Skip to main content
. 2018 Jul 24;41(12):1355–1369. doi: 10.1007/s40264-018-0699-2

Table 2.

Overview information on the considered combinations of reference sets and datasets

Reference set Positive controls Negative controls Dataseta Positive controls
N ≥ 1b
n (%)
Positive controls
N ≥ 3b
n (%)
Negative controls
N ≥ 1b
n (%)
Negative controls
N ≥ 3b
n (%)
Harpaz 62 75 VigiBase 41 (66) 29 (47) 36 (48) 24 (32)
Social 0.4 13 (21) 5 (8) 17 (23) 8 (11)
Social 0.5 8 (13) 5 (8) 8 (11) 2 (3)
Social 0.6 3 (5) 2 (3) 2 (3) 2 (3)
Social 0.7 3 (5) 1 (2) 2 (3) 2 (3)
Social 0.8 3 (5) 1 (2) 2 (3) 2 (3)
Social 0.9 3 (5) 1 (2) 2 (3) 2 (3)
Social 0.99 3 (5) 1 (2) 2 (3) 2 (3)
WEB-RADR 200 5332 VigiBase 197 (98) 180 (90) 5072 (95) 3853 (72)
Social 0.4 98 (49) 75 (38) 2527 (47) 1879 (35)
Social 0.5 85 (42) 56 (28) 2294 (43) 1653 (31)
Social 0.6 46 (23) 26 (13) 1461 (27) 879 (16)
Social 0.7 42 (21) 20 (10) 1345 (25) 772 (14)
Social 0.8 37 (18) 19 (10) 1267 (24) 679 (13)
Social 0.9 35 (18) 17 (8) 1216 (23) 624 (12)
Social 0.99 34 (17) 14 (7) 1176 (22) 585 (11)
Forum posts 61 (30) 28 (14) 1657 (31) 886 (17)

a‘Social 0.X’ means social media data from Twitter and Facebook, with a post-level threshold on the indicator score of 0.X. For forum posts, an indicator score threshold of 0.7 was used

bThese figures refer to the specific time points at which data were extracted for positive and negative controls for the purposes of receiver operating characteristic analysis