Normalizing Reporting Ratios in Adverse Event Data Mining
*Stuart Chirtel, FDA 


Adverse event scoring algorithms which are based on the relative reporting ratio rely on the assumption that signal scores involving the vast majority of compound by symptom combinations behave as though they are members of a uniform population, where the distribution of the reporting ratio (RR) is a function of the frequency of the events occurrence and expectation. When log(RR) is plotted versus the frequency of occurrence of the event, a funnel-shaped pattern is observed, where both the mean and variance of the distribution decrease with increasing frequency. A signal is viewed as a statistical outlier from this distribution. One must adjust for the changes in the mean and variance in order to obtain the “statistical distance” of the log(RR) in question from the “expected” score.