Re-evaluation of population-level protection conferred by a rotavirus vaccine using the ‘fried-egg’ approach in a rural setting in Bangladesh

N/A

Background: Vaccine herd protection assessed in a cluster-randomized trial (CRT) may be masked by disease transmission into the cluster from outside. However, herd effects can be unmasked using a ‘fried-egg’ approach whereby the analysis, restricted to the innermost households of clusters, ‘yolk’, creates an insulating ‘egg-white’ periphery. This approach has been demonstrated to unmask vaccine herd protection in reanalyses of cholera and typhoid vaccine CRTs. We applied this approach to an earlier CRT in Bangladesh of rotavirus vaccine (RV) whose overall analysis had failed to detect herd protection. Herein we present the results of this analysis. Methods: In the study area, infants in 142 villages were randomized to receive two doses of RV with routine EPI vaccines (RV villages) or only EPI vaccines (non-RV villages). We analyzed RV protection against acute rotavirus diarrhoea for the entire cluster (P100) and P75, P50, P25 clusters, representing 75%, 50% and 25% of the innermost households for each cluster, respectively. Results: During 2 years of follow-up, there was evidence of 27% overall (95 %CI: 7, 43) and 42% total protection (95 %CI: 23, 56) in the P100 cluster, but it did not increase when moved in smaller yolks. There was no evidence of indirect vaccine protection in the yolks at any cluster size. Conclusion: Our reanalysis of the CRT using the fried- egg approach did not detect RV herd protection. Whether these findings reflect a true inability of the RV to confer herd protection in this setting, or are due to limitations of the approach, requires further study.

The trial was conducted in Matlab, a rural field study area of the icddr,b (formerly known as the International Centre for Diarrhoeal Disease Research, Bangladesh) close to Dhaka, Bangladesh. The population under analysis has been observed under a Health and Demographic Surveillance System (HDSS) since 1966 and is well-described through routinely updated population census activities [10]. The study area comprises 142 villages geographically divided into two administrative areas, the icddr,b service area (ISA) made up of 67 villages, and the government service area (GSA) made up of 75 villages. In ISA, icddr,b provides enhanced services, including maternal, child health, and family planning services and routine immunization; in GSA, the Bangladesh Ministry of Health and Family Welfare provides standard public health and routine immunization services. Geographic coordinates are recorded for each household. The residents of Matlab predominantly belong to low- and middle-income communities and small-scale farming and fishing are the main occupations. The routine Expanded Programme on Immunization (EPI) vaccines were Bacillus Calmette-Guerin (BCG), Oral poliovirus Vaccine (OPV), diphtheria-tetanus-pertussis (DTP) and measles, administered at 6 weeks, 10 weeks, 14 weeks and 9 months of age during the study period [8]. The cluster-randomized trial evaluated Rotarix (GSK Biologicals, Rixensart, Belgium), an attenuated monovalent RV [11], which was delivered by the EPI staff in two oral doses separated by 4 weeks. Infants 6–20 weeks of age at the time of vaccination and residing in randomly assigned villages were offered the first dose of RV, given simultaneously with DTP1 or DTP2 vaccine delivered by the EPI in Bangladesh at 6 and 10 weeks of age. The second dose of RV was given 4 weeks after the first dose but before 20 weeks of age. Parents or legal guardians of participating infants provided written informed consent. Rotavirus vaccination, which was started in GSA from November 1, 2008 and in ISA from April 1, 2009, was carried out continuously during the study period. Vaccination status was recorded in the EPI cards as well as in the study forms. Infants in the non-RV villages received only the routine EPI vaccines [8]. A total of 142 villages within the Matlab HDSS were randomly assigned in a 1:1 ratio to either the RV study villages (total population 116,649) or non-RV study villages (total population 105,569). Infants who were 6–20 weeks of age on or after the study initiation date (November 1, 2008 in the GSA and April 1, 2009, in the ISA) and were included in the HDSS database were eligible for participation. Demographic and socioeconomic data were collected from this database. Individual vaccination records were also retrieved and linked to the database. The geographic linear distances (km) from the child’s residence to the Matlab icddr,b hospital, was calculated using geographic information system (GIS) data collected as part of the HDSS. Diarrhoea surveillance was conducted at the main icddr,b hospital in Matlab as well as in two community-operated treatment centers run by icddr,b in Matlab, the only sources of care for diarrhoea for the trial population. Children less than 2 years of age presenting to any of these health facilities with diarrhoea were assessed and treated accordingly. Diarrohea was defined as having three or more loose stools within 24 h of presentation; severe dehydration was defined as per WHO criteria [12]. Stool samples were collected from all patients coming from the trial area and were tested for rotavirus at the icddr,b laboratory. An acute rotavirus diarrhoea (ARD) was defined as a diarrhoea patient with symptoms less than 7 days in duration and whose stool specimen yielded rotavirus antigen (group A rotavirus specific VP6 proteins) using a qualitative enzyme immunoassay (Prospect Rotavirus Microplate Assay, Oxoid, Basingstoke, UK). Clinical characteristics of illnesss, treatment history and health outcome were recorded in the surveillance system [8]. Acute rotavirus gastroenteritis can also be characterized only by emesis (≥ 1 episode of forceful vomiting within 24 h), but this is rare in Bangladesh. Our analysis is a reanalysis of an earlier trial of the GSK’s RV, which, like other trials of this vaccine, targeted diarrheal disease as the primary outcome. The trial was completed on March 31, 2011. We used a ‘fried-egg’ approach [9] to reanalyze the Matlab CRT dataset. In this approach, progressively smaller central ‘yolks’ within the original allocation clusters demarcated the study population are analysed for evidence of vaccine herd protection. The logic is that as the yolks become smaller and are hence surrounded by wider ‘whites’ of surrounding populations, these surrounding populations would progressively insulate the yolk populations from the inward transmission of rotavirus, thereby revealing true vaccine herd protective effects. We computed the distance (linear) from the child’s residence to the nearest village boundary. To define the different size of yolks, we first sorted the distances from the village boundaries in descending order (furthest from to closest to the nearest village boundary) within the cluster. We then assembled successive proportions of the study population, beginning with the household furthest from the village boundary and proceeding to include households progressively closer to the nearest village boundary until the desired fraction of households in the cluster was achieved. We specified four fractions of households for analysis: 25% households (innermost yolk), 50% households, 75% households, and 100% households (outermost including the entire cluster) referred to, respectively as P25 clusters, P50 clusters, P75 clusters, and P100 clusters [13]. We hypothesized that if herd protection was attenuated by the transmission of rotavirus into the clusters from the outside, this protection would be most evident in the innermost households. Fig. 1 shows households that were included in analyses of overall rotavirus protection for the P25 clusters. Distribution of study area households for P25 clusters (left:entire study area; right:magnified view of P25). Three different measurements of population-level vaccine protection were estimated: overall protection (protection of all children in the RV clusters relative to control residents), total protection (protection of RV recipients relative to control residents who had received OPV), and indirect protection (protection of non-RV recipients in RV clusters relative to control participants). All measures of vaccine protection against ARD were expressed as the proportionate reduction of rotavirus disease incidence in RV versus non-RV villages. For overall protection, children who were 6–20 weeks old at study initiation, migrated in at 6–20 weeks of age during the study period, or became 6 weeks old during the study period were considered for analysis regardless of whether they received a vaccine (RV/ OPV) or not. The start date for follow-up was defined as the earliest of these three dates for each analyzed child. For total protection, among all children who were eligible for overall protection, we analyzed only those who received at least one dose of RV in RV villages and received at least one dose of OPV in non-RV villages and did not receive the first dose prior to the study initiation. Start date of follow-up was the date of vaccination with RV or OPV. For indirect protection, children who were aged 20 weeks to 24 months at study initiation or who migrated in between 20 weeks and 24 months of age and who but who did not receive RV as per record in the EPI card were included in the analysis. The start date of observation was defined as the date of initiation of the study, date of migration into the study villages, or date of aging into the targeted age range, whichever came first. The end of follow-up for all measures of protection was considered as the date of turning 23.9 months of age, migration out of the study area, death, the study end date, or development of ARD, whichever came first. Person-years were calculated as the sum of individual periods of observation for the population under analysis. To estimate vaccine protection, we analyzed the time to first episode of ARD, fitting Cox proportional hazards regression models with village vaccine assignment, together with potentially confounding covariates, including sex, mother’s highest education, house wall construction with bricks, TV ownership, reported usage of a modern toilet at home, use of a tubewell as the source of drinking water (a type of water well in which a long, 100–200 mm-wide, stainless steel tube or pipe is bored into an underground aquifer and water is pumped out through a handle present on it), and distance of household to the icddr,b Matlab hospital (km), as independent variables. Covariates were identified by backward selection with a cut-off p-value of <0.05 when compared between RV and non-RV villages for differing yolk sizes. Hazard ratios (HR) were estimated by exponentiation of the coefficient for the village vaccine assignment variable in models and vaccine protection was estimated as [(1 − HR) × 100%]. Standard errors for the coefficients were used to estimate two-tailed p-values and 95% confidence intervals for the HRs. All statistical analyses were performed using SAS version 9.4. The initial trial was funded by Gavi, the Vaccine Alliance (GAV.1141–02) through PATH's Rotavirus Vaccine Program (RVP). GSK donated the vaccine for the study. None of the funding agencies were involved in the study design, data collection, analysis, interpretation or writing of the manuscript. The corresponding author had full access to all the data sets. This publication was made possible through a grant from the Bill & Melinda Gates Foundation (INV-025388). The original protocol was approved by the Instutional Review Board of icddr,b and the Western Institutional Review Board. The protocol was registered at ClinicalTrials.gov ({"type":"clinical-trial","attrs":{"text":"NCT00737503","term_id":"NCT00737503"}}NCT00737503).

• Dima Health