tific trials—practitioners are likely to be blamed for deviating from the protocol when other factors, such as a different culture in the target population, may be responsible.
The reality that most who use evidence to support obesity prevention policy or practice are likely to face is that the target population for an obesity prevention intervention will be more diverse than the populations in the original studies (Cohen et al., 2008). This situation often leads to the fidelity problem because users need to adapt the protocol, but it also raises issues related to generalizability that characterize much research; in particular, the implementation of the intervention is highly controlled, and the study population settings, procedures, and circumstance often are not representative of the typical settings or target populations at large.
The generalizability problem relates in part to a distinction between function and form with respect to each component of an intervention (Hawe et al., 2004). Functional components are those regarded as essential to successful implementation wherever and with whomever the intervention might be applied, whereas those components of the intervention that are a matter of form usually lend themselves to successful adaptation to different populations, settings, and circumstances. While form should in general follow function, it is necessary to decide which components of an intervention are essential for implementation and which can bear modification. Published studies usually do not offer guidance on these decisions. The research community needs to provide more guidance to implementers who are faced with the need to tailor components of an intervention to individuals or to various population segments (cultures, genders, age groups, etc.), settings, and times.
Decision makers should also consider whether the research process, including recruitment and informed consent procedures, screening, and attrition, led to an unrepresentative pool of subjects on which the studies’ conclusions were based. This issue is akin to that of the representativeness of the cases on which the conclusions were based. It stems, however, not from the homogeneity or unrepresentativeness of the original population sampled. Rather, sampling is altered by the experimental conditions; the study protocol; and/or dissatisfaction with questionnaires, blood draws, weighing, or other research procedures—all of which could produce an unrepresentative final sample after the initial representative population was properly sampled.
Finally, decision makers should consider that obesity is unlike many of the health-related problems to which the usual canons of evidence-based medicine apply, especially those problems that have a singular causal agent, a straightforward mechanism of causation, or a relatively consistent result for all who would receive a standardized intervention. Obesity should be seen as belonging in a class of problems that are influenced by a constellation of complex social and political factors, some of which change during the process of solving the problem. Such a problem is likely to be viewed differently by different populations, practitioners, and vendors of services and products that influence it, depending on the perspectives and biases of those with a stake in the problem (Kreuter et al., 2004). Therefore, such problems are less amenable to the usual methods of randomized experiments than most of the more strictly