Table 4.

Performance of the classifier per source of data for the 28 GBD categories

	Exact-matching (% n/N)				Weighted average across 28 GBD categories
Source	All trials	One GBD category	No GBD category	Two or more GBD categories	Sensitivity	Specificity
Emdin 2015	66.7 (346/519)	66.4 (300/452)	68.2 (45/66)	100.0 (1/1)	71.5	96.4
Viergever 2013	82.2 (1045/1271)	85.3 (925/1085)	64.5 (120/186)	0.0 (0/0)	86.6	97.8
On going work	77.9 (758/973)	88.5 (700/791)	32.9 (51/155)	25.9 (7/27)	81.3	97.2

Exact-matching and weighted averaged sensitivities and specificities for the classifier to the 28 GBD categories for each source of data. The version of the classifier used was: using the Word Sense Disambiguation server, the expert-based enrichment database and the priority to the health condition field. Exact-matching corresponds to the proportion (in %) of trials for which the automatic GBD classification is correct. Exact-matching was estimated over all trials, trials concerning a unique GBD category, trials concerning 2 or more GBD categories, and trials not relevant for the GBD. The weighted averaged sensitivity and specificity corresponds to the weighted average across GBD categories of the sensitivities and specificities for each GBD category plus the “No GBD” category (in %)