Predictive models for kidney disease: improving global outcomes (KDIGO) defined acute kidney injury in UK cardiac surgery

Table 9 Calibration: Hosmer Lemshow tests and results of linear regression analysis of observed versus expected calibration plots

	Development sample Bristol and Birmingham n = 16,527				Validation sample Wolverhampton n = 4,468
	Any-stage AKI		Stage-3 AKI		Any-stage AKI		Stage-3 AKI
Hosmer-Lemeshow P -values ^a	P		P		P		P
Initial model (P <0.001)	0.490		0.001		0.192		<0.001
More inclusive model (P <0.05)	0.784		<0.001		0.406		<0.001
Euroscore	<0.001		<0.001		0.333		<0.001
Cleveland clinic	0.112		<0.001		0.141		<0.001
Metha score	0.009		<0.001		0.136		<0.001
Ng score	0.013		<0.001		0.174		<0.001
	Any-stage AKI		Stage-3 AKI		Any-stage AKI		Stage-3 AKI
Slope and intercept ^b	Slope	Intercept	Slope	Intercept	Slope	Intercept	Slope	Intercept
Initial model (P <0.001)	1.045	-0.007	1.094	-0.002	1.152	-0.009	1.529	0.009
More inclusive model (P <0.05)	1.050	-0.008	1.334	-0.007	1.140	-0.001	2.088	-0.003
Euroscore	1.050	-0.009	1.145	-0.004	1.152	-0.038	1.789	-0.016
Cleveland clinic	1.283	-0.048	1.145	-0.004	1.927	-0.138	3.376	-0.033
Metha score	1.144	-0.029	1.513	-0.013	1.348	-0.040	2.199	-0.010
Ng score	1.130	-0.021	1.378	-0.008	1.338	-0.010	2.196	-0.004

^aHigher values indicate better calibration. ^bSlope and intercept from linear regression analysis of observed versus expected values analysed by decile, as plotted in Figure 2. The closer the slope is to 1, and the closer the intercept is to 0, indicates better calibration.

ISSN: 1364-8535