Kilifi District, the second poorest in the country, is located on the Indian Ocean Coast of Kenya [10]. In the 1999 Kenya Government census it had a population of 544,303. A demographic surveillance study (DSS) was established in 2000 in the sub-population of 210,000 living closest to Kilifi District Hospital, with 4–6 monthly enumeration and registration of vital events. This was the target population for our surveys. Ethical approval for the study was granted by the Kenya Medical Research Institute (KEMRI) National Review Ethical Committee. Verbal consent was sought from participants during invitation to take part in the study. Pentavalent vaccine was distributed rapidly in Kilifi District, and all old DPT vaccine stock recalled, during October 2001. Immunizations are recorded on vaccine cards but children who attend for vaccination without their cards are immunized nonetheless.
Package volume
We measured the dimensions of the DPT package and the two packages of the pentavalent vaccine that replaced it.
WHO cluster coverage survey
We simultaneously conducted two WHO-EPI cluster coverage surveys among children of different ages during November-December 2002. The first survey sampled children who were old enough to have completed the 3-dose DPT series prior to introduction of pentavalent vaccine (born between 25th June 2000 and 24th June 2001); the second sampled younger children who should have received 3 doses of pentavalent vaccine (born after 19th September 2001 and at least 14 weeks of age at time of survey).
We followed the WHO-EPI Coverage Survey Manual [11], using the DSS to select at random the starting household for each of 30 clusters. Fieldworkers continued sampling adjacent households until 7 appropriately-aged children were identified in each cluster. Vaccine records or parents'/guardians' (usually mothers') histories were transcribed to WHO-designed forms at the household by trained fieldworkers and double-entered onto a WHO-designed computer programme (WinCosas, Vaccines & Biologicals, WHO, Geneva).
SRS coverage survey
In March-April 2004 we did a second survey by simple random sampling (SRS) of the DSS database. We ranked >9000 resident children aged 9–23 months (born between 1st Jun 2002 and 31st Jul 2003) by applying random numbers. Starting with the highest ranked child we invited children to take part sequentially until 100 of each sex had been recruited. We asked participants to attend appointments at the research unit where their immunization records were transcribed and digitally scanned. If no immunization record was available or the record was incomplete, the mothers were interviewed using a standard questionnaire. Site of immunization was used to identify the antigens from recall data e.g. only pentavalent vaccine is given into the left lateral thigh. To evaluate the sensitivity of mothers' recall we also interviewed two mothers each day whose children had complete immunization records. The very high response by participants ensured minimal sample bias. Card and questionnaire data were double-entered in Epi-Info (Centers for Disease Control and Prevention, Atlanta). We chose a simple random sample because households in Kilifi are dispersed not clustered in villages [12]; because a DSS register was available; and because a SRS of similar size to the cluster surveys would provide more precise estimates of coverage [13].
Coverage by clinic register and administrative methods
The numbers of pentavalent vaccine doses delivered to Kilifi District between October 2001 and December 2003 were obtained from the Kenya Expanded Programme on Immunization (KEPI) national stores in Nairobi and provincial stores in Mombasa. The monthly returns to KEPI of the number of doses recorded as given at vaccine clinics in Kilifi District were also collated for the years 2002–3.
Pentavalent vaccine is packaged as a two-dose vial and unused vials must be discarded at the end of each day. We assumed that the daily probability of wasting a single dose in a functioning clinic is 0.5. With 30 vaccination clinics working 20 days a month, there are 7200 vaccine-clinic days leading to 3600 wasted doses per year. As 134,800 doses were delivered to the district in 2002–2003, this suggests a minimum wastage of 5.3%. To adjust for losses due to distribution problems we estimated a total wastage of 10%.
Population denominator and Geographic Information Systems (GIS) analysis
Annual population growth in rural Kenya in 2003 was 3% and the crude birth rate was 38.6/1000 [14]. We estimated the population of Kilifi District to be 594,774 and 612,618 in the years 2002 and 2003, with birth cohorts of 22,958 and 23,647 respectively. After adjustment for a neonatal mortality rate of 45/1000 live births, this yielded vaccine cohorts of 21,925 and 22,583 children. We digitally mapped subjects' residences in the DSS area and the vaccination clinics in the whole of Kilifi District. The shortest distance from each household to the nearest vaccine clinic was determined directly using ArcGIS v9.0 (ESRI, Redlands, CA).
Data was analysed using STATA v8.2 (Stata Corp, College Station TX). Immunization prevalence was determined for the date of the survey. Timeliness of pentavalent vaccine uptake was determined in the SRS survey using an inverse Kaplan-Meier survival function [15]. In unvaccinated children the time to the next eligible dose was censored on the date of survey. Precision estimates for the cluster survey were modified by a design effect of 2.6 calculated from the data [13]. Administrative coverage was the annual number of doses delivered to vaccine clinics (adjusted for 10% wastage), or administered at clinics, divided by the yearly vaccine cohorts.
A Cox proportional hazards model was fit to recurrent vaccination data to determine the contributions of sex, clinic distance, family size, mother's age and rainfall season to immunization rates. Survival periods were synchronized to start 28 days before each dose was due and separate survival times were entered into the data for each of the three doses for each child, tied by child in a shared-frailty model [16]. Variables were excluded if they did not contribute significantly to model fit (i.e., Likelihood Ratio Test p value >0.05). We tested the proportional hazards assumption using Schoenfeld residuals and by examination of Cox predicted and log-log curves. The resultant hazard ratios are here called age-specific immunization rate ratios or, for brevity, immunization ratios.
Kilifi has rainy seasons from April to July and October to November during which the mean precipitation measured in daily rainfall records in Kilifi from 1993–2004 is = 3 mm. In the other six months mean daily precipitation is <3 mm. A survival period was categorised as rainy if it began during a rainy season month. We also constructed, for each immunization event, the mean daily rainfall for 14 days before each dose was due.