Missing data are a common problem in randomized clinical trials. In the presence of missing data, inference about treatment effects relies on untestable assumptions about the distribution of missing outcomes. To address this issue, a recent National Academy of Sciences report recommends the use of sensitivity analysis anchored at a benchmark assumption. In this talk, we develop, in the context of a randomized tuberculosis study, a sensitivity analysis scheme for non-monotone missing data centered around a benchmark assumption that leverages all the available data. We develop a likelihood-based procedure to multiply impute the missing solid culture conversion data. This is joint work with Maria Abraham and Andrea Rotnitzky.