Table 3.

Comparison of performances of the constructed models to predict severe injuries after falls

A.Development dataset

Model	Risk Criteria	Event +ve	Event -ve		Total		Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)	F-measure
1	High	26	5 205		5 231
	Low	17	23 862		23 879		60.47	82.09	0.50	99.93	0.0099
	Total	43	29 067		29 110
2	High	25	4 282		4 307
	Low	18	24 785		24 803		58.14	85.27	0.58	99.93	0.0115
	Total	43	29 067		29 110
3	High	18	3 118		3 136
	Low	25	25 949		25 974		41.86	89.27	0.57	99.90	0.0113
	Total	43	29 067		29 110
STRATIFY	High	23	6 372		6 395
	Low	20	22 695		22 715		53.49	78.08	0.36	99.91	0.0071
	Total	43	29 067		29 110
FRAX™	High	32	10 438		10 470
	Low	11	18 629		18 640		74.42	64.09	0.31	99.94	0.0061
	Total	43	29 067		29 110
No screening		43	29 067		29 110				0.15
B. Test dataset

Model	Risk Criteria	Event +ve	Event -ve		Total		Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)	F-measure
1	High	14	4 685		4 699
	Low	4	14 902		14 906		77.78	76.08	0.30	99.97	0.0059
	Total	18	19 587		19 605
2	High	14	3 879		3 893
	Low	4	15 708		15 712		77.78	80.20	0.36	99.97	0.0072
	Total	18	19 587		19 605
3	High	9	2 337		2 346
	Low	9	17 250		17 259		50.00	88.07	0.38	99.95	0.0076
	Total	18	19 587		19 605
STRATIFY	High	10	3 327		3 337
	Low	8	16 260		16 268		55.56	83.01	0.30	99.95	0.0060
	Total	18	19 587		19 605
FRAX™	High	17	10 522		10 539
	Low	1	9 065		9 066		94.44	46.28	0.16	99.99	0.0032
		Total	18	19 587		19 605
No screening		18	19 587		19 605				0.09

After three models had been constructed to predict severe injuries after falls, we compared the performances of the models by applying the models to the development dataset (A) and the test dataset (B). Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and F-value were calculated. Event +ve, event positive; event -ve, event negative.