The cluster survey showed that immunization coverage for three doses of DPT/pentavalent vaccine in Kilifi in 2001–2002 was very high (91–100%). The results were underpinned by good retention of vaccine cards in younger children (88%) and reasonable retention in older children (75%). Among older children, only three of 153 children with a vaccine card did not have written evidence confirming receipt of DPT1; among the mothers of these three children, and the 50 children who did not have a vaccine card, every single one reported that her child had received this dose. Results for pentavalent 1 in the younger sample were similar. In both samples an extremely high proportion of mothers reported that their children had received doses two and three.
It is clear from the vaccine card evidence alone that immunization coverage was high but we were concerned about the validity of the mothers recall data which was used to estimate coverage more accurately. The SRS survey attempted to reduce the reliance on mothers' histories and introduced detailed interview about the immunizations. We reasoned that mothers given an appointment at the hospital would have time to find misplaced cards  and that interviews conducted at hospital, closely monitored by the investigator, would be superior to field interviews. In the cluster-survey many children were vaccinated long after the target date and some appeared to have been vaccinated too early. In the SRS survey potential transcription errors among outlier dates were counterchecked against a stored digital image of vaccine document. In a previous study from Egypt, data errors were more frequently attributable to interviewer factors and data processing than to maternal recall .
Immunization coverage in the SRS survey was also high; 88% of children aged 9–23 months had received three doses of pentavalent vaccine. Despite differences in the methodology, and the fact that the median age of children in the SRS survey was lower, the proportion of children with cards was no greater than in the first survey. As anticipated, however, coverage estimates among those without vaccine cards were slightly lower using hospital-based interviews. The differences between the two surveys, however, may have resulted from changes in coverage over time.
Validation of mothers' histories against complete card entries suggested that mothers in Kilifi have some loss of recall for second and third vaccine doses, indicating that we have slightly underestimated coverage for doses two and three by including recall data. However, in an area with incomplete vaccine card retention coverage is estimated most accurately by combining both card and history data . To analyse only written evidence of immunization from the whole sample would significantly underestimate coverage since vaccine cards are not necessarily retained after the vaccine schedule is completed; to analyse only those who retained a card would overestimate coverage by excluding children who never attend hospital facilities. We included mother's recall data in the analysis of vaccination timeliness too because we expected that mothers who did not retain vaccine cards would be less likely to adhere punctually to the vaccine schedule. This turned out to be the case (analysis not shown).
To examine small changes in immunization coverage over time, evidence from mother's memories is invalidated by recall bias. Restricting to card data, and controlling for the potentially confounding effect of age by analysing coverage at 14 or 18 weeks of age we found that coverage was significantly greater for doses 1 and 3 after introduction of Pentavalent vaccine than before. This improvement occurred despite the fact that logistic requirements for Pentavalent vaccine distribution were greater because the per dose package size was more than six times greater than for DTP vaccine alone.
Administrative coverage estimates were slightly higher than those obtained from direct surveys of children, as has been observed elsewhere in Africa [18, 19]. The low coverage estimate from clinic registers suggests incomplete reporting at clinic level. Reliance on reported data has more frequently led to overestimation of immunization coverage by developing countries, when compared to survey data . Administrative coverage calculations are particularly sensitive to estimates of vaccine wastage. Assuming that the SRS survey coverage is accurate, and that inaccuracies in the national and provincial administrative estimates are due entirely to vaccine wastage, we can derive indirect estimates of vaccine wastage of 10.9% and 12.9% respectively, slightly over twice the acceptable level for the vaccine.
For a disease like Hib which peaks early in infancy (Cowgill et al, submitted), timeliness of immunization will increase the directly protective effect of immunization. Two doses of Hib-conjugate vaccine are considered sufficient to induce protection against disease . The median age of immunization with two doses in the SRS survey was 3 months, and it was not until after 9 months of age that 90% of children were immunized with a second dose of vaccine. Clearly the protection of a substantial risk group aged <3 months depends entirely on indirect vaccine protection.
In Egypt immunization coverage declined with increasing distance from the vaccination clinics . We observed a similar effect with distance in Kilifi, but the size of this effect was small probably because the median distance to the nearest vaccine clinic was 4 km and therefore vaccination is geographically accessible to the majority of children. Immunization was more strongly associated with annual patterns of rainfall than with the actual precipitation records from the 14 days before each dose was due. This suggests that longer term seasonal changes, such as the requirement to plant crops at a remote small-holding or the increase in costs of public transport fares, which occur at predictable calendar points in relation to anticipated rainfall patterns, are greater impediments to immunization than the rain itself. Parents of large families are less likely to immunize their children. This may simply reflect an association between use of family planning and use of immunization services, or it may indicate that mothers at home are unable to bring their infant for immunization because of the practical difficulties and expense of having other children at home. Kilifi is a poor district in Kenya with high infant mortality rate; better coverage would be expected in other districts and Kenya as a whole.