In general the difference is likely due to the selection of only de facto respondents (those who stayed in the household the previous night) for the individual women and men datasets. You could confirm this by linking the AR data to the PR data, and seeing if the cases that are dropped are non de facto (did not stay in the household - hv103). For Burundi, in addition, there are more than 5000 children who were tested (see variable slage in the AR dataset), and who would not match with the individual women and men's datasets.