Estimating the population direct effect in cluster randomized trials of targeted interventions using contemporaneous counts in a non-target group.
Moore Kerryn A KA, Cheung Yin Bun YB, Fogelson Ari A, Milligan Paul P
In cluster randomized trials of interventions that target a subset of the at-risk population, contemporaneous event counts in a non-target group can be used to estimate the population direct effect if indirect effects are unlikely, but the application of this approach has been limited. We describe the rationale for using the contemporaneous counts approach in the WHO Malaria Vaccine Program Evaluation and how it was applied, and in a re-analysis of a human rotavirus vaccine trial, we demonstrate its application and compare results with conventional methods using event rates per unit of person-time. Indirect effects were considered unlikely because the target group represents a very small proportion of the at-risk population. The population direct effect of intervention introduction in the target group is estimated by comparing between arms the ratio of event counts in the target group to the non-target group. Expanding surveillance into a non-target group may often be possible without significant additional costs. Confounding due to chance imbalance is reduced implicitly insofar as there is better balance in the relative difference between the target and non-target groups than in the variable itself within the target group alone. The approach can improve efficiency when target group event counts are more strongly correlated with non-target-group counts than they are with target group person-time, which is more likely as heterogeneity increases. The contemporaneous counts approach can increase the efficiency and internal validity of cluster trials and reduce their operational complexity because person-time denominators are not required.