# Identifying confounders and estimating the causal effect of antenatal care on age-specific childhood vaccination

## TL;DR

This study examines how the number of antenatal care visits affects childhood vaccination rates, identifying key confounding factors like maternal age and household wealth.

## Contribution

The paper introduces a novel comparison of significance testing and change-in-estimate methods for confounder identification in causal analysis of antenatal care and vaccination.

## Key findings

- The zero-inflated Poisson model best modeled covariate-exposure associations.
- Significance testing methods outperformed change-in-estimate methods for confounder identification.
- Key confounders include maternal age, region, and household wealth.

## Abstract

Immunization is an efficient and cost-effective public health program. It averts millions of child deaths per year. It is taken as one of the main interventions that can be used to achieve the third Sustainable Development Goal, which is to end preventable deaths of newborns and under-five children by 2030. The study was done with the aim of identifying appropriate confounder identification methods and examining confounders for the causal effect of a number of antenatal care visits on age-specific childhood vaccination.

A family of generalized linear models with log link functions was used to model the covariate and the number of antenatal care association. A cumulative link model was used to model the number of antenatal care and covariate-age-specific childhood vaccination associations. AIC and BIC values were used to compare models. Significance testing methods and change in estimate methods were used to identify covariates that confound the effect of a number of antenatal care on age-specific childhood vaccinations.

A zero-inflated Poisson model was selected to model covariate–exposure association, and a proportional odds model with a log link was selected to model the outcome variable. Among significance testing methods, the common cause approach yielded smaller values of BIC and a smaller number of covariates. However, the likelihood ratio test showed no difference between the common cause and other approaches. A change in the estimate method is more conservative at a 10% cut point, which selects a smaller number of confounders. However, the significance testing method was better performed than the change in estimate method.

The significance testing method with a p-value of less than or equal to 0.2 performed better than a change in estimate method at a 10% cut point of effect change for confounder identification. Mothers’ age at first birth, region, place of residence, education status of mothers, presence of radio and television in the household, religion, household size, wealth status, total children ever born, and birth order number are identified as confounders.