The counterintuitive effect of multiple injuries in severity scoring: a simple variable improves the predictive ability of NISS

Table 2 Models' Performances

Model	C statistics (95% CI)	P of C statistics comparison*	Hosmer-Lemeshow C statistics	Hosmer-Lemeshow H statistics	Akaike's information criterion
Simple
MaxAIS	0.729 (0.699-0.758)	/	11.52 p = 0.24	228.68 p < 0.01	1712
NISS	0.755 (0.726-0.784)	0.02	14.69 p = 0.14	7.12 p = 0.52	1635
NISS + num_inj	0.775 (0.745-0.804)	0.03	9.03 p = 0.52	10.32 p = 0.24	1602
Augmented†
MaxAIS	0.841 (0.820-0.862)	/	11.96 p = 0.28	18.66 p = 0.04	1542
NISS	0.865 (0.844-0.886)	<0.01	7.47 p = 0.68	17.51 p = 0.06	1352
NISS + num_inj	0.874 (0.855-0.894)	0.01	7.21 p = 0.72	10.27 p = 0.41	1331
Complete‡
MaxAIS	0.890 (0.872-0.909)	/	10.69 p = 0.38	12.71 p = 0.24	1234
NISS	0.898 (0.880-0.916)	0.06	5.50 p = 0.85	15.87 p = 0.10	1174
NISS + num_inj	0.901 (0.884-0.919)	0.09	4.00 p = 0.94	9.05 p = 0.52	1167
Complete‡, NISS>15
MaxAIS	0.888 (0.868-0.907)	/	7.22 p = 0.70	20.79 p = 0.02	1165
NISS	0.897 (0.879-0.916)	0.03	6.92 p = 0.73	19.14 p = 0.03	1105
NISS + num_inj	0.901 (0.883-0.919)	0.05	5.76 p = 0.83	13.68 p = 0.18	1098

* comparison with the preceding model in the table
num_inj = an indicator variable expressing the number of injuries (1,2,3+)
†augmented with age, gender and mechanism of injury
‡completed with the above variables plus systolic blood pressure and motor component of Glasgow Coma Scale

ISSN: 1757-7241