Table 4.

Cross‐Validated AUCs for Various Predictive Models

	Binomial			Multinomial
	All fractures			Trochanteric fracture		Neck fracture
Model	AUC	95% CI	DEV	AUC	95% CI	AUC	95% CI	DEV
m	0.653	0.59–0.71	425	0.482	0.40–0.57	0.695	0.62–0.76	562
m + DXA	0.783	0.72–0.83	369a	0.714	0.63–0.78	0.762	0.69–0.83	508a
m + QCT	0.763	0.70–0.82	365a	0.731	0.64–0.80	0.732	0.65–0.80	498a
m + CBM	0.787	0.73–0.84	361a	0.777	0.68–0.84	0.818	0.75–0.87	466a
m + DXA + CBM	0.790	0.73–0.84	362	0.767	0.68–0.84	0.817	0.75–0.87	471c
m + QCT + CBM	0.794	0.74–0.85	343b,c	0.765	0.67–0.83	0.834	0.77–0.89	446b,c
m + all DXA	0.782	0.73–0.83	372a	0.699	0.61–0.78	0.777	0.69–0.84	508a
m + all QCT	0.756	0.69–0.81	370a	0.704	0.62–0.79	0.719	0.63–0.79	513a
Combined model 1	0.779	0.72–0.84	359a	0.689	0.60–0.76	0.760	0.69–0.82	497a
Combined model 2	0.783	0.72–0.83	355a	0.700	0.62–0.77	0.757	0.68–0.82	498a

AUCs calculated using binomial (all fractures) or multinomial (specific fractures) logistic regression. The base model m includes age + height + site. Significance (based on DEV) is given for p < 0.005, based on the model compared to ^am, ^bm + CBM, or ^cthe same model without CBM variables. The last 4 models were selected postanalysis: “all DXA” includes all 4 listed DXA variables; “all QCT” includes all 9 listed QCT variables; “Combined model 1”28 is m + DXA FnBMD + QCT trabecular FnBMD; and “Combined model 2”18 is m + DXA FnBMD and ThBMD + all 3 QCT trabecular BMD values.

AUC = area under the receiver‐operating characteristic curve; DEV = deviance.