Sample and data collection
At the time of the sample selection, Lasbela district consisted of five tehsils and had 22 union councils - 5 urban and 17 rural (the sampling and data collection took place before the addition of newly created tehsils, now totalling nine). We selected a stratified random cluster sample of communities to give representation of the situation in the different tehsils or talukas. First, union councils were randomly selected from each tehsil, reflecting urban/rural spread and with the number according to the population in each tehsil. We included a minimum of four union councils per tehsil to allow for tehsil level findings if needed for district level planning purposes. The official list of union councils provided by the district government was used as the sampling frame for the selection of union councils. From each union council we randomly selected one community (village or mohalla) from the list of communities in the union.
We drew a stratified random sample of 23 rural and 9 urban enumeration areas in Lasbela district to allow adequate representation of the heterogeneity across the district - particularly to allow tehsil level representation and urban/rural differentiation. In each selected community, the sample included a group of 100 contiguous households with children under five years, spreading out from a random starting point. There was no sampling within the site; all the eligible households were included.
Data collection instruments included a household questionnaire, which asked about household demographics and socio-economic status, and a questionnaire for the mothers or caregivers of children under 60 months old, which asked about the mothers' education, vaccine related knowledge, attitudes and practices, and about vaccination and illnesses of the children. The field teams also completed community profiles for each village of each community, by means of discussion with knowledgeable people and their own observations, including information about the location of facilities providing vaccination services and visits from mobile vaccination teams. It is possible for there to be different results within each community for this data as the community profiles were completed at the village level. For some indicators (i.e. visits by a vaccination team) there was missing data in the community profile for some villages, therefore reported denominators for these variables are smaller than variables from the household questionnaire.
Field teams comprising mainly women interviewers undertook the household survey in March and April 2005. After preliminary analysis of the household data, the teams returned to all of the sample communities in July 2005 and discussed the findings. Separate male and female focus group were conducted in each sample community [11].
Analysis
Data entry used the public domain software package EpiInfo [28]; double data entry with validation reduced keystroke errors. Analysis relied on CIETmap open source software [29, 30]. Although the sample drawn from each tehsil reflected its relative population size, this was not exact. To take into account under- and over-sampling of tehsils, we calculated population weights and applied these when making district level estimates. All the district level estimates reported in this article are weighted.
We examined associations between measles vaccination (among children aged 10-59 months), and related factors using the Mantel Haenszel procedure [31]. We first tested crude associations in a sequential analysis (stratifying by one factor at a time) and then used a multiple stratification - analogous to logistic regression analysis [32] - stepping down from an initial saturated model. Final results are presented as adjusted Odds Ratios (OR) and 95% confidence interval. Initial sequential stratification revealed that the associations between many of the variables and measles vaccination were different between urban and rural communities. We therefore built separate models for urban and rural settings.
In order to adjust for clustering, we applied Gilles Lamothe's robust variance estimator for cluster-correlated data to the Mantel-Haenszel stratification. Based on the odds ratio, the Lamothe estimator weights the effect rather than simply the in-cluster correlation. The adjustment works for medium and large data sets, where zero margins are not an issue.
Measurements of trend of vaccination uptake used the Mantel extension [33] calculated using the Statcalc module in Epi Info.
We used measles vaccination, as reported by the mother, as an indicator of vaccination coverage. In addition to uptake of measles vaccination as our primary outcome, we considered intermediate outcomes based on a behaviour change model called CASCADA, first developed in a study of HIV and AIDS prevention in 2001 [34] and subsequently used in developing an intervention to improve vaccination rates [24]. This model extends the knowledge, attitudes and practice model, adding more intermediate outcomes between knowledge and action. These include conscious knowledge (able to correctly identify an illness preventable by vaccination), attitudes (think it is worthwhile to vaccinate), subjective norms (neighbours think it is worthwhile to vaccinate), intention to change (willing to take time away from daily activities to vaccinate), agency (mother is involved in decisions about vaccination), discussion (discuss vaccinations within the household), and action (uptake of measles vaccination).
We defined several vulnerability variables to describe inequities between households and children that might be relevant to the uptake of measles vaccine.
Access
We divided children according to whether they lived within 5 km of a government facility offering vaccinations, and whether they lived in a community that was visited by a mobile vaccination team.
Type of roof
As a proxy for economic status we used roof quality, grouping roofs made of reinforced concrete, iron, asbestos or T-iron as good quality, and roofs that were thatched, mud or wood as poor quality.
Occupation of main breadwinner
Keeping in view the problems that are faced in asking directly about the household income, we used occupation of the main breadwinner as a proxy to the household economic status. We then grouped the households into those where the main breadwinner had an occupation with a potential of better yield in terms of income (such as skilled workers and office work) and those with a relatively poor occupation (such as unskilled worker or unemployed).
Household crowding
We calculated room occupancy by dividing the number of household members by the number of rooms in the household. We classed households with room occupancy of four or more as crowded.
Education of the mother
The education and literacy of women in Lasbela is low. We categorised mothers and caregivers according to whether they had any formal education or not.
Household visits from a lady health worker (LHW)
LHWs in Pakistan are an important source of preventive education and information for mothers. LHWs in Pakistan are considered as the prime source of preventive education and information to the households. They also counsel and motivate caregivers and household decision makers to immunize their children. We defined access to LHWs as mothers who had been visited by an LHW and who received information about vaccinations from the LHW.
Higher level variables
We generated higher level variables to test the combination of equity-related risk factors when these factors did not have a significant effect on their own. For example, in the rural multivariate model, we considered those who had the double disadvantage of poor access (further than 5 km from a government facility offering vaccination) and poor quality roofs.