Which Study Design Is Best?
Decisions regarding which study design to use rest on a number of factors including::
- Uncommon Outcome: If the outcome of interest is uncommon or rare, a case-control study would usually be best.
- Uncommon Exposure: When studying an uncommon exposure, the investigators need to enroll an adequate number of subjects who have that exposure. In this situation a cohort study is best.
- Ethics of Assigning Subjects to an Exposure: If you wanted to study the association between smoking and lung cancer, It wouldn't be ethical to conduct a clinical trial in which you randomly assigned half of the subjects to smoking.
- Resources: If you have limited time, money, and personnel to gather data, it is unlikely that you will be able to conduct a prospective cohort study. A case-control study or a retrospective cohort study would be better options. The best one to choose would be dictated by whether the outcome was rare or the exposure of interest was rare.
There are some situations in which more than one study design could be used.
Smoking and Lung Cancer: For example, when investigators first sought to establish whether there was a link between smoking and lung cancer, they did a study by finding hospital subjects who had lung cancer and a comparison group of hospital patients who had diseases other than cancer. They then compared the prior exposure histories with respect to smoking and many other factors. They found that past smoking was much more common in the lung cancer cases, and they concluded that there was an association. The advantages to this approach were that they were able to collect the data they wanted relatively quickly and inexpensively, because they started with people who already had the disease of interest.
The short video below provides a nice overview of epidemiological studies.
However, there were several limitations to the study they had done. The study design did not allow them to measure the incidence of lung cancer in smokers and non-smokers, so they couldn't measure the absolute risk of smoking. They also didn't know what other diseases smoking might be associated with, and, finally, they were concerned about some of the biases that can creep into this type of study.
As a result, these investigators then initiated another study. They invited all of the male physicians in the United Kingdom to fill out questionnaires regarding their health status and their smoking status. They then focused on the healthy physicians who were willing to participate, and the investigators mailed follow-up questionnaires to them every few years. They also had a way of finding out the cause of death for any subjects who became ill and died. The study continued for about 50 years. Along the way the investigators periodically compared the incidence of death among non-smoking physicians and physicians who smoked small, moderate or heavy amounts of tobacco.
These studies were useful, because they were able to demonstrate that smokers had an increased risk of over 20 different causes of death. They were also able to measure the incidence of death in different categories, so they knew the absolute risk for each cause of death. Of course, the downside to this approach was that it took a long time, and it was very costly. So, both a case-control study and a prospective cohort study provided useful information about the association between smoking and lung cancer and other diseases, but there were distinct advantages and limitations to each approach.
Hepatitis Outbreak in Marshfield, MA
In 2004 there was an outbreak of hepatitis A on the South Shore of Massachusetts. Over a period of a few weeks there were 20 cases of hepatitis A that were reported to the MDPH, and most of the infected persons were residents of Marshfield, MA. Marshfield's health department requested help in identifying the source from MDPH. The investigators quickly performed descriptive epidemiology. The epidemic curve indicated a point source epidemic, and most of the cases lived in the Marshfield area, although some lived as far away as Boston. They conducted hypothesis-generating interviews, and taken together, the descriptive epidemiology suggested that the source was one of five or six food establishments in the Marshfield area, but it wasn't clear which one. Consequently, the investigators wanted to conduct an analytic study to determine which restaurant was the source. Which study design should have been conducted? Think about the scenario, and then open the "Quiz Me" below and choose your answer.
Case-control studies are particularly efficient for rare diseases because they begin by identifying a sufficient number of diseased people (or people have some "outcome" of interest) to enable you to do an analysis that tests associations. Case-control studies can be done in just about any circumstance, but they are particularly useful when you are dealing with rare diseases or disease for which there is a very long latent period, i.e. a long time between the causative exposure and the eventual development of disease.