Statistical analyses

The distribution of cohort characteristics between exposed and un-exposed groups were compared using Pearson’s chi-square test. Fisher’s Exact test was used in the case of smoking, since less than 5 women smoked. Participants with missing values were excluded from the analysis.

Logistic regression was performed to estimate odds ratio (OR) with 95% confidence interval between IPV and premature termination of EB. We first estimated the crude OR and then adjusted for potential confounders, which were selected based on a priori assessment suggested by a directed acyclic graph (DAG),[29] including maternal age, education, HIV-status, alcohol use during pregnancy and parity(S1 Fig). As shown in the supporting information, we assume postpartum depression to be part of the pathway to reduced breastfeeding, and as such it is not adjusted for in the regression model. Instead it is addressed in the stratified analysis.

In order to determine if any groups were more vulnerable to the effects of IPV than others, we conducted a stratified analysis comparing women who were exposed to at least one kind of IPV at any point in their relationship to those who were not. We chose the variables maternal age, education, EPDS Score, HIV status and alcohol consumption as well as child sex for the stratification. The selection was made a priori based on clinical experience.

Statistical analyses were performed using STATA software package (version 15).