Background: A cluster randomized trial (CRT) provides an ideal framework to evaluate mass drug administration (MDA) intervention strategies for control of trachoma, a leading infectious cause of preventable blindness. Heterogeneity is a crucial consideration in the design and analysis of CRTs, as it can occur as clustering of the outcome and clustering of not receiving treatment (nonparticipation during MDA). Methods: Data from a CRT of MDA interventions for trachoma control in The Gambia (NCT00792922) were used to investigate clustering of, and risk factors for non-participation. Simulation studies investigated implications of non-participation occurring independently of baseline infection status (homogeneously, analogous to MDA coverage<100%) and amongst those with infection at baseline (heterogeneously), on power to detect pre-specified effect sizes in intention-to-treat (ITT) analysis in CRTs. An ITT analysis evaluated population-level effectiveness of azithromycin MDA on a secondary outcome of the CRT in The Gambia of all-age all-cause mortality. A pragmatic bootstrapping approach, simultaneously adjusting for clustering of both mortality and non-participation to minimise bias, was used to estimate a complier average causal effect (CACE), as an indication of efficacy in those who receive treatment. Results: Non-participation clustered repeatedly in the same households over three annual MDA rounds. Increased numbers of clusters in CRTs are required to allow for non-participation amongst individuals infected at baseline. The ITT rate ratio for two annual MDA rounds versus no MDA on all-age all-cause mortality was 1.11 (95% CI: 0.85-1.44). The CACE rate ratio was 1.36 (0.86- 2.79). Conclusions: CRT design should include consideration of likely prevalence of non-participation amongst infected individuals, rather than relying solely on expected coverage, during mass treatment rounds. Results were inconclusive about whether large-scale azithromycin MDA could yield important reductions in mortality in The Gambia. Estimating efficacy from CRTs of MDA interventions without bias remains a challenge.