The role of biostatistics in the prevention, detection and treatment of fraud in clinical trials

Abstract

Recent cases of fraud in clinical trials have attracted considerable media attention, but relatively little reaction from the biostatistical community. In this paper we argue that biostatisticians should be involved in preventing fraud las well as unintentional errors), detecting it, and quantifying its impact on the outcome of clinical trials. We use the term 'fraud' specifically to refer to data fabrication (making up data values) and falsification (changing data values), Reported cases of such fraud involve cheating on inclusion criteria so that ineligible patients can enter the trial, and fabricating data so that no requested data are missing. Such types of fraud are partially preventable through a simplification of the eligibility criteria and through a reduction in the amount of data requested. These two measures are feasible and desirable in a surprisingly large number of clinical trials, and neither of them in any way jeopardizes the validity of the trial results. With regards to detection of fraud, a brute force approach has traditionally been used, whereby the participating centres undergo extensive monitoring involving up to 100 per cent verification of their case records. The cost-effectiveness of this approach seems highly debatable, since one could implement quality control through random sampling schemes, as is done in fields other than clinical medicine. Moreover, there are statistical techniques available (but insufficiently used) to detect 'strange' patterns in the data including, but no limited to, techniques for studying outliers, inliers, overdispersion, underdispersion and correlations or lack thereof. These techniques all rest upon the premise that it is quite difficult to invent plausible data, particularly highly dimensional multivariate data. The multicentric nature of clinical trials also offers an opportunity to check the plausibility of the data submitted by one centre by comparing them with the data from all other centres. Finally, with fraud detected, it is essential to quantify its likely impact upon the outcome of the clinical trial. Many instances of fraud in clinical trials, although morally reprehensible, have a negligible impact on the trial's scientific conclusions. Copyright (C) 1999 John Wiley & Sons, Ltd.