Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
Randomized Controlled Trial
, 141 (11), 2281-2290

Methodological Considerations for Disentangling a Risk Factor's Influence on Disease Incidence Versus Postdiagnosis Survival: The Example of Obesity and Breast and Colorectal Cancer Mortality in the Women's Health Initiative

Affiliations
Randomized Controlled Trial

Methodological Considerations for Disentangling a Risk Factor's Influence on Disease Incidence Versus Postdiagnosis Survival: The Example of Obesity and Breast and Colorectal Cancer Mortality in the Women's Health Initiative

Elizabeth M Cespedes Feliciano et al. Int J Cancer.

Abstract

Often, studies modeling an exposure's influence on time to disease-specific death from study enrollment are incorrectly interpreted as if based on time to death from disease diagnosis. We studied 151,996 postmenopausal women without breast or colorectal cancer in the Women's Health Initiative with weight and height measured at enrollment (1993-1998). Using Cox regression models, we contrast hazard ratios (HR) from two time-scales and corresponding study subpopulations: time to cancer death after enrollment among all women and time to cancer death after diagnosis among only cancer survivors. Median follow-up from enrollment to diagnosis/censoring was 13 years for both breast (7,633 cases) and colorectal cancer (2,290 cases). Median follow-up from diagnosis to death/censoring was 7 years for breast and 5 years for colorectal cancer. In analyses of time from enrollment to death, body mass index (BMI) ≥ 35 kg/m2 versus 18.5-<25 kg/m2 was associated with higher rates of cancer mortality: HR = 1.99; 95% CI: 1.54, 2.56 for breast cancer (p trend <0.001) and HR = 1.40; 95% CI: 1.04, 1.88 for colorectal cancer (p trend = 0.05). However, in analyses of time from diagnosis to cancer death, trends indicated no significant association (for BMI ≥ 35 kg/m2 , HR = 1.25; 95% CI: 0.94, 1.67 for breast [p trend = 0.33] and HR = 1.18; 95% CI: 0.84, 1.86 for colorectal cancer [p trend = 0.39]). We conclude that a risk factor that increases disease incidence will increase disease-specific mortality. Yet, its influence on postdiagnosis survival can vary, and requires consideration of additional design and analysis issues such as selection bias. Quantitative tools allow joint modeling to compare an exposure's influence on time from enrollment to disease incidence and time from diagnosis to death.

Keywords: breast cancer; colorectal cancer; methods; mortality; obesity; survival.

Figures

Figure 1
Figure 1. Analyses of “Cancer Mortality from Enrollment” versus “Post-Diagnosis Survival”
An analytic sketch that contrasts cancer mortality (left panel) and post-diagnosis cancer survival (right panel) with respect to cohort and time-scale. Each line or “study ID” represents a participant, where the survivor cohort (right panel) includes only participants with a cancer diagnosis, a subset of the cohort that was enrolled (left panel). The time to cancer mortality from enrollment can be partitioned into two parts: time from enrollment to incidence of a cancer, and time from cancer diagnosis to death as indicated by n-th participant.
Figure 2
Figure 2. Breast Cancer Incidence, Mortality and Survival in the Women’s Health Initiative: Multivariable Adjusted Hazard Ratios by BMI at Enrollment, compared to women of normal weight (BMI < 25)
Refer to Figure 1 for terminology definitions. Bolded summary statistics for T1, T2 & T3 were computed from the usual time-to-first-event Cox regression models with baseline normal weight (BMI < 25 kg/m2) as the referent group. The un-bolded HRs for T3 were obtained by multiplying the corresponding estimates of risk for incidence (T1) with survival (T2); HRT3 = HRT1 × HRT2 (see appendix for mathematical derivation). The p-values correspond to a 1 degree of freedom test for trend. To highlight the influence of time-scale (time from enrollment vs. time from diagnosis), all time-to-first event models were adjusted for age, race/ethnicity, education, bilateral oophorectomy, Gail 5-year risk of breast cancer, estrogen-alone use and duration, estrogen and progesterone use and duration, smoking status, diabetes mellitus, alcohol consumption, self-reported health at baseline, and stratified by baseline age group, WHI study (OS vs. CT), HT trial randomization group, dietary trial randomization group, hysterectomy status, Calcium/Vitamin D Randomized Trial randomization group (time-dependent) and extended follow-up (time-dependent).
Figure 3
Figure 3. Colorectal Cancer Incidence, Mortality and Survival in the Women’s Health Initiative: Multivariable Adjusted Hazard Ratios by BMI at Enrollment, compared to women of normal weight (BMI<25)
Refer to Figure 1 for terminology definitions. Methods correspond to those described in Figure 2 for breast cancer, but models were adjusted for age, race/ethnicity, education, estrogen-alone use and duration, estrogen and progesterone use and duration, smoking status, diabetes mellitus, alcohol consumption, self-reported health at baseline, family history of colorectal cancer, occurrence of colonoscopy/ sigmoidoscopy/flexible sigmoidoscopy ever, physical activity, total dietary energy, fiber, fat, fruits, vegetables, red meat and stratified by baseline age group, WHI study (OS vs. CT), HT trial randomization group, Calcium/Vitamin D Randomized Trial randomization group (time-dependent) and extended follow-up (time-dependent).

Similar articles

See all similar articles

Cited by 3 articles

Publication types

Feedback