‘mixed’ model than to the others used in the study. Typically, these involved both year-round services performed by staff at their static facilities and mobile or outreach services provided in tents or peripheral health facilities, respectively. In keeping with national age prioritization, they emphasized recruitment of men 15–49 years [19], but circumcised any age-eligible boy or man on request and often held school-based campaigns.

During the baseline year, November 2016-October 2017, partners collected routine programmatic data regularly reportable to PEPFAR, including client numbers and ages. Notably, the baseline year was also the service startup year under new CDC/PEPFAR cooperative agreements for the partners working in both counties. Partners and the county MoHs also maintained routine internal expenditure records on the county without disaggregation to the study cluster levels. These expenditures captured by activity type using internal categorization systems were reported to the project evaluation team and retrospectively aligned with the study’s activity categories and designated as recurrent vs. investment by the investigators. Recurrent expenditures included administration, medical and non-medical supplies, program supplies, communication, vehicle maintenance and renovations, sensitization meetings and supervision, training of service providers including refresher trainings, salary for contracted staff, transport and personnel costs, utilities (e.g. water bills, electricity bills, etc.), expenditures on technical working groups, allowances and expenditures on County Health Management Teams (CHMTs) and Sub-County Health Management Teams (SCHMTs), supportive supervision, logistical coordination with schools and other sources of adolescents for recruitment, transportation, and backup supply chain support. Investment expenditures included renovating or improving minor theatres in the county health facilities, renovating central sterile services department (CSSD) in the health facilities, building incinerators, improving water supply and management and distribution systems for VMMC commodities, information technology infrastructure, office equipment and medical equipment.

At the study launch in November 2017, each study cluster was assigned one of the three service delivery models targeting boys 10–14, replacing routine service delivery: Each model was implemented by a team operating in partnership with its county health department according to the Kenya national VMMC guidelines [20] and following WHO standards [21]. Each team included a clinical officer or nurse VMMC performing the surgery, an assistant, a VMMC counsellor and an infection prevention officer. Standard national VMMC demand creation strategies were implemented. These included referrals from satisfied clients and peer mobilisers, distribution of education and communication materials in communities and schools, and interpersonal communication approaches to both clients and their parents, all focused on young adolescent clients.

Data collection and outcomes

Programmatic data was captured using paper-based data collection tools which were later transferred by a trained staff member into a spreadsheet using standard study forms and categories on a monthly basis, and then provided to study coordinators based in the counties. Expenditure data was directly collected by economic study team staff in collaboration with MoH and implementing partner staff retrospectively, reviewing their primary expenditure data together once per project year for the full period. Metrics were collected separately for every study cluster in each county. These, and their collection processes, were:

• Primary performance outcome: Number of VMMCs achieved among 10-14-year-old residents of study clusters, as a percentage of the cluster’s annual target. Client age was collected in routine intake data, and providers were trained to document self-reported client residence (inside vs. outside the study cluster) through chart flagging. This was based on where they reported spending the most nights in the past six months, excluding boarding school. These data, and VMMC numbers, were summarized and submitted through a monthly paper study form for each cluster. Annual targets for study clusters were calculated based on each cluster’s 10-14-year-old male population projected to 2017 from the 2009 national census [22]. The goal was to achieve 90% coverage over a two-year period (essentially completing the scaleup phase in the model areas among the target population), from a baseline of 68% in Migori and 76% in Siaya obtained through expert consensus. Annual targets for all clusters ranged from 1277 to 3302.
  Routine program data from the baseline period was not stratified by client residence, and therefore the comparator when considering changes in overall performance from baseline is the total number of 10-14s circumcised in Year 1 (Y1).
• Secondary performance outcome: Percentage of 10-14-year-old VMMC clients who were residents of the model area. This was collected to provide context, in the event of a cluster performing well numerically but without achieving its coverage targets because a large proportion of clients lived outside the cluster.
• Primary expenditure outcome: Unit expenditure (UE) per VMMC. The numerator was total VMMC expenditure from all sources of support on all clients, because it was not feasible to attribute expenditures to specific age categories. The denominator was total MCs across all ages, to avoid an artificial UE decline from the increasing proportion of MCs done in 10-14s by Y1. Decreases noted in UE by Y1 were therefore expected to represent a combination of newfound efficiencies and existing inherent efficiencies from targeting adolescents. Activity-specific data were collected in both fine and broad categories, and sources of funds were attributed at baseline and for each cluster and activity. A financial approach was taken where actual expenditures were available; an economic approach was used for equipment already purchased or donated prior to the baseline period.
• Secondary expenditure outcome: Percent of total expenditure in each broad recurrent spending category supplied by the county Ministry of Health. The fine categories used in initial data collection were collapsed by the local economic coauthors into prespecified broad categories in reporting MoH contributions to percent recurrent expenditure, to more intuitively display which areas MoH was preferentially investing in. Categories were: human resources for routine services; human resources for technical assistance and quality assurance; commodities/ consumables; facility operation, transport, waste management; and demand creation. Annual targets for rising Ministry contributions were set for each category, ranging from 10–80% in Y1 and all reaching 100% by year 5. MOH expenditure that was supported by the IP was collected at the IP level (classed as IP expenditure).
• Additional expenditure outcome: Though not prespecified as a major outcome, identification of major expenditure drivers and comparison across models was also of interest in understanding impact of model type on financial aspects of sustainability.
• Key respondent perception outcome set: mean Likert scale scores across all respondents on each question in a given questionnaire, covering perceptions of multiple aspects of program sustainability. Respondents were grouped into three different levels, with a different self-administered anonymous key respondent survey given to each level, after consent. These were: leadership level (National VMMC MoH staff, county-level health leadership, UN agency representatives on the national VMMC task force), project site level (VMMC site staff and managers, facility in-charges), and community level (a school administrator, a teacher, a religious leader, a local leader and five residents per ward). For leadership respondent recruitment, questionnaires were distributed to all staff of county health management teams within the HIV domain and all national MoH program managers. For site-level recruitment at all participating facilities, questionnaires were distributed to facility in-charges at health centers and medical superintendents at higher-level facilities. For community member recruitment, village elders, teachers and church leaders were convenience sampled, one per ward, based on willingness to participate, seniority and accessibility. Thus except for community-level respondents, the study did not target specific numbers but rather sought to obtain a census of all eligible respondents.
  Surveys covered respondent perceptions of VMMC services in the study cluster across the appropriate subset of the four key PEPFAR domains of sustainability for each respondent group. Domains included multiple elements and each element potentially contained multiple questions. Domains are defined and their contents summarized in Table 1; full questionnaires are in S2–S4 Appendices. The Civil Society Engagement subdomain, for example, includes questions about the degree of commitment of diverse community groups to the program, the proactive efforts of program leadership to communicate updates to community groups, and community engagement efforts in planning and evaluating the program. The questionnaires were distributed at study launch for the baseline period, and a year later for the Y1 period. Individual respondents could be the same individuals at both time points if they met the recruitment criteria at both, or could change between time points. Respondents received questionnaires and instruction sheets from study coordinators, who collected the completed questionnaires after two working days. Non-responders were reminded three times, and continued non-response led to automatic replacement. Responses were along a Likert scale from 1 (“never” or “strongly disagree”) to 5 (“always” or “strongly agree”).
• Additional outcomes: To gather information on the predetermined policy and strategic metrics of sustainability (summarized in S5 Appendix) based on the PEPFAR SID, the evaluation team also reviewed relevant national and county-level VMMC documents including policies, strategic plans, scopes of work and others.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Sustainability domains assessed through key informant surveys: Definitions, elements, and levels assessed (selected from PEPFAR SID tool).

https://doi.org/10.1371/journal.pone.0252725.t001

Finally, changes in implementation approaches over time were reported through biweekly to monthly calls between study leadership and implementer leadership; and implementers, lead evaluators, and investigators together generated consensus on challenges, refinements and lessons learned through dedicated discussions.

Data management, storage and analysis

Completed data collection tools were verified by research coordinators and electronically transferred to the database by the evaluation data officer. Monthly routine database backups were automatically scheduled. Paper forms and records were kept under lock and key. A random sample of 10 records per model was verified by the research officer to ensure accuracy of the details captured. Qualitative data was entered into the REDCap software, which was programmed to include logic and consistency checks to minimize data entry errors. Summary statistics were also used to ensure no missing data; any missing data was resolved via calls with persons responsible for the initial reporting. Data was stored by Jhpiego on the Kenya National AIDS & STI Control Programme (NASCOP) encrypted secure database.

Y1 data analysis addressed:

• Performance: Monthly and monthly cumulative achievements among target-aged residents, as a percent of target, were tracked for each model area in Siaya and Migori. Additionally, though not a major performance outcome, mean total monthly VMMCs (total VMMCs performed in all clients during the project year, divided by number of months included in that year) were compared between the baseline and Y1 years. This analysis was done to provide an intuitive check on the reasonability of performance, as a large absolute drop from baseline to Y1 would have required explanation even if performance were adequate against targets. The monthly means rather than annual totals were used to adjust for the different lengths of the two “years” compared, as there were 12 months in the baseline year and 10 in Y1. The comparison used a simple bivariate linear regression with generalized estimating equations with exchangeable correlation to account for clustering at the level of the facility. The p-value generated was for the t statistic testing the hypothesis that the coefficient for the project year is equal to 0, with a significance threshold of p < .05.
• Expenditure: A program-level approach was taken. Key descriptive analyses done on the model-area level included baseline and Y1 unit expenditure, and percent contribution by MoH to expenditure in total and stratified by major expenditure domain. Assumptions used in expenditure estimation in each county, and mapping to the categories displayed in the results section, are in S6 Appendix.
• Key respondent perceptions: For each model area and for all model areas combined, mean Likert scores for each subdomain (calculated as the mean of scores across all questions in the domain) were tested for change between baseline and Y1 using an unpaired t-test for indicators with sample size over 30. For indicators with sample size less than 30, the Wilcoxon rank test was used to test for differences. Descriptive analysis was also done to identify the individual questions with the five highest and lowest scores for each questionnaire at each time point.
• Challenges, refinements and lessons learned: These were distilled from weekly and one-off meeting notes and summarized by the primary investigator.

Ethics statement

This study was approved by the Johns Hopkins University School of Public Health Institutional Review Board, and the Maseno University Ethics Review Committee. This project was reviewed in accordance with CDC human research protection procedures and was determined to be research, but CDC investigators did not interact with human subjects or have access to identifiable data or specimens for research purposes.

Performance outcomes

Targets and performance of study models are shown in Table 2, covering the primary and secondary performance outcomes.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Performance in total and against targets, by county and model, baseline and Year 1.

https://doi.org/10.1371/journal.pone.0252725.t002

Performance against targets was higher in Migori than in Siaya in the mixed and mobile models. In both counties, the mixed model outperformed the others against targets. Mean monthly MCs performed in 10-14-year-olds (not shown), compared between project year periods, increased from baseline in the static and mixed models, either significantly despite the shorter year (Siaya) or non-significantly (Migori), but decreased significantly in both counties’ mobile models.

Substantial variations in monthly performance over the Y1 period were also noted. Fig 1 shows monthly number of MCs in the target population performed by model in Migori and Siaya, with key influencing events marked. The overall pattern was characterized in both areas by spikes in performance during campaigns in April, August and December, but also by gradually increasing base volume. Performance also dropped at the beginning of the PEPFAR year due to staff diversion from clinical work to planning activities by the implementing partners.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Monthly performance (10-14-year-old residents) by model with key influencing events, Siaya and Migori, Year 1.

https://doi.org/10.1371/journal.pone.0252725.g001

Expenditure outcomes

Total VMMC expenditure in Migori rose from approximately $906,000 at baseline to $1,009,011 in Y1 despite the shortened year (approximately $1.2 million if corrected for the shortened year). In Siaya it fell from $2.4 million to $2.1 million (approximately $2.5 million if also corrected). Total investment expenditure in Siaya rose from $11,800 at baseline to $32,000 in Y1, all by the PEPFAR-supported implementing partner. Total investment expenditure dropped in Migori from $53,000 at baseline to none in Y1.

Unit expenditures for each model and county in the baseline and Y1 periods are shown in Table 3, along with Y1 volumes, covering the primary expenditure outcome. The key columns for comparison between baseline and Y1 are shaded.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Volume and unit expenditures at baseline and in Year 1.

https://doi.org/10.1371/journal.pone.0252725.t003

Baseline unit expenditures were essentially the same between counties. Unit expenditures dropped from baseline to Y1 in the mixed model (both counties) and the Migori static model; they rose substantially in the mobile model and the Siaya static model. Comparison with achieved volumes shows that models with higher volumes had lower unit expenditures, except that the Migori static and Siaya mixed models had low expenditures with low volumes.

Primary expenditure drivers in Migori were personnel (both years), travel (both years), commodities (baseline) and administrative costs (Y1). Primary expenditure drivers in Siaya were personnel, travel and commodities (both years).

Primary Y1 expenditure drivers in the model areas with substantial expenditure increase were personnel (Siaya static) and commodities, personnel and travel (Migori mobile). Primary Y1 expenditure drivers in areas with substantial decrease were personnel and travel (Migori mixed) and personnel and commodities (Siaya mixed). Indirect personnel were a more important cost driver in Migori than in Siaya; in Migori these costs were paid mostly to implementing partner HQ staff, and increased in Y1 as additional HQ support was added, where in Siaya they were paid mostly to county health management team staff. Full expenditure category breakdowns are in Fig 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Expenditure breakdowns, baseline and Year 1, Migori and Siaya counties.

https://doi.org/10.1371/journal.pone.0252725.g002

Percents of total expenditures covered by the MoH for each county by category (secondary expenditure outcomes for the study) are shown in Table 4. MoH contributions dropped from baseline to Y1 in most categories and in total, except for small increases in facility expenditures (Migori) and assumption of responsibility for demand creation (Siaya). Total MoH expenditures also dropped. In Siaya they dropped from $147,700 (14.77 million KSh) at baseline to $95,000 (9.5 million KSh); at the same spend rate over a full 12 months, corrected total Y1 expenditure would have been $114,000. In Migori they dropped from $116,000 (11.6 million KSh) at baseline to $62,300 (6.23 million KSh); corrected 12-month Y1 expenditure would have been $66,600. The shortened Y1 period should not affect MoH percent contributions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Percent of recurrent VMMC expenditure provided by MoH by spending category, baseline and Year 1.

https://doi.org/10.1371/journal.pone.0252725.t004

Key respondent perception outcome

Mean key informant response scores for baseline and Y1 are displayed in Table 5, covering the key respondent perception outcome. Mean baseline and Y1 scores were similar across models (maximum differences of 0.64 and 0.61 respectively). Higher scores signify stronger mean respondent agreement with a set of positive statements across aspects of the domain or subdomain.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Mean key informant response scores across sustainability domains, baseline and Year 1.

https://doi.org/10.1371/journal.pone.0252725.t005

On the national level, greatest strengths (highest mean individual question scores) for Y1 were in stakeholder understanding of the project goals, communication between the project team and county health teams, routine convening of all stakeholders by county health teams, county health team commitment to maintaining VMMC coverage, and alignment of staffing goals with VMMC targets. Greatest weaknesses were lack of public expenditure reporting, lack of sufficient resource allocation for VMMC in county health budgets, and lack of county health team engagement with civil society for VMMC program planning and feedback. Also notably, this was the only level on which scores decreased significantly in Y1, with respondent assessments of the “planning and coordination” and “civil society engagement” subdomains dropping substantially.

On the site level, for all models, common greatest strengths for Y1 were in reliable regular provision of VMMC services, regular review of service quality against standards, conduction of VMMC without interfering with other health facility services, community acceptability of services, and staff training and certification in service provision and quality improvement. Key weaknesses included lack of diverse community groups committed to success of VMMC sustainability (mixed and static models), and lack of appropriate resource allocation in county budgets (mobile and static models). There was also a perceived lack of task-shifting to the lowest permitted cadre (static model), but this task-shifting was in fact in place. Nearly all subdomain scores increased significantly in Y1, and none decreased significantly.

On the community level, for all models, strengths for Y1 were in ease of timely access to services, communication about services to community and incorporation of community feedback. Weaknesses were in community representation in meetings with the MoH and involvement in the service planning stage. A notable weakness identified at baseline but not in Y1 was in keeping service provision nondisruptive to other community services (e.g., schooldays). Both subdomain scores increased significantly in Y1. Detailed key informant responses are available in S7 Appendix.

Additional outcomes: The documentation review found multiple enablers already in place from baseline in both counties. These included, among others:

• documentation of annual targets and division of responsibilities between county and partner staff
• an enabling environment free of policies discriminating against potential clients outside the typical target population
• full task-shifting already in place: of surgical MC from doctors to nurses, and of infection prevention from nurses to high school graduates with 6 months of training in health and infection prevention.
• institutionalized continuous quality improvement practices

Challenges, refinements and lessons learned

The most important early challenge, across multiple model areas, was aligning demand creation with the target population. Mobilization is conducted not only by implementing partner staff, but also by trained community health volunteers (CHVs) employed by the MoH to perform outreach on multiple topics. The CHVs initially continued their prior practices of mobilizing for VMMC across model area boundaries, including in traditionally circumcising communities. When this was noted after the first month, expansion trainings were held and coupled with field supervisory follow-up visits, to align mobilization work with the study models. This also became the key refinement identified by partners for continued use after the study period: ensuring that mobilizers prioritized adolescent clients from non-circumcising communities to maximize the contribution to increased coverage. Implementers also observed that all models required continued investment in demand creation work. This was true even in the static model, where it had been initially hoped that since services there were at consistent locations and times, this predictability and the intrinsic high demand for VMMC in 10-14s might make continued substantial investment in demand creation unnecessary.

The models’ age focus made close coordination with schools and local Ministry of Education (MoE) leadership crucial. On the leadership level, VMMC program staff engaged with MoE county and subcounty management teams, resulting in school heads adding VMMC promotion messages into existing packages of school-based preventive health talks. On the school level, a pool of interested teachers emerged who committed to be trained and mentored as VMMC champions and went on to lead regular VMMC health talks in their schools. However, stakeholders were unable to codify coordination in a formal Memorandum of Understanding between Ministries, which proved to be a complex process to navigate, requiring national-level engagement.

An ongoing challenge across all models was staff turnover, necessitating frequent trainings and sometimes creating temporary staffing gaps. Similarly, periodic local circumcision campaigns required temporary staff reassignment between or away from model areas, though eligible VMMCs generated by these campaigns were counted toward model achievements. In addition, goal staffing levels initially calculated as sufficient for maintenance-phase services were not always enough to achieve local targets. Model area programs had to actively reassess staffing levels and adjust them several times during the study period.

However, the staffing realm also underwent a crucial sustainability development during Y1. Both county MoHs deliberately made substantial shifts toward ownership of the human resources and management aspects of VMMC service delivery, working with the implementing partners to plan and execute these shifts. The Migori county MoH spent most of Y1 planning this transition, but in Siaya, service delivery staff began the move from IPs into MoH payrolls and oversight structures, reporting to facility in-charges and discharging some additional health responsibilities along with their VMMC work. Partners there continued to provide the salary funding through a sub-agreement with the MoH, but salaries and benefits were aligned with MoH standards. In both counties, county/sub-county health management teams continued working with IP supervision teams to conduct regular supportive supervision activities.

Another crucial advance emerging from the development of these transition plans was that both counties added VMMC to their annual operations plans in 2018. However, these plans are not supported by dedicated budgets, and the lack of domestic financial support for the program was seen by all as the primary obstacle to sustainability. The county health directors also began work with the county health assemblies’ Committees for Health, to legislate and allocate funds for VMMC in the county HIV budgets, to advance domestic financial commitment to the program through platforms such as the next 5-year County Integrated Development Plan. Another avenue identified for increasing domestic contributions was incorporating costs currently borne by partners for types of support services already funded by MoHs for other health care programs—e.g., autoclaving and waste management—into MoH budgets. A third was incorporating VMMC service delivery into provider (especially nurse) job descriptions with associated benchmarks, implicitly increasing domestic human resources investment.

Implementers and MoHs also identified and executed additional opportunities for more efficient service delivery. These included implementing a national policy replacing universal HIV testing of 10-14-year-old MC clients with risk-based testing; shifting surges in service delivery to times of year when 10-14s were free from school; and the national MoH’s restocking sites where its mapping had identified instrument stockouts as an obstacle to maintaining services with VMMC reusable kits.

Finally, some challenges were model- or area-specific. For the static models these included refitting three facilities previously used only as outreach sites with the dedicated space and equipment to support daily VMMC services, in their new role as static sites. Service volumes were constrained until these renovations were completed, after three months. For mobile models, a key challenge was long travel times from base facilities to some service sites, causing service delivery to begin late in the day with clients waiting longer than expected. Finally, in some model areas, medical resources were temporarily diverted due to a cholera outbreak, and in others facility availability was interrupted during doctors’ and nurses’ strikes.