Testing the proportional hazards assumption in case-cohort analysis

Table 3 Assessment of proportional hazards for each variable in the multivariate cox model for the example case-cohort study of colorectal cancer risk, using insulin levels as the primary exposure variable

Covariates in the model	Corr with event time	Corr with rank of time	Corr with KM estimates
Insulin	0.154	0.176	0.164
Mets for physical activity: (0,3.75)-ref
(3.75,10)	0.389	0.384	0.397
(10,20)	0.045	0.040	0.044
> = 20	0.629	0.686	0.645
Ethnic: white -ref
Black	0.305	0.304	0.294
Hispanic	0.167	0.166	0.176
Others	0.820	0.586	0.704
Family history of colorectal cancer	0.628	0.598	0.624
History of colonoscopy	0.099	0.076	0.089
Smoking: none-ref
Former	0.962	0.858	0.887
Current	0.613	0.594	0.598
Alcohol consumption: none-ref
(0,3)	0.491	0.326	0.411
> = 3	0.219	0.256	0.255
NSAID	0.059	0.105	0.075
Age group continuous	0.322	0.181	0.261

Note: Abbreviation: Mets metabolic equivalent tasks per hour per week, NSAID use of nonsteroidal anti-inflammatory drugs in the preceding year.
Multivariate Cox model was applied to the WHI case-cohort study with insulin as the primary exposure variable, categorized as quartile groups based on the distribution in the subcohort and then treated as a linear trend. The model included age group categorized as (a) 50 to 54 years of age (referent), (b) 55 to 59, (c) 60–64, (d) 65 to 69, (e) 70 to 74, (f) 75–79.

ISSN: 1471-2288