Causality Assessment of Adverse Events in Clinical Trials: I. How Good is the Investigator Drug Causality Assessment?

Abstract

On-therapy adverse events reported in clinical trials are customarily rated by investigators as not-, possibly-, or probably-related. Conventional causality factors evaluated are temporality, known association with the drug class, alternative etiology, and result of de-challenge and re-challenge. In terms of the proportion being rated as possibly- or probably-related, we evaluated the investigator causality assessment (ICA) in six double-blind, placebo-(3) or active-(3) controlled studies of six different drugs. Despite the fact that ICA was not systematic and not standardized, we found that the proportion of adverse events rated as possibly- or probably-related by ICA was significantly higher among patients who received active drugs than those who received placebos, and as dose increased, ICA tended to rate more adverse events as related. Furthermore, the number of adverse events rated as not-related by ICA was found to be equally distributed among treatment groups. Greater relatedness was associated with events that had no alternative etiology, were mentioned in Physicians' Desk Reference (PDR) or investigator brochure, led to discontinuation of treatment, were the more severe on-therapy events, and occurred early during dosing. Multivariate analyses indicated that major determining factors of ICA were alternative etiology, mentioned-in-PDR, and adverse discontinuation, all of which accounted for 55% of the total variation in ICA rating. The remaining 45% were unexplained by data available in case report forms. It was also clear that rather than wait for the result of de-challenge, ICA coincided with their decision to discontinue therapy due to an adverse event. In conclusion, investigators' assessments were, on average, consistent with conventional causality expectation and investigators did use all known causality factors, which were treatment- and dose-related. There were also indications that unwanted variability in the individual assessment were substantive, but could be reduced.