Statistical methods in pharmacoepidemiology. Principles in managing error

To successfully appraise the significance of epidemiological data on drug risk and safety requires a good understanding of the errors involved in the design and analysis of pharmacoepidemiological studies. A proper comprehension of the repercussions of these errors and of the strengths and limitations of the tools used to measure their magnitude are essential to sound decision making by the regulatory, industry or clinical consumers of these data. In this paper, we examine the role of statistics in managing the quantifiable errors present in pharmacoepidemiological data analysis and interpretation. Some epidemiological principles on the measurement of risk are first introduced. The influences of controllable systematic error and random error on our assessment of epidemiological data are then presented, along with the prevailing statistical principles and measures necessary to control these errors. To illustrate the various issues addressed, published data on the risks of NSAIDs, focusing particularly on upper gastrointestinal bleeding (UGIB), the risks of replacement estrogens for endometrial cancer and the safety of allopurinol for cataracts are used as examples throughout.