Prognostic Value of Quality-of-Life Scores During Chemotherapy for Advanced Breast Cancer By Alan Coates, Val Gebski, David Signorini, Paul Murray, Don McNeil, Michael Byrne, and John F. Forbes for the Australian New Zealand Breast Cancer Trials Group Purpose: We observed that quality-of-life (QL) scores, collected to evaluate treatment in a randomized trial in advanced breast cancer, predicted survival duration. This report explores the prognostic associations between QL and survival in more detail. Patients and Methods: In a randomized clinical trial comparing intermittent and continuous therapy policies for patients with advanced breast cancer, QL was measured by linear analog self-assessment (LASA) and the Quality-of-Life Index (QLI). Baseline scores and subsequent changes were included in statistical models of survival duration, with and without other prognostic factors. Results: Physician assessment of QLI and patient LASA scores for physical well-being (PWB), mood, nausea and vomiting, appetite, and overall QL (but not pain) at the commencement of treatment were significant predictors of subsequent survival. Scores for PWB and QLI were

independent of other prognostic factors. Changes in scores were also prognostically important. Both baseline and change in scores for PWB, mood, pain, and QLI after the first three treatment cycles, but before an arbitrary 180day time point, were significantly predictive of survival beyond that time. Both QLI and PWB were prognostically independent of tumor response. Although QL improvement was correlated with tumor response, continuous therapy yielded significantly better QL scores, even in nonresponders. Conclusion: These findings support the validity of the simple QL measures used in the trial. They are compatible with the simple explanation that patients perceive disease progression before it is clinically evident, but also with a causal relationship between QL and survival duration. J Clin Oncol 10: 1833-1838. © 1992 by American Society of Clinical Oncology.

IMPROVING quality of life (QL) is a major objective

after three cycles, and were recommenced on the same combination for a further three cycles when their disease progressed. This sequence was repeated until increasing disease occurred on treatment (Fig 1). Evaluation of response categories has been described.' Performance status was recorded using the Eastern Cooperative Oncology Group scale.

1 in the treatment of advanced breast cancer, and a number of methods have therefore been developed to allow reliable and valid measurement of aspects of QL affected by the disease and its treatment. 2-6 Such measures have usually been used to compare different treatments.1,4,7 As part of the validation of instruments used by the Australian New Zealand Breast Cancer Trials Group, it was noted that changes in some QL scores were significantly associated with prognosis.1,8 The present report analyzes these associations in greater detail, and confirms a strong predictive value for survival duration both of baseline QL scores obtained at study entry and of the subsequent changes in QL scores.

PATIENTS AND METHODS ClinicalTrial Details of the clinical trial from which the present studies derive have been presented elsewhere.' Briefly, 308 patients with measurable or assessable metastatic breast cancer were stratified by institution, performance status, and prior adjuvant chemotherapy, then randomized in a 2 x 2 factorial design to continuous or intermittent therapy, using one of two chemotherapy combinations: doxorubicin plus cyclophosphamide (DC) or cyclophosphamide, methotrexate, fluorouracil, and prednisone (CMFP), as previously described.' Informed consent was obtained according to standards required by research ethical committees of the participating institutions. Patients with objectively increasing disease during treatment went off study. Those without increasing disease at 18 months were observed without therapy. In the absence of increasing disease, patients assigned to intermittent therapy arms ceased treatment

Evaluation of QL Patients' self-evaluations of QL comprised five linear analog self-assessment (LASA) scales that measure physical well-being (PWB), mood, pain, nausea and vomiting, and appetite, 3,6 and a 5 single LASA scale that measures overall QL. The physician completed a Quality-of-Life Index (QLI) questionnaire. 5 The study schema and OL assessment schedule are shown in Fig 1. Comparisons of QL scores between treatment groups, based on changes in 1 scores within individual patients, have been presented elsewhere. The score used in analyses was obtained by measurement in millimeters from the good end of the line, so that scores ranged from 0 (good) to 100 (bad). Negative coefficients in survival regression therefore indicate that higher scores (worse QL) were associated with shorter survival duration. Scores for the QLI ranged from 5 (best) to 15 (worst). We did not measure QL after patients discontinued protocol treatment, so that all values re-

From the AustralianNew Zealand Breast Cancer Trials Group. Submitted September 23, 1991; acceptedJuly 9, 1992. Supported by grants from the National Health and Medical Research Council of Australia, the Royal Australasian College of Surgeons, and the Anti-CancerCouncil of Victoria. Address reprint requests to Alan Coates, MD, Department of Oncology, Royal Prince Alfred Hospital, Camperdown NSW 2050, Australia. © 1992 by American Society of Clinical Oncology. 0732-183X/92/1012-0003$3.0010

Journalof Clinical Oncology, Vol 10, No 12 (December), 1992: pp 1833-1838

Downloaded from ascopubs.org by University of Newcastle Upon Tyne on April 6, 2019 from 128.240.208.034 Copyright © 2019 American Society of Clinical Oncology. All rights reserved.

1833

1834

COATES ET AL Landmark PD

R -A N

CONTINUOUS

D O M

PD 3 CYCLE

•' 3 CYCLES

_

PD PD 3 CYCLE---

model that also contained the corresponding baseline score, but which was limited to patients for whom both a baseline and a change in the score were available. Finally, to determine whether the baseline and changes in QL scores were independent of other variables (for example, performance status), we analyzed models including all patient and disease variables that were significant in analysis of overall survival from randomization. All quoted P values are two-sided.

RESULTS

IL OL

QL

IL QL

QL

Fig 1. Study schema. Treatment was allocated at random in a 2 x 2 design that compared continuous therapy with an intermittent therapy policy in blocks of three cycles, and the combination of CMFP with DC. PD, progressive disease; QL, quality-of-life assessment. ported were obtained before ultimate disease progression, although some patients on the intermittent therapy treatment arms had QL assessments at the time of disease progression, before therapy was restarted. Prognostic Value of QL Scores The present analysis is concerned with associations between QL scores and survival duration. This work arose during exploration of the validity of the QL instruments used. Historically, the first approach was a landmark analysis in which the change in QL score in an individual patient was tested for its predictive value for subsequent survival in a proportional hazards model., 8 It became apparent that the proportional hazards model was not optimal for this analysis, as several prognostic factors significantly departed from the proportional hazards assumption. 9 Accordingly, an alternative statistical model was required that did not require the proportionality of hazards. We chose a multiple linear regression 0 model with censored response variable.1 Various transformations of survival duration were examined by the Shapiro-Wilk statistic" for normality. The cube root of survival duration proved to be approximately normally distributed, and this transformation of survival was therefore used as the dependent variable. Multiple regression of this type allows assessment of the independent prognostic information contributed by each indicator, while allowing for other indicators included in the regression model. Next, baseline QL scores obtained at the time of randomization were used as predictors of overall survival from randomization. Each QL variable was tested singly, then in a multivariate censored linear regression model that made simultaneous allowance for other non-QL prognostic factors, and selected the best set of patient and disease variables established by exhaustive search12 using the SPIDA (Statistical Package for Interactive Data Analysis) software package.' 3 The two remaining significant QL scores (for PWB and QLI) were then included together, to test whether they contributed independent prognostic information. A further landmark analysis was then performed using censored linear regression techniques as above to examine survival beyond the 180-day landmark, testing the change in each QL score from a value obtained at 60 to 120 days after randomization (which was the time the treatment strategies diverged'), to the average value observed in the period 120 to 180 days from randomization. Again, the change in each QL variable was examined first alone, then in a

Forty-three patients from one institution were excluded entirely from the present analysis, since most QL forms from that institution were not completed. Institution was a stratification variable, so this exclusion should not lead to bias in the remaining data. Of the remaining 262 patients, 226 (86%) completed baseline selfassessment, and 240 (92%) had a baseline OLI completed by their physician. Patient QL measures after three cycles of chemotherapy were available as 168 forms from 146 patients, and physician QLI data at this time were based on 178 forms from 154 patients. Of these, at least one form was completed between 120 and 180 days after randomization by 66 patients, 65 of whom survived for more than 180 days to contribute to the landmark analysis. Similar analysis of the physician QLI was based on 70 patients, of whom 68 survived beyond 180 days. A single average value was calculated if multiple forms for a patient were available at either time period. Patients for whom QL data were delinquent did not differ from the rest of the group in terms of survival duration. Follow-up of the study is virtually complete, with only two of 243 patients involved in these analyses alive at last follow-up. QL Scores at Randomization All baseline QL scores, except those for pain, were significant predictors of overall survival in univariate analyses (Table la). Survival curves divided by baseline scores for PWB, the QLI, and performance status are depicted in Fig 2. In each case, the group was divided into three approximately equal sets, and the differences in survival were all highly significant (Fig 2). We next looked to see whether the QL scores gave additional information independent of standard prognostic criteria. The best subset of patient and disease characteristics for predicting survival was as previously published': performance status and liver, brain, and node metastases were adverse prognostic factors for survival. As previously reported 8 with a longer follow-up than that of our original publication,' intermittent therapy assignment in the present analyses was also significantly associated with shorter survival duration (Table

Downloaded from ascopubs.org by University of Newcastle Upon Tyne on April 6, 2019 from 128.240.208.034 Copyright © 2019 American Society of Clinical Oncology. All rights reserved.

1835

QUALITY-OF-LIFE SCORES PREDICT SURVIVAL Table 1. Prognostic Significance of QL Scores at Randomization Coefficient

SE

(a) Univariate analyses of QL scores* PWB -0.019 0.004 Mood -0.014 0.005 Pain -0.004 0.004 N&V -0.019 0.006 Appetite -0.014 0.004 Uniscale -0.014 0.005 QLI -0.263 0.053 (b) Significant other prognostic factors Liver metastases -1.137 0.244 PS 0,1 v2,3,4 -0.995 0.232 Brain metastases -1.172 0.491 Node metastases -0.652 0.239 Intermittent tmt -0.545 0.227 (c) QLI added to model (b) QLI -0.169 0.071 Liver metastases -1.149 0.258 PS 0,1 v 2,3,4 -0.323 0.339 Brain metastases -0.933 0.541 Node metastases -0.602 0.255 Intermittent tmt -0.564 0.240 (d) PWB score added to model (b) PWB -0.012 0.005 Liver metastases -1.113 0.278 PS 0,1 v 2,3,4 -0.570 0.302 Brain metastases -1.293 0.621 Node metastases -0.579 0.275 Intermittent tint -0.468 0.257 (e) QLI and PWB score added to model (b) QLI -0.144 0.064 PWB -0.011 0.005 Liver metastases -1.107 0.269 Brain metastases -1.185 0.580 Node metastases -0.528 0.265 Intermittent tmt -0.520 0.250

P

Prognostic value of quality-of-life scores during chemotherapy for advanced breast cancer. Australian New Zealand Breast Cancer Trials Group.

We observed that quality-of-life (QL) scores, collected to evaluate treatment in a randomized trial in advanced breast cancer, predicted survival dura...
500KB Sizes 0 Downloads 0 Views