study of a disease that deals with how many people have it, where they are, how many new cases develop, and how to control the disease.
What is pharmacoepidemiology?
For the purposes of this chapter, the terms epidemiology and pharmacoepidemiology will be used interchangeably even though purists will object to this since the two are not entirely the same.
Of necessity, pharmacoepidemiology uses numbers and statistical methods to summarize large volumes of data into interpretable information. It is the study of populations as opposed to the study of individuals. Thus, it is used to answer questions about groups of people rather than about individual patients. It is also used to extrapolate and generalize from individuals to groups and populations. Thus, it would answer questions like “Are the women who live on Long Island, New York, at greater risk for breast cancer than those who live elsewhere?” or “Is the use of drug X associated with a higher incidence of atrial fibrillation in elderly men?” as opposed to questions like “Did Ms. Jones have breast cancer because she lives on Long Island?” or “Did drug X produce atrial fibrillation in 79-year-old Mr. Jones?”
In the world of drug safety, epidemiology is used to answer questions about adverse events (AEs), and, in particular, serious adverse events (SAEs), in populations after (usually) a signal has been generated based on one or more individual case reports. The purpose is to confirm and quantify the risk associated with the signal or to rule it out. Such studies can rarely answer questions about causality but rather give information on probabilities and statistical associations.
In this chapter, we give a very high-level view of the handful of concepts that continually appear in the drug safety and pharmacovigilance literature and for which a passing knowledge (at least) is useful. Pharmacoepidemiology is an area of much research and interest and it plays a greater role in drug safety as risk-based pharmacovigilance becomes a mainstay of drug safety. This has been made possible with the development and population of better and bigger databases and the “cloud”.
This is the type of experiment that most people are familiar with. It is an experimental study, not an observational study, because a protocol, used by the investigators, determines who receives what drug treatment. The protocol may differ from the standard practice of medicine. In an observational study, one merely observes and records what happens in the normal course of medical practice and treatment. An observational study may have a protocol, but it does not dictate which treatment a subject will receive, which is left up to the treating physician.
An RCT is prospective in time. It involves two or more groups of subjects who have a disease (or are at risk of a disease) and they receive different treatments. For example, one group may get drug A and the other group may get drug B or placebo. It may be single blinded (the subject does not know what the treatment is) or double blinded (neither the subject nor the investigator knows what the treatment is). The individuals participating may also be randomized between the treatment groups to minimize known and unknown biases (factors other than the drugs tested that may alter or explain the drug effects). These studies are often complex to implement and resource-intensive. They represent the gold standard of clinical research: the double-blind RCT. These studies are usually done during phases I, II, and III of drug development, and for many reasons (ethical, availability of subjects, etc.), may not be feasible after the drug is marketed. The results are usually clear and easily understandable with the calculation of a risk difference between groups. For example, the group receiving drug A had a 4.1% incidence of AEs and the placebo group a 2% incidence of AEs, a difference of 2.1%. These trials, especially before approval for marketing, are usually prepared primarily to examine drug efficacy. The subject numbers and design are done to maximize the likelihood of finding a meaningful clinical and statistical result with the primary efficacy endpoints. The RCTs are usually statistically “powered” to show this one way or another. Statistical power refers to the likelihood that the trial and statistical test will reject a false null hypothesis, or, to put it another way, power is the probability that one will observe a treatment effect in the trial population when such effect would really occur.
The safety data from the trials are rarely sufficient to draw conclusions because rare adverse drug reaction (ADR) will not be observed with only a few hundred to several thousand subjects studied. The studies are not powered to pick up safety information and one might falsely conclude that a drug is “safe” or, more precisely, that doses do not differ from the comparator drug in terms of safety. Thus, the safety information is just presented as tables or listings without statistical tests. This is known as “descriptive statistics”.
An emerging methodology in clinical trial design is being used more frequently. This methodology is referred to as “adaptive clinical trials”. They have been used for some years in phase I in sequential dose tolerance studies in which three or four subjects are treated with a fixed dose. If well tolerated, another three subjects are then treated with a higher dose, and so on until a toxicity occurs that precludes further dose escalations. Many different methodologies have been developed, all of which have in common the use of the early data to determine what changes to make (to adapt) in the next subjects being treated in that trial.
Bayesian techniques are used to determine how the trial will be adapted. The goals of these trials are to obtain efficacy information and minimize toxicity by eliminating cohorts or treatments that do not work or are toxic. When successful, efficacy information can be obtained more rapidly and with fewer subjects or cohorts. In terms of safety, however, fewer subjects will be exposed to different treatments or doses (as the unsuccessful ones are rapidly abandoned). If this means that efficacy is determined with fewer subjects treated, a clear upside, the number of subjects examined for AEs will drop and the infrequent AEs will be less likely to be found, a clear downside. This design is particularly applicable for the study of new therapies to treat life-threatening disease where there are few existing treatment options. How this methodology, which the health agencies support in general plays out in terms of safety remains to be seen.
A cohort study is a basic type of epidemiologic study used in pharmacoepidemiology. This is an important concept because nearly all epidemiology designs have an underlying cohort concept. This type of study is relatively easy to grasp intuitively. Two groups of subjects are