A framework for classifying study designs to evaluate health care interventions.

Researchers who are trained in epidemiology recognise and use the standard range of designs, i. e. retrospective and prospective cohort studies and case control studies, for studying aetiological questions. The application of study designs to health care interventions is more complex. Researchers may modify familiar designs, in response to specific problems posed by the interventions being evaluated, and an unambiguous nomenclature has not been established. Also, researchers who set out to evaluate interventions are not always trained in epidemiology and often use familiar study design labels in an ambiguous or inappropriate manner. The susceptibility to bias of different study designs is a critical consideration for users of research evidence. Sources of bias are the same for all studies, i. e. selection, performance, detection and attrition bias, but the susceptibility of studies is likely to vary depending on their design. Evidence of associations between features of study designs, specific biases and their consistent influence on effect estimates is lacking. A framework for classifying study design will be proposed, based on key features of study designs, i. e. what researchers actually did. This framework may (a) help to reduce the ambiguity about study design labels, and uncertainty about how a study was actually carried out, and (b) help methodological researchers to gather evidence about associations between different study design features and susceptibility to bias.