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)

Charles M. Beasley’s reply to Donald F. Klein’s comment

 

          We thank Don Klein for his comments concerning “What We Know and What We Do Not Know by Conventional Statistical Standards About Whether a Drug Does or Does Not Cause a Specific Side Effect.” Dr. Klein has underscored what we believe to be an important proposition in our work, that sample sizes can be minimized by selecting the appropriate experimental design combined with the optimal inferential statistical method best suited to the experimental design and the anticipated observations in the experiment. Expert statistical consultation can be extremely beneficial. However, for highly infrequent or rare events robust proof of effect or lack of effect remains an impractical goal.

          Dr. Klein has suggested the utility of large databases in the early and accurate identification of adverse drug reactions (ADRs) and specifically mentioned the Scandinavian countries as repositories of such databases. We concur with this opinion as briefly mentioned in Section 7 of our work. The utility of such databases is highly dependent on the accuracy and completeness of the information they contain. Unfortunately, even hospital based medical records (e.g., discharge summaries) may contain inaccuracies on occasion. There is good reason that when the primary objective of a clinical trial or development program involves the identification of an adverse event (AE) that might be an ADR, it is customary for the sponsor of trial or development program to rely on an Event Assessment Committee. The Committee would review, blinded to treatment, all clinical information made available regarding an AE reported clinically as the AE of interest and make the final determination as to whether the reported AE is the AE of interest. For example, such a committee would generally be used in studies of Major Adverse Cardiovascular Events (MACE) in the development of new anti-diabetic medications.

          In the US, the Food and Drug Administration (FDA) has clearly recognized the importance of such a large data base as a major advance over simply receiving reports of possible ADRs. FDA has developed a program to create such a database and more information on this effort can be reviewed at www.fda.gov/safety/fdas-sentinel-initiative.

          Dr Klein has wondered about the potential utility of the application of machine learning technology, and by implication other artificial intelligence processes, to such a database as resulting in a further advancement in speed and accuracy of identification of ADRs. He is not optimistic about the utility of such computational methods. We are cautiously optimistic. However, to be maximally useful artificial intelligence technology will likely need to progress to the point that the programs can recognize novel patterns and association against a large background of noise. The programs will need to achieve something close to human creativity and imagination.

 

November 14, 2019