The methodological ‘revolution’: caution accepted

Excerpt

Porta and Bolúmar address the issue of ‘overusing’ new method developments in epidemiology at the expense of more classical and transparent methods [1]. They, thus, follow an old tradition of senior epidemiologists who question the need for and usefulness of new designs and new ways of analyzing data. Such a reaction is needed now as it has been in the past and it is warranted. This does not mean we should disregard these technical advancements—and Porta and Bolúmar do not state such a point of view themselves, but rather use a statement from one of the reviewers to advance this view. We should, however, use all the methods with care, and do all we can to understand the limitations of these methods—new and old. The New England Journal of Medicine (NEJM) was, for example, very quick to demand the use of multiple imputation rather than simple (old) methods to handle problems of missing data. The old methods were obviously limited and subject to bias but they were easy to understand and were usually interpreted with great care. More complicated methods may give the impression that selection bias is avoided by using a large number of fictitious data modelled using strong and often completely unrealistic assumptions. …