A study of adverse reaction algorithms in a drug surveillance program

To improve agreement among observers, several investigators have recently proposed methods (algorithms) to standardize assessments of causality for presumed adverse drug reactions. We evaluated one such method in the context of an intensive pediatric drug surveillance program. Four observers rated 50 randomly selected case reports drawn from the program, first using only general guidelines and then, several months later, using the strict criteria of the algorithm. Agreement among observers was poor in both study phases. The presence of selected characteristics of adverse events (e.g., major severity) did not improve agreement in either phase of the study. We conclude that routine use of such algorithms in drug surveillance programs is not likely to be of benefit.