Evaluation of community level interventions to prevent HIV infection presents significant methodological challenges. Even when it is feasible to randomly assign a treatment versus control level of the intervention to each community in a sample, measurement of incident HIV infection remains difficult. In this talk we describe an experimental design developed for the SEARCH Trial, a large community randomized trial that will evaluate the impact of expanded treatment on incident HIV and other outcomes. Regular community-wide testing campaigns are conducted and a random sample of community members who fail to attend a campaign are tracked. The data generated by this experiment are subject to non-monotone missingness; however, the missing at random assumption is known to hold by design, and the missingness mechanism is known. We present two-stage targeted minimum loss-based estimator (TMLE) of the randomized intervention on incident HIV infection using these data. The method described can also be applied, under additional non-testable assumptions, to estimate the effects of non-randomized community level interventions in the setting of incomplete tracking success.