Geographic remoteness, area-level socioeconomic disadvantage and inequalities in colorectal cancer survival in Queensland: a multilevel analysis

Table 2 Measures of model fit and estimates of geographical variations in all-cause and colorectal cancer-specific survival in Queensland, 1997–2007

Model	description¹	DIC²	Area-variance (95% CrI)³	p value	% reduction variance⁴	MOR (95% CrI)⁵
	All-cause survival
1	Null (no explanatory variables)	57769.71	0.025 (0.014, 0.039)	<0.001	-	1.16 (1.13, 1.21)
2	Individual-level covariates⁵	46072.45	0.011 (0.006, 0.018)	0.042	56	1.10 (1.08, 1.14)
3	Individual-level covariates and area-remoteness	46058.44	0.007 (0.003, 0.014)	0.101	72	1.08 (1.05, 1.12)
4	Individual-level covariates and area-disadvantage	46054.03	0.006 (0.001, 0.014)	0.078	76	1.08 (1.03, 1.12)
5	Individual- and both area-level covariates	46046.57	0.005 (0.001, 0.012)	0.118	80	1.07 (1.03, 1.11)
6	All covariates with area-level interactions	46051.12	0.005 (0.001, 0.012)	0.128	80	1.07 (1.03, 1.11)
	Colorectal cancer-specific survival
7	Null (no explanatory variables)	48609.89	0.021 (0.011, 0.034)	0.001	-	1.15 (1.10, 1.19)
8	Individual -level covariates	37013.72	0.008 (0.003, 0.012)	0.044	61	1.09 (1.05, 1.11)
9	Individual-level covariates and area-remoteness	36987.44	0.003 (0.001, 0.007)	0.231	86	1.05 (1.02, 1.08)
10	Individual-level covariates and area-disadvantage	36995.14	0.004 (0.001, 0.011)	0.173	81	1.06 (1.02, 1.10)
11	Individual- and both area-level covariates	36979.49	0.002 (0.001, 0.008)	0.231	91	1.04 (1.02, 1.09)
12	All covariates with area-level interactions	36993.31	0.003 (0.001, 0.011)	0.298	86	1.05 (1.02, 1.10)

CrI: credible interval for posterior distributions from Markov Chain Monte Carlo multilevel models.
¹ All except null models adjusted for time (years after diagnosis), patient age at diagnosis, sex, occupation, marital and Indigenous status, cancer stage, site, differentiation and surgical margins. Models 3 and 9 also adjusted for area remoteness; Models 4 and 10 also adjusted for area disadvantage; Models 5 and 11 also adjusted for area remoteness and area disadvantage; Models 6 and 12 include area-level interactions and all explanatory factors in Models 5 or 11 respectively.
² DIC Deviance Information criterion. Model with a smaller DIC (Difference ≥ 7 units) is better supported by the data while those with difference of 3–5 units can be weakly distinguished.
³ The residual area-level variance from the multilevel models.
⁴ The percentage reduction in variance reflects the change in area-level variance relative to the null model.
⁵ MOR (median odds ratio) translate the area-level variance to the odds ratio scale. MOR is the median value of the distribution of the odds ratio between an individual in the area at highest risk and an individual in the area at lowest risk when randomly selecting two individuals from two different areas. In absence of variation the MOR is 1 with a larger value indicating greater geographical variations.

ISSN: 1471-2407