Evaluating the impact of community health volunteer home visits on child diarrhea and fever in the volta region, ghana: A cluster-randomized controlled trial

Background Although there is mounting evidence demonstrating beneficial effects of community health workers (CHWs), few studies have examined the impact of CHW programs focused on preventing infectious diseases in children through behavior changes. We assessed the preventive effects of community health volunteers (CHVs), who receive no financial incentive, on child diarrhea and fever prevalence in Ghana. Methods and findings We conducted a cluster-randomized controlled trial in 40 communities in the Volta Region, Ghana. Twenty communities were randomly allocated to the intervention arm, and 20 to the control arm, using a computer-generated block randomization list. In the intervention arm, CHVs were deployed in their own community with the key task of conducting home visits for health education and community mobilization. The primary outcomes of the trial were diarrhea and fever prevalence at 6 and 12 months among under-5 children based on caregivers’ recall. Secondary outcomes included oral rehydration treatment and rapid diagnostic testing for malaria among under-5 children, and family planning practices of caregivers. Generalized estimating equations (GEEs) with a log link and exchangeable correlation matrix were used to determine the relative risk (RR) and 95% confidence intervals (CIs) for diarrhea, fever, and secondary outcomes adjusted for clustering and stratification. Between April 18 and May 4, 2015, 1,956 children were recruited and followed up until September 20, 2016. At 6 and 12 months post-randomization, 1,660 (85%) and 1,609 (82%) participants, respectively, had outcomes assessed. CHVs’ home visits had no statistically significant effect on diarrhea or fever prevalence at either time point. After a follow-up of 12 months, the prevalence of diarrhea and fever was 7.0% (55/784) and 18.4% (144/784), respectively, in the control communities and 4.5% (37/825) and 14.7% (121/825), respectively, in the intervention communities (12-month RR adjusted for clustering and stratification: diarrhea, RR 0.73, 95% CI 0.37–1.45, p = 0.37; fever, RR 0.76, 95% CI 0.51–1.14, p = 0.20). However, the following were observed: improved hand hygiene practices, increased utilization of insecticide-treated bed nets, and greater participation in community outreach programs (p-values < 0.05) in the intervention group. In a post hoc subgroup analysis, the prevalence of diarrhea and fever at 6 months was 3.2% (2/62) and 17.7% (11/62), respectively, in the intervention communities with ≥70% coverage and a ≥30-minute visit duration, and 14.4% (116/806) and 30.2% (243/806) in the control communities (RR adjusted for clustering, stratification, baseline prevalence, and covariates: diarrhea, RR 0.23, 95% CI 0.09–0.60, p = 0.003; fever, RR 0.69, 95% CI 0.52–0.92, p = 0.01). The main limitations were the following: We were unable to investigate the longer-term effects of CHVs; the trial may have been underpowered to detect small to moderate effects due to the large decline in diarrheal and fever prevalence in both the intervention and control group; and caregivers’ practices were based on self-report, and the possibility of caregivers providing socially desirable responses cannot be excluded. Conclusions We found no effect of CHVs’ home visits on the prevalence of child diarrhea or fever. However, CHV programs with high community coverage and regular household contacts of effective duration may reduce childhood infectious disease prevalence. We conducted a cluster-randomized controlled trial in 40 communities (villages) of the Ketu South District in the Volta Region, Ghana, between February 1, 2015 and September 20, 2016 (ISRCTN49236178) [14]. The baseline survey, and recruitment of caregivers (mother or primary female caregiver) and under-5 children, began on April 18, 2015, and the trial was registered on June 16, 2015. Randomization was performed in June 2015 using the baseline survey results, and the intervention started on August 15, 2015. The estimated population of the Ketu South District in 2015 was 181,881, and the number of children aged under 5 years was 36,376. Communities in the study area had a range of 130–254 households. A phase-in design was adopted for the trial, wherein the CHVs were recruited and activated in the 20 intervention communities during the trial period, while the 20 communities in the control arm received CHVs after completion of the end line survey of the trial. Ethical approval for the trial was obtained from the Ghana Health Service Ethics Review Committee (GHS-ERC:07/01/15), and the evaluation was supplementarily approved by the Harvard T.H. Chan School of Public Health (IRB17-2051). We assured allocation concealment for participants by selecting them before randomizing clusters [15]. The study recruited participants from April 18 to May 4, 2015. The inclusion criteria for the trial were households located in the trial catchment area that had at least 1 child under 5 years of age. The best estimate of each child’s age was determined by health card, insurance card, or caregiver’s report. All caregivers of the participants were informed of their right to withdraw from the study at any time. Children of caregivers who declined to participate in the surveys were excluded from the study. Study participants were enrolled for 16 months, and outcome data collection occurred at 6 and 12 months after study initiation. In households with more than 1 child, we recruited the youngest child. Children were not censored at the following surveys if they became greater than 60 months of age. A cluster randomization was chosen to prevent contamination in this study. The community level, where people interact with one another most closely, was taken as the randomization unit. To minimize the possibility of selection bias, we identified and recruited clusters before randomization [16]. Rural communities in the Ketu South District were identified for potential participation in the trial. Among the 57 rural communities that were identified, 40 were randomly selected using probability-proportionate-to-size methods to be included in the trial. In the second stage, stratified randomization was used, based on the baseline survey results, to assign communities to either intervention or control and reduce the risk of baseline imbalances. The 40 selected communities were stratified into 8 strata based on the estimated diarrheal prevalence among under-5 children, the economic status of the community, and the proportion of caregivers who had skilled delivery for their youngest child. To estimate the household economic status of a community, the proportion of housing structures with wattle and daub construction was used as a proxy indicator representing low-economic-status communities (cutoff point: 55.0%). Within the 8 strata, 20 communities were randomly allocated to the intervention arm, and 20 to the control arm, using a computer-generated block randomization list (by YM and SC). On the basis of a preliminary survey undertaken in 2013, the prevalence of both child diarrhea and malaria was estimated to be 25%, and was assumed to be reduced by 25% by the intervention based on a previous study [6,9,10]. We estimated the coefficient of variation to be 0.16. With a 10% attrition rate, the required sample size was 950 households across 20 clusters per arm, with a study power of 80% [17]. Details of the CHV intervention and characteristics of the CHVs are summarized in S1 Text. We incorporated multiple activities into the program that have been recommended to promote CHVs’ effectiveness (e.g., training and retraining, selecting CHVs among those most respected by the community, household education with visual aids, material rewards as incentives, regular supervision by health professionals, collaboration with the existing health system, and community awareness of CHVs) [3,18]. All CHVs’ activities were delivered within the community setting. The CHVs’ core task was to carry out home visits to each household every 2 months. They were recommended to spend at least 30 minutes on each household visit. While visiting households, CHVs were instructed to provide health education based on 10 key messages using visual aids. The messages covered actionable recommendations to improve the household members’ knowledge and behavior related to maternal and child health. They included information on the prevention of diarrhea and malaria, more specifically, that proper handwashing and improved sanitation and hygiene can prevent diarrhea, and sleeping under insecticide-treated bed nets (ITNs) can prevent malaria. CHVs also provided messages about how to manage diarrhea and malaria in children, more specifically, that children with diarrhea should be given oral rehydration salts (ORS) and taken to a health facility and that suspected malaria cases should be properly diagnosed and treated at a health facility. The key messages also included the benefits of using contraceptive methods and information on types of contraception, as well as the importance of participating in the community outreach program (child welfare clinic). Another major task of CHVs during home visits was to give ORS to children with diarrhea and to perform a malaria test if any child had a fever. CHVs were also recommended to support CHNs and to mobilize community members for child welfare clinics that were held monthly in the community. After the communities were randomly allocated to the intervention and control arms, CHVs were recruited from the intervention communities. The CHV selection process was coordinated by the Ketu South District Health Management Team (DHMT) in cooperation with community committees. The committees were requested to nominate candidates based on their literacy level, volunteerism, and experience. Regional and district health officials trained the CHVs for 5 days, including 2 days of field-based training to practice home visit skills and outreach support. The content of the training curriculum was prepared through discussions among the regional and district health teams and the project team. The objective of the CHVs’ training was to help them to understand the concept of CHPS, their roles and responsibilities, and essential maternal and child health services. The training also aimed to teach skills to properly conduct home visits and provide health education using 10 key messages, malaria tests using a rapid diagnostic test (RDT), ORS, and referral of patients to their supervisor. Refresher training was undertaken every month by the DHMT members and CHNs [14]. After the training, community leaders, together with DHMT officials, held community meetings to introduce the CHVs and to publicly declare the CHVs’ roles and position. The CHVs, who were the implementers of the program, were mainly community-based men and women who had graduated from junior high school, and they were rewarded with material compensation in the form of cell phone minutes and food items. CHNs treated patients who visited the CHPS compound and undertook community outreach programs to provide vaccinations, nutrition, and health education activities on a monthly basis. Community people and the project team were not blinded to the intervention because of the distinctive nature of CHV home visits. Three rounds of household surveys were conducted to evaluate the effect of the CHV program. The data collection team consisted of 14 data collectors and 2 supervisors, and all were blinded to whether a community was randomized to the intervention or control arm. All data collectors were trained for 2–3 days before every round of the survey, and participated in daily review sessions during the data collection period for quality assurance. The baseline survey was conducted from April 18 to May 4, 2015, before the start of the CHV intervention on August 15, 2015. The first follow-up survey was administered from February 6, 2016, after 6 months of intervention, and the end line survey was conducted from September 5, 2016, after 12 months of intervention. A set of questionnaires was developed and used to conduct the survey. The survey instrument primarily consisted of questions assessing the socioeconomic status of households and the prevalence of children’s diarrhea and febrile illness based on caregivers’ reports. Another key element was questions on CHVs’ activities reported by caregivers, including the frequency and duration of their home visits. The survey also contained a series of questions that gathered information about caregivers’ experiences of family planning, such as the use of various types of contraceptive methods, as well as information about caregivers’ most recent delivery, such as the place of delivery and utilization of antenatal and postnatal care. We applied systematic sampling to approach households in the communities. The data collectors visited households in each community using the interval method, in which the total number of households in a community is divided by the cluster size. They started by visiting a household located nearest to the main road, from which they continued to visit the next nth household based on the interval (e.g., the next fifth household if the interval was 5). Data collectors asked any household members if they had at least 1 under-5 child. If the household did not meet the eligibility criterion of having at least 1 child under 5 years old, the data collectors visited the next house and continued visiting the next nth one. At each visited household, the primary female caregiver of the youngest child in the household was asked to participate in the survey, and signed the consent form for enrollment if she agreed. Among the 1,956 households enrolled for impact evaluation, 408 (21%) were randomly selected and registered for process evaluation after obtaining separate informed consent for this additional survey. The components of the process evaluation were developed based on the framework of Steckler and Linnan [19]. Data on the indicators of each step of the intervention process were collected through a combination of methods, including household surveys, documentation review, and direct observation. Four rounds of household surveys were carried out at 3-month intervals. For the sample size, we referred to previous studies [20,21] of process evaluation in randomized controlled trials. A combination of self-reporting and direct observation was used to assess process indicators. Direct observation accompanied the household surveys for health behaviors such as appropriate use of ITNs and proper handwashing. Those who responded that they had used ITNs for their child during the previous night at the time of the survey were only considered to have actually done so if an ITN was hung up inside their house. For diarrhea, after administering standard questions about caregivers’ handwashing practices in the previous 24 hours at the time of the survey, direct observations were made. Only when caregivers were observed to wash their hands with running water and soap, upon request of demonstration, were they considered to practice appropriate hand hygiene. Caregivers’ participation in community outreach program (child welfare clinic) and child growth checks was surveyed. Among the 10 key messages delivered by the CHVs through home visits, we examined which specific messages were recalled by caregivers. In addition, various process indicators of intervention fidelity were assessed, including the results of 4 rounds of CHVs’ tests, the CHV monthly review meeting attendance rate, the proportion of CHVs with regular recording of logbooks, CHVs’ self-reported coverage of home visits, CHVs’ participation in community-wide health activities, and the rate of CHVs’ retention in their duties at 3, 6, 9, and 12 months. For these indicators, CHVs’ logbooks, monthly review meeting minutes, CHN outreach records, referral records, and the CHPS inventory logs were investigated. The primary outcomes of the trial were caregiver report of 14-day diarrhea and fever prevalence among under-5 children. Diarrhea was defined as having 3 or more instances of watery stools within 24 hours in the past 14 days, and we used febrile episodes in the last 14 days as a proxy indicator of malaria prevalence. Family planning practices and case management for child diarrhea and fever were investigated as secondary outcomes. For family planning practices, we examined long-term and short-term contraceptive methods (i.e., female/male sterilization, intrauterine devices, injectable contraceptives, implants, pills, female/male condoms, the standard-days method, the rhythm method, the lactational amenorrhea method, and withdrawal). For child diarrhea, we investigated whether the sick child was administered ORS, and for fever, we investigated whether the child was tested for malaria with a RDT kit. When designing the trial, we planned to examine the effect of the intervention on antenatal and postnatal care, as well as on case management of malaria of pregnant women as secondary outcomes. However, effects on these outcomes were not analyzed because they were extremely underpowered due to the small number of pregnancies in the registered households during the intervention (S3 Table). Baseline caregiver, child, and household characteristics were assessed for comparability between randomized arms. The statistical analyses for the primary and secondary endpoints were based on the intention-to-treat analysis principle. Generalized estimating equations (GEEs) with a log link and exchangeable correlation matrix were used to assess the relative risk (RR) and 95% confidence intervals (CIs) for diarrhea, fever, and secondary outcomes adjusted for clustering and stratification. Fixed-effect covariates were used to account for variables used in the stratified randomization design. We also conducted sensitivity analyses controlling for potential baseline imbalances in factors such as baseline diarrhea or fever prevalence, household income quintile, caregivers’ education, child age and sex, access to an improved water source, and improved sanitation. If the RR of the baseline-covariate-adjusted model differed from that of the unadjusted model by greater than 10%, we presented the adjusted model as the primary model. In addition, we examined associations of CHVs’ home visit coverage and intensity with primary outcomes. We presented observational relationships stratified by coverage (whether 70% or more of households received CHV home visits at least once in the past 3 months) and duration of visit (whether the average time spent on each home visit in a community was ≥30 minutes), as well as the combination of these factors. For the coverage and duration of household visits, we used caregivers’ reports gathered from household surveys. Due to the observational nature of these analyses, the analyses were adjusted for baseline fever or diarrhea, income quintile, caregiver education, child sex, and child age. All statistical analyses were conducted using STATA version 13. p-Values < 0.05 were considered to indicate statistical significance.