Effectiveness of a live oral human rotavirus vaccine after programmatic introduction in Bangladesh: A cluster-randomized trial

Background Rotavirus vaccines are now globally recommended by the World Health Organization (WHO), but in early 2009 WHO’s Strategic Advisory Group of Experts on Immunization reviewed available data and concluded that there was no evidence for the efficacy or effectiveness of a two-dose schedule of the human rotavirus vaccine (HRV; Rotarix) given early at 6 and 10 wk of age. Additionally, the effectiveness of programmatic rotavirus vaccination, including possible indirect effects, has not been assessed in low-resource populations in Asia. Methods and findings In Bangladesh, we cluster-randomized (1:1) 142 villages of the Matlab Health and Demographic Surveillance System to include two doses of HRV with the standard infant vaccines at 6 and 10 wk of age or to provide standard infant vaccines without HRV. The study was initiated November 1, 2008, and surveillance was conducted concurrently at Matlab Diarrhoea Hospital and two community treatment centers to identify children less than 2 y of age presenting with acute rotavirus diarrhea (ARD) through March 31, 2011. Laboratory confirmation was made by enzyme immunoassay detection of rotavirus antigen in stool specimens. Overall effectiveness of the HRV vaccination program (primary objective) was measured by comparing the incidence rate of ARD among all children age-eligible for vaccination in villages where HRV was introduced to that among such children in villages where HRV was not introduced. Total effectiveness among vaccinees and indirect effectiveness were also evaluated. In all, 6,527 infants were age-eligible for vaccination in 71 HRV villages, and 5,791 in 71 non-HRV villages. In HRV villages, 4,808 (73.7%) infants received at least one dose of HRV. The incidence rate of ARD was 4.10 cases per 100 person-years in non-HRV villages compared to 2.8 per 100 person-years in HRV villages, indicating an overall effectiveness of 29.0% (95% CI, 11.3% to 43.1%). The total effectiveness of HRV against ARD among vaccinees was 41.4% (95% CI, 23.2% to 55.2%). The point estimate for total effectiveness was higher against ARD during the first year of life than during the second (45.2% versus 28.9%), but estimates for the second year of life lacked precision and did not reach statistical significance. Indirect effects were not detected. To check for bias in presentation to treatment facilities, we evaluated the effectiveness of HRV against acute diarrhea associated with enterotoxigenic Escherichia coli; it was 4.0% (95% CI, −46.5% to 37.1%), indicating that bias likely was not introduced. Thirteen serious adverse events were identified among recipients of HRV, but none were considered related to receipt of study vaccine. The main limitation of this study is that it was an open-label study with an observed-only control group (no placebo). Conclusions The two-dose HRV rotavirus vaccination program significantly reduced medically attended ARD in this low-resource population in Asia. Protection among vaccinees was similar to that in other low-resource settings. In low-resource populations with high rotavirus incidence, large-scale vaccination across a wide population may be required to obtain the full benefit of rotavirus vaccination, including indirect effects. Trial registration ClinicalTrials.gov NCT00737503

The main purpose of this document is to describe the planned statistical analyses of this study so as to avoid "data dredging" at the time of preparing analyses to inform policy and for publication. Some background information and sample size calculations are copied here from the protocol dated 21.02.11 for ease of reference.
This document will guide analyses and may be given to journal editors for clarification of analyses and their timing. This document is being finalized months before the end of data collection, without any knowledge of treatment arm-specific results, and only a general idea of enrollment and total numbers of events.

BACKGROUND
A safe and effective vaccine is needed to reduce the enormous public health burden associated with rotavirus illness, especially in developing countries. About 40,000 children with rotavirus diarrhoea are treated each year at the ICDDR,B hospitals. Globally, there are an estimated 527,000 deaths annually. Unanticipated adverse events (intussusception) experienced with a rhesus rotavirus vaccine have accelerated efforts to develop and evaluate alternative vaccine candidates so that a safe and effective public health tool would become available. GSK Biologicals" live attenuated human rotavirus vaccine (HRV), Rotarix has been tested in clinical studies and was shown to be efficacious, safe and well tolerated among infants. The vaccine has been licensed in more than 80 countries, including Bangladesh.

2.1
Study Overview The study will be conducted in the Matlab Health and Demographic Surveillance System (HDSS) field area of the International Centre for Diarrhoeal Disease Research, Bangladesh (ICDDR,B) to determine the population effectiveness of Rotarix in Bangladeshi children. Villages in both the Maternal and Child Health and Family Planning intervention area (hereafter referred to as the ICDDR,B service area) and the comparator Government service area will be included in this evaluation. We propose to introduce Rotarix into half of the villages of the Matlab HDSS. In villages randomized to receive the vaccine, all eligible children will be offered Rotarix during their first two Expanded Programme on Immunization (EPI) visits, as would routinely be done if Rotarix were included in the Bangladesh EPI schedule. If a child residing in a village assigned to Rotarix vaccine did not receive Rotarix during the first EPI visit, the child may still receive the Rotarix during subsequent EPI visits so long as the vaccine is given within age guidelines as defined by the Rotarix package insert. In villages randomized not to receive Rotarix, children will receive their EPI vaccinations exactly as they would have in the absence of this study. Administration of Rotarix will be conducted by regular EPI staff, but ICDDR,B study staff will be present to document informed consent and collect study-specific information. The Ministry of Health will be an active partner in this evaluation since they will be the agency which may follow up with any subsequent vaccine programme. Vaccination with Rotarix will be recorded on the infant"s immunization card which is normally used by the EPI programme, but also on a separate study-specific data collection form.
Rotarix vaccination started in the government service area villages in September 2008 (although was not fully operational until November 1 of that year) and in the ICDDR,B serivce area villages on April 1, 2009. The delayed start in the ICDDR,B service area was to allow for completion of follow-up in the RotaTeq individually-randomized trial which was carried out in the 67 villages of the ICDDR,B service area.
Vaccination with Rotarix and surveillance activities for trial purposes will continue from study initiation through March 31, 2011. Surveillance for rotavirus gastroenteritis among children less than 2 years of age will occur at Matlab Diarrhoeal Hospital and the two community treatment centres of the Matlab HDSS continuously throughout the study period. Diarrheal illness information collected through surveillance will be linked to Rotarix study-specific data using the subject"s HDSS identification numbers. The primary study endpoint will be the occurrence of an illness episode of acute diarrhoea, among infants and children presenting to a diarrhea treatment centre, determined to be caused by wild-type rotavirus found in a stool specimen. At the end of the surveillance period, rates of this primary study endpoint among age-eligible infants will be compared for villages randomized to receive Rotarix versus for villages randomized not to receive Rotarix. We expect that the rates of rotavirus diarrhoea will be significantly lower among children from the vaccinated villages. "Age-eligible" children is the cohort of children whose birthdate indicates that they could have received rotavirus vaccine if they resided in the villages assigned to vaccine, regardless of whether they actually received it (e.g. an intention to treat analysis comparing rates of rotavirus diarrhea among the cohort of children who were eligible to receive vaccine compared to age eligible children in the comparison villages who did not receive vaccine.)

Randomization and Blinding
There are 75 villages in the government comparison area and 67 villages in the intervention area. The villages in each area will be randomized separately. Infants in half of the villages in each area will be eligible to receive Rotarix during their EPI doses. In each area, villages will be randomly selected for Rotarix introduction after pre-stratifying on village size and baseline hospitalization rates/numbers for diarrhoea at the Matlab Diarrhoeal Hospital and community treatment centres during 2005 and 2006. Hospital charts for diarrhoea admissions will be reviewed to determine the number/rates of hospitalizations from those villages. The villages will be arranged sequentially based on diarrhoea admission number/rates to the Matlab and Nayergaon and Kalirbazaar treatment centres. For each stratum of villages, half of the villages will be randomly selected for introduction of Rotarix and the other half will for no introduction of Rotarix.
Although not a blinded evaluation, the introduction of bias will be reduced in several ways. First, the vaccine teams located in the EPI centres that are completely separate from the clinical settings where clinicians will be conducting surveillance. Second, the clinicians who treat the children when they present to a treatment facility are separated from those staff who carry out the vaccinations or conduct the laboratory testing. Third, the illness episodes are widely separated in time from the vaccination, and thus the illness episode is not easily associated with receipt of vaccine. Fourth, laboratory assays for rotavirus will be carried out in the laboratory in Dhaka using coded specimens and are not conducted at the time of illness (since they have little clinical value). Although a parent could inform the clinician as to his/her child"s participation in the study, this is believed to be an unlikely or important source of bias.

Patient Evaluations
Study personnel will record the child's symptoms and clinical history, and collect a fecal specimen for laboratory testing for rotavirus. The identity of the child will be confirmed through the HDSS subject identification number system. All patients coming to the Matlab Diarrhoea Hospital and Nayergoan community treatment centre from these villages are already included in the Matlab diarrhoea surveillance system, and faecal samples are routinely tested for rotavirus, vibrios, salmonellae, shigellae, and enterotoxigenic E coli, with results linked to demographic and health databases of the HDSS. Kalirbazaar treatment centre is not currently included in the current surveillance system, but we will establish surveillance at this centre during this study. All children from the Matlab HDSS who present with acute gastroenteritis have stool specimens collected for testing for rotavirus by EIA. Additionally, all children have illness information recorded on a standard format Diarrhoea Treatment Record which contains information on length of illness, dehydration status, associated symptoms, outcome of illness and lab results. [This record contains sufficient clinical data to calculate a "Vesikari score" for most illnesses (Ruuska and Vesikari, 1990).Symptom data was collected using an illness form with some data abstracted from the patient record (See GE Guidelines document, attached).
Although most cases are discharged after complete recovery of diarrhoeal illnesses, there may be few cases in which the episode has not resolved fully at the time of discharge. In order to treat subjects from vaccinated and unvaccinated villages equivalently, there will be no home visits to obtain final outcomes in order to have all necessary information to score their illness according to the Vesikari score.

Data Capture
As is currently being done as part of disease surveillance in the hospital, when a child is admitted for acute gastroenteritis, the child is identified by his/her HDSS number and link the child to his/her village of residency. In all cases this can be accomplished.  (Table  3a); among infants 0 to 11 months of age from the intervention area it was 37.3 per 1000 and among infants 12 to 23 months of age it was 24.5 per 1000 (Table 3b). Because we will conduct surveillance among age-eligible infants continuously from study initiation, infants enrolled early will fall under longer surveillance while infants enrolled later will fall under brief surveillance (for example, only a few months). Additionally, because enrolment and surveillance for rotavirus diarrhoea will occur in the government comparison area and intervention area along different timelines, we used modeling to estimate a total study-period rotavirus diarrhoea hospitalization rate of 33 per 1000 infants (3.3%) (data not shown). With inclusion of infants admitted for rotavirus gastroenteritis at the Nayergaon and Kalirbazaar Community Treatment Centres, this rate may be slightly higher at 3.5%.

Sample Size Considerations
Because this is a cluster-randomized design, we must account for the intracluster (intra-village) correlation of acute rotavirus gastroenteritis. At the time of study planning, this value is not available for the study population. However, given the ubiquitous nature of rotavirus, we expected this value to be low (we assumed 0.02). Additionally, because we predefine clusters as villages for this study, we cannot set the cluster sizes to be equal; villages from the government comparison and intervention areas vary substantially in size (numbers of annual births). According to the timelines for enrolment for the 142 villages in from each area, the average cluster size is expected to be 65 with a coefficient of variation of cluster sizes is 0.96. Factoring in these values, along with the assumed intracluster correlation coefficient, into the calculation of the design effect for a cluster-randomized trial (Eldridge, 2006), we calculate a design effect of 3.48. Later estimations of the design effect based on rotavirus admission data for Matlab hospital place the design effect in the range of 2.9 to 4.5.
We calculated sample size and power based upon a fixed number of events design and conditional on the total number of cases observed (Blackwelder, 1993;Chan 1998). The number of cases in each group was assumed to follow a Poisson distribution, with parameters  V for a vaccine villages and  U for the unvaccinated villages. Under this assumption, the number of vaccine cases, x V , has a binomial distribution (X, P) conditional on X, the total number of cases, with P =  V /( V + U ). For the primary objective, we assumed 50% vaccine population (overall) effectiveness. With no design effect (DE=1), observation of at least 77 admissions for rotavirus gastroenteritis among all age-eligible infants from vaccine and unvaccinated villages would ensure that the study has a minimum power of 80% to rule out a lower bound of 0% overall effectiveness. However, with a design effect of 3.48, a minimum of 268 admissions for rotavirus diarrhoea must be observed. Assuming a 3.5% total study-period attack rate in the unvaccinated villages, 268 admissions translates into an estimated minimum sample size of 10,210 infants (5105 in each group, vaccinated villages and unvaccinated villages) ( Table 6).

Sample size for the immunogenicity subset
We assume that the seropositive rate of the immunized children with Rotarix would be 60% and for the non immunized children be 35%. A sample size of 90 in each group would allow  During the planned course of the study (through March 2011), it is estimated that a total of approximately 11,650 infants will be born into both groups (5825 per group, section 2.3.3). While this estimated total study population is higher than that required to achieve 80% power (10,210), if population effectiveness or the underlying attack rate of were lower than 50% or 3.5%, respectively, the study would quickly become underpowered. Because of these uncertainties we plan to continue enrollment and follow-up through March 2011 (which is practically to the end of the "winter" rotavirus season) in order to assure the study will have sufficient power.

General Considerations
Data analyses will be conducted using Stata Version 11 (StataCorp, College Station, Texas) and SAS  System Version 9.2 (SAS Institute Inc., Cary, NC) in a Microsoft  Windows  environment.

Exclusions:
Except for the Total population, there will be no late exclusions from the analyses based on the study"s inclusion/exclusion criteria for immunization. This is to preserve the effectiveness goals of the study.
In the rest of this SAP, "November 1, 2008" and "April 1, 2009" will be used as shorthand for the dates of first study administration of Rotarix in the government service area and ICDDR,B service area, respectively. Regardless of the analysis, infants are assigned as their village of residence the village in which they were residing when they turned 6 weeks or age, or when they in-migrated (if they inmigrated at an age greater than 6 weeks).
The following participant populations will be used for analyses: Follow-up period: Person-time will be counted from 6 weeks of age or age of in-migration until two years (731 days) of age or March 31, 2011, whichever comes first.

Secondary analysis
Those included: Identical to those in the "overall effectiveness-primary analysis" population plus those infants residing in study area villages who are between 6 and 20 weeks of age on Apr 1, 2009 (ICDDR,B service area) or on Nov 1, 2008 (Govt service area). Inclusion will then be restricted to those who received OPV1 (exclusive of birth dose) in a Matlab village after Apr 1, 2009 (ICDDR,B service area) or Nov 1, 2008 (Govt service area).
Follow-up period: Person-time will be counted from the date of OPV1 (exclusive of birthdose) until two years (731 days) of age or March 31, 2011, whichever comes first. "Total effectiveness" populations Modified Intent-to-treat (mITT) populations for total effectiveness

First priority analysis
Those included: Infants in the "overall effectiveness-secondary analysis" population who 1) received at least one dose of Rotarix and received OPV1 (exclusive of birthdose) while residing in a Rotarix village between 6 and 20 weeks of age after Apr 1, 2009 (ICDDR,B service area) or after Nov 1, 2008 (Govt service area) or 2) received OPV1 (exclusive of birthdose) while residing in a non-Rotarix village between 6 and 20 weeks of age after Apr 1, 2009 (ICDDR,B service area) or after Nov 1, 2008 (Govt service area).

Second priority analysis
Those included: Infants in the "overall effectiveness-primary analysis" population who 1) received at least one dose of Rotarix while residing in a Rotarix village between 6 and 20 weeks of age after Apr 1, 2009 (ICDDR,B service area) or after Nov 1, 2008 (Govt service area) or 2) resided in a non-Rotarix village between 6 and 20 weeks of age after Apr 1, 2009 (ICDDR,B service area) or Nov 1, 2008 (Govt service area).
Follow-up period: Person-time will be counted from 6 weeks of age or age of in-migration until two years (731 days) of age or March 31, 2011, whichever comes first. According-to-Protocol (ATP) populations for total effectiveness

First priority analysis
Those included: Infants in the "mITT total effectiveness first priority" population who complete both the second dose of Rotarix and OPV2 (exclusive of birthdose) in Rotarix villages and OPV2 (exclusive of birthdose) in non-Rotarix villages.
Follow-up period: Person-time will be counted from date of OPV2 (exclusive of birth dose) plus 2 weeks until two years (731 days) of age or March 31, 2011, whichever comes first.

Second priority analysis
Those included: Infants in the "mITT total effectiveness second priority" population who complete the second dose of Rotarix in Rotarix villages and all those from non-Rotarix villages.
Follow-up period: Person-time will be counted from 26 weeks of age until two years (731 days) of age or March 31, 2011, whichever comes first.

"Direct effectiveness" populations mITT analysis
Those included: Infants in the "overall effectiveness-primary analysis" population from Rotarix villages plus those infants residing in Rotarix study villages who are between 6 and 20 weeks of age on Apr 1, 2009 (ICDDR,B service area) or on Nov 1, 2008 (Govt service area), regardless of receipt of Rotarix.
Follow-up period: Person-time will be counted from six weeks of age until two years (731 days) of age or March 31, 2011, whichever comes

ATP analysis
Those included: Infants in the "overall effectiveness-primary analysis" population from Rotarix villages plus those infants residing in Rotarix study villages who are between 6 and 20 weeks of age on Apr 1, 2009 (ICDDR,B service area) or on Nov 1, 2008 (Govt service area), regardless of receipt of Rotarix.
Follow-up period: Person-time will be counted from 26 weeks of age until two years (731 days) of age or March 31, 2011, whichever comes first.

"Indirect effectiveness" populations
Three groups are definable for analysis of indirect effects. Indirect Population 1 Those included: Infant and child residents of Matlab who are at least 20 weeks of age as of Apr 1, 2009 (ICDDR,B service area villages) or Nov 1, 2008 (Govt service area villages) or in-migrants to the study area who are between 20 weeks and 2 years of age during the study period. Treatment group assignment for these children will be determined by village of residency at the time that first allows them to be included in this population.
Follow-up period: Person-time will be counted from Apr 1, 2009 (ICDDR,B service area villages) or Nov 1, 2008 (Govt service area villages) or date of in-migration until two years of age (731 days) of age or March 31, 2011, whichever comes first.

Indirect Population 2
Those included: Infants who resided in Matlab between 6 and 20 weeks of age after Apr 1, 2009 (ICDDR,B service area) or after Nov 1, 2008 (Govt service area).
Follow-up period: Person-time will be counted from Apr 1, 2009 (ICDDR,B service area villages) or Nov 1, 2008 (Govt service area villages) or date of in-migration until receipt of OPV1, the child turns two years of age (731 days), or March 31, 2011, whichever comes first.

Indirect Population 3
Those included: All infants or children resident of the study area aged between 6 weeks and 2 years at some point during the study period.
Follow-up period: All person-time contributed during the study period. The analysis will account for receipt of vaccine (this will include any indirect effects that also accrue to those who receive Rotarix, which is not 100% protective).

Baseline Characteristics
Summaries will be provided to describe the subject populations in this study by study arm. Only subjects with data will be included in the summaries, i.e. missing data will be counted as missing.
We will report baseline characteristics in two ways: 1) Data on infants and children who are aged 6 weeks to 2 years residing in the service areas on November 1, 2008 (April 1, 2009) 2) Data on infants who are part of the Overall cohort of those who turn 6 weeks of age over the course of the entire study period, or who in-migrated at an age less than 24 weeks Also by age groups: 6 weeks to 1 year, 1 to 2 years. Also by sex. Also, the proportions of children aged 12-23 months within each randomization unit with 1, 2, and 3 OPV doses. Also, by SES quintile…

3.4.1
 Seeking care for treatment of diarrhea at a treatment facility in Matlab, Nayargaon or Kalibazar.
 Age less than 2 years of age.
 Diarrhoea: three or more looser than normal stools in a 24 hour period.
 Acute rotavirus diarrhoea: an acute diarrhoeal illness episode in which rotavirus is identified in a stool sample collected as soon as possible but no later than 7 days after admission to the medical facility.
 Fever: rectal temperature ≥ 38C Severe rotavirus Diarrhea: Defined as a case of acute rotavirus diarrhea, as above, with a Vesikari score of 11 or greater; next most important will be for 15 greater.
Vesikari score (slightly modified for lack of full follow up of cases): The completion of the diary card for each GE episode will allow the assessment of the intensity by using a 20-point scale. In this system, points will be assigned according to duration and intensity of diarrhoea and vomiting, the intensity of fever, presence of dehydration or hospitalisation for each episode of GE as shown below: * The highest temperature recorded during the episode will be scored. ** Dehydration will be assessed clinically based on World Health Organisation criteria.

Adverse Events and Serious Adverse Events
Overall mortality in the two types of study areas will be compared with Kaplan-Meier curves among those in the Overall population, up to earliest of: out-migration, 24 months of age, Mar 31, 2010. Statistical inference will proceed via Cox models with gamma frailty.

Statistical Analyses
Here we describe the analyses according to the study aims (in italics) listed in the protocol; Vaccine population effectiveness parameters (with 95% CI) will be calculated as [1 -(incidence rate of rotavirus among vaccine villages/incidence rate among the villages without vaccine)] x 100.
For all analyses, we will handle migration of infants among study villages as follows: Regardless of the analysis, an infant or child"s village of residence will be the one in which he or she first enters the Overall population. This village assignment will be used regardless of migration within any of the study areas.

Primary Aim
To evaluate the population (overall) (Halloran, et al., 1997)  Note that this overall effectiveness is a function of vaccine efficacy and vaccine coverage. Since over 90% coverage is expected, this will not be much different from a total effects analysis. Also, if there are any indirect effects that act to reduce transmission, these may exert a differential effect over time, as it is not until near the end of the study period that the entire population of children under 2 years of age will be saturated with vaccine.
Additional analysis: same as main analysis, but including all episodes of rotavirus diarrhea, not just first episodes. The scale parameter at the randomization unit level also accounts for all within-child correlation, so the same analysis approach can be used. , whichever comes first. Also included are those who in-migrate between ages 20 weeks-2 years. Poisson regression as described above will be used; all cases (first and subsequent) will be used.

Secondary
b) Those in the Overall population, but with censoring at first OPV dose. Because of the high coverage, this will not comprise many person-years. Poisson regression as described above will be used; all cases (first and subsequent) will be used. c) All person-time contributed by infants or children who were between 6 weeks and 2 years of age at some point during the study period.
A detailed, yet "black box" analysis will use a Cox regression model with calendar timeline, with baseline hazard stratified by age at entry: 6wks-6 months, 6 months-1 year, 1yr-2yrs. The first step is to determine, for each village, for each day of the study, the time-varying proportion of infants and young children (6wks to 2 years) who have received at least 1 dose of OPV. Then, for each village, we find the 3 dates corresponding to when the village first achieves the proportions: 0.25, 0.50, 0.75. Thus, at any point in time, we can categorize each village into one of four immunization coverage categories: 0-24%, 25-49%, 50-74%, and 75-99%. There may be some misclassification in that a village may reach 27% one day and drop to 20% a month later, but these variations will be too difficult to track, and should not have much effect.
Crossing these four categories with whether a village is a Rotarix or control village yields 8 timevarying categories, which can be represented by 7 covariates: where R and C correspond to Rotarix and control villages, respectively, i is for the ith infant, and t is for the infant"s age in days. For example, if a given infant living in a Rotarix village is 87 days old on a given day, and on that day the village has already reached 25% coverage but not yet 50% coverage as defined above, then 25 49 it R  =1 and the other 7 covariates are coded as 0. Note the reference category is 0-24% coverage in a control village. In addition, we will track the immunization status of each infant with the variable Vac it which starts as 0 and becomes 1 upon first OPV. This variable is not of much interest-its coefficient will estimate any protective effect due to selection for vaccination. Crossing this term with village type of residence, denoted by X i (=1 for Rotarix, 0 for control) yields an interaction term that indicates the individual-level protection afforded by Rotarix. Putting this all together into a specification of the hazard function, we get: