Table 3

Predictive performance of the four instruments.

					Positive predictive values, % (n)

	AUC-I ‘original’	AUC-II ‘3 categories’	Cut-offs (n), sensitivity (%), and specificity (%)		Low risk	Medium risk	High risk
Risk estimation by GP (n = 283^a)	0.59 (0.52 to 0.66)	0.58 (0.51 to 0.65)	Cut-off ≥1 Sensitivity = 86 Specificity = 17	Cut-off ≥4 Sensitivity = 28 Specificity = 87	33.3 (45)	33.0 (185)	56.6 (53)

Örebro questionnaire (n = 296^a)	0.61 (0.54 to 0.67)	0.58 (0.51 to 0.65)	Cut-off >68 Sensitivity = 79 Specificity = 26	Cut-off >99 Sensitivity = 35 Specificity = 81	32.4 (71)	32.7 (150)	52.0 (75)

	Linton (2002)^b	0.61 (0.54 to 0.68)	Cut-off ≥90^b Sensitivity = 52 Specificity = 66	Cut-off >105^b Sensitivity = 28 Specificity = 89	30.3 (175)	39.1 (69)	59.6 (52)

Low Back Pain Perception Scale (n = 298^a)	0.59 (0.52 to 0.66)	0.57 (0.50 to 0.64)	Cut-off ≥2 Sensitivity = 80 Specificity = 27	Cut-off ≥4 Sensitivity = 30 Specificity = 81	30.1 (73)	36.1 (155)	48.6 (70)

Prediction rule (n = 267^a)	0.75 (0.69 to 0.81)	0.72 (0.66 to 0.79)	Cut-off ≥0.28 Sensitivity = 79 Specificity = 55	Cut-off ≥0.41 Sensitivity = 57 Specificity = 81	18.6 (113)	33.3 (66)	63.6 (88)

↵a Incidental missing values 9.9% (GP estimation), 5.7% (Örebro questionnaire), 5.1% (Low Back Pain Perception Scale), and 15% (prediction rule).
↵b Cut-off scores recommended by Linton.27 AUC = area under the curve.