Methods of Signal Detection and Refinement

METHODS OF SIGNAL DETECTION AND REFINEMENT

DATA MINING

‘Data mining’ is a technique for extracting mean-ingful, organized information from large complex databases and has been used to identify hidden patterns of associations or unexpected occurrences (‘signals’) in spontaneous reporting system databases. One goal of this developing technology is auto-mated signal generation within spontaneous AERS databases (Bousquet et al., 2005). If this can be reli-ably achieved, data mining might serve as a poten-tially useful adjunct to traditional pharmacovigilance practices (Almenoff et al., 2005). Since 1998, FDA has explored new automated and rapid Bayesian data mining techniques to enhance its ability to monitor the safety of drugs, biologics and vaccines (Szarfman, Machado and O’Neill, 2002; Szarfman, Tonning and Doraiswamy, 2004).

Importantly, data mining cannot prove or refute causal associations between drugs and events. Data mining simply identifies disproportionality of drug-event reporting patterns in databases. The absence of a signal does not rule out a safety problem. Simi-larly, the presence of a signal is not a proof of a causal relationship between a drug and an adverse event. Hands-on review of the cases is critical to the evaluation of potential signals identified through data mining. Data mining does not quantify the magni-tude of risk, and caution should be exercised when comparing drugs. The potential limitations of data mining include those inherent to spontaneous report-ing databases such as under-reporting, influences by media/publicity and litigation. Results obtained from data mining technique should be interpreted with caution and with the knowledge of the weaknesses of the spontaneous reporting system (Anon. Guidance for Industry. E2E Pharmacovigilance Planning. ICH. April 2005; http://www.fda.gov/OHRMS/DOCKETS/ 98fr/04d-0117-gdl0002.pdf).