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)
1. Definition of Terms Used in this Document


·         Adverse Event: (AE)– an adverse or untoward medical event (complaint, symptom, sign, syndrome, disorder, disease) that occurs or worsens in temporal association with a study treatment (investigational drug or control [placebo or active drug]) or during any period of observation without treatmentin a randomized clinical trial (RCT). An AE might be etiologically related to atreatment or an incidental observation with an etiology other than treatment.

·         Adverse Drug Reaction: (ADR)an AE where there is “reasonable evidence” that the AE was etiologically related totreatment (investigational drug or control).  To the best of our knowledge, “reasonable evidence” has never been operationally defined or even quantified by any regulatory entity or drug safety organization, including:

o   U.S. Food and Drug Administration (FDA) or other national regulatory agencies;

o   International Conference on Harmonization of Technical Requirements for Registration of Pharmaceuticals for Human Use(ICH – a group of major worldwide drug regulatory agencies and pharmaceutical manufacturers’ associations);

o   Council for International Organizations of Medical Sciences (CIOMS – a nongovernmental organization set up by WHO and UNESCO that works with ICH to establish standards and methods of evaluating drug safety.

“Reasonable evidence” might be the medical equivalent of the legal standard of “preponderance of evidence” that is quantitatively well defined (>50%).  However, it might be some quantity ≤50%.ADRs are identified based on the totality of relevant available data.  The most robust data are provided by placebo-controlled RCTsand meta-analyses of multiple such RCTs.  But, prospective and retrospective epidemiological studies, post-marketing surveillance and multiple other sources of data contribute to sponsors’ and regulatory bodies’ decisions about what AEs are ADRs and should be identified as such in product labeling.  Even if “reasonable evidence” was quantitatively well-defined, the judgment of the magnitude of the totality of data and analyses relevant to whether an AE is or is not an ADR would remain a subjective opinion, at least for “uncommon” AEs (see definition below).  In some cases, an ADR can be attributed to a drug treatment (or the potential for a specific ADR is considered a strong possibility)in product labeling even if the AE has not been observed with that drug treatment (e.g., all dopamine antagonist antipsychoticsarepotentiallyassociated with the ADR ofneuroleptic malignant syndrome [NMS]).  The potential for this ADR will appear in product labeling, in the Warnings and Precautions Section of a US label for all drugs in this class. If NMS had not been observed at the time of approval, the Warnings and Precautions text related to NMS is likely to include that caveat.Pharmacological class effect (a supposition rather than an empirical finding) is the basis for believing that there is “reasonable evidence” that adopamine antagonist causes or contributes to the development of NMS.

·         Incidence categories of ADRs (and AEs observed in a clinical trial):

o   Very common: ≥ 1/10, 10%, 0.1000

o   Common (frequent): ≥1/100, 1%,0.0100 to<1/10, 10%, 0.1000

o   Uncommon (infrequent): ≥1/1,000, 0.1%, 0.0010 to<1/100,1%, 0.0100

o   Rare: ≥1/10,000, 0.01%, 0.0001 to<1/1,000, 0.1%, 0.0010

o   Very rare: <1/10,000, 0.01%, 0.0001

·         “Proof” of a drug effect (and proof of absence of a specific effect):The standard of proof for a binary categorical outcome (in our case of interestthe occurrence of an AE that might be an ADR) isbased on a difference in incidences or a ratio of incidences observed in well designed, prospective, RCTs (or meta-analysis of multiple RCTs).If the difference or ratio, analyzed with proper statistical methods,is significant (p≤0.05), the results are interpreted as ‘proof’ of an effect.For “proof”’ of efficacy the regulatory standard, at least that of FDA for potential drugs intended to treat non-life-threatening disorders, is generally two RCTs with inferential results of p≤0.05.  If statistical significance is overwhelming in a single trial (e.g., p<0.001 in the single trial/analysis and/or the trial could be randomly split into two trials/analyses multiple times,and analyses of the split samples would consistently result in p<0.05) one trial might be sufficient.



December 13, 2018