Algorithms Used in Adverse Drug Event Reports: A Comparative Study

Abstract

Objective

To determine if and how the Kramer and Karch algorithms differ in assigning a probability that a published case was actually an adverse drug event (ADE), and to determine if these algorithms could be used to assess published ADEs.

Design

Open, single-rater comparison of Karch and Kramer algorithms in 200 published ADE reports.

Main Results

The algorithms were not significantly different regarding the proportion of cases deemed definite (p = 0.5204) or probable (p = 0.2972) ADEs. The Kramer instrument was more likely to assign a possible risk of ADE (p = 0.0001), while the Karch instrument was more likely to assign a risk of unlikely (p = 0.0001). The algorithms agreed in 41% of the cases and could be used to assess published ADEs.

Conclusions

The Karch and Kramer algorithms may disagree in how they assign a probability of risk to a potential ADE. This may be due to how algorithms are applied, as well as to structural differences.