|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Journal of Clinical Oncology, Vol 22, No 20 (October 15), 2004: pp. 4184-4192 © 2004 American Society of Clinical Oncology. DOI: 10.1200/JCO.2004.12.091 Health-Related Quality of Life and Psychosocial Status in Breast Cancer Prognosis: Analysis of Multiple VariablesFrom the Departments of Medicine and Surgery, Division of Clinical Epidemiology at the Samuel Lunenfeld Research Institute; Mount Sinai Hospital; Toronto-Sunnybrook Regional Cancer Centre; St Michael's Hospital; and University of Toronto, Toronto, Ontario, Canada Address reprint requests to Pamela J. Goodwin, MD, Mount Sinai Hospital, 1284-600 University Avenue, Toronto, Ontario M5G 1X4, Canada; e-mail: pgoodwin{at}mtsinai.on.ca
PURPOSE: Evidence that psychosocial status and health-related quality of life (HRQOL) are associated with breast cancer (BC) outcomes is weak and inconsistent. We examined prognostic effects of these factors in a prospective cohort study. PATIENTS AND METHODS: Three hundred ninety-seven women with surgically resected T1 to T3, N0/N1, M0 BC completed the European Organization for Research and Treatment of Cancer Quality of Life Questionnaire (Core 30 items), Profile of Mood States, Psychosocial Adjustment to Illness Scale, Impact of Events Scale, Mental Adjustment to Cancer Scale, and the Courtauld Emotional Control Scale 2 months after diagnosis and 1 year later. Data on tumor-related factors, treatment, and outcomes were obtained prospectively from medical records, and Cox survival analyses were performed.
RESULTS: Mean age was 52.0 ± 9.9 years. Two hundred twenty-five women had T1, 136 women had T2, 16 women had T3, and 20 women had TX tumors; 127 were N1. One hundred thirteen women received adjuvant chemotherapy, 130 received hormone therapy, 45 received both, and 109 received neither. We investigated 140 prognostic associations; four were found to be statistically significant at a P value of CONCLUSION: HRQOL and psychosocial status at diagnosis and 1 year later are not associated with medical outcome in women with early-stage BC.
Breast cancer diagnosis and treatment are frequently associated with psychologic distress and reduced health-related quality of life (HRQOL).1 This distress can influence mood and ability to cope; it may also influence marital, family, and social relationships; functioning at home and work; sexual activity; and HRQOL. Considerable research has focused on the development and evaluation of a variety of interventions designed to reduce this distress2 and on the potential association of this distress with cancer outcomes. Several mechanisms have been proposed for an effect of psychologic factors and HRQOL on cancer outcomes.3 Many of these mechanisms fall in the realm of psychoneuroimmunology, an evolving discipline that investigates the complex relationships between psychologic state and physiologic processes within the body. In breast cancer, the key areas of interest have been the effect of psychologic distress on hormones in the hypothalamic-pituitary-adrenal axis (notably cortisol and melatonin) and on immune function (notably reduced natural and lymphokine activated killer cell activity). Evidence exists that psychologic distress, stress, and inadequate social support can be associated with changes in these factors in breast cancer patients. For example, Andersen et al4 reported that high levels of stress after surgical treatment for breast cancer were associated with reduced natural-killer cell activity, whereas Turner-Cobb et al5 reported that certain aspects of interpersonal support were associated with salivary cortisol levels. Unfortunately, causality has not been established for any of these potential mechanisms; that is, none has been clearly associated with breast cancer outcome. Reduced compliance with effective medical interventions in distressed women has been suggested as an alternate mechanism for prognostic effects of psychologic factors in breast cancer. Empiric studies of the prognostic effect of psychosocial status and HRQOL in breast cancer have yielded inconsistent and largely negative results. Early reports6 suggesting that fighting spirit and denial (avoidance) coping styles were associated with longer survival have not been replicated.7 ,8 Recent research suggests that minimization9 or a helplessness-hopelessness coping style7 is associated with outcome. Social support, emotional expressiveness, and low levels of psychologic distress have also been inconsistently associated with improved outcomes.8 ,10 ,11-14 Coates et al15 made the intriguing observation that HRQOL was associated with outcome in advanced, but not early, breast cancer. They suggested that any prognostic significance of HRQOL scores in early-stage breast cancer were minimal or were obscured by effects of adjuvant chemotherapy, whereas patient perception of the severity of underlying illness may have determined HRQOL scores after metastases developed, thereby leading to readily identifiable prognostic effects of HRQOL in advanced breast cancer. These inconsistent results may be due to the presence of weak effects, inadequately powered studies, different effects in different stages of breast cancer or limitations in measuring psychosocial attributes, all of which might lead to false-negative results. They may also reflect the play of chance. The latter may lead to false-positive associations, particularly when large numbers of variables are studied and hypothesized effects, analytic cutoffs and subgroups are not defined in advance. Failure to adjust for recognized prognostic factors may also lead to false-positive associations. Women who have involved axillary nodes who are told that they are at increased risk of recurrence might develop psychologic distress as a result of this knowledge; their poor breast cancer outcomes resulting from the axillary node involvement could be misinterpreted as being caused by the ensuing psychologic distress. We evaluated the prognostic effects of psychosocial and HRQOL attributes measured at diagnosis and 1 year later in a prospectively studied cohort of women with newly diagnosed locoregional breast cancer. We hypothesized that worse HRQOL,15 greater psychologic distress,7 ,10 lower emotional expressiveness and a less active coping style,6 ,7 ,9 ,10 as well as a greater adverse impact of breast cancer on functioning and social/family/vocational adjustment,10 ,11 ,13 would be associated with increased risk of relapse and death after adjustment for tumor and treatment-related prognostic variables. We planned to treat each variable as continuous in our prognostic analyses. Specific cut points were evaluated in a limited number of secondary analyses in an attempt to replicate previously reported associations. Although we have used the psychoneuroimmunologic model discussed above as the basis for formulating our hypothesis, potential physiologic mechanisms for prognostic effects were not investigated.
Population Assembly Three hundred ninety-seven women diagnosed with breast cancer at participating University of Toronto teaching hospitals between October 1991 and May 1996 were enrolled. They represent women recruited during the final 5 years of a larger prospective cohort study examining prognostic effects of a number of lifestyle-related factors.16 ,17 Women younger than 75 years of age with T1 to T3, N0/N1, M0 breast cancer treated by lumpectomy and axillary dissection or mastectomy were eligible. Women were excluded if they met any of the following criteria: prior malignancy (except nonmelanoma skin cancer or carcinoma-in-situ of the cervix); serious coexisting medical conditions; use of medications that could influence diet, lipids, or glucose; or inability to speak English. The refusal rate was 16.7%. Participating women could receive adjuvant chemotherapy and/or tamoxifen as deemed necessary by their treating physician. Subjects provided written informed consent to participate in this study as approved by the Human Subjects Committee at the University of Toronto.
Measurement HRQOL was measured using the European Organization for Research and Treatment of Cancer Quality of Life Questionnaire–Core 30 (EORTC QLQ-C30),18 a 30-item standardized measure that includes a global health status scale (two items), five functioning scales (physical, role, emotional, cognitive, and social), three multi-item symptom scales (fatigue, pain, and emesis) and six single items (dyspnea, sleep disturbance, appetite, diarrhea, constipation, and financial impact). Higher scores on the global health status and functioning scales reflect better HRQOL. Mood was measured using the Profile of Mood States (POMS),19 a 65-item adjectival list with each item scored on a scale of 0 to 4 (not at all to extremely). This questionnaire yielded scores for depression/dejection, tension/anxiety, anger/hostility, fatigue/inertia, vigor/activity, and confusion/bewilderment, as well as a Total Mood Disturbance (TMD) score. Higher scores reflect greater amounts of each attribute. The Impact of Events Scale (IES)20 measures symptoms of the stress response syndrome. The identified stressor was breast cancer diagnosis. Fifteen items were scored on a four-point scale from not at all to often, grouped into two subscales (intrusion and avoidance), and combined into a total score. Higher scores reflect greater distress. Coping was measured using the Mental Adjustment to Cancer scale, (MAC)21 which includes 40 items scored one to four (definitely doesn't apply to me to definitely does apply to me). Responses were grouped into five subscales: fighting spirit, hopelessness/helplessness, anxious/preoccupation, fatalism (previously labeled stoic acceptance), and avoidance (previously labeled denial). Higher scores reflect greater use of each coping style. The Courtauld Emotional Control Scale (CECS)22 includes 21 items scored 1 to 4 (almost never to almost always). Responses were grouped into three categories: anger, anxiety, unhappiness, and a total control scale. Higher scores represent greater emotional control. The Psychosocial Adjustment to Illness Scale Self-Report (PAIS-SR)23 evaluates the impact of breast cancer and its treatment on adjustment. This scale includes 46 items, each having a four-level scale. Items are grouped into seven subscales: healthcare orientation, vocational environment, domestic environment, sexual functioning, extended family relationships, social environment, and psychologic distress. A total score can be obtained by summing the standardized T scores of the individual subscales. Higher scores reflect greater adverse impact of the cancer and its treatment on functioning.
Follow-Up and Outcomes
Statistical Analysis All P values reported are two-tailed likelihood ratio P values calculated for the psychosocial variable treated as a continuous variable after adjustment for the prognostic variables. CIs are calculated from Wald error estimates.
Characteristics of the Study Population On average, study patients were 52.0 ± 9.9 years old, just under 50% were postmenopausal, and the majority had undergone lumpectomy for T1 or T2 tumors and had negative axillary nodes. The majority had hormone receptor–positive tumors. Most received adjuvant chemotherapy, hormone therapy, or both (Table 1). Any postsecondary education was associated with improved OS (adjusted P = .02) but not DDFS (adjusted P = .06). Marital status (currently married or living common law v not married or living common law v unknown marital status) was not associated with either outcome. Baseline questionnaires were completed by 378 women; follow-up questionnaires were completed by 323 women. Median follow-up of the survivors was 5.8 years (range, 0.3 to 8.9 years). Only three women were followed-up for less than 1 year; 50% were followed  6 years. Fifty-seven women experienced distant recurrences, and 34 women died.
Prognostic Associations of Psychosocial Questionnaire Variables Prognostic associations are listed in Tables 2 through 5. The P values shown in these tables were obtained from multivariate Cox proportional hazards models as described above. We have provided point estimates of hazard ratios (HRs) of distant recurrence and death. These HRs were calculated by comparing the hazard of distant recurrence or death at the 75th percentile of scores to the hazard at the 25th percentile of scores. The choice of these two points was arbitrary. Had HRs been calculated to compare more extreme percentiles, the magnitude of these HRs would have been greater. HRs of greater than 1.00 reflect an increased risk of distant recurrence or death.
None of the baseline variables was significantly associated with either DDFS or OS. The HRs that were calculated for the extremes in scores described above were modest, the majority being between 0.75 and 1.5; no HRs greater than 2.0 or less than 0.5 were identified.
There was little evidence of significant prognostic associations of 1-year psychosocial measurements. The associations that were significant at a P
Targeted Analysis To Confirm Prognostic Associations Reported by Other Investigators
Interpretation of Our Results
We also compared our observations to our a priori hypotheses. Two of the significant associations (role functioning on the EORTC and domestic environment on the PAIS with OS) were in the direction we hypothesized, and two (cognitive functioning on the EORTC QLQ-C30 with DDFS and avoidance on the IES with OS) were in the opposite direction. This distribution of findings is consistent with chance. The magnitude of the HRs for the significant associations was modest, representing at most a potential two-fold effect, within the range that could be due to chance or confounding. Next, we evaluated the timing of measurements. None of the baseline measurements (least likely to be impacted by systemic treatment or undiagnosed recurrence) was significantly associated with either outcome. The four potentially significant associations were seen for follow-up measurements which are most likely to be impacted by adjuvant treatment. Adjuvant treatment was given to women at greatest risk of recurrence, thus its use reflects prognosis. Its use may also impact psychosocial status and HRQOL. Because of this, these follow-up associations may represent confounding by treatment. We also sought evidence of consistency and coherence of our findings. We expected a variable that predicted DDFS would also predict OS. This was not the case for any of our variables. We sought evidence of similar associations among variables measuring similar attributes. For example, the statistically significant association of cognitive functioning (EORTC QLQ-C30) at 1 year with DDFS (HR = 1.76) would ideally have been reflected in a similar association for confusion/bewilderment (POMS). Because the latter is scaled in an opposite direction to the former, we would have expected to see an HR significantly less than 1.00. The observed HR was 1.11 (P = .59). This is consistent with the association of cognitive functioning with DDFS being caused by chance. Finally, we examined the magnitude of our P values. The smallest P value was .014, a P value that would arise by chance in one of every 71 comparisons; we performed twice that many. Had we used a statistical corrections for multiple comparisons, for example the Bonferoni correction,24 which assumes that prognostic associations of all of our variables were unrelated (an assumption that is probably not valid), we would have divided a desirable overall study P value of .05 x 140 to obtain a P value of .00036 as our upper boundary for considering individual associations significant. None of our P values were this small. Taking a more liberal approach, assuming some of our variables had correlated prognostic associations (ie, a smaller number of independent variables were studied), we could have divided our desirable study-wide P value of .05 by the number of questionnaires leading to a P value of .0083 to define significance (.05 ÷ 6). None of our observed P values were at this level. We would have had to assume that there were three or fewer independent attributes to arrive at an adjusted P value that would have permitted even one of our observed associations to be considered significant. Once again, this suggests that chance may explain the associations we identified.
We have found little convincing evidence that HRQOL or the psychosocial variables studied have important associations with DDFS or OS in women with newly diagnosed locoregional breast cancer. Although four significant associations were identified, seven would have been expected because of chance alone. The modest magnitude of the associations, the lack of consistency across outcomes and over time, the lack of coherence of results across different questionnaires, and the fact that all of our significant associations were for follow-up measurements (most likely to be influenced by treatment) and that half of our significant associations were in a direction opposite to our a priori hypothesis all suggest that our findings may be due to chance or confounding by disease- and treatment-related variables. For our baseline measurements (which are least susceptible to confounding by treatment), we had planned to have 80% power to detect HRs in the range of 2.0 to 2.5 for comparison of 75th and 25th percentiles, depending on the variable. Post hoc power calculations using Cox proportional hazards models with continuous predictors demonstrated we had 80% power to detect slopes that correspond to the following HRs for comparison of the 75th to the 25th percentile: POMS TMD, HR = 1.79; PAIS total score, HR = 1.85; IES total score, HR = 1.78; MAC fighting spirit, HR = 1.86; CECS confusion, HR = 1.83; EORTC QLQ-C-30 global health scale, HR = 2.08. We had 80% power to detect the following HRs for OS: POMS TMD, HR = 2.13; PAIS total score, HR = 2.21; IES total score, HR = 2.13; MAC fighting spirit, HR = 2.26; CECS confusion, HR = 2.19; EORTC QLQ-C30 global health scale, HR = 2.65. For comparison, observed HRs for key traditional clinical and prognostic variables were as follows: nodal stage (N0 v N1), HR = 4.5 (95% CI, 2.6 to 7.9) for DDFS and HR = 4.3 (95% CI, 2.1 to 8.8) for OS; tumor stage (T1 v T2, T3, TX), HR = 2.7 (95% CI, 1.5 to 4.7) for DDFS and HR = 4.9 (95% CI, 2.2 to 10.9) for OS; nuclear grade (3 v 1, 2), HR = 2.4 (95% CI, 1.4 to 4.3) for DDFS and HR = 2.2 (95% CI, 1.1 to 4.5) for OS. Thus we believe we had adequate power to detect clinically important prognostic effects of psychosocial variables, particularly for DDFS. As discussed earlier, psychologic status and HRQOL have been postulated to influence breast cancer prognosis through a variety of mechanisms. We have not examined any of these physiologic or compliance-related mechanisms in our research, nor have we measured all potentially relevant psychologic attributes. Notably, we have not measured social support, a factor that has been associated with breast cancer outcome by Hislop et al10 Furthermore, although our measure of mood (POMS) provides scores for depression and anxiety, these scores are not intended to be measures of clinical depression or anxiety; instruments such as the HADS (used by Watson et al7) would provide a better measure of depressive symptoms. Thus our observations should not be generalized to psychosocial attributes that we have not measured. They should also not be generalizable to other cancers or to other stages of breast cancer. The patients we studied represent a relatively homogenous group of women who presented to teaching hospitals with early-stage breast cancer. The majority were white, married, with low-to-moderate levels of psychologic distress, fairly good HRQOL, and minimal impact of the illness and its treatment on psychosocial adjustment. Our findings may have been different had we studied a more diverse group of patients or had our patients reported greater baseline psychologic distress. In general, we have not performed subset analyses to examine the impact of extremes of psychologic distress because of our desire not to further increase our study-wide type I error. However, we have performed a limited number of such analyses in an attempt to replicate observations made by others. We did not confirm an early report by Greer et al6 that women with locoregional breast cancer who showed greater evidence of fighting spirit and denial (avoidance) coping during a structured interview lived longer at 5, 10, and 15 years after diagnosis than women who used these coping styles to a lesser extent. Nor did we confirm the finding of Hislop et al10 that low anger was an independent prognostic factor for survival, whereas low cognitive disturbance was associated with longer DDFS. We did not replicate the suggestion of Cassileth et al8 that breast cancer patients scoring in the middle third of the hopelessness scale had longer remissions. We also failed to replicate Watson's subgroup analyses showing an increased risk of death in breast cancer patients with marked depression or with high helplessness/hopelessness coping scores, although, as noted above, depression scores of the POMS may measure a somewhat different attribute than depression score of the HADS used by Watson et al.7 Some of the inconsistency in published results may be related to stage of disease; in the metastatic breast cancer setting, HRQOL14 and psychologic distress11 have been associated with poor outcomes. Recent advances in measuring HRQOL and psychosocial attributes have made it possible to include a spectrum of questionnaires and/or variables in a single study. However, the larger the number of variables, the greater the likelihood that chance associations will be falsely interpreted as significant. Incorrect conclusions can negatively impact patient care and/or misdirect research as other investigators attempt to replicate or build on the earlier chance findings. For example, if we had concluded that a true association of cognitive functioning (EORTC QLQ-C30) at 1 year with OS was present, women reporting high levels of cognitive functioning might be advised their risk of dying was doubled. Taken to the extreme, this might have led to the development of interventions to reduce cognitive functioning, clearly an ill-advised undertaking. We have discussed several approaches that can be used to reduce the likelihood that chance associations will be called significant. We also discussed the potential confounding effects of prior treatment on prognostic associations of HRQOL and psychosocial variables. We also recommend that unplanned, data-driven, subgroup analyses should be avoided. If performed, these analyses should be viewed as hypothesis generating. Replication of such analyses and most, if not all, statistically significant associations that are not felt to be due to chance would minimize false-positive conclusions. Caution should be exercised in recommending that all unplanned data-driven subgroup analyses yielding positive results be replicated. The majority will be chance findings, and scarce resources need not be allocated to their replication. Negative results, reflecting a failure to identify a hypothesized association, are as important as positive results. Although they may be more difficult to publish, true-negative results allow researchers to move on to new areas of inquiry that may lead to important advances. Apparently significant results should be interpreted cautiously when multiple variables and/or outcomes are studied, effect sizes are modest, and statistical significance borderline. The potential role of chance in such findings should be carefully scrutinized to avoid false-positive conclusions.
The authors indicated no potential conflicts of interest.
This research was funded by the Canadian Breast Cancer Research Alliance. Authors' disclosures of potential conflicts of interest are found at the end of this article.
1. Spiegel D: Psychosocial aspects of breast cancer treatment. Semin Oncol 24:S1-36-S1-47, 1997 2. Goodwin PJ: Psychosocial support for women with advanced breast cancer. Breast Cancer Res Treat 81: 5103-5110, 2003 3. Greer S: Mind-body research in psychooncology. Adv Mind Body Med 15: 236-244, 1999[Medline]
4. Andersen BL, Farrar WB, Golden-Kreutz D, et al: Stress and immune response after surgical treatment for regional breast cancer. J Natl Cancer Inst 90: 30-36, 1998
5. Turner-Cobb JM, Sephton SE, Koopman C, et al: Social support and salivary cortisol in women with metastatic breast cancer. Psychosom Med 62: 337-345, 2000 6. Greer S, Morris T, Pettingale KW: Psychological response to breast cancer: Effect on outcome. Lancet 2: 785-787, 1979[Medline] 7. Watson M, Haviland JS, Greer S, et al: Influence of psychological response on survival in breast cancer: A population-based cohort study. Lancet 354: 1331-1336, 1999[CrossRef][Medline] 8. Cassileth BR, Lusk EJ, Miller DS, et al: Psychosocial correlates of survival in advanced malignant disease? N Engl J Med 312: 1551-1555, 1985[Abstract]
9. Butow PN, Coates AS, Dunn SM: Psychosocial predictors of survival: Metastatic breast cancer. Ann Oncol 11: 469-474, 2000 10. Hislop TG, Waxler NE, Coldman AJ, et al: The prognostic significance of psychosocial factors in women with breast cancer. J Chron Dis 40: 729-735, 1987[CrossRef][Medline] 11. Derogatis LR, Abeloff MD, Melisaratos N: Psychological coping mechanisms and survival time in metastatic breast cancer. JAMA 242: 1504-1508, 1979[Abstract] 12. Neale AV, Tilley BC, Vernon SW: Marital status, delay in seeking treatment and survival from breast cancer. Soc Sci Med 23: 305-312, 1986 13. Waxler-Morrison N, Hislop TG, Mears B, et al: Effects of social relationships on survival for women with breast cancer: A prospective study. Soc Sci Med 33: 177-183, 1991 14. Goodwin JS, Hunt WC, Key CR, et al: The effect of marital status on stage, treatment and survival of cancer patients. JAMA 258: 3125-3130, 1987[Abstract] 15. Coates A, Gebski V, Signorini D, et al: Prognostic quality-of-life scores during chemotherapy for advanced breast cancer. J Clin Oncol 10: 1833-1838, 1992[Abstract]
16. Goodwin PJ, Ennis M, Pritchard KI, et al: Fasting insulin and outcome in early-stage breast cancer: Results of a prospective cohort study. J Clin Oncol 20: 42-51, 2002
17. Goodwin PJ, Ennis M, Pritchard KI, et al: Diet and breast cancer: Evidence that extremes in diet are associated with poor survival. J Clin Oncol 21: 2500-2507, 2003
18. Aaronson NK, Ahmedzai S, Bergman B, et al: The European Organization for Research and Treatment of Cancer QLQ-C30: A quality-of-life instrument for use in international clinical trials in oncology. J Natl Cancer Inst 85: 365-376, 1993 19. McNair DM, Lorr M, Droppleman LF: EdITS Manual for the Profile of Mood States (revised edition). San Diego, CA, Educational and Industrial Teaching Service, 1992
20. Horowitz M, Wilner N, Alvarez W: Impact of Event Scale: A measure of subjective distress. Psychosom Med 41: 209-218, 1979 21. Watson M, Greer S, Young J, et al: Development of a questionnaire measure of adjustment to cancer: The MAC Scale. Psychol Med 18: 203-209, 1988[Medline] 22. Watson M, Greer S: Development of a questionnaire measure of emotional control. J Psychosom Res 27: 299-305, 1983[CrossRef][Medline] 23. Derogatis LR: The psychosocial adjustment to illness scale (PAIS). J Psychosom Res 30: 77-91, 1986[CrossRef][Medline] 24. Ingelfinger JA, Mosteller F, Thibodeau LA, et al (eds): Biostatistics in Clinical Medicine (ed 2). New York, NY, Macmillan, 1987, p 161 Submitted December 13, 2003; accepted August 5, 2004. Related Correspondence
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 2004 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
TOP
ABSTRACT
INTRODUCTION
.05 (three fewer than expected by chance). Two were in the hypothesized direction of effect, and two were in the opposite direction. All arose from measurements 1 year after diagnosis, which were most susceptible to confounding by treatment. There was no evidence of consistency of associations across outcomes or questionnaires. These results are in keeping with chance as the explanation for our statistically significant findings. 
