A Matter of Prospective
When you begin a study in Radiology, you will often characterize your study as either ‘prospective’ or ‘retrospective.’ This distinction should seem obvious to most researchers: it simply denotes whether the study design, population, and exposures were determined before the observation of outcomes. It couldn’t be simpler, right?
In many cases, wrong. While RCTs and many longitudinal cohort studies are clearly prospective, there are many types of study that are somewhat of a grey area, and the field of Radiology has some interesting examples of these issues and controversies. These issues may be important to your study, especially since you do not want to fall into the trap of adding the term ‘prospective’ in dubious circumstances simply to make the study seem stronger. Here are our quandaries:
Issue #1: Prospective but Not Longitudinal
A common association of terms is that longitudinal/cohort studies are prospective, while case-control studies are usually retrospective. However, the opposite is possible: if a researcher looks at an existing dataset, finds a cohort that was examined longitudinally, and creates a study design examining this population, that would be a retrospective cohort study. A case-control experiment could also be done with prespecified study design and exposures, though this would introduce risks of confounders that are not usually present in prospective studies.
Also, what about cross-sectional studies? The idea of a prospective cross-sectional study is slightly problematic, because exposures and outcomes are recorded simultaneously. However, the determination between ‘prospective’ and ‘retrospective’ is also largely based upon pre-specification of study design, hypotheses, and populations, so many pre-specified, cross-sectional studies done in Radiology are prospective in every way except the necessity of exposures preceding outcomes. This has led some medical researchers to declare that ‘prospectivity’ is irrelevant to cross-sectional studies. Despite this, these pre-specified studies still have the advantage of being able to collect data about possible confounders and avoid confirmation bias in their study design, so they deserve a separate designation from those in which study design postdates data examination. It may not be ‘prospective,’ if the ‘exposure precedes outcome’ definition is accepted, but it is completely understandable that the question ‘Can a cross-sectional study be prospective?’ keeps coming up.
Issue #2: ‘A Retrospective Study of Prospectively Collected Data’
This is a phrase seen in many Radiology papers, and it probably seems bizarre to you. In truth, it should, because it essentially means that the study design was created after the collection of a dataset for a different study or even simply from a database. If the study in question is a secondary analysis that was not pre-specified and that was designed with the knowledge of outcomes, then the description roughly fits, but the study design and population selection should be examined very carefully, because it can still be plagued by sampling bias and confounding factors in the manner of any other retrospective analysis. If it is longitudinally collected data (as many prospective datasets are), it would also have the risk of bias from attrition or changes in methodology. Essentially, such studies are retrospective but with the caveat that they can fall prey to some problems that usually affect only prospective studies. Therefore, though the strange phrase ‘retrospective study of prospectively collected data’ is nonsensical, its underlying idea still introduces an even greater need for discussion of limitations, confounders, and biases, rather than eliminating it.
Issue #3: Predicting the Past
A similar type of study leads to an even more fraught question about ‘prospectiveness.’ If you were to create a study design, including sample sizes, exposures or interventions, and hypotheses for an experiment, and then apply it to data that had been collected in the past–but that you had never seen before? This question is roughly analogous to the maddening logical dilemma of Schroedinger’s cat, for it seems that the data collection concerning outcomes both precedes and follows the study design and exposures. Unlike the other issues, this one is almost up to individual preference, since both arguments have substantial, but contradictory, validity. Describing such a study as ‘prospective’ is correct in the technical elements of the experiment, but contradicts the more general meaning of ‘prospective’ (‘looking forward’), may have issues with the lack of knowledge of data collection methodology by the researcher, and introduces an element of moral hazard (because it trusts researcher ignorance). So, what do you think? If I flip a coin, and immediately cover it with my hand, can I then ‘predict’ whether it is heads or tails? There is no completely satisfactory answer, and we will just have to continue wrangling with the idea of ‘predicting the past.’
How Do I Address These Issues?
When push comes to shove, most of these issues can be reduced to arguments over the definition of ‘prospective.’ Especially, the conflict comes between the requirements of pre-specification and exposure/outcomes, which can exist independently of each other. Therefore, the important message should be to examine study design rather than being dependent on a single word to characterize a study. In cross-sectional studies, recognize the advantages of pre-specification, and in ‘retrospective analyses of prospectively collected data’ (if you accept this label as non-ludicrous), scrutinize the limitations due to bias and confounders. Also, keep considering the grey areas and internal conflicts inherent in these definitions, to keep from being tricked. The misuse of ‘prospectivity’ will certainly continue in medical literature, but you can avoid being fooled by others and ensure your own correct usage by focusing on the reasons pre-specification of design and exposures preceding outcomes are important, rather than on semantics. Good luck!