Table 3.

Multivariate logistic regression analysis for CDKN2A methylation or CDKN2A (p16) loss (as an outcome variable) in colorectal cancer

Variables in the final model for CDKN2A methylation (as an outcome variable)	Multivariate OR (95% CI)	P value
CIMP-high (vs. CIMP-0)	39.6 (20.6-76.1)	<0.0001
CIMP-low (vs. CIMP-0)	5.30 (3.52-8.00)	<0.0001
PTGS2 (COX-2) expression	0.45 (0.28-0.72)	0.0008
BRAF mutation	2.62 (1.41-4.90)	0.0024
High tumor grade (vs. low tumor grade)	2.50 (1.33-4.70)	0.0045
Age (10-year increment as a unit)	1.38 (1.09-1.74)	0.0067
TP53 expression	1.70 (1.15-2.52)	0.0077
Body mass index ≥30 kg/m2 (vs. <30 kg/m2)	1.41 (0.87-2.27)	0.17
Variables in the final model for CDKN2A (p16) loss (as an outcome variable)	Multivariate OR (95% CI)	P value
CDKN2A methylation	12.7 (7.84-20.7)	<0.0001
CIMP-high (vs. CIMP-0)	2.23 (1.20-4.18)	0.012
Disease stage III-IV (vs. stage I-II)	1.52 (0.98-2.36)	0.061
KRAS mutation	0.67 (0.43-1.05)	0.082
LINE-1 methylation	1.76 (0.90-3.46)	0.099

The multivariate logistic regression model for CDKN2A methylation initially included sex, age at diagnosis, tumor location, body mass index, family history of colorectal cancer, disease stage, tumor grade, CIMP, MSI, KRAS, BRAF, LINE-1 methylation, TP53, CDKN1A, CDKN1B, CCND1, CTNNB1, PTGS2 and FASN. A backward stepwise elimination with a threshold of p=0.20 was used to select variables in the final models. The multivariate model for CDKN2A (p16) loss initially included CDKN2A methylation in addition to the above variables, and a backward stepwise elimination was performed.

CI, confidence interval; CIMP, CpG island methylator phenotype; MSI, microsatellite instability; OR, odds ratio.