The role of data mining in pharmacovigilance

Abstract

A principle concern of pharmacovigilance is the timely detection of adverse drug reactions that are novel by virtue of their clinical nature, severity and/or frequency. The cornerstone of this process is the scientific acumen of the pharmacovigilance domain expert. There is understandably an interest in developing database screening tools to assist human reviewers in identifying associations worthy of further investigation (i.e., signals) embedded within a database consisting largely of background 'noise' containing reports of no substantial public health significance. Data mining algorithms are, therefore, being developed, tested and/or used by health authorities, pharmaceutical companies and academic researchers. After a focused review of postapproval drug safety signal detection, the authors explain how the currently used algorithms work and address key questions related to their validation, comparative performance, deployment in naturalistic pharmacovigilance settings, limitations and potential for misuse. Suggestions for further research and development are offered.

Quantitative signal detection for vaccines: Effects of stratification, background and masking on GlaxoSmithKline’s spontaneous reports database

“…5,14 Although we did not assess whether all scores that moved from above to below the thresholds (Quadrant D) represented false signals, their relatively low magnitude seems to suggest this. Other researchers have made similar observations using Gamma Poisson Shrinker on other adverse events database.…”

“…In a review, Hauben et al 5 argue that small databases can be especially prone to the effect of masking because of the relative lack of diversity of events or drugs; however none of the quoted studies had used MGPS. Our results show that this effect is not pronounced for the single vaccine that represents 40% of the reports in our database and any effect is limited to events that are disproportionately reported with that vaccine.…”

“…This entails using the total number of reports for the drug/vaccine of interest as a denominator to calculate the proportion of the adverse event of interest; this proportion is compared with the corresponding proportion for all drugs/vaccines, giving a measure called the relative reporting ratio, 5 which can be illustrated in a 2 x 2 contingency table ( Table 1). The relative reporting ratio is one of several disproportionality measures which include the information component 5,6 and the proportional reporting ratio, [5][6][7] among others. The observed value is n = A; the expected value is E = (A + B)(A + C)/(A + B + C + D) and the relative reporting ratio RR= N/E.…”

Analysis of the Interaction between Clopidogrel, Aspirin, and Proton Pump Inhibitors Using the FDA Adverse Event Reporting System Database

“…35) Although research on the performance, accuracy, and reliability of different data mining algorithms is in progress, there is no recognized gold standard methodology. 25,36) In the Table 3 (for hemorrhagic events) and Table 6 (for embolic/thrombotic events), the crude ROR offered a rough indication of the signal strength for the influence of PPIs on combined administration of clopidogrel and aspirin. As shown in Table 3, the reporting of the hemorrhagic events decreased when PPIs were used along with concomitant clopidogrel and aspirin.…”

Mining Safety Signals in Spontaneous Reports Database Using Concept Analysis

“…Signal detection has been approached with a variety of analysis tools [7] including statistical methods for disproportionality assessment, deviation detection, etc. However, none of these proposes a way, both automated and feasible, for generating all potential signals from the srs db.…”