(although cohort studies may also be prospective and conducted in the field). The first group is those who took the drug in question for a certain amount of time, and the second group (the comparator “cohort”) is the group that did not take the drug. The investigator attempts to choose (“match”) a cohort that is as comparable to the drug group as possible, demographically and in other relevant aspects.
The subjects’ medical records are reviewed and the incidence rate of the AE in question is calculated for each group. These studies are prospective, as the investigator picks a point in time and studies the two groups as they move forward in time to see whether they develop the AE. This study may be done on historical data that have already been collected and stored in a database or it may be done on data that are being collected de novo and moving forward. The advantage of the cohort over the case-control study (see Table 1) is the ability to calculate the true risk ratios, excess risk, the absolute risk difference, but also incidence rates and incidence rate ratios. The excess risk is useful in determining the number needed to harm, that is, the number of patients who need be exposed to a drug to produce a specified AE.
After the data are filled in, the absolute and relative risks are calculated (Table 2).
This tells us that the drug group is more than two times at risk for this AE compared with the comparator group. If the relative risk were 1, the likelihood of getting the AE is the same in the drug group and in the cohort group. If the relative risk is less than 1 (but greater than zero — the value in this calculation is always greater than zero), then the risk of getting the AE in the drug group is less than that in the cohort group.
Table 1. Organization of Data Cells and Formula to Calculate Relative Risk and Absolute Risk.
Note: Relative risk = [a/(a + c)] ÷ [b/(b + d)] or the rate in exposed divided by the rate in the unexposed. Absolute risk = [a/(a + c)] − [b/(b + d)] or rate in exposed minus rate in unexposed.
In the example, the absolute risk indicates that there were 343 additional cases of AE per 1,000 patients exposed to the drug. Therefore, the number needed to harm is about 3; that is, every third patient on the average will develop the AE (something the clinician should keep in mind).
For an excellent discussion of risk ratios and odds ratios, their differences, when they are close to being the same (i.e., when the AEs are very rare), and other complex issues.1 The entire Bandolier site is worth looking at for, as they put it, “evidence-based thinking about healthcare”.
This concept is sometimes hard to grasp intuitively. The study may be prospective or retrospective. In a retrospective study, some of the events already happened by the time the cases are classified as cases and controls.
This type of study determines the chance (“odds” or “probability”) of having taken the suspect drug in a group of subjects already suffering from an AE and compares that with the odds of having taken the drug in a group of subjects who did not have the AE. To put it more simply, take a group of subjects who had the AE (cases) and another group who did not have the AE (non-cases or controls) and see how many in each group had taken the drug in question. A large database, e.g., a claims database, a hospital database, etc., is ordinarily used.
Table 2. Example of Mock Data used to Calculate Relative Risk and Absolute Risk.
Note: Relative risk = [90/(90 + 50)]/30/(30 + 70)] = 0.643/0.300 = 2.14 times more cases of AE, where as absolute risk = 0.643 − 0.3 = 0.343.
Usually, the investigator attempts to select groups of cases and non-cases that are comparable, and can, for example, match the two groups as closely as possible based on demographic characteristics, e.g., age, sex, race, indication treated, concomitant diseases, and other medications, so that the groups are identical in all other aspects other than the exposure to the drug of interest. Both groups should have the same medical profile and opportunity to receive the drug. For example, if a hospital database is used, the drug should be on the pharmacy formulary for the entire time period examined in the study. This is a retrospective study because the AE has occurred and the investigator is looking back in time into the medical history of the subject, to before the AE occurred.
The subjects’ medical histories are then examined to see which subjects in each group used the drug in question. The simplified data are then organized into a 2 × 2 table as depicted in Table 3.
If the following mock data were used to populate the above table, the data would appear as in Table 4.
Table 3. Organization of Case Control Data and Formula for Calculation of an Odds Ratio.
Table 4. Example Calculation of an Odds Ratio from Mock Case Control Data.
The odds ratio is calculated by dividing the odds of having the AE in the group that took the drug (90 of 140 patients or 90:50 = 1.8) by the odds in the group that had the AE but did not take the drug (30 of 100 patients or 30:70 = 0.43). The division is 1.8/4.3 = 4.2. Using the above formula:
A value greater than 1.0 is suggestive of an association between the drug and the AE. As a rule of thumb, anything greater than 2.0 is believed to be fairly strong and quite suggestive of an association. In our example, this is a high value (4.2) and suggests that there is an association between taking the drug and having the AE.
Advantages of case-control studies are that they are useful for studying very rare AEs because the investigator seeks out a database where this AE is found in a large enough number of subjects. Obviously, the subjects also had to have the opportunity to take the drug if the identified database is to be used. For rare AEs, the odds ratio can be considered approximate to the relative risk or risk ratio of a conceptual cohort study. It is also possible to use logistic regression for adjustment of covariates and risk factors that may confound the association, e.g., severity of the disease treated, or concomitant medications and possible interactions between drugs.
These studies are fast and relatively inexpensive. However, they are potentially liable to significant bias both in the selection of the subjects for the two groups and in the amount and quality of the drug exposure and medical history data.
This type of case-control study is inside, or nested in, a cohort design. The nested case-control design uses estimates from a sample of the cohort rather than the whole cohort. It permits the collection of less data than in a full cohort study with an acceptable statistical analysis and may save time and resource.