. 2013 Dec 28;88(3):701–723. doi: 10.1007/s00204-013-1156-8

Table 10.

Resampling probabilities of 10,000 theoretical chemicals according to UN GHS/EU CLP criteria

True class	Database	Endpoint	Sample size	Predicted class
True class	Database	Endpoint	Sample size	No Cat	Cat 2	Cat 1
No Cat	RCD		606	99.9	0.1	0.0
	NCD		4,611	99.9	0.1	0.0
Cat 2	RCD	All^a	162	3.9	95.5	0.6
		1 ≤ CO^{Maj b} < 3	124	2.3	97.2	0.6
		CR^Maj/CC^{Maj c} ≥ 2 (and CO^{Maj b} < 1)	38	6.0	94.1	NA
	NCD	All	582	11.7	88.2	0.1
		1 ≤ CO^{Maj b} < 3	315	5.0	94.8	0.2
		CR^Maj/CC^{Maj c} ≥ 2 (and CO^{Maj b} < 1)	252	9.7	90.3	NA
		1 ≤ IR^{Maj b} ≤ 1.5 only	15	7.6	92.4	0.0
Cat 1	RCD	All	187	0.0	10.8	89.2
		CO^{Maj b} ≥ 3	43	0.0	1.7	98.3
		CO = 4^d (and CO^{Maj b} < 3)	61	0.0	13.0	87.0
		IR^{Maj b} > 1.5 only	7	0.0	14.2	85.8
		Persistence only	76	0.1	13.5	86.4
	NCD	CO^{Maj b} ≥ 3 + IR^{Maj b} > 1.5 only^e	63	0.0	7.2	92.8
		CO^{Maj b} ≥ 3	48	0.0	1.9	98.1
		IR^{Maj b} > 1.5 only	15	0.0	8.2	91.8

Proportions in bold represent agreement between the predicted classification and the true classification

RCD Reference Chemicals Databases, NCD European New Chemicals Database, CO corneal opacity, CR conjunctiva redness, CC conjunctiva chemosis, IR iritis, NA not applicable

^aNone of the chemicals in this group was classified based on the driver 1 ≤ IR^Maj ≤ 1.5 only

^bsecond (n = 3 or 4) or third (n = 5 or 6) highest mean score

^cThe largest of the second (n = 3 or 4) or third (n = 5 or 6) highest mean CR and CC scores

^dObserved any time during the observation period (before day 21)

^eIndividual animal data only accessible for the mean scores of days 1–3. Information on persistence of effects is only available in the form of comments, often not specifying in which tissues and in how many animals the persistence of effects was observed. Consequently, it was not possible to estimate the resampling probabilities for the chemicals that were classified based only on CO = 4 (observed any time during the observation period, before day 21) or persistence of a tissue effect. The resampling probabilities reported for the classification drivers CO^Maj ≥ 3 and IR^Maj > 1.5 should therefore not be considered as representative of the overall resampling probabilities for chemicals classified as Cat 1 in the NCD