Public health application of predictive modeling: an example from farm vehicle crashes

Table 4 Validation of the predictive models

Validation data years (Training data years)	Total validation data occupants (training data occupant)	Observed injuries or deaths in validation data (N)	Non-modifiable (Model 1)		Non + semi-modifiable (Model 2)		Non + semi + modifiable (Model 3)
Validation data years (Training data years)	Total validation data occupants (training data occupant)	Observed injuries or deaths in validation data (N)	Avg. pred. Prob.	Expected injuries/ deaths (N)	Avg. pred. Prob.	Expected injuries/ deaths (N)	Avg. pred. Prob.	Expected injuries/ deaths (N)
2008-‘10 (‘05-‘07)	7624 (7210)	1095	0.1407^b	1073	0.1445^e	1102	0.1437^h	1095
2009-‘10 (‘05-‘08)	5216 (9618)	779	0.1383^c	721^a	0.1405^f	733	0.1397ⁱ	729
2010 (‘05-‘09)	2615 (12,219)	372	0.1424^d	372	0.1410^g	369	0.1401^j	366

Abbreviations: Avg. pred. Prob. Average predicted probability
^ap-value = 0.0327 (Chi Sq = 4.56, df = 1), suggesting that expected injuries and deaths (n = 721) were significantly different than the observed (n = 779). All other expected to observed differences were non-significant
^bAUC = 0.69 based on training data 2005–2007
^cAUC = 0.70 based on training data 2005–2008
^dAUC = 0.69 based on training data 2005–2009
^eAUC = 0.75 based on training data 2005–2007
^fAUC = 0.75 based on training data 2005–2008
^gAUC = 0.74 based on training data 2005–2009
^hAUC = 0.77 based on training data 2005–2007
ⁱAUC = 0.77 based on training data 2005–2008
^jAUC = 0.77 based on training data 2005–2009

ISSN: 2197-1714