Table 1.

For both detrimental health outcomes (mortality from CeVD and heart diseases) the final non-nested models that are included into Occam’s group are shown with their final deviances (dev), difference in final deviances (Δdev) with respect to the model with the smallest final deviance, number of model parameters (N _par), AIC-values, difference in AIC-values (ΔAIC) with respect to the model with the smallest AIC-value, and AIC-weights

	dev	Δdev	N par	AIC	ΔAIC	AIC-weight
CeVD (ICD-9 430–438)
Streamlined baseline model	13,422.27		15	13,452.27
ERR-LNT model	13,417.53	3.38	16	13,449.53	1.76	0.2412
ERR-Q model	13,415.77	1.61	16	13,447.77	0	0.5823
ERR-two-line spline model, D th = 0.14 Gy	13,414.15	0	18	13,450.15	2.39	0.1765
Heart diseases (ICD-9 393–429, excluding 401, 403, 405)
Streamlined baseline model	13,163.17		20	13,203.17
ERR-LNT model	13,152.52	1.29	21	13,194.52	0	0.3089
ERR-Q model	13,154.54	3.32	21	13,196.54	2.02	0.1123
ERR-smooth step model, D th = 1.52 Gy	13,153.66	2.44	23	13,199.66	5.14	0.0236b
EAR-LNT modela	13,151.22	0	22	13,195.22	0.71	0.2171
EAR-Q modela	13,151.78	0.55	22	13,195.78	1.26	0.1647
EAR-LTH model, D th = 2.36 Gy	13,152.29	1.07	22	13,196.29	1.78	0.1271
EAR-smooth step model, D th = 2.54 Gy	13,152.31	1.09	23	13,198.31	3.79	0.0464b

The Akaike Information Criterion is denoted by AIC (AIC = dev + 2 × N _par). For CeVD, the deviance related to MMI is 13,415.46. As a comparison, the results from fitting the streamlined baseline models are also provided

^aContains an age-dependent dose–effect modifier

^bSee Online Resource, page 10, for an explanation why for heart diseases models with AIC-weights < 0.05 were included into the set of non-nested models that was used for MMI