Mobile phone-delivered reminders and incentives to improve childhood immunisation coverage and timeliness in Kenya (M-SIMU): a cluster randomised controlled trial

Background As mobile phone access continues to expand globally, opportunities exist to leverage these technologies to support demand for immunisation services and improve vaccine coverage. We aimed to assess whether short message service (SMS) reminders and monetary incentives can improve immunisation uptake in Kenya. Methods In this cluster-randomised controlled trial, villages were randomly and evenly allocated to four groups: control, SMS only, SMS plus a 75 Kenya Shilling (KES) incentive, and SMS plus 200 KES (85 KES = USD$1). Caregivers were eligible if they had a child younger than 5 weeks who had not yet received a first dose of pentavalent vaccine. Participants in the intervention groups received SMS reminders before scheduled pentavalent and measles immunisation visits. Participants in incentive groups, additionally, received money if their child was timely immunised (immunisation within 2 weeks of the due date). Caregivers and interviewers were not masked. The proportion of fully immunised children (receiving BCG, three doses of polio vaccine, three doses of pentavalent vaccine, and measles vaccine) by 12 months of age constituted the primary outcome and was analysed with log-binomial regression and General Estimating Equations to account for correlation within clusters. This trial is registered with ClinicalTrials.gov, number NCT01878435. Findings Between Oct 14, 2013, and Oct 17, 2014, we enrolled 2018 caregivers and their infants from 152 villages into the following four groups: control (n=489), SMS only (n=476), SMS plus 75 KES (n=562), and SMS plus 200 KES (n=491). Overall, 1375 (86%) of 1600 children who were successfully followed up achieved the primary outcome, full immunisation by 12 months of age (296 [82%] of 360 control participants, 332 [86%] of 388 SMS only participants, 383 [86%] of 446 SMS plus 75 KES participants, and 364 [90%] of 406 SMS plus 200 KES participants). Children in the SMS plus 200 KES group were significantly more likely to achieve full immunisation at 12 months of age (relative risk 1·09, 95% CI 1·02–1·16, p=0·014) than children in the control group. Interpretation In a setting with high baseline immunisation coverage levels, SMS reminders coupled with incentives significantly improved immunisation coverage and timeliness. Given that global immunisation coverage levels have stagnated around 85%, the use of incentives might be one option to reach the remaining 15%. Funding Bill & Melinda Gates Foundation. The M-SIMU study was a four-arm, cluster-randomised controlled trial done within the Health and Demographic Surveillance System (HDSS) overseen by the Kenyan Medical Research Institute (KEMRI) and Centers for Disease Control and Prevention (CDC) in Siaya County, Nyanza Province. Clusters were villages, as defined by HDSS, and were randomly assigned and evenly allocated to one of four study groups: control, SMS reminders only (SMS only), SMS reminders plus a 75 Kenyan Shillings incentive (KES; SMS plus 75 KES, where 85 KES = USD$1 as of August, 2015) and, SMS reminders plus a 200 KES incentive (SMS plus 200 KES). A cluster-randomised approach was preferred to an individually randomised approach to minimise potential discord if neighbouring participants were randomised to different study groups. The conduct, analysis, and reporting of results were done in accordance with the Consolidated Standards of Reporting Trials (CONSORT) guidelines adapted for cluster-randomised trials.23 The M-SIMU study was done in rural, western Kenya, an area with high prevalence of HIV, tuberculosis, and malaria.24 Clusters (ie, villages) were included in the trial if they were located within Gem or Asembo districts and were within the HDSS boundaries. Clusters were excluded if they had ongoing special health programmes or immunisation activities that could bias the study outcomes. The M-SIMU trial recruited HDSS village reporters to identify eligible caregivers and their infants. Village reporters used mobile phones to send birth notification text messages to the RapidSMS server. Birth notifications were relayed to field-based community interviewers who then screened caregivers of newborns for eligiblity into the study. Inclusion criteria for participation included being a caregiver of an infant aged 0–34 days and being a current resident of one of the randomised study villages. Exclusion criteria included being a caregiver planning to migrate from the study area in the next 6 months, if the infant received vaccination other than Bacillus Calmette–Guérin (BCG) or polio birth dose, or if the caregiver was not willing to vaccinate the child at an M-SIMU staffed clinic. Caregivers of infants aged 35 days and older were excluded because of the close temporal proximity to the first pentavalent visit scheduled at 6 weeks (ie, 42 days).26 M-SIMU health facility recorders were stationed at 24 clinics whose catchment area overlapped with all study villages.24 Mobile phone ownership was not an inclusion or exclusion criterion. Participants only needed to have access to a mobile phone, whereby access was defined by the caregiver. For those participants did not own a phone, the enrolment was paused until the participant confirmed with the owner of the shared phone that text messages and incentives, as applicable, could be sent to the mobile phone. The protocol received ethical clearance from the Center Scientific Committee, Scientific Steering Committee (SSC), and the KEMRI-Nairobi Ethical Review Committee (ERC; SSC#2409). Johns Hopkins University Bloomberg School of Public Health and CDC deferred ethical clearance to KEMRI-ERC. A detailed description of the methods and protocol, including content of text message reminders and map of randomly assigned villages, has been reported.25 This trial is registered with ClinicalTrials.gov, number 01878435. A baseline survey of vaccination coverage, phone ownership, and geographical and demographic characteristics was done on March 13 and April 29, 2013, to provide data for the randomisation.27 A constrained randomisation28 was done with GAUSS Mathematical and Statistical System by one of the study investigators, which randomly generated 1000 allocations that met the following criteria for balance across study arms: within a relative 10% for the means of full immunisation coverage, phone ownership, distance to the nearest clinic, and village population of children 12–23 months old; within a relative 25% within each district (Gem or Asembo) for the means of full immunisation coverage and phone ownership. The randomisation was stratified on district so that each study group contained exactly 30 villages from Gem and eight villages from Asembo. The 1000 sequences were labelled with three-digit numbers, 000 to 999, each one assigning 38 villages to each of the four groupings (A–D). At a public randomisation ceremony on Sept 12, 2013, village chiefs determined the final randomisation outcome by picking numbered balls from a cloth sack to select one of these 1000 sequences, then picking labelled (study group) balls to assign the interventions to the chosen allocation.24 All caregivers and their infants (hereby referred to as infant-caregiver pairs) were allocated to the same study group as the randomised village in which they resided. If a caregiver moved during the follow-up period, the infant–caregiver pair retained their initial study group allocation. Due to the nature of the intervention and study design, study participants were not masked to their study group allocation. Field staff were not informed of a village’s allocation, but this could be inferred from some enrolment and follow-up survey questions. Data cleaning was done by a statistician blinded to the allocation. Participants provided written informed consent and were enrolled into the study by community interviewers after villages were randomly assigned. After obtaining consent, community interviewers sent an enrolment SMS to the RapidSMS server that contained the caregiver’s phone number, the infant’s birthdate, the preferred language to receive SMS reminders (English, Kiswahili, or Dholuo), and the infant’s name. All caregivers received a single text message at enrolment welcoming them to the study. For the three intervention groups, SMS reminders were sent three days and the day before scheduled immunisation visits at ages 6 weeks, 10 weeks, and 14 weeks for the three doses of pentavalent vaccine and age 9 months for measles vaccine using the free and open-source RapidSMS platform. Health facility recorders (HFR) were present at M-SIMU clinics to document immunisation. For immunised children, HFRs sent an SMS with the date of immunisations received and any change in caregiver’s phone number to the RapidSMS server. For pentavalent 2 and pentavalent 3 vaccines, their respective due dates were recalculated to be 4 weeks from the texted pentavalent date (interval-appropriate schedule)26 and reminders were sent accordingly. Children who either went undocumented by the HFR or who did not receive a pentavalent vaccine had reminders sent at 6 weeks, 10 weeks, and 14 weeks. In addition to receiving SMS reminders, caregivers were sent either 75 KES (group 3) or 200 KES (group 4) to their mobile phone for each timely dose of pentavalent and measles vaccine received, defined as vaccination within 2 weeks of the Expanded Programme on Immunisations (EPI) scheduled date (ie, pentavalent1 at 6 weeks, pentavalent2 and pentavalent3 4 weeks after the previous pentavalent dose, and measles at 9 months).26 Incentives were sent to participant’s mobile phones using a mobile-money programme22 and through their preferred mobile network (eg, Safaricom, Airtel). For children who were in the two incentive groups, the RapidSMS system log was downloaded daily and mobile-money delivered to caregivers whose children were timely vaccinated. Community interviewers did household follow-up visits when children reached 12 months of age to document the child’s immunisation status with the maternal and child health (MCH) booklet. If the MCH booklet was not available, a verbal report of immunisation history was taken. The primary outcome was the proportion of fully immunised children by 12 months of age, defined as receiving BCG, three doses of polio vaccine, three doses of pentavalent vaccine, and measles vaccine. Polio birth dose was excluded from the primary outcomes definition because this vaccine is recommended to be received within the first 2 weeks of life. BCG was included because it is recommended up to 59 months of age and could be received at any of the pentavalent or measles vaccination visits.26 Vaccination coverage at 12 months of age and timely vaccination for pentavalent, polio, and measles vaccines were predetermined as secondary outcomes. Vaccination timeliness was defined as receiving vaccination within 2 weeks of the EPI due date for individual vaccines. A timely fully immunised child was defined as being fully immunised within 2 weeks of the measles EPI due date. Sensitivity analyses of pentavalent timeliness used an interval-appropriate schedule for calculating due dates of pentavalent2 and pentavalent3 vaccines, where the new pentavalent due date was calculated to be 4 weeks after the receipt of the previous pentavalent vaccine. Data for primary and secondary outcomes came from written immunisation records found on the child’s MCH booklet at 12-month follow-up visits. As available, MCH records were compared with prospectively collected immunisation records by HFRs. If there were discrepancies between MCH and prospectively-collected data, the health facility immunisation registries were consulted. If the registries did not resolve the discrepancy, MCH data were used. The MCH booklet was chosen as the primary source document because it was independent of the HFR-collected data. Infant–caregiver pairs were considered lost to follow-up if they outmigrated or the infant died before 12 months of age. Verbal reports of immunisation at 12-month household follow-up visits, in the absence of written documentation, were excluded from the analytic sample. Sample size calculations found that 152 villages were needed to detect a 15% absolute difference in the proportion of fully immunised children at 12 months of age between the control group and any given intervention group. A priori, we selected a 15% difference as a meaningful outcome as this represents an effect size that would motivate policy makers to adapt the interventions. The following assumptions were made for sample size calculations: a baseline coverage of fully immunised children of 70% at 12 months, village birth cohort with harmonic mean of 16 newborns, a between-cluster coefficient of variation (k) of 0·25, a 25% loss to follow-up, a type I error (α) of 0·05, and power (1-beta) of 0·80. Primary analyses were done with modified intention-to-treat analyses at the participant level so that participants’ outcomes were analysed regardless of the degree of exposure to study interventions. The term modified refers to the requirement of being able to determine the 12-month immunisation outcomes. Risk ratios for primary and secondary outcomes were calculated for the intervention groups compared with the control group using log-binomial regression29 and General Estimating Equations (GEE) to account for correlation within clusters. As a secondary analysis of the primary outcome, time-to-immunisation curves were constructed with the Kaplan–Meier method. To assess the heterogeneity of treatment effects by various risk factors at baseline, the log-binomial models were extended and interaction terms were tested. Subgroup analyses by phone ownership and time to clinic were prespecified. An additional six subgroups were analysed by study group to explore potential effect modification. Based on the 18 post-hoc subgroup interaction analyses, about one statistically significant test of interaction (p<0·05) would be expected on the basis of chance alone. Socioeconomic status was derived from multiple correspondence analysis of household possessions to produce quintile scores and then dichotomised into bottom 40% and upper 60%. Years of maternal education were collected as a continuous variable and dichotomised to align with the number of years needed to complete primary school in Kenya (8 years). Per-protocol analyses of SMS reminders delivered were done for primary and secondary outcomes; per-protocol was defined as being sent the appropriate number of reminders per vaccine. Sensitivity analyses of vaccination coverage and timeliness were done with an interval-appropriate schedule of the pentavalent series. Analyses were done with STATA/SE (version 14.1; Stata Corp, College Station, TX, USA). An α of 0·05 was assumed for all statistical tests of significance. The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.