Strengthening nonrandomized studies of health communication strategies for HIV prevention.

BACKGROUND: There is growing interest in impact evaluations of health communication (HC) interventions for HIV prevention. Although cluster randomized trials may be optimal in terms of internal validity, they are often unfeasible for political, practical, or ethical reasons. However, a common alternative, the observational study of individuals who do and do not self-report HC intervention exposure, is prone to bias by confounding. Cluster-level quasi-experimental study designs offer promising alternatives to these extremes.
METHODS: We identified common rollout strategies for HC initiatives. We mapped these scenarios against established quasi-experimental evaluation designs. We identified key issues for implementers and evaluators if these designs are to be more frequently adopted in HC intervention evaluations with high internal validity.
RESULTS: Stronger evaluations will document the planned intervention components in advance of delivery and will implement interventions in clusters according to a predefined systematic allocation plan. We identify 4 types of allocation plan and their associated designs. Where some places get the HC intervention, whereas others do not, a nonrandomized controlled study may be feasible. Where HC is introduced everywhere at a defined point in time, an interrupted time series may be appropriate. Where the HC intervention is introduced in phases, a nonrandomized phased implementation or stepped-wedge design may be used. Finally, where there is variation in strength of implementation of HC, a nonrandomized, dose-response study can be planned. DISCUSSION: Our framework will assist teams planning such evaluations by identifying critical decisions for the implementers and for the evaluators of HC interventions.