Table 4

Assessment of linearity and non-linearity on subsite–molecular relationship in colorectal cancer by multivariate logistic or linear regression model

Multivariate regression model	Outcome variable (molecular feature)	Bowel subsite variable (from rectum to ascending colon)	Squared subsite variable		Cubic subsite variable		LRT
Multivariate regression model	Outcome variable (molecular feature)	p Value (Wald test)	Included	p Value (Wald test)	Included	p Value (Wald test)	Degrees of freedom	p Value^*
Logistic	CIMP	<0.0001	No	–	No	–	–	Referent
		0.0017	Yes	0.17	No	–	1	0.17
		0.68	Yes	0.15	Yes	0.096	2	0.098
Logistic	MSI	<0.0001	No	–	No	–	–	Referent
		0.020	Yes	0.56	No	–	1	0.56
		0.93	Yes	0.48	Yes	0.42	2	0.61
Logistic	BRAF mutation	<0.0001	No	–	No	–	–	Referent
		0.0041	Yes	0.26	No	–	1	0.26
		0.56	Yes	0.76	Yes	0.63	2	0.47
Logistic	KRAS mutation	0.66	No	–	No	–	–	Referent
		0.48	Yes	0.42	No	–	1	0.42
		0.16	Yes	0.19	Yes	0.23	2	0.35
Logistic	PIK3CA mutation	0.0034	No	–	No	–	–	Referent
		0.070	Yes	0.20	No	–	1	0.20
		0.096	Yes	0.23	Yes	0.30	2	0.26
Linear	LINE-1 methylation level	0.020	No	–	No	–	–	Referent
		0.0070	Yes	0.0006	No	–	1	0.0036
		0.036	Yes	0.0022	Yes	<0.0001	2	<0.0001

A multivariate regression model included age, sex, year of diagnosis, family history of colorectal cancer, body mass index, physical activity, smoking, alcohol consumption and the bowel subsite variable with or without the squared and cubic subsite variables, as indicated in the table.
↵* A significant p value by the LRT indicates a non-linearity, and a combination of insignificant p values by LRT and a significant p value by the Wald test on the subsite variable in the model without the squared or cubic location variable indicates a linear relationship.
CIMP, CpG island methylator phenotype; LRT, likelihood ratio test; MSI, microsatellite instability.