Data Mining Analysis

We conducted empirical Bayes data mining using the Multi-Item Gamma Poisson Shrinker algorithm in the Oracle Empirica™ Signal system to assess for disproportional reporting of AEs reported for DigiBind and DigiFab. The Empirica Signal system utilizes a case-matching algorithm applied to all FAERS reports that systematically identifies likely duplicate reports [12]. Analyses were limited to US reports, undertaken separately for DigiBind and DigiFab, and adjusted for year of report received at the FDA, sex, and age. The main statistical score computed was the empirical Bayes geometric mean (EBGM) and the 90% confidence interval, with the lower and upper 95% confidence bounds represented as EB05 and EB95, respectively [12]. The EBGM reflects the relative reporting rate after Bayesian smoothing for a drug/biologic–event pair relative to all other drug/biologic–event pairs in the FAERS database [13]. An EB05 ≥ 2 is the threshold commonly used by the FDA as a criterion for considering an AE a potential signal to be further investigated. This threshold is associated with a high probability of a drug/biologic–event pair being reported at least twice as often as expected under the assumption that drug/biologic–events are randomly paired [14]. Data mining findings are used for signal detection but do not imply a causal association between the drug/biologic–event pair identified.