Mobile nudges and financial incentives to improve coverage of timely neonatal vaccination in rural areas (GEVaP trial): A 3-armed cluster randomized controlled trial in Northern Ghana

Background Despite progress in vaccination coverage, timeliness of childhood vaccination remains a challenge in many settings. We aimed to assess if mobile phone-based reminders and incentives to health workers and caregivers could increase timely neonatal vaccination in a rural, low-resource setting. Methods We conducted an open-label cluster randomized controlled 1:1:1 trial with three arms in 15 communities in Northern Ghana. Communities were randomized to 1) a voice call reminder intervention; 2) a community health volunteer (CHV) intervention with incentivized rewards; 3) control. In the voice call reminder arm, a study staff member made voice calls to mothers shortly after birth to encourage vaccination and provide personalized information about available vaccination services. In the incentive arm, CHVs promoted infant vaccination and informed women with recent births about available vaccination opportunities. Both CHVs and women were provided small monetary incentives for on-time early infant vaccination in this arm, delivered using mobile phone-based banking applications. No study activities were conducted in control communities. A population-based survey compared vaccination coverage across arms in the pre-intervention and intervention periods. The primary endpoint was completion of at least one dose of Polio vaccine within 14 days of life and BCG vaccination within 28 days of life. Results Six-hundred ninety births were identified; 106, 88, and 88 from pre-intervention and 150, 135, and 123 in the intervention period, in the control, voice call reminder and CHV incentive arms, respectively. In adjusted intent-to-treat analysis, voice call reminders were associated with 10.5 percentage point (95% CI: 4.0, 17.1) higher coverage of on-time vaccination, while mobile phone-based incentives were associated with 49.5 percentage point (95% CI: 26.4, 72.5) higher coverage. Conclusion Community-based interventions using mobile phone technologies can improve timely early vaccination coverage. A CHV approach with incentives to community workers and caregivers was a more effective strategy than voice call reminders. The impact of vaccination “nudges”via voice calls may be constrained in settings where network coverage and phone ownership are limited. We conducted an open-label cluster randomized controlled trial with three arms in 15 communities (clusters) in Northern Ghana. Communities were randomized to three groups: 1) a voice call reminder intervention (Intervention Group A, 5 communities); 2) a CHV intervention with incentivized rewards (Intervention Group B, 5 communities); 3) control (5 communities). A cluster randomized design was used for evaluation because the intervention was fundamentally delivered at the cluster level by local community volunteers, and thus individual randomization was not feasible. A checklist of criteria for compliance with the Consort Statement for cluster randomized controlled trials is included (S1 File). The Ghana Early Vaccination Program (GEVaP) study was conducted in rural Karaga District in Ghana’s Northern Region. The population in the region generally has substantially worse socioeconomic and health status than the country overall. Approximately 66% of women 15–49 in the region have never attended school, as opposed to 19% on average nationally; 49% have not accessed any media sources (tv, newspaper or magazine or radio) in the previous week, compared with 31% on average nationally. Approximately 35% of women in the Northern Region delivered their last child in a health facility, whereas nationally 73% on average are delivered in facilities. Preventive child health services including vaccinations are coordinated by the Ghana Health Services (GHS) and District Authorities and are available at selected facilities via fixed clinics or via “outreach” services delivered to communities at routine intervals. Vaccinations are usually offered at fixed clinics only during designated days and times at facilities, usually on a weekly or twice-weekly basis. “Outreach” vaccination services are typically delivered once per month in communities. Child births commonly occur outside of health facilities [13]. Although 97% of children in the region receive BCG vaccination by two years of life, less than half are vaccinated within the recommended first month of life. Less than 60% of children in the region receive the first dose of Polio vaccination and only 41% receive all age-appropriate vaccinations by two years of life [13]. One major barrier to timely vaccination in the region is timely and complete birth registration. Community Health Volunteers (CHV) trained by the GHS are tasked with documenting births, providing health promotion and health education activities in their communities, but are not a formal cadre within the GHS and are not remunerated. As a result, few births are reported to or officially registered with local authorities; registration is often delayed and civil registration systems are rarely or incompletely linked with health service information systems [13,14]. In each intervention community, teams of 2–4 CHVs previously selected and trained by the GHS for health and development work were invited to participate in the GEVaP program. If no previously trained CHVs were available, community leaders were asked to appoint new CHVs for the program. CHVs received standardized training. Many CHVs were not able to read and to write, thus training, tools and documentation materials were adapted to be appropriate for limited literacy and numeracy. CHVs were expected to document and report births in their communities within a week to a central study personnel and were provided with mobile network/data credit via “mobile money” to cover the costs of communicating with and sending verification materials to study staff. CHVs were given a 2 cedis reward (USD 0.50) for each birth reported within the first week of life. Vaccine records booklets were distributed to GHS staff in intervention and control communities to support documentation of services. A TiDIER checklist listing the location of items describing intervention components is included (S2 File) [15]. In intervention arm A (“voice call reminders”) CHVs documented and reported births to study personnel as described above. CHVs invited caregivers who had delivered a child in the previous week to participate in the program and obtained informed consent from eligible women. Enrolled women provided a phone number at which they could be contacted, either a personal or shared phone or phone to which they had access. CHVs were not instructed to provide specific follow-up to caregivers in this arm to encourage or track vaccination after enrollment, in addition to any contact or services delivered as part of their routine community health activities. A central study staff made voice calls to congratulate caregivers on the birth, highlighted the importance of early vaccinations and provide information on where and when vaccine services were available for young infants in that community. Study staff communicated closely with local health workers to ascertain accurate and up-to-date information on the schedule and availability of vaccination services in communities and provided tailored, personalized information on the schedule (dates, times, locations) for upcoming vaccine outreach services and local clinic-based services available to caregivers in their community. Up to three initial attempts to make contact with participants were made and additional communication continued up to 28 days of newborn life if young infants had incomplete vaccination. Participants in this arm did not receive any compensation, incentive, or reward, and CHVs were not incentivized for timely vaccinations in their communities. In intervention arm B (“CHVs and incentives”), CHVs similarly identified, documented, and reported births to central study staff and enrolled women who had recently given birth. CHVs were responsible for encouraging enrolled women in their communities to vaccinate young infants, and for providing information about the availability of local vaccination services. The timing, frequency and method of follow-up with participants was not specified and was ultimately at the discretion of the CHVs. CHVs took photos of the vaccination card and/or record with the date and location that each vaccination was received, and reported vaccinations to the central study personnel via the WhatsApp mobile phone application. The mother and the CHV were each provided a 1 Ghana cedi reward for verified, on-time vaccination with the first dose of Polio and with BCG (maximum of 2 Ghana cedis (0.50 USD) per CHV and percaregiver). Incentives were transferred to CHVs and participating caregivers via a “mobile money” phone-based banking application. A local research staff member provided supervision to CHVs, and tracked enrollment, birth and vaccination reporting and verification in all communities. Field visits were conducted to monitor CHV activities, motivate engagement and troubleshoot challenges, and frequent communication was made with CHVs via WhatsApp to encourage participation and reporting. CHVs found not to be performing (failure to report expected number of births; failure to report vaccinations) were followed up with by local research staff and were replaced if performance problems continued. Interventions in both treatment arms were launched on November 1, 2018; enrollment was phased out by March 31, 2019, and follow-up with enrolled particpants was completed April 30, 2019. No study-initiated activities were conducted in control communities during the intervention phase. Vaccine services continued to be available as per routine GHS health services. The GHS District Authority provided a list of all communities in Karaga and provided approval for implementing the project in the District. Communities were eligible for inclusion if they had a population size of at least 1,000 and were accessible from the main road within two hours using motorized vehicles and/or walking, during the dry season. The 15 villages with the largest populations sizes were selected from a list of all communities in the District. To avoid spillover, a minimum distance of 5 km between any two villages was imposed. Sensitization meetings with community leaders were held prior to initiation and community leaders provided permission to work in the community. Prior to randomization, research staff conducted field visits to each community to conduct a brief community survey to collect information on key characteristics considered a priori to be potential confounding factors. Stratified block randomization was used to achieve relative balance between communities in potential confounding factors including distance from/access to the nearest health facility; vaccine and health service delivery platform (static clinics or outreach services); population size; and distance from a main road. Random allocation to the control and the two intervention arms was assigned at the community level using max-min randomization conducted by the Principal Investigator in Switzerland [16]. All study investigators and the local research supervisor were aware of village allocation arm prior to study implementation. Following allocation in each of the ten intervention communities, the local research supervisor conducted visits with community leaders and District health workers, and hired and trained CHVs, based on the appropriate protocol for the allocation assignment. Due to the nature of the intervention, blinding of participants, investigators, and data collectors was not possible. CHV, enrolled participants, and community members were not informed of the respective research activities in other intervention arm communities, but no attempt at concealment was made. Data collectors and interviewers for the endline assessment were not engaged in program implementation and were unaware of intervention allocation, but were not blinded to the allocation. All residents of intervention communities who delivered a live-born, surviving infant during the intervention period were eligible to participate in the intervention program if the birth was identified by or reported to CHVs within the first week of life. Informed consent was obtained from all mothers or primary caregivers for program participation. Participants in each intervention arm were followed through to infant receipt of both vaccines (study endpoint) or a maximum of 28 days of newborn life. We conducted an endline population-based survey in the 15 study communities from 30 May 2019 to 9 July 2019 to assess vaccination coverage before and after the intervention launch. Study enumerators visited every household and structure in the communities to identify all households in which a live birth had occurred in the previous year (May 2018 through May 2019) and enumerated all infants up to 12 months of age residing in the community, or born in the previous 12 months and deceased. All households with children born during the relevant time periods currently residing in the communities, for whom a caregiver provided informed consent, were eligible for inclusion in the survey. Interviewers conducted a subsequent visit to eligible households to collect informed consent and conduct a detailed interview at the household with caregivers. Interviewers returned up to two additional times if the potential participant was not available or could not be located. Vaccination history and date of birth were based on documentation in nationally standardized child health booklets and vaccination cards if available, and caregiver report if written documentation was unavailable. Photographs of child vaccination records were taken for verification. Classification into pre-intervention period or intervention period was based on the child’s documented date of birth. Field supervisors conducted audits to ensure no households were missed during the census and and re-visited approximately 10% of eligible households identified for the survey for quality assurance purposes. Interviewers conducting the endline survey were not involved in the interventions and not aware of the treatment status of communities. The primary endpoint was on-time completion of both early vaccinations, defined as receiving at least one dose of Polio vaccine within 14 days of life and BCG vaccination within 28 days of life. Secondary endpoints were the on-time receipt of each vaccine separately; and receipt of each vaccine and both vaccines at any age. An additional secondary outcome was the proportion of births identified via the endline survey that was documented and reported by CHVs to central research staff. We hypothesized either intervention could result in 80% coverage of the primary endpoint, or an absolute difference of 30 percentage points relative to the 50% of children receiving both BCG and Polio vaccines within the first month after delivery from the 2014 DHS [13]. Assuming an intra-class correlation coefficient (rho) of 0.05 and 20 women per cluster, a sample size of 100 infants was required per arm to achieve a power of 90% at α of 0.05, for tests comparing each active treatment arm to control. We anticipated with a 12-month reporting period (6 months prior to intervention and 6 months post-intervention initiation), we would identify approximately twice the number of births expected in the 6-month period, or 600 total births. To account for additional women not identified during the intervention period, who chose not to participate or moved into the community after intervention initiation, we anticipated interviewing approximately 750 women for the survey. The total number of births identified in the initial enumeration was larger than expected. To avoid exceeding the target sample size, after completing the initial population enumeration, we modified the eligibility criteria for the endline survey to limit to births in the previous 9 months. The final analysis population excluded births from April 1, 2019, onwards, which occurred after intervention enrollment had concluded (S1 Fig). The primary outcome of interest was the likelihood of young infants receiving both the first dose of the Polio and the BCG vaccine on-time. Linear regression models were used to estimate the difference in the probability of getting vaccinated on time in each of the intervention arms compared with the control arm, adjusting for baseline vaccination coverage at the community level. Base models controlled only for pre-intervention cluster-level early vaccination coverage. Adjusted models included additional controls for maternal phone ownership and phone access; mobile network coverage; time to childbirth location; location of childbirth; maternal education, household electricity, and television ownership. Primary analyses were intent-to-treat based on community-level randomized treatment assignment. Data collected as part of program implementation activities during the active intervention in intervention arm communities was not used in the primary intent-to-treat analysis. Generalized linear log-binomial regression was additionally used to estimate prevalence ratios comparing the likelihood of complete on-time early vaccination in each intervention arm relative to the control arm during the post-intervention period. In sensitivity analyses, we estimated the difference in the change in the likelihood of timely vaccination from the pre-intervention to intervention periods, in each treatment arm compared with control, among the population of births in the pre-intervention and intervention periods. Generalized linear regression models were used with control terms for the month of birth, treatment group, period (pre-intervention or intervention), an interaction term for treatment and period and other adjustment covariates as the primary model. We also conducted an exploratory per-protocol analysis to evaluate the effect of the intervention among program participants and those successfully reached by the intervention program activities. We linked women and infants identified in the endline survey with program data, matching on the community, mother’s name, and child date of birth. In the voice call reminder arm, we defined intervention fidelity with two classifications: births documented but never successfully reached, and births documented and reached by phone at least once. In the CHV and incentives arm, intervention fidelity was defined as a birth documented by a CHV for which the caregiver agreed to participate in the program. We did not capture data on the amount, level or type of interaction of the CHV with the enrolled participants in this arm. In the per-protocol analysis, we used linear and negative binomial regression models adjusting for baseline vaccination in the intervention period only, adjusted for the same covariates previously specified. When log binomial models did not converge, we used negative binomial regressions [17]. Standard errors in all models were adjusted for correlation within clusters using robust variance estimates [18,19]. Point estimates and 95% confidence intervals were reported for all main outcomes. We did not adjust for multiple comparisons. Stata version 15 was used for analyses (StataCorp. 2017. Stata Statistical Software: Release 15. College Station, TX: StataCorp LLC.) The trial was retrospectively registered at ClinicalTrials.gov as {“type”:”clinical-trial”,”attrs”:{“text”:”NCT03797950″,”term_id”:”NCT03797950″}}NCT03797950 (submission 7 November 2018, publication 9 January 2019). Publication of the registration occurred after randomization due to delays in the administrative review of the submission. This is unlikely to bias results because the endline survey and all outcome assessment data were collected after publication, and endpoints were from objective, documented records. The Ghana Health Services Ethics Review Committee approved this trial (Protocol ID 008/07/18) and Ethikkommission Nordwest-und Zentralschweiz in Switzerland determined the trial conformed to research standards for studies in Switzerland (Request 2018–00548).