Fig. 2.

The rate of spuriously highlighted drug–event combinations by disproportionality analysis (defined as IC₀₂₅ > 0) for different types and sizes of VigiBase subsets. a shows box plots for randomly generated subsets of sizes between 500 and 100,000 reports. Each box is based on 500 subsets, except for those with 500 reports (118 included subsets) and 750 reports (461 included subsets). b shows results for country-specific subsets. Of 131 countries, 21 (16%) were excluded, all with fewer than 500 reports. c displays results for subsets based on ATC groups; 2%, 6% and 20% of subsets were excluded at level 2, 3 and 4, respectively. The horizontal lines at 0.10 indicate an empirical threshold for normal spuriousness rates derived from large subsets; individual points in (b, c) above and below this threshold are drawn as red squares and blue circles, respectively. Note that all x axes are logarithmic and adjusted to the data of the individual panels