Predicting Anaphylactoid Reactions Based on Costart Terms

Abstract

An epidemiological study was designed to compare three nonsteroidal antiinflammatory drugs in terms of whether or not anaphylactoid reactions were coded using the same COSTAR T terms. These three drugs were chosen because they are (a) in the same drug class, (b) are essentially prescribed for the same indications, (c) have been widely prescribed, and (d) have a similar marketing history. Anaphylactoid reaction reports submitted to the Food and Drug Administration (FDA) over the most recent 6 month period prior to study initiation were selecterd for each of the three study drugs. These records are to be carefully reviewed by an FDA medical team. A decision will be made as to whether each patient did or did not have an anaphylactoid reaction. Based on this decision and each patient's COSTART data there are two immediate analytical goals. These are to establish a method to determine the “best” algorithm for predicting anaphylactoid reactions based on COSTART terms alone and to examine whether or not this “best” algorithm is drug specific. Only the first goal, ie, proposing an approach to selecting a best algorithm, will be considered here. This algorithm is based on using stepwise logistic regression. The goodness of the result will be checked by making use of a sensitivity/specificity plot, and by making comparisons to other algorithms used by the FDA in searching for cases of anaphylaxis.