Poor physical health predicts time to additional breast cancer events and mortality in breast cancer survivors

Authors


Abstract

Background: Health-related quality of life has been hypothesized to predict time to additional breast cancer events and all-cause mortality in breast cancer survivors.

Methods: Women with early-stage breast cancer (n=2967) completed the SF-36 (mental and physical health-related quality of life) and standardized psychosocial questionnaires to assess social support, optimism, hostility, and depression prior to randomization into a dietary trial. Cox regression was performed to assess whether these measures of quality of life and psychosocial functioning predicted time to additional breast cancer events and all-cause mortality; hazard ratios were the measure of association.

Results: There were 492 additional breast cancer events and 301 deaths occurred over a median 7.3 years (range: 0.01–10.8 years) of follow-up. In multivariate models, poorer physical health was associated with both decreased time to additional breast cancer events and all-cause mortality (p trend=0.005 and 0.004, respectively), while greater hostility predicted additional breast cancer events only (p trend=0.03). None of the other psychosocial variables predicted either outcome. The hazard ratios comparing persons with poor (bottom two quintiles) to better (top three quintiles) physical health were 1.42 (95% CI: 1.16, 1.75) for decreased time to additional breast cancer events and 1.37 (95% CI: 1.08, 1.74) for all-cause mortality. Potentially modifiable factors associated with poor physical health included higher body mass index, lower physical activity, lower alcohol consumption, and more insomnia (p<0.05 for all).

Conclusion: Interventions to improve physical health should be tested as a means to increase time to additional breast cancer events and mortality among breast cancer survivors. Copyright © 2010 John Wiley & Sons, Ltd.

Introduction

Approximately 2.5 million US women are currently living with a history of breast cancer 1. Advances in detection and treatment have improved breast cancer prognosis, so the 5-year relative survival rates for women with breast cancer is 89.3% overall 2; however, a significant risk of recurrence continues through 15–20 years post-diagnosis 3. Cancer characteristics are highly predictive of early recurrences but their predictive role diminishes over time 4–6.

Considerable research has examined how diagnosis and treatment influence health-related quality of life (HRQOL) among breast cancer survivors 7, 8; however, fewer studies have explored the association between HRQOL and breast cancer recurrence and overall mortality 9. Of the two HRQOL domains (mental health and physical health), research suggests that physical health may be more important for prognosis and mortality. Physical health has been associated with survival in the general population and in head and neck cancer patients 10–12. In the EPIC–Norfolk study of 10 000 female participants, a 142% higher death rate was reported for women in the lowest compared with the highest quintile of physical functioning (a subscale of physical health) 12; however, three earlier studies of breast cancer patients did not find an association between physical health and survival 13–15.

A recent review 9 of other psychosocial variables and prognosis in breast cancer patients emphasized the considerable inconsistency in published studies. For example, depressive symptoms were associated with outcomes in six studies, while nine found no association; social support was associated in four studies, while six reported null findings; and anger/hostility was associated in four, while four others found no association 9. In a sample of 397 breast cancer survivors, Goodwin et al. 15 tested 146 different hypotheses relating psychosocial variables to health outcomes and reported that all associations could be explained by chance. However, that study and the majority of other previous studies included relatively small sample sizes with limited events (and short follow-up periods); thus, there is a need for studies with adequate power to address the role of HRQOL and other psychosocial variables on prognosis following breast cancer diagnosis.

The Women's Healthy Eating and Living (WHEL) Study provided the opportunity to assess whether HRQOL and other psychosocial variables predicted time to additional breast cancer events or all-cause mortality in a large sample of breast cancer survivors. This study, a randomized, controlled trial designed to test the effects of a dietary intervention, followed 3088 early-stage breast cancer survivors for a median of 7.3 years. We previously described baseline HRQOL for this population and noted that participants reported generally high HRQOL that was comparable to norms for women in the general population and other women with breast cancer, but they reported better social functioning and more role limitations due to physical problems. Neither the physical health nor the mental health summary scores were related to characteristics of the original cancer 16.

Subjects and methods

Study design and sample

Details of eligibility criteria, data collection, and assessment of cancer outcomes in the WHEL Study have been reported previously 17. Briefly, 3088 participants were randomized into the controlled diet trial at 7 study sites between 1995 and 2000. Major eligibility criteria for this volunteer cohort included: breast cancer diagnosis within the past 4 years of primary operable invasive stage I (≥1 cm), II, or IIIA breast carcinoma; age 18–70 years at the time of diagnosis; no current or planned chemotherapy; no evidence of recurrent disease or new breast cancer since completion of initial treatment; no other cancer in the past 10 years, and no discernible threats to life from co-morbidities during the study follow-up period. WHEL participants were mailed a series of questionnaires for completion before or at a baseline clinic visit. The Human Subjects Committee of the University of California, San Diego, and all participating institutions approved the study procedures.

Quality of life assessments

At baseline, all participants were asked to complete a 147-item Thoughts and Feelings Questionnaire 18 that included the SF-36-Item Health Survey 19. The SF-36 summarizes four physical subscales (general health perceptions, physical functioning, bodily pain, and role limitations due to physical health problems) and four mental subscales (mental health index, vitality, role limitations due to emotional problems, and social functioning) 16, 20. All eight subscales have been shown to be reliable (Cronbach's α = 0.75–0.91) and to have adequate construct validity in a variety of diseased populations 19. Physical and mental health summary scores can be derived from the eight subscales (range = 0–100; higher scores indicate better health).

Social support was assessed using the 9-item Medical Outcome Study social support scale (α = 0.75) 21; optimism, by the 6-item the Life Orientation Test—revised scale (α = 0.75) 22; hostility, by the 13-item Cook–Medley Cynicism subscale (α = 0.74) 23. Depression was assessed by the 8-item Center for Epidemiologic Studies Depression screen (α = 0.73), which has been validated in cancer patients 24, 25: the sum of the items yield a total score that can be converted to a logarithmic scale. A value of ≥0.06 in the logarithmic scale suggests clinical levels of depressive symptoms and the possibility of a diagnosable mood disorder.

Other assessments

Standardized questionnaires administered at baseline ascertained demographic, behavioral, and lifestyle characteristics, including age (<44, 45–54, 55–64, and ≥65 years), race/ethnicity (non-Hispanic white, African-American, Hispanic, Asian-American, and other), education (college graduate: yes, no), employment and marital status (yes, no), menopausal status (pre, peri, and post), smoking status (current, past, and non-smokers), drinking status (none, 1–19, and ≥20 g alcohol/day), and hot flash status (yes, no). Weight and height were measured at baseline with the participants wearing light clothing and no shoes. Body mass index (BMI) was calculated as kg/m2 and was categorized (<25, 25–29.9, 30–34.9, and ≥35 kg/m2).

Physical activity was assessed using a 9-item measure of physical activity from the Personal Habits Questionnaire 18, adapted from the Women's Health Initiative. This scale was validated in the WHEL Study against a standard physical activity recall and accelerometer reading (21), and responses were converted to metabolic equivalent tasks in minutes per week (20).

The insomnia score was the sum of scores for trouble falling asleep, waking several times, waking early, trouble resuming sleep, and overall sleep quality; adapted from 5-item WHI Insomnia Rating Scale 26. A score of 9 was used as the cut-point for clinical insomnia 27, 28.

Data on the pre-trial tumor characteristics and treatment were abstracted from medical records. Specific variables included prior chemotherapy use (yes, no), anti-estrogen use (yes, no), tumor type (either or both lobular and ductal invasive, none), tumor differentiation (well, moderate, poor), cancer stage, and estrogen and progesterone receptor status.

Outcome ascertainment

Primary outcomes were the time to additional breast cancer events and all-cause mortality. Additional breast cancer event-free survival is the time from study enrollment (1995–2000) to the development of an additional breast cancer event, defined as a recurrence from the original cancer or developing a new breast cancer, or the end of follow-up. Follow-up time was censored at the earliest of the following: (a) time of a participants' non-breast cancer death (for breast cancer events analysis) (b) the last documented staff contact date, or (c) study completion (June 2006). For time to death, follow-up time was censored at the earlier of: (a) time of the last documented staff contact date or (b) study completion (June 2006).

Statistical analyses

Data for baseline quality of life and/or psychosocial functioning were missing for 121 participants (3.9%) resulting in an analytic sample of 2967. Quality of life and psychosocial variables did not differ by randomization assignment. Additionally, there was no treatment group by time interaction with study outcomes. Hence, we consider all study participants as a single cohort. For ease of interpretation, variables were categorized as quintiles. Cox proportional hazard regression was used to determine the unadjusted association of quality of life and psychosocial functioning with time to additional breast cancer event and all-cause mortality. Hazard ratios (and associated 95% confidence intervals) were the measure of association. In addition, we assigned each quintile an ordered score (e.g. 0,1,2,3,4) and tested for a linear trend across quintiles.

Univariate associations were also examined for the following variables known to affect additional breast cancer events and all-cause mortality: (1) patient characteristics (age, BMI, race/ethnicity, education, marital and employment status), (2) lifestyle (physical activity, smoking, and alcohol consumption), (3) menopausal status and presence/absence of hot flashes, and (4) tumor characteristics (type, grade, stage, estrogen (ER), and progesterone receptor status). We started building the statistical model by including all variables that were univariately associated with outcomes at p-value of ⩽0.1. As the main study was a randomized trial and there is always the possibility that the intervention could affect other variables, group assignment was included in the model. We reduced the variables in the model with a standard backwards selection procedure. Finally, to develop the most parsimonious models we excluded variables that did not affect the β coefficient of the exposure variables by more than 10%.

The effect size and the associated confidence interval of the quintiles (Q) of the physical health score supported categorizing participants into two groups that were internally homogeneous: ‘Poor’ = Q 1–2 and ‘Better’ = Q 3–5. The poor and better physical health groups were compared and contrasted with baseline tumor characteristics and cancer treatment using χ2 tests.

Finally, a multivariate logistic regression model was used to identify lifestyle factors associated with being in the poor physical group. Likelihood ratio tests were used to test whether the modifiable factors were independently associated with poor physical health. The Hosmer–Lemeshow goodness-of-fit statistic 29 was used to check the model fitness. All tests were two-sided and analyses were conducted in SAS version 9.2 (Cary, NC).

Results

In this sample of 2967 breast cancer survivors, the mean (SD) baseline age was 53.3 (8.9) years and BMI was 27.3 (6.1) kg/m2; 55% were college graduates and 85% were white. Further details regarding the WHEL Study sample have been published elsewhere 30. The majority of tumors were Stage I (38.6%) or Stage IIA (32.9%) and poorly (35.9%) or moderately (39.9%) differentiated. Most participants had received some form of adjuvant therapy as part of their initial treatment (67.0% tamoxifen; 69.7% chemotherapy, and 61.5% radiation therapy) and 75.3% of tumors were ER positive. At the end of the trial, there were 492 (16.5%) additional breast cancer events and 301(10.1%) deaths from all causes.

As shown in Table 1, poor physical health was associated with shorter time to additional breast cancer events (p trend = 0.004) and all-cause mortality (p trend<0.0001). Mental health also appeared to negatively affect both outcomes in the unadjusted model (p trend = 0.05 and 0.03, respectively). Hostility predicted time to additional breast cancer events (p trend = 0.005) but not all-cause mortality, while none of the other psychosocial variables predicted either outcome (Table 2).

Table 1. Unadjusted hazard ratios (HR) of time to additional breast cancer events and all-cause mortality in relation to quality of life in a cohort of US breast cancer survivors followed for a mean of 7.3 years (n = 2967)
Variablec (Quintile)Additional breast cancer eventsaAll-cause mortalityb
 nEventHR (95%CI)p trendEventHR (95%CI)p trend
  • a

    aAdditional breast cancer events is the time from study enrollment to development of an additional breast cancer event.

  • b

    bAll-cause mortality is the time from enrollment to reported/confirmed death from all causes.

  • c

    cSee text for details regarding the measurement scales.

Physical health summary score
0−57.5588121Reference0.00487Reference<0.0001
57.6−76.36101160.88 (0.68, 1.14) 710.74 (0.52, 1.02) 
76.4–85.6595880.66 (0.50, 0.87) 590.62 (0.44, 0.86) 
85.7–91.9578710.53 (0.40, 0.72) 400.42 (0.29, 0.62) 
92−100594960.73 (0.56, 0.95) 440.46 (0.32, 0.67) 
Mental health summary score
0–63.3593114Reference0.0574Reference0.03
63.4−76.75931060.89 (0.68, 1.16) 690.89 (0.64, 1.24) 
76.8–85603870.71 (0.53, 0.94) 460.58 (0.40, 0.84) 
85.1–90.5592910.76 (0.57, 1.00) 560.72 (0.51, 1.02) 
90.6–100584940.80 (0.61, 1.05) 560.73 (0.52, 1.04) 
Table 2. Unadjusted hazard ratios (HR) of time to additional breast cancer events and all-cause mortality in relation to psychosocial variables in a cohort of US breast cancer survivors followed for a mean of 7.3 years (n = 2967)
Variablec (Quintile)Additional breast cancer eventsaAll-cause mortalityb
 nEventHR (95%CI)p trendEventHR (95%CI)p trend
  • a

    aAdditional breast cancer events is the time from study enrollment to development of an additional breast cancer event.

  • b

    bAll-cause mortality is the time from enrollment to reported/confirmed death from all causes.

  • c

    cSee text for details regarding the measurement scales.

Social support
11–3257687Reference0.2158Reference0.69
33–376521091.11 (0.84,1.48) 681.04 (0.73, 1.47) 
38–41596901.00 (0.74,1.34) 590.98 (0.68, 1.41) 
42–445521041.26 (0.95,1.68) 681.23 (0.86, 1.74) 
45568981.15 (0.86,1.53) 470.82 (0.56, 1.21) 
Optimism
0–1450282Reference0.7952Reference0.73
15–178561511.07 (0.81,1.40) 890.99 (0.70, 1.40) 
18395640.98 (0.70,1.36) 431.04 (0.69, 1.55) 
19–20540780.86 (0.63,1.17) 480.84 (0.57, 1.25) 
21–246651161.06 (0.80,1.41) 690.99 (0.69, 1.43) 
Hostility
060886Reference0.00559Reference0.19
1884711.11 (0.81,1.52) 420.94 (0.63, 1.40) 
2–37121241.06 (0.80,1.39) 780.95 (0.68, 1.33) 
4–54601091.38 (1.04,1.83) 601.08 (0.75, 1.55) 
6–13290991.41 (1.06,1.89) 611.23 (0.86, 1.76) 
Depression score
045977Reference0.6048Reference0.89
15951041.05 (0.78,1.41) 540.85 (0.57, 1.25) 
2−38391380.99 (0.75,1.31) 951.09 (0.77, 1.54) 
4−5505800.94 (0.68,1.28) 450.84 (0.56, 1.26) 
6−20558890.97 (0.71,1.32) 581.01 (0.69, 1.49) 

In fully adjusted multivariate models that included physical health, mental health and hostility (Table 3), being in better physical health remained significantly protective while mental health no longer predicted either outcome, whereas hostility remained significantly associated with time to additional breast cancer events, but not with all-cause mortality. Among the four individual scales that comprise the physical health summary score, role limitations due to physical health problems was the only significant predictor of additional breast cancer-free survival (p = 0.0007), whereas both physical functioning and role limitations were predictive of all-cause mortality (p-value 0.03 and 0.07, respectively) (data not shown).

Table 3. Adjusteda hazard ratios (HR) of additional time to breast cancer events and all-cause mortality in relation to quality of life in a cohort of US breast cancer survivors followed for a mean of 7.3 years (n = 2967)
Variabled (Quintile)Additional breast cancer eventsbAll-cause mortalityc
 nEventHR (95%CI)p trendEventHR (95%CI)p trend
  • a

    aThe additional breast cancer events and all-cause mortality model were both adjusted for age at randomization, race/ethnicity, menopausal status, tumor type, tumor grade, tumor stage, anti-estrogen use, clinical site, time between cancer diagnosis and study entry, physical activity, hot flashes, randomization group, and interaction between hot flashes and randomization group. All-cause mortality model was additionally adjusted for marital status.

  • b

    bAdditional breast cancer events is the time from study enrollment to development of an additional breast cancer event.

  • c

    cAll-cause mortality is the time from enrollment to reported/confirmed death from all causes.

  • d

    dSee text for details regarding the measurement scales.

Physical health
0–57.5588121Reference0.00587Reference0.004
57.6–76.36101160.94 (0.71,1.23) 710.85 (0.61,1.19) 
76.4–85.6595880.67 (0.50,0.91) 590.71 (0.49,1.04) 
85.7–91.9578710.53 (0.37,0.75) 400.51 (0.33,0.80) 
92–100594960.71 (0.50,1.00) 440.58 (0.36,0.92) 
Mental health
0–63.3593114Reference0.4174Reference0.32
63.3–76.75931061.05 (0.79,1.39) 691.03 (0.73,1.45) 
76.8–85603870.91 (0.67,1.23) 460.82 (0.55,1.23) 
85.1–90.5592911.07 (0.78,1.47) 561.16 (0.78,1.74) 
90.6–100584941.21 (0.85,1.72) 561.26 (0.81,1.98) 
Hostility
060886Reference0.0359Reference0.94
1884711.04 (0.76,1.44) 420.86 (0.57,1.29) 
2–37121241.05 (0.79,1.39) 780.91 (0.64,1.29) 
4–54601091.40 (1.05,1.88) 600.95 (0.65,1.38) 
6–13290991.24 (0.92,1.68) 610.93 (0.64,1.36) 

Compared with the participants with a better physical-health score, participants with a poor score had a 42% higher risk of experiencing additional breast cancer events (HR = 1.42: 95% CI = 1.16, 1.75, p-value = 0.0002) and a 37% higher risk of death from any cause (HR = 1.37: 95% CI = 1.08, 1.74, p-value = 0.009). Kaplan–Meier plots of time to additional breast cancer events and all-cause mortality showed that the poor physical-health group had a worse prognosis than the better physical-health group [log rank p<0.0001] (Figures 1(a) and (b)).

Figure 1.

(a) Time to additional breast cancer events by poor vs better physical health (PH) in a cohort of US breast cancer survivors followed for a mean of 7.3 years (n = 2967). (b) Time to all-cause mortality by poor versus better physical health (PH) in a cohort of US breast cancer survivors followed for a mean of 7.3 years (n = 2967). Q - Quintile

The two physical-health groups (poor vs better) were not statistically different in their initial cancer characteristics, such as poorly differentiated cancer (37 vs 35%, p = 0.20), advanced stage of cancer (AJCC VI stage IIIA and IIIC = 17.8 vs 14.6%, p = 0.07) or with both lobular and ductal invasive cancer (4.7 vs 5.9%, p = 0.41). Nor were these two groups different in the treatment that they received such as those who received any chemotherapy (70 vs 69%, p-value = 0.34); or those who were prescribed Tamoxifen® therapy (65 vs 68%, p = 0.11). The associations of potentially modifiable behavioral and lifestyle factors with physical health appear in Table 4. Logistic regression indicated that women in the poor physical-health group had higher BMI, lower physical activity, lower alcohol consumption, and more insomnia (p<0.05 for all) than those in better physical-health group.

Table 4. Behavioral and lifestyle factors associateda with poor physical health in a cohort of US breast cancer survivors followed for a mean of 7.3 years (n = 2967)
VariableNOR (95% CI)p trend
  • a

    aLogistics regression model was also adjusted for age at randomization, mental health summary score, and number of co-morbid conditions at baseline.

  • b

    bMetabolic equivalent task minutes per week.

  • c

    cScore of ≥9 was used as the cut-point for clinical insomnia.

Body mass index (kg/m2)
<251274Reference<0.0001
25–29.99171.4 (1.1, 1.7) 
30–34.94572.2 (1.7, 2.8) 
≥353192.9 (2.1, 3.9) 
Alcohol intake (drinks/month)
None932Reference0.003
1–5919100.8 (0.6, 0.9) 
≥601190.5 (0.3, 0.8) 
Physical activity (MET-min/week)b
0–904881.5 (1.2, 1.9)0.005
91–4496951.4 (1.2, 1.7) 
≥4501784Reference 
Insomnia (scale)c
<91805Reference<0.0001
≥911511.6 (1.3, 1.9) 

Discussion

In this large cohort of early-stage breast cancer survivors, poor physical health, as assessed by the SF-36 quality of life measure, was significantly associated with both a shorter time to additional breast cancer event and death from all causes. Breast cancer survivors who appear to be at greatest risk for both of these outcomes were in the lower two quintiles of the physical-health score. For both additional breast cancer events and death, this subgroup had ∼40% higher risk than those in the upper three quintiles of the physical-health score.

The increased risk for women with poorer physical-health scores is not explained by either sociodemographic variables or characteristics of the original cancer, as neither of these variables predicted the physical-health score 16. However, women with poorer physical-health scores were much more likely to be obese and to be physically inactive. In the recently reported RENEW trial, Morey et al. 31 demonstrated that a diet and physical activity intervention can be effective in reducing the rate of decline of the physical-health score that was observed in their population of elderly cancer patients. There would appear to be a strong justification for testing a similar intervention in a sample of breast cancer patients, because our data suggest that improving physical-health score could lead to an improvement in breast cancer outcomes.

In this study, the mental health summary score, which was marginally associated with study outcomes when considered alone, showed no relationship in the multivariate model. This finding is in accordance with most other studies that reported either a null or weak association between mental domain of HRQOL and survival 10, 11, 32, 33. Further, we confirmed the findings by Goodwin et al. 15 that various psychosocial variables (including social support, depression, insomnia, optimism, and hostility) were not associated in the multivariate analysis with either additional breast cancer events or all-cause mortality.

This analysis has several strengths, including the use of a broad range of validated and standardized scales for quality of life and psychosocial functioning, objective measures of height and weight, and a physical activity scale that has been validated in this study population 34. Treatment and tumor characteristics variables were obtained from patients' medical reports and charts and verified by an oncologist. Study outcomes triggered a medical record request and the diagnosis was confirmed by oncologist review. Reported deaths were verified with death certificates and vital status was checked with the Social Security Death Index. Detailed vitality status was available on 96% of all those enrolled in the study 30. The study had both a large sample size of women with invasive early-stage breast cancer and 7.3 years of follow-up so that it had the power to identify associations with study outcomes and to control for potential confounders, including demographic and lifestyle variables.

It may be that this study result could indicate that poor physical-health scores are a marker for participants with more advanced disease (clinical or subclinical) at the time of enrollment. A limitation of this study is that it used a single, well-validated, but general, quality of life measure (SF-36), rather than multiple measures. Finally, our physical-health measure came from a single point in time. Serial measures of physical health will enable a determination of how changing physical-health scores are associated with prognosis. The importance of physical health to outcomes needs to be further tested in a randomized trial with an intervention that improves physical health, such as the RENEW trial. 31

In summary, among breast cancer survivors, self-perceived physical health was a robust, positive predictor of time to additional breast cancer events and all-cause mortality. Women ranked in the lower 40% of the physical-health distribution had an increased risk for both outcomes. Initial data suggest that a lifestyle intervention may be able to improve physical health and reduce this risk. However, further studies that focus on changes in physical-health scores over time are needed to assess whether such an intervention might lead to improved health outcomes.

Acknowledgements

The Women's Healthy Eating and Living (WHEL) Study was initiated with the support of the Walton Family Foundation and continued with funding from the National Cancer Institute (Grant CA 69375). Some of the data were collected from General Clinical Research Centers, NIH grants M01-RR00070, M01-RR00079, and M01-RR00827.

Ancillary