|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
© 1999 American Society for Clinical Oncology
Science, Language, Intuition, and the Many Meanings of Quality of LifeMcMaster University, The Program in Evidence-Based Care, Cancer Care Ontario, Hamilton, Ontario, Canada "I BELIEVE THAT people believe what they believe they believe."—Ogden Nash. A patient I know demanded during a discussion of the plans for her chemotherapy that her quality of life (QOL) be considered a high priority. She then added, "And I don't mean a questionnaire." This patient has tacit knowledge about what QOL is, or at least what it means to her. In a treatise on the subject, Michael Polanyi introduced the topic of tacit knowing with the principle that "we can know more than we can tell."1 QOL researchers are interested in being able to tell what we think we know. Because QOL is such an abstract construct, the ability to define it explicitly requires meticulous attention to scientific principles so that we can be confident we are measuring what we think we know. This is done through the use of instruments (ie, questionnaires) cleverly designed and rigorously tested to probe the various dimensions now acknowledged to reflect adequately the construct of health-related QOL.2-6 Thus, to the researcher, QOL takes on an operational definition based on the scores arising from valid instruments. But, in their own way, researchers cannot help but also have some sort of tacit knowledge of QOL, which may sometimes intervene as they interpret the findings of their research. This is reflected in the intuitive judgments and inferences they make about their research findings. The clinician also has a stake in QOL research and its applications. A recent survey of academically based Canadian oncologists found that the vast majority identified QOL as more appropriate than survival as an outcome measure for most randomized controlled trials in cancer.7 The same proportion (approximately 75%) indicated they would be willing to use published QOL results to change their practice patterns.7 Despite this, more than half the respondents in the survey exhibited their unawareness of a fundamental principle of QOL measurement. They identified nurses, rather than patients, as the most reliable source of such information. A recently published (1998) exhaustive review (106 citations) of drug therapy in the treatment of breast cancer revealed the barriers still faced in incorporating QOL considerations into our practices.8 QOL information was virtually ignored in favor of the more traditional outcomes of tumor response and survival for informing treatment recommendations by the author. What are the implications of such observations for QOL measurement and use in oncology practice? They suggest that the same term is understood in many different ways by the key participants affected by this field, and perhaps in different ways by the same participant at different times. Perhaps this is because the language of the science of QOL measurement is so accessible to the lay person in all of us. To the patient QOL is an implicit state of being (something known that cannot be told), to the researcher it is a difficult measurement problem, to the biostatistician, an interesting analytical problem, and to the clinician, just one of many other equally relevant inputs into a clinical judgment. For QOL to become a principal component of clinical oncology research and practice, we must become more explicit about what we expect to do with the information we obtain about an individual's QOL.9 We also need to ask what additional knowledge QOL measures give us that we could not otherwise have. The different perspectives suggest that when discussing QOL, we're reading from different texts. In this issue of the Journal of Clinical Oncology are two articles about formal QOL assessments in the context of randomized clinical trials (RCT). The trials tested palliative treatment alternatives for hormone-resistant metastatic disease (prostate cancer and breast cancer), where treatment is known not to improve survival.10,11 These articles provide a useful backdrop to highlight some of the challenges we face in reconciling the scientific and intuitive perspectives of QOL assessment to ensure that QOL research is taken seriously and used with confidence. They demonstrate how scientific rigor may be interpreted as an impediment to making the "obvious" (ie, intuitive) inferences the data seem to be telling us. The most valid application of QOL data within the context of an RCT is when they are used to compare the effects of the randomized interventions. Here, average QOL scores for the randomized groups are obtained, and because of the randomization, the assumptions of the formal statistical comparisons are not violated. Patients and providers can use the scores to make inferences about which treatment, on average, is likely to be associated with better QOL, in the same way that collective efficacy data related to tumor response and survival are used. Valid QOL scores can also be used to follow changes over time in the average QOL of a similarly treated group of patients, but the usefulness of an individual patient's QOL score as a valid input to clinical decisions has been hotly debated.12 The articles by Osoba et al10 and Bernhard et al11 present the initial QOL findings in the context of comparisons between randomized groups receiving different treatments. However, the authors come up against serious methodologic challenges related to the problem of dropouts over time. Bias lurks behind the analyses because of selection factors, a problem acknowledged by the authors of both articles. But, they soldier on. Each of the articles goes on to apply the QOL data in different ways. Osoba et al10 attempt to overcome the problem of selection bias caused by uneven dropouts in longitudinal data by examining change scores within the treatment groups. They are able to demonstrate that the combination of prednisone and mitoxantrone improves many of the domains of QOL more than prednisone alone, and that the duration of benefit is longer for the combination. Of course, any comparisons between groups regarding change scores at this time are also subject to bias because of selection. Furthermore, the use of significance levels to make inferences about which treatment is better based on within-group comparisons is flawed because of the wide difference in attrition rates, which will tend to allow significant differences to be more easily detected in the larger remaining group. Finally, failure to correct for multiple testing (two treatments, several items within several domains, and several time points within each of the items and domains) renders the findings suspicious from the biostatistician's and the research methodologist's perspectives. However, because the findings resonate with our intuitions and because we want to do the best we can with the limited data at hand, we may be less critical of such scientific violations, or more accepting of the inherent flaws of the analyses. Osoba et al recognize the limitations of their analysis, attempt some compensating but suboptimal analytic solutions, and try to make their inferences despite the difficulties. At least the authors restrict their inferences to comparisons involving the randomized groups. Bernhard et al11 go further. They attempt to examine differences in QOL domain scores across subgroups that were not initially randomized. They analyze the pooled data from the randomized groups for QOL as a function of the response level achieved by patients (responders, stable disease, and progressive disease). Like Osoba et al,10 they engage in unadjusted multiple significance testing that includes collapsing some response categories into more than one analytical comparison (eg, responders v stable disease; stable disease grouped with responders v progressive disease; and stable disease grouped with progressive disease v responders). Furthermore, all of these comparisons are done across several domains independently and over multiple time points independently without any adjustments. However, the findings of the data are consistent across analyses and once again resonate with human intuitive logic. Bernhard et al11 report an association between QOL scores and response category achieved. This is important information because of the known lack of difference in survival between some response categories (responders and stable disease). The authors also claim that baseline QOL predicts for subsequent scores (which may simply reflect a separate bias). Of great concern in this study is the important difference in the proportion of patients at baseline with bone metastases favoring responders (42% responders v 71% progressive disease and 72% stable disease), which could account for many of the other associations. Even when not statistically significant, unequal distribution of a powerful prognostic factor at baseline between groups can have a profound effect on the outcomes of interest.13 The statistical difficulties faced by these authors threaten the validity of the findings they present. However, given the nature of the data available, and the limited solutions to these theoretical problems, what are the alternatives? When is a problem such as unadjusted multiple statistical testing just a bookish retreat to scientific prudishness that gets in the way of progress, and when is it a fundamental challenge to validity that threatens the foundation of the knowledge we think we have? How should we interpret the different solutions to the statistical problem of missing data that plagued Osoba et al,10 who reverted to intuitive solutions because of the drawbacks they perceived with other approaches? There may be empirical answers to these questions. In the meantime, researchers, as they always have, seem prepared to allow their own form of tacit knowledge to influence the way in which their scientific data are interpreted and presented. How can we advance the state of a difficult science toward clinical application when the same words used by scientists, clinicians, and patients mean such different things to each of them? How can we overcome strong intuitive beliefs to make a compelling case for the importance of scientific rigor in QOL research? And should we? One promising approach comes from a recent review by Osoba14 entitled "Lessons learned from measuring health-related quality of life in oncology." One of Osoba's observations was that QOL research can occasionally produce findings that are unexpected or counterintuitive. The superior QOL findings for mitoxantrone plus prednisone versus prednisone alone in this issue of the Journal of Clinical Oncology provide evidence that aggressive therapy may result in improved QOL. Other examples are provided.14 The finding that individuals' perceptions of a health state may actually change when they enter that state is intuitive in retrospect, but might not have been before the data were generated.15 Provided that the scientific methods underpinning such findings can hold up to scrutiny, such counterintuitive examples may help convince those with strong belief systems that their tacit knowledge needs to be harmonized with empirical, rigorous research. Perhaps then QOL research will provide the new knowledge we need that otherwise could not be predicted from other approaches. REFERENCES 1. Polanyi M: The Tacit Dimension. Garden City, NY, Doubleday & Co, Inc, 1966 2. Osoba D (ed): Effect of Cancer on Quality of Life. Boca Raton, FL, CRC Press Inc, 1991 3. Quality of life in cancer clinical trials. J Natl Cancer Inst Monogr 20:1996
4.
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 trials in oncology. J Natl Cancer Inst 85:365-376, 1993 5. Ganz PA: Quality of life measures in cancer chemotherapy: Methodology and implication. Pharmacoeconomics 5:376-388, 1994[Medline] 6. Cella DF: Methods and problems in measuring quality of life. Support Care Cancer 3:11-22, 1995[Medline] 7. Bezjak A, Taylor KM, Ng P, et al: Quality-of-life information and clinical practice: The oncologist's perspective. Cancer Prev Control 2:230-235, 1998[Medline]
8.
Hortobagyi GN: Drug therapy: Treatment of breast cancer. N Engl J Med 339:974-84, 1998 9. Consensus Development Conference: Assessment of the quality of life in cancer clinical trials—Italian Psycho-Oncology Society (SIPO). Tumori 78:151-154, 1992[Medline]
10.
Osoba D, Tannock IF, Ernst DS, et al: Health-related quality of life in men with metastatic prostate cancer treated with prednisone alone or mitoxantrone and prednisone. J Clin Oncol 17:1654-1663, 1999
11.
Bernhard J, Thürlimann B, Hsu Schmitz S-F, et al: Defining clinical benefit in postmenopausal patients with breast cancer under second-line endocrine treatment: Does quality of life matter? J Clin Oncol 17:1672-1679, 1999 12. Sutherland HJ, Till JE: Quality of life assessments and the levels of decision making: Differentiating objectives. Quality Life Res 2:297-303, 1993 13. Altmann DG, Dore CJ: Randomization and baseline comparisons in clinical trials. Lancet 335:149-153, 1990[Medline] 14. Osoba D: Lessons learned from measuring health-related quality of life in oncology. J Clin Oncol 12:608-616, 1994[Abstract] 15. Llewellyn-Thomas HA, Sutherland HJ, Thiel EC: Do patients' evaluations of a future health state change when they actually enter that health state? Med Care 31:1002-1012, 1993[Medline]
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 1999 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
