Comparison of three algorithms used to evaluate adverse drug reactions

The consistency of three algorithms in evaluating adverse drug reactions (ADRs) was studied. As part of a hospital's ADR protocol, doctor of pharmacy students were required to collect and summarize all ADR data. Algorithms by Kramer, Naranjo, and Jones were used to evaluate all ADRs between January and May 1984. Kramer's algorithm was used for every reported ADR; Naranjo's and Jones' algorithms were used to check consistency in scoring among ADRs already scored with the Kramer algorithm. The two numerical scales (Kramer and Naranjo) were compared using linear regression. The results of all three algorithms were translated into categories of suspicion (A = definite or probable; B = probable; C = possible; and D = unlikely, doubtful, or remote) and evaluated for consistency with a weighted kappa (k(w)) statistical test. A total of 28 ADRs were evaluated, and the correlation (r = 0.87) between the total numerical scores of the Kramer and Naranjo algorithms was significant. Comparison of the Kramer and Naranjo algorithms showed 67% agreement with a k(w) value of 0.43 (-1 = perfect disagreement and +1 = perfect agreement). Similarly, there was 67% agreement (k(w) = 0.48) between Kramer's algorithm and Jones' algorithm. Agreement between Naranjo's and Jones's algorithms was 64%, but the k(w) value was only 0.28. The simpler and less time-consuming Naranjo algorithm compared favorably with the Kramer algorithm in scoring ADRs; more data are needed to support the use of the Jones algorithm.