Drug Information Journal - March 2009 - (Page 172) 172 BIOSTATISTICS Ross berg and colleagues (6), where a series of observational studies of the risk of ovarian cancer associated with oral contraceptives were meta-analyzed. The individual patient data from studies were available, and odds ratios for cancer risk, using both hospital controls and community controls, were computed. For the main analysis question—that is, what is the risk of ovarian cancer?—the results were similar in direction, magnitude, and precision regardless of whether IPD or aggregate data were used. Potentially important practical and analytical distinctions do exist, however. It has been estimated that an IPD meta-analysis costs at least five times as much as an aggregate data meta-analysis. In addition, it is often impossible to obtain full patient-level data sets on all eligible studies, so analysts run the risk of running meta-analyses on an incomplete study set if they insist on IPD from every study. So why try to do IPD meta-analyses at all, given the high cost in terms of time and money? The main reason why IPD meta-analyses are sometimes worth it is that analysts can assess the influence of patient-level covariates on all collected outcomes and measured time points of interest (not all of which are always reported in the aggregated reports in the literature). In addition, time-toevent analyses can be performed (7). IPD metaanalyses permit these kinds of assessments; traditional, aggregate data meta-analyses do not. It behooves industry sponsors of drug and device studies to perform IPD meta-analyses, ideally with a prospective plan to pool study results. Increasingly we are seeing examples of these prospective meta-analyses in the literature, but to date, they stem primarily from academic research institutions, not industry. Yet it is the industry sponsor who controls the IPD database for every premarketing study, so feasibility is less an obstacle than corporate imagination and will. The inescapable investment of time and resources is more than offset by the potential advantages of prospectively planned IPD metaanalyses. The return on investment could be huge, in terms of expediting a portfolio of clinical trials intended for regulatory purposes, and in terms of preparing the most intelligent and most useful integrated summary of efficacy and integrated summary of safety for new drug applications (NDAs) (8). This idea brings us to another advance in the field, cumulative metaanalysis. C U M U L A T I V E M E T A - A N A LY S I S Q. Can we monitor real-time results of all relevant studies as soon as they are available? (And can we make course corrections if needed?) A. Yes, you can, and yes, you should be doing so. Fifteen years ago Tom Chalmers and colleagues published their landmark article on cumulative meta-analysis (CMA), showing how the opinions of experts lagged behind what was knowable from the literature (9). CMA is a metaanalysis that is updated after each new study is completed. As soon as the meta-analyzed treatment effect reaches statistical significance, it is apparent. Analyzing study results this way, in real time, permits identification of winners and losers as early as possible. CMA also permits assessment of how changes in the summary measure are impacted by differences in the treatment, procedures, or subjects over time (10). Industry sponsors, among others, should be clamoring for this information as soon as it is available. If they are not already doing so, they should start performing CMA on every new drug in their portfolios, for both efficacy and safety outcomes. An example of how this technique might have been used to identify a highly publicized and costly safety problem was provided by Juni et al. (1 They showed that the relative risk 1). of myocardial infarction with rofecoxib could have been known much earlier than it was, if rofecoxib studies had been subjected to CMA. So how would sponsors do this? For every new drug entering phase 2 trials in a particular indication, and continuing after marketing, sponsors would prepare a prospective plan to meta-analyze efficacy and safety outcomes of all their studies, as soon as each study was completed. This plan would include a statement of specific hypotheses or objectives, the general eligibility criteria of trials and of patients within trials, the
For optimal viewing of this digital publication, please enable JavaScript and then refresh the page. If you would like to try to load the digital publication without using Flash Player detection, please click here.