Charles M. Beasley, Jr and Roy Tamura: What We Know and Do Not Know by Conventional Statistical Standards About Whether a Drug Does or Does Not Cause a Specific Side Effect (Adverse Drug Reaction)

8. Practical alternatives to “proof” of presence or absence of an ADR : the need for best assessment possible as  quickly as possible of the AE / ADR  profile of a marketed drug.


          Statisticians and data scientists, both academic and regulatory, have developed and are continuing to refine methods for working with data from sources other than RCTs.  These sources include retrospective and prospective epidemiological studies (especially retrospective studies employing “big data” from evolving large databases possible with electronic medical records), large simple studies including those without a control group, and spontaneous adverse event reporting databases maintained by regulatory agencies where precise knowledge of total persons treated is not available but can be estimated, among other data sources.  It can be hoped that these methods result in the reduction in failure to find true ADRs and reduce false attribution of an ADR to a drug.  These methods are the ones that generally result in the discovery of very “infrequent,” “rare” and “very rare” ADRs associated with a given treatment.  However, these methods are more subject to error than those methods used to evaluate efficacy and lack of efficacy.  All interested parties should keep in mind the nature of the analyses that lead to the attribution of all but “common” ADRs to a given drug and the potential uncertainty of such attribution.  Also, all interested parties should clearly understand the virtual impossibility of “proving” by a conventional gold standard what is or is not an ADR associated with a drug.

            It cannot be emphasized enough that for AEs that might or might not be ADRs but of low incidence, it can be impossible to “prove” that a drug is or is not associated with the potential ADR based on the RCTs that are conducted to prove that the drug is efficacious.  Probably the best that we can do in the future is to develop stronger prospective epidemiological studies that are initiated soon after a drug is launched.  By stronger, we mean studies with exceptionally large numbers of subjects, extended exposure time frames and rigorous prospective methods for identifying with clinical certainty AEs of interest.  An important and interesting question is: What entity would fund such studies?  They would be expensive.  Advances in data sciences might make such studies more practical and reduce their costs.  Such studies are our best chance of ruling in or ruling out a rare but important potential ADR in a faster time frame with a lower probability of false positive and false negative attribution.


March 21, 2019