Sampling designs for xerophthalmia prevalence surveys.

BACKGROUND The purpose of this study was to estimate the bias and design effects associated with the Expanded Program on Immunization's (EPI) sampling design when estimating xerophthalmia prevalence, and to estimate the savings associated with EPI in terms of distance travelled within selected clusters. METHODS Computer simulation of the EPI sampling strategy was done using maps from a xerophthalmia survey of 40 wards in Sarlahi district, Nepal. Samples of fixed cluster sizes of 7, 10, 15, 20 and 25 were compared. The estimated prevalence using the EPI design was compared with the true prevalence in the 40 wards to estimate the bias. The design effect was estimated by taking the ratio of the variance under EPI sampling to that of stratified random sampling (SRS) with fixed cluster sizes. The EPI was also modified by increasing the distance between selected houses from nearest neighbour to skipping 1-4 houses between selected ones. The difference between the distance travelled within clusters using SRS compared with EPI was weighed against the bias and increased variance. RESULTS The prevalence of xerophthalmia was 2.8%. The EPI design overestimated xerophthalmia prevalence by between 0.27% and 1.16%. The design effects of EPI cluster sampling within wards varied between 0.73 and 1.35. Neither the bias nor the design effect was related to distance between households or cluster size. Distance travelled within wards was always less for EPI designs with cluster sizes of 7 or 10. There was no saving in terms of distance travelled for designs with cluster sizes from 15 to 25 if there were two or more houses between selected ones. For fixed cluster sizes of 15 or fewer, the EPI sampling design using nearest or next nearest neighbours is a better choice than SRS in terms of minimizing the distance travelled and the mean square error. CONCLUSIONS The choice of an optimum method would need to account for the density of clusters and difficulty of travel between clusters, as well as the costs of travel within clusters. Based on certain assumptions, EPI with 15 children per cluster would be favoured over examining all children in selected wards unless the travel time between wards was more than 2 days.