*Geek Box: Retrospective Analysis
If you’ve heard of a retrospective study, most likely it is in the context of a retrospective cohort study, in which the study is conducted after participants have developed a particular outcome of interest. Such a study tends to look back at records or other data to try to identify any relationship between an exposure and an outcome.
For example, let’s say the investigators think there may be a relationship between exposure to lead in paint factories and cancer incidence. They could look at back records of people who died from cancer, that also worked in paint factories, and compare them to an unexposed control group. They could identify a higher prevalence of cancer incidence. As you can probably see, as data is not collected at baseline, and as the study participants are all identified and compared after the outcomes have occurred, retrospective studies are more prone to bias than other observational research.
But what about a randomised controlled trial, can there be a ‘retrospective RCT’? Well, yes: all this means is that a retrospective analysis is conducted with data from an RCT. This means that it is not part of the direct hypothesis tested in the initial intervention, and the analysis is conducted to explore further relationships in the data. The difference may also be defined as explanatory vs. exploratory. In an explanatory RCT, the intervention is designed with the hypothesis of testing whether Exposure A results in Outcome B, and ‘explaining’ any such cause-effect relationship.
Exploratory study means an initial exploration of a theory, from which a hypothesis could be generated. In the present study, the analysis was retrospective: the initial study was completed, but the data was available to explore whether there was any relationship between the primary intervention [B-vitamins] and plasma omega-3 fatty acid levels. Finding a relationship in turn generates further hypotheses to be tested directly in further study. The key distinction in interpreting this retrospective study is that it is not demonstrated any cause-effect relationships, and is in effect an observation.