Making Sense of Biostatistics: Meta-Analysis and the Odds Ratio

Last month’s column discussed therapeutic confirmatory trials. These trials play an important role in determining whether a regulatory agency will grant marketing approval to the investigational drug. However, such trials do have limitations. Consider the case of a rare adverse drug event. The typical sample size of preapproval clinical trials means that there is a low probability of observing a rare adverse event. The “rule of threes ” (e.g., see Strom) tells us that to be 95 % confident that a single case of an identified adverse event of interest would be seen, the sample size must be roughly three times the reciprocal of the frequency of the event in the general population. That is, for an event that occurs in 1/1,000 individuals, a sample size of around 3,000 subjects would be needed. Based on this rule, the sample size that would be needed to be 95 % confident that at least one adverse event would be observed when the frequency of the event is 1/10,000 (i.e., on the order of 30,000) would be much larger than is feasible in a single trial. Therefore, we need much larger databases and strategies to examine them. One such strategy is meta-analysis. Meta-analysis “combines ” data from multiple studies to conduct a new analysis that was not possible with the smaller individual data sets. The larger data set makes it much easier t