Five-year retention of volunteer community health workers in rural Uganda: a population-based retrospective cohort

Community health workers (CHWs) effectively improve maternal, newborn and child health (MNCH) outcomes in low-to-middle-income countries. However, CHW retention remains a challenge. This retrospective registry analysis evaluated medium-term retention of volunteer CHWs in two rural Ugandan districts, trained during a district-wide MNCH initiative. From 2012 to 2014, the Healthy Child Uganda partnership facilitated district-led CHW programme scale-up. CHW retention was tracked prospectively from the start of the intervention up to 2 years. Additional follow-up occurred at 5 years to confirm retention status. Database analysis assessed CHW demographic characteristics, retention rates and exit reasons 5 years post-intervention. A multivariable logistic regression model examined 5-year retention-associated characteristics. Of the original cohort of 2317 CHWs, 70% were female. The mean age was 38.8 years (standard deviation, SD: 10.0). Sixty months (5 years) after the start of the intervention, 84% of CHWs remained active. Of those exiting (n = 377), 63% reported a ‘logistical’ reason, such as relocation (n = 96), new job (n = 51) or death (n = 30). Sex [male, female; odds ratio (OR) = 1.53; 95% confidence interval (CI): 1 · 20-1 · 96] and age group (<25 years, 30-59; OR = 0.40; 95% CI: 0.25-0.62) were significantly associated with 5-year retention in multivariable modelling. Education completion (secondary school, primary) was not significantly associated with retention in adjusted analyses. CHWs in this relatively large cohort, trained and supervised within a national CHW programme and district-wide MNCH initiative, were retained over the medium term. Importantly, high 5-year retention in this intervention counters findings from other studies suggesting low retention in government-led and volunteer CHW programmes. Encouragingly, findings from our study suggest that retention was high, not significantly associated with timing of external partner support and largely not attributed to the CHW role i.e. workload and programme factors. Our study showcases the potential for sustainable volunteer CHW programming at scale and can inform planners and policymakers considering programme design, including selection and replacement planning for CHW networks. We conducted a retrospective operational review of an existing population-based database comprising all CHWs in two districts (Bushenyi and Rubirizi) in rural southwest Uganda, prospectively designed to track CHW characteristics, retention rates and characteristics associated with retention during and following an MNCH project intervention. The Ugandan health system consists of two levels of administration—the national (central government) level and the district level (local government) (World Health Organization, 2017). Below the district level, health subdistricts are responsible for primary healthcare services, including the coordination, delivery and monitoring of health services for the local population at health centres and through CHWs (World Health Organization, 2017). In Uganda, CHWs provide health facility outreach at the village and household levels to support health promotion activities and facilitate referrals to the health facility (World Health Organization, 2017). Bushenyi district (projected population 250 400 in 2021) includes 563 villages clustered into 64 parishes, and Rubirizi district (projected population 146 600 in 2021) consists of 294 villages and 53 parishes (Uganda Bureau of Statistics, 2016). In both districts, most (80%) families are rural, mainly subsistence farmers with limited access to essential health services (Uganda Bureau of Statistics, 2016). Moreover, the terrain is hilly, tarmac roads are few, and many families must walk long distances to access primary healthcare services. In 2012, MNCH promotion-focused volunteer CHWs were recruited, trained and supervised from all villages in Bushenyi and Rubirizi districts. Operationalizing CHW networks (i.e. Village Health Team National Strategy) was one component of a comprehensive MNCH initiative locally known as ‘MamaToto’ (Swahili for mother-baby), implemented by District Health Teams in collaboration with ‘Healthy Child Uganda’. As a government-embedded package, the CHW component was designed for sustainability, where CHWs were expected to continue in their roles beyond the externally funded ‘MamaToto’ intervention period. Between July 2012 and August 2014, the ‘MamaToto’ package was implemented, first in Bushenyi District (2012–13) and then in Rubirizi district (2013–14). Selected and trained CHWs within the ‘MamaToto’ cohort were followed prospectively for 2 years until project end (September 2014). The ‘active status’ of the CHWs was then assessed again at 5 years post-intervention (2017–18). In 2012, initial sensitization meetings in each parish involved orientating community members and local leaders to the national CHW programme and describing expected CHW roles and conditions. Participants selected CHWs in each village according to government CHW recruitment guidelines and each community’s process and criteria. Community-informed selection criteria included characteristics such as being a parent, active community involvement, demonstrating voluntary spirit and being a trusted and respected community member. Selected CHWs attended a 5-day initial training workshop conducted by government health facility-based trainers, using a participatory training approach to emphasize core skills and knowledge related to MNCH promotion, leadership, communication and role-specific expectations. CHWs within each parish were organized into parish teams who were, in turn, assigned to a designated health facility-based supervisor who supported the team during quarterly meetings. One 2-day refresher workshop occurred in the second year of intervention. CHWs did not receive monetary remuneration for their role. Financial and non-financial incentives included: Within their first year together, almost all CHW teams self-initiated some form of income-generating activity. While such initiatives were not directly supported by the intervention, self-dependency, problem-solving and teamwork were reinforced and practised during training, and group-driven initiatives were encouraged as sustainability strategies. CHW team income-generating activities included saving and loan groups by most groups and small businesses by some groups (e.g. catering, animal husbandry and handicrafts). CHWs supported health promotion within their communities, especially related to MNCH. Specific tasks included home visits, health talks, mobilizing peers for health outreach, reporting, and assessing/referring ill patients to nearby health facilities. CHW did not provide medications or supplies unless they were part of a small percentage of CHWs supporting integrated community case management programming (Year 4–5 post-training, initiated by the district and partners outside of ‘Healthy Child Uganda’ programming). Additional details of the ‘MamaToto’ intervention, CHW programming and CHW training package are available online: http://www.healthychilduganda.org/resources/. Study participants were all CHWs who completed an initial ‘MamaToto’ training workshop in target districts in 2012 or 2013. CHW demographics were collected upon initial workshop completion, and CHW retention was reported prospectively for 2 years (after which the funded and partner-supported intervention ended). Retention was evaluated at 5 years post-intervention by follow-up of CHWs and CHW Supervisors by project staff. The primary study outcome was retention. Analyses were done for retention at 5 years for the initial CHW cohort. The CHW ‘Start date’ was defined as the last day of initial ‘MamaToto’ training. ‘Exit date’ was estimated according to when their supervisor deemed a CHW to have ‘exited’ and required replacement. In cases where no formal ‘exit date’ was prospectively recorded, the exit date corresponded to the CHW replacement date. If no replacement date was available, we used the date of the last formal activity attended. ‘Time to exit’ for those who left by 5 years (60 months) was calculated in days between ‘start date’ and ‘exit date’. All initial CHWs had the potential for 60-month participation (i.e. 5 years) and thus contributed to the denominator for 60-month retention proportion calculations. An ‘exit reason’ was recorded based on CHW self-report or via the peer/supervisor. During data cleaning (starting at 2 years), categories were created. ‘Exit reason’ responses were clustered into categories according to the most common reasons provided. From then onward, database entries were either recorded as one of the categories noted or ‘other’ with details specified. Categories included ‘logistical’ reasons, which were not directly related to the CHW role (i.e. death, moves, new job/workload change, poor personal health, family duties/family health, divorce/separation and further study) or ‘non-logistical’ where a potential direct link to the CHW role was considered more likely (i.e. too busy, too much work/difficult, no longer interested, community rejection, peer/supervisor rejection and spouse opposed). All CHWs were prospectively registered in an operational Microsoft Excel database during the intervention, recording characteristics and retention details. Demographic data, including date of birth, sex, highest formal education completed and village location, were collected during initial training workshop registration and transcribed into the database by a trained records clerk upon workshop completion. Quarterly, CHW supervisors reported details of any retention, including exit date and reason. At 24 and 60 months, a researcher visited CHW teams to confirm their status and clarify any missing/conflicting data. Database cleaning verified outlier values, duplicates, and identified missing data through phone calls, report reviews and field visits. The analysis used STATA® version 13 (StataCorp LP., 2013; College Station, TX, USA). Descriptive statistics [means, SD, median, interquartile range, frequencies and percentages] summarized the sample and described retention variables. Multivariable logistic regression was used to assess the characteristics that were associated with retention. As part of checking linearity assumptions with age with the log of odds, age was binned in 5-year increments/group and plotted against the observed log of odds. These showed nonlinear behaviour and suggested the age (in years) categorizations used in the logistic regression, which are <25, 25–29, 30–59 and >60. Multivariable logistic regression examined associations of factors with CHW retention using Hosmer and Lemeshow’s purposeful selection of variables (Bursac et al., 2008). The most complex model was fitted based on a priori literature and experience-based selection of variables and plausible interaction terms were included. Interaction terms were tested and removed if not significant, coefficients were checked for changes greater than 10% when removing non-significant variables, and a final reduced model was compared with the fuller model with use of the likelihood ratio test for assessing removal of variables. The final adjusted model retained variables with a P-value of <0.05 and any potential confounder as per the 10% change criteria. Potential independent confounders included in the multivariable model were: sex (male or female), enrolment age (<25 years, 25–29 years, 30–59 years or >60 years), education (secondary school completion vs. non-completion), community type (rural or urban/mixed), and the interaction between community type with district (Bushenyi or Rubirizi). Apart from the logistic regression, as an exploratory exercise, the Kaplan–Meier method (Kaplan and Meier, 1958) was used to graphically explore the exiting patterns from start date. Where CHWs did not experience an event (exiting the programme) or for which follow-up was lost, we could not observe if the event happened within the 5 years as the initial training dates were right censored (exit status is greater than or equal to the last available follow-up status).