Determining Risk Factors of Antenatal Care Attendance and its Frequency in Bangladesh: An Application of Count Regression Analysis

Standard Poisson and negative binomial regression models are the common count regression analysis tools for modelling the number of antenatal care (ANC) visits. Two-part (zero and count) models like zero-inflated and hurdle regression models are recommended for modelling ANC visits with excess zeros. The intra-cluster correlation (ICC) can be accounted by incorporating cluster-specific random intercepts in the corresponding standard and two-part models. The existence of excess zeros in the distribution of ANC visits in Bangladesh raises the issue of identifying a proper count regression model for the number of ANC visits covering the issues of overdispersion, zero-inflation, and ICC in determining the risk factors of ANC use and its frequency. The data have been extracted from the 2014 Bangladesh Demographic and Health Survey. The hurdle negative binomial regression model with cluster-specific random effects at both zero- and count- parts is found as the best fitted model. Women who have poor education status, live in poor households, have less access to mass media, and belong to Sylhet and Chittagong divisions are less likely to use prenatal care and to have more ANC visits. In addition, women who live in rural areas, depend on other family members’ decision for taking health care, and have unintended pregnancies had lower tendency to more ANC visits. The findings recommend incorporation of random community effects along with overdispersion and zero-inflation in modelling the ANC data of Bangladesh, and model selection should be model-driven rather than data-driven since practically assumption of structural zeros is tough to meet.