*2.4. Statistical Analysis*

The weighted study outcomes were, (1) the percentage of children who received the MR vaccine during the SIA, (2) the percentage of measles zero-dose children who received the MR vaccine during the SIA, and (3) MCV1 and MCV2 coverage before and after the SIA. In addition, the impact of the MR-SIA was estimated by calculating the increase in MCV1 and MCV2 coverage after the SIA. For children whose vaccination cards or records were available, we assessed routine vaccination status before the SIA to estimate the proportion of Gavi-defined zero-dose, under-immunized and fully vaccinated children [13]. Gavi defined "zero-dose children" as children who have not received the first dose of the DTP vaccine. "Under-immunized" children were defined as those who missed a third dose of the DTP vaccine. Together, zero-dose and under-immunized children formed missed communities. To assess MCV coverage, we defined "measles zero-dose" children as children who did not receive MCV1. Unvaccinated was defined as children whose parents or caregivers could not recall their children's vaccination status. Sensitivity analyses were conducted by (1) treating these children as vaccinated and (2) restricting the analysis to children whose vaccination status was card-confirmed.

Log-binomial regression analysis was performed to assess factors associated with missing MR-SIA vaccination. Participant characteristics for enrolled children (sex, setting, age, and DTP vaccination status) and household characteristics (relationship of head of household with enrolled children, maternal education level, maternal COVID-19 vaccination status, and travel time) were included in the univariable analysis to identify social determinants associated with missing MR-SIA vaccination. The prevalence ratio (PR) and the corresponding 95% confidence intervals (95% CI) were calculated. Age-adjusted logbinomial regression was used to address heterogeneity between age categories. Forward and backward stepwise selection methods were used to select the best-fit model. The final multivariable model included variables with a *p*-value < 0.05 from age-adjusted univariable analyses and variables that were of public health importance. Log-likelihood and Akaike's information criteria (AIC) were used to determine the goodness of fit, and the model with the lowest AIC was selected as the best-fit model. All statistical analyses were performed in R version 4.1.3, and the model was run using the survey package to account for sampling weights [14,15]. Figures were generated using the ggplot2 package [16].
