Tumor and Treatment Characteristics and Mortality
The average follow-up interval between original breast cancer diagnosis and censor point was 8.7 years, with an average of 6.7 years of follow-up after baseline data collection. As of December 1, 2005, this study group (n = 1,490) had accumulated 9,665 person-years of observation, during which time there were 236 breast cancer events and 135 deaths: 118 were attributable to breast cancer, 10 were attributable to other cancers, and only seven were attributable to noncancer mortality.
In the univariate analysis (), as expected, tumor stage and grade were strongly related to mortality. Age, hormone receptor status, type of chemotherapy and adjuvant tamoxifen were not associated significantly with mortality.
Mortality by Age and Clinical Characteristics of Original Tumor and Treatment, Using Unadjusted Categorical Cox Models
Lifestyle Factors and Mortality
and present the univariate analyses of mortality for all lifestyle factors considered in our model. Less than 5% of participants were current smokers, averaging 11 cigarettes/d, and nearly 60% had never smoked. Smoking status was not associated with mortality (P = .45). Less than one third (29%) of these breast cancer survivors were current drinkers, with 7% classified as heavy drinkers (60 drinks/mo). Alcohol consumption showed an inverse association with mortality (P = .03).
Univariate Associations of Baseline Lifestyle Factors With Subsequent Mortality in WHEL Study Comparison Group, Using Categoric and Continuous Cox Models: Smoking, Alcohol, and BMI (N = 1,490)
Univariate Associations of Baseline Lifestyle Factors With Subsequent Mortality in WHEL Study Comparison Group, Using Categoric and Continuous Cox Models: Diet and Physical Activity (N = 1,490)
BMI was related to mortality (P = .06) with an apparent J-shaped relationship and no significant linear trend. Women categorized as low BMI (BMI < 20; n = 77) had twice the mortality (15.6% v 7.1%) as women with BMI 20 to 24.99 or categorized as overweight (BMI = 25 to 29.99). Women categorized as obese (BMI ≥ 30) had a higher mortality rate than normal-weight participants (12.4% v 7.1%; P = .01).
The median intake of VF was 4.93 servings/d. Consumption of VF was related to mortality (P = .02), although the trend was not linear (Ptrend = .08). Women who consumed fewer than 3.43 VF servings/d (quartile 1) experienced the highest mortality rate (12.4%). Mortality seemed to decrease with increasing intake to 5 VF servings/d.
The mean reported energy intake was 1,722 kcal/d and was not associated with mortality. The median intake of fat was 28.5% of energy intake, with the highest quartile consuming more than 33% energy from fat and the lowest quartile consuming less than 24% energy from fat; however, mortality rates did not differ significantly across quartiles (P = .6), and a significant trend was not observed (Ptrend = .10).
The lowest quartile of fiber consumption was less than 16 g/d and the highest was more than 25 g/d. The median fiber consumption was 20.2 g/d. Mortality did not differ across quartiles (P = .15) nor was there a linear trend (Ptrend = .12).
The median level of PA was 636 MET-min/wk. Women who reported no PA (7.7%) had a mortality similar to other women in the lowest quartile and were combined with them. Mortality rates showed a linear trend (Ptrend = .02), with the highest mortality in the lowest quartile (< 225 MET -min/wk) and reduced mortality, particularly in the highest two quartiles.
Composite Dietary Pattern-PA Variable
The correlation between the two continuous lifestyle variables, PA and VF, was 0.19. First, we made binary categories of each variable. For VF consumption, we used the recommendation of 5 servings a day as the cut point. In accordance with Holmes et al,3
we chose 540 MET -min/wk to for the PA categorization. A total of 30% of study participants were in the high VF/high PA category (mean, 7.6 VF servings/d; 1,513 MET -min/wk) with a mortality of 4.8%. Twenty-two percent of the sample was classified as low VF/high PA (mean, 3.4 VF servings/d; 1,478 MET-min/wk) with a mortality of 10.4%. Another 18% were classified as high VF/low PA (mean, 7.2 VF servings/d; 224 MET-min/wk) with a mortality of 10.7%. Finally, 30% of the sample was classified as low VF/low PA (mean, 3.1 VF servings/d; 221 MET-min/wk) with a mortality of 11.5%. A difference in mortality observed across categories was statistically significant (P
= .01), although the lifestyle variable did not differ by time since diagnosis.
Multivariate Regression Model of Mortality
Using our a priori criteria, an initial multivariate model included stage, grade, BMI, VF consumption, PA, and the combination VF-PA variable (). In the preliminary multivariate model, VF-PA interaction term reached statistical significance although neither main effect was significant. Accordingly, our final model included only the composite VF-PA variable from . Stage of original breast cancer was the major predictor of mortality: women with stage III cancers were 4.5 times more likely to die during the study period than were women with stage I disease. Tumor grade was also significantly associated with mortality. Adjusted for other variables, having low BMI (BMI < 20) at study entry (average 20 months postdiagnosis) was associated with almost a two-fold increase in mortality, although the subsample size was small and this was only marginally significant (P = .08). Overweight (BMI = 25 to 29.99) was not associated with an increased risk (P > .9) but obese women (BMI > 30) tended to have a higher death rate, although this relationship did not reach statistical significance (P = .16). No differences were observed in mortality for the first three combined VF-PA categories; however, compared with women in the low VF/low PA group, the hazard ratio for women in the high VF/high PA was 0.56 (95% CI, 0.31 to 0.98).
Final Cox Proportional Hazard Ratios for Mortality by Four Diet and Physical Activity Categories: WHEL Comparison Group (N = 1,490)
The strong protective association of high VF/high PA with improved survivorship is also evident in the Kaplan-Meier survival curves (): survival data from study enrollment through the December 2005 follow-up illustrates an estimated 10-year survival rate postdiagnosis of 93% in the high VF/high PA group, and 86% to 87% in the other three groups. Thus, the estimated absolute unadjusted mortality risk reduction was 6% to 7% at 10 years for women who achieved this pattern, compared with women who consumed fewer daily VF servings and/or performed less PA.
Fig 1 Kaplan-Meier survival after Women’s Healthy Eating and Living (WHEL) Study enrollment by four diet and physical activity categories. Low vegetables-fruits (VF), less than 5 servings/d; high VF, ≥ 5 servings/d; low physical activity (PA), (more ...)
presents the mortality experience for categories of BMI and the composite VF-PA variable, comparing nonobese with obese women. For the nonobese group, there was no difference in mortality in the study period for women in any group other than the high VF/high PA group with estimated mortality percentage between 9.1% and 9.9%. Mortality among women in the high VF/high PA group (4.9%; 95% CI, 2.7% to 7.1%) was approximately half that of the other groups. For women who were obese, there was no difference in mortality in the same three study groups, with estimates from 12.7% to 14.2%. Again, there were significantly fewer deaths in the 16% of obese women classified in the high VF/high PA group (4.7%; 95% CI, 0% to 9.8%). Within each VF/PA category, women who were obese had an apparent increase in mortality compared with those who were not obese, except in the high VF/high PA group, where the observed mortality rates were comparable.
Fig 2 Mortality by diet and physical activity (PA) in Women’s Healthy Eating and Living Study comparison group: body mass index (BMI) categories. Bars show proportion (SE) for all-cause mortality by baseline BMI category. Low vegetables-fruits (VF), (more ...)
We also assessed whether the high VF/high PA lifestyle effect on survival was limited to women with hormone receptor–positive tumors (), as might be expected if the mechanism of action was mediated by gonadal reproductive hormones. Univariate analysis showed no survival advantage for healthy lifestyle in estrogen receptor–negative, progesterone receptor–negative (ER negative, PR negative) group (P = .4), a borderline advantage for ER-negative, PR-positive group (P = .09), and significant advantages for ER-positive, PR-negative (P = .04) and ER-positive, PR-positive groups (P = .01).
Fig 3 Mortality by tumor hormone receptor status and lifestyle in Women’s Healthy Eating and Living Study comparison group. Bars show proportion (SE) for all-cause mortality by baseline tumor estrogen receptor (ER) and progesterone receptor (PR) status (more ...)
Finally, we investigated cause of death and saw no difference in breast cancer versus other-cause mortality when stratified by the lifestyle variable (P = .37).