Data relating to DRAEs were extracted from the FDA AERS database up to 31 December 2006. The FDA AERS database, which replaced the earlier FDA Spontaneous Reporting System, contains more than 2.5 million spontaneously reported adverse events (AEs) submitted by patients, healthcare professionals and pharmaceutical manufacturers. In this extensive database, longitudinal data are available whereby one report of an event can be linked to follow-up reports. Submitted AE reports (using a MedWatch form) undergo clinical review by trained safety evaluators in the Center for Drug Evaluation and Research and the Center for Biologics Evaluation and Research who assess each event according to defined principles. Events are subsequently entered into the AERS database using the Medical Dictionary for Regulatory Activities (MedDRA) coding system, which encompasses approximately 10,000 terms to define AEs within the database. MedDRA search terms for 24 DRAEs related to metabolic effects of atypical agents, as used in the current analysis, are listed in Table 1 .
Seven widely used antipsychotics in the USA were analyzed for drug-event associations. These were the atypical antipsychotics, clozapine, risperidone, olanzapine, quetiapine, ziprasidone and aripiprazole. Haloperidol was also included in the analysis as a representative of the older, typical antipsychotic group. A separate analysis of the entire AERS database was also carried out to assess drug-event associations between diabetes mellitus and therapeutic agents using the Anatomical Therapeutic Chemical (ATC) Classification System. Drug-event associations by pharmacological subgroup (ATC Level 4, e.g., indole derivatives) and generic drug name (ATC Level 5, e.g., ziprasidone) were analyzed.
Systematic disproportionality analysis (also referred to as data-mining analysis) is commonly used to extract information from large drug-safety databases. A commonly used measure of disproportionality of a targeted drug-DRAE combination is the Reporting Ratios (RRs) defined by RR = Observed Rates/Expected Rates, where Observed Rate = Number of reports of targeted DRAE with targeted drug/Number of all reports for targeted drug, and Expected Rate = Number of reports of targeted DRAE in AERS/Number of all reports in AERS. The further the RR is from 1, the stronger the indication of a degree of association.
It was observed that when the reports for a particular drug-event combination are few, the estimate of RR becomes unstable. However, more robust methods such as the Multi-item Gamma Poisson Shrinker (MGPS) data-mining algorithm have been developed to ascertain a stable estimate of RR for a particular drug-event combination in large safety databases. The MGPS method adjusts or shrinks the estimate of RR towards 1, which is referred to as 'adjusted' RR. Although this adjustment biases the estimate of RR, it provides a more precise estimate by reducing the volatility of the estimates when there are a low number of reported drug events. Generally, this approach is referred to bias-variance trade-off (see Hastie et al). Therefore, using the adjusted RR helps to focus on the drug-event combinations that have a stable adjusting ratio >1. The resulting adjusted RRs are denoted by the Empiric Bayes Geometric Mean (EBGM) and corresponding 90% confidence intervals (CI EB05-EB95). The EB05 is interpreted as a value such that there is about a 5% probability that the true value of RR (i.e., Observed/Expected) lies below it. Similarly, EB95 is a value such that there is about a 5% probability that the true value of RR (i.e., Observed/Expected) lies above it.
Typically, the identification of a potential signal is based on the EB05 values and corresponding pre-specified threshold value of 2 for identifying a potential signal, as described in previous studies.[34,35]
In this study, stratified MGPS analysis of association between antipsychotic drugs' generic name and MedDRA search terms for 24 DRAEs was carried out to control for background differences in relative reporting rate by age, gender and FDA year. Additionally, in order to understand when drug-event combination first appeared as a potential signal, we performed the analysis of cumulative subsets of the data by year of report submission from January 1968 to December 2006.
MGPS analysis was performed utilizing WebVDME™, version 6.0 (Phase Forward, Lincoln Technologies).
In addition to MGPS analysis, logistic regression (LR) analysis was used to explore the association between antipsychotic drugs and the most commonly reported AE, diabetes mellitus. Logistic regression controls for other drugs showing association with diabetes mellitus. Multiple logistic regression is the standard statistical method for modeling the probability of occurrence of an event as a function of many covariates. Covariates considered in the LR model were: age group, gender, year of report, presence/absence of antipsychotic drug, and presence/absence of 100 other drugs showing association with diabetes mellitus and at least 10 reports in the database. The results are reported as logistic regression odd-ratios (LROR) and the corresponding 90% confidence interval (CI LR05-LR95). The identification of a potential signal is similar to EBGM with a threshold for LR05 of 2, which is chosen to indicate a potential signal.
Psychopharmacol Bull. 2009;42(1):1-21. © 2009 MedWorks Media Global
Cite this: Atypical Antipsychotic Drugs and Diabetes Mellitus in the US Food and Drug Administration Adverse Event Database: A Systematic Bayesian Signal Detection Analysis - Medscape - Jan 01, 2009.