Introduction
Observational studies in clinical research can be classified as either analytic or descriptive (Table 12–1). Analytic observational studies are similar to randomized, controlled clinical trials in that the goal is to estimate the causal effect of an exposure on an outcome. Also similar to trials, analytic observational studies always include some type of comparison group, against which the experience of the exposed group is compared. Well-designed analytic studies can generate strong evidence for or against a stated hypothesis. Descriptive studies, on the other hand, aim to describe the characteristics or experiences of a particular patient group. Even well-designed descriptive studies cannot be used to draw strong conclusions about the effect of an exposure on an outcome. Instead, these studies are often used to generate study questions that can then be tested by more rigorous methods.
Analytic designs | Descriptive designs | ||||
---|---|---|---|---|---|
Case-control study | Cohort study | Prevalence study | Ecologic study | Case report/case series | |
Study purpose | Hypothesis testing | Hypothesis testing | Hypothesis generating | Hypothesis generating | Hypothesis generating |
Primary study characteristics | Selection of study population by outcome status Need to ascertain prior exposures | Selection of study population by exposure status Need to ascertain subsequent outcomes | Single point-in-time survey of exposures and outcomes | Exposure and outcome data are aggregate measures | Detailed information regarding one or multiple interesting medical cases |
Typical study population size | Small | Large | Can be small or large | (Not person level) | Very small |
Temporal relationship between exposure and outcome | Difficult to establish | Easier to establish | Unknown | Difficult to establish | Usually established |
Best use of study design | When outcome is rare | When exposure is rare | To establish prevalence of a disease in the sampled population | When it is infeasible to measure exposure at the individual level (e.g., environmental exposures) | To report initial results or potential safety signals for a new therapy or procedure |
Primary challenge | Selection of appropriate controls Avoiding differential recall bias when assessing exposure | For prospective designs, following study participants over time and waiting for events to occur | Not usually possible to discern timing of exposures and outcomes | No assurance that exposed persons are the same as those who experienced the outcome | No control group for comparison Very limited population size |
Although many observational study designs are available to researchers (1), a few are most widely used and will be described below. The analytic study designs presented are the case-control study and the cohort study. The descriptive study designs presented are the ecologic study, the cross-sectional prevalence survey, and case reports or case series.
Analytic Study Designs
In a case-control study, the study population is selected based on a person’s outcome status (2). For most case-control studies, the outcome is a disease. Cases are those that have or have had the outcome. Controls are those that lack the outcome. Cases are usually identified when they initially seek treatment for the condition of interest. Health system databases and disease-specific registries can be helpful for identifying cases. Controls are identified in any number of ways, including selecting them at random from the population or from within the same healthcare delivery system or geographic area as the cases. The principle that guides the selection of controls is that they must be representative of the underlying population that generated the cases. At a minimum, this means the controls must have been eligible to have had the outcome of interest. Once the cases and controls have been selected, the prior exposure status of each study member is ascertained; then, statistical analysis is used to determine whether or not the exposure rates differ between the case group and the control group.
There are multiple advantages of this design. Most case-control studies are modestly sized and can be completed relatively quickly. Because there is no need to wait for events to occur, the major time costs incurred are those related to identifying controls and interviewing selected cases and controls about their exposure status. The case-control study design is also well suited to the study of rare outcomes. Rare types of cancer, for example, may be difficult to observe in any given cohort, but a registry of patients who developed that cancer might be a suitable starting point for a case-control study.