Selecting end point variables for a community intervention trial

Abstract

Purpose: Community interventions, while aimed at affecting individual health behavior, are typically applied at a level of organization higher than the individual, such as schools or worksites. Evaluation of a community intervention trial may likewise involve either direct measurement of the environment or aggregation of individual-level end points. We review some statistical considerations that enter into the choice of end points for a community trial. Methods: Statistical precision is what ultimately determines the power of a trial to detect an effect of intervention. Random-effects analysis of variance provides an apt model for assessing statistical precision in the presence of several levels of between-unit and within-unit variation, as are typically present in a community trial. Results: Three examples of multilevel variation are provided, one based on a simple measurement problem and the other two on data from the Child and Adolescent Trial for Cardiovascular Health. The examples demonstrate that choosing to conduct measurements at a higher or lower level of organization can profoundly influence the power, cost, and interpretability of a trial. Cconclusions: The best end point for a community trial is one that is sensitive and specific to the intervention. The level of measurement need not match the level of intervention exactly; however, too great a discrepancy may be deleterious, lowering statistical power and blurring interpretation. Careful assessment of variance at each level can help an investigator choose the best level for evaluation and the optimal sample size at each level.