Application of CUSUM technique and beta-binomial model in monitoring adverse drug reactions

The cumulative sum (CUSUM) technique is applied to monitor spontaneous reports of adverse drug reactions (ADRs). The beta-binomial model, as suggested by Moussa (2), was used to fit the observed proportion of upper gastrointestinal bleeding, perforation, and ulcer (UGIBPU) for drug A, a nonsteroidal anti-inflammatory drug (NSAID). ADRs reported directly by physicians and manufacturers from the third quarter of 1982 through the fourth quarter of 1985 were analyzed. The model estimates the proportion of UGIBPUs for drug A, expressed as a fraction of all UGIBPUs for NSAIDs. To illustrate the procedure, estimated mean proportion of UGIBPUs for drug A and all NSAIDs was computed for directly reported physician and manufacturer ADRs. The beta-binomial model leads to estimates of 0.096 for directly reported physician ADRs compared to 0.039 for directly reported manufacturer ADRs. The difference is statistically significant (p < 0.025) by the method of maximum likelihood applied to the beta-binomial model. Moussa also suggested the use of this model for signaling important new ADRs. For these data on NSAIDs, the CUSUM method is less likely to signal an alarm than the approaches suggested by Mandel (5) and Patwary (4).