As a reminder, note that the single most important hallmark of a scientific theory is that its hypotheses are capable of being disproved. Experimental studies, when done, can be excellent tests of hypotheses that we could not otherwise examine were we to simply try to observe naturally occurring events. But it is equally important to note that the kind of questions we ask in research cannot all be answered using the same exact research design; the design we use when trying to understand the factors that lead to, say, low back pain is very different from the design we use when trying to see which of two interventions is more effective at decreasing the pain seen with that same low back pain. Thus, knowing a bit about design is very helpful when you read journal articles, because the design of the study is a key to looking at what the study is about.
Generally speaking, we can classify research designs into two broad categories: descriptive (or analytic) designs and experimental. This is a broad classification, because there are also quasi-experimental designs, and non-experimental designs which can be used in clinical research, but this larger classification breaks the study designs down nicely. Descriptive studies are done when for whatever reason we cannot have control over the independent variable. For example, we might look to see if vibration from driving a truck leads to a higher incidence of low back pain- but the interval to do so here might run into years between exposure and outcome, and we would not want to set this up as a clinical trial where we expose some people to vibration over the course of a year and compare them to controls who were not exposed. In experimental designs, the intervention is under the control of the researcher, and this helps limits threats to validity.
The following is by no means all-inclusive, but is an introduction to common research designs. Descriptive designs include cross-sectional studies, case-control studies and cohort studies. Experimental designs include randomized clinical trials and cross-over designs.
Cross-Sectional: This assesses health status and exposure levels of individuals within a population at one point in time. By this, I mean that we are collecting this from each person at one point in time, not that we gather all the data at a single point in time. The classic example of a cross-sectional study is a survey. What we are looking for is a potential association between some causal factor and a condition of interest. For example, we might ask to see if people who have more than 3 drinks per day have higher rates of liver disease than those who do not. We cannot say if one causes the other, only that they are associated with one another.
Case-Control: In a case-control study, we look backward in time (retrospectively) to examine exposure factors between a group of people with the condition of interest (the cases) and those without (the controls). So we are likely looking at patient records in doing so, and we are trying, as much as possible, to ensure our populations are matched on all factors save for the presence of the condition of interest. These kinds of studies cannot determine the risk of developing a disease, but can determine the odds of doing so. An odds ratio can be calculated providing this information- this was discussed in an earlier blog post. Here, the OR is simply the ratio of odds of the cases being exposed divided by the odds of the controls being exposed to some factor.
Cohort: In a cohort study we follow patients forward in time and compare outcomes after one group is exposed to some suspected factor of disease while the other group is not. Such studies can last for many years, and the benefit of this design is that it is capable of detecting whether an exposure precedes an outcome; that is, does drinking 3 glasses of alcohol per day lead to a higher rate of liver disease? Thus, in a cohort study, we can determine a risk level, in this case a relative risk of developing the condition of interest between the group exposed and the group not so exposed. This is a particularly strong epidemiological design to use.
Randomized Clinical Trial: This is the classic pre-test/post-test randomized experimental design used in clinical trial research. It begins with two groups that are as similar in all important demographics as possible, and then subjects one to an intervention the other does not get, and finally compares the outcomes between the two groups. This design permits the statistical comparison of the two groups. It can be set as a factorial design when there are several explanatory variables involved.
Cross-Over: This design provides a treatment to one group while the other receives either a placebo or alternative treatment, and then switches those assignments at some specified point in time. There is usually a wash-out period involved to allow for time so that the initial treatment does not influence the alternative treatment. This is a complicated design that carries a high risk for drop-outs due to the time necessary to conduct the study.
As always, the idea is to use the proper design to answer the question, the best tool for the job, so to say. This is but an overview of the topic, but may help you as you read journal articles.