During the past 10 to 15 years, we have experienced the phenomenal development of a new scientific technology that allows for the reliable and valid assessment of patient outcomes in research and clinical practice [1-3]. The technology of patient outcomes relies heavily on patient self-report of various aspects of health and well-being, as well as symptoms associated with disease. The tools for measuring these outcomes are derived from social science research and the field of psychometrics. The science of measuring quality of life and the criteria for evaluation of candidate tools have been described extensively by others [4-6].
In this paper, we assume that oncologists will soon have a choice of several validated general cancer and site-specific quality of life (QOL) instruments for use in their practices. This may allow future clinicians to order and utilize "quality of life tests" much as they would order a serum glucose test or an electrocardiogram. When this is the case, what will be the impact of quality of life outcomes on clinical practice?
Clinical practice and treatment decisions are influenced by a variety of factors based on the physician's training and clinical experience, synthesis of the relevant literature, and familiarity with consensus opinions developed by scientific bodies, treatment guidelines, and the results of randomized clinical trials. While clinical trials are considered the "gold standard" for the scientific evaluation of new treatments, relatively few therapies have been subjected to such rigorous evaluation. More important, even when a new therapy has been demonstrated to be efficacious, there may be very slow diffusion of the findings and limited adoption of the effective therapy. For example, two randomized trials comparing breast conservation therapy with mastectomy were published in a prestigious journal during the early 1980s [7,8]; however, several years after those results, there was only limited adoption of breast conservation, and extraordinary geographic variation in the use of this approach to the management of primary breast cancer [9,10].
If clinicians have so much difficulty changing their clinical practice in the face of data from well-controlled scientific studies that use familiar biomedical outcomes (eg, survival and recurrence), how can we expect them to accept without question the findings from quality of life studies, whose measures may be unfamiliar and whose outcomes are at times contrary to clinical opinion?
We have only recently begun to see the systematic incorporation of QOL outcomes in randomized clinical trials [11,12]. This process began over a decade ago in European and Canadian trials, and has been gradually introduced into US treatment trials during the past 5 years. Some pharmaceutical companies began to incorporate QOL measures into their studies because of their potential importance in the drug approval process . Many current phase III trials within the US cooperative groups include QOL assessment; however, it will be many years before these studies mature and the QOL data are examined. Therefore, at present, only a limited amount of QOL outcome information from randomized clinical trials is available for clinicians. Quality of life data from some phase III trials may be useful to clinicians in their practice, eg, the observation that patients being treated with chemotherapy for metastatic breast cancer had a better quality of life if treated on a continuous rather than intermittent schedule  (discussed below). As current phase III trials mature, this important source of QOL data will expand rapidly.
Much of the information that we have about QOL in cancer patients comes from well-designed research studies. Although not phase III randomized clinical trials, these studies have the advantage of being comprehensive and often use multiple tools for QOL assessment-something that is not feasible within the clinical trial setting. The major advantage of a research study is the adequacy of support staff for collection of QOL data, which usually results in less missing data. Limitations of research studies include the lack of treatment randomization and use of volunteer samples. Nevertheless, the high quality and comprehensiveness of data from research studies make it possible to draw important conclusions about the quality of life impact of some therapies.
For example, in a descriptive study of a community-based sample of women treated for early-stage breast cancer, Ganz et al  found no significant differences in the major dimensions of health-related quality of life between women choosing mastectomy and those choosing breast conservation (discussed below). The clinical implications of these findings are that we can describe these two surgical treatments for breast cancer in terms of their impact on survival and quality of life, and use this information to help patients make treatment decisions.
In situations in which there are no data from clinical trials or the research literature, clinicians must rely on the clinical interview. Unfortunately, few clinicians have been trained to incorporate questions about health-related quality of life into their usual patient interviews. Instead, physicians usually focus on physical symptoms and findings. Even under these circumstances, more subjective symptoms such as pain are often ignored. Clinical interviews are seldom multidimensional (unlike a QOL instrument) and are usually unsystematic with regard to psychosocial concerns [16,17]. The incorporation of several open-ended questions that address physical, emotional, and social functioning could enhance the standard clinical interview. Other tools, such as short screening instruments (eg, the Duke Health Profile  or the Dartmouth COOP charts ), could facilitate the clinical identification of QOL concerns of patients in the clinical setting. Unfortunately, most of these tools have been evaluated only in general medical settings and have not been tested with oncology patients.
Despite the expanding interest in measurement of QOL outcomes in cancer clinical trials, there are few data available to guide the clinician. Although many trials with QOL endpoints are in progress, these are fewer than might be hoped, due to the complexity of appending these studies to clinical trials and the limitation of resources available to conduct these additional studies . The National Cancer Institute of Canada has been an exception in requiring the use of QOL endpoints in all of its phase III clinical trials . Thus far, this policy has been implemented successfully and with high rates of compliance to data collection. As these trials mature, we can expect valuable information to emerge that will likely have an impact on clinical care.
As with treatment trials, QOL outcomes data may sometimes be counterintuitive (eg, lack of significant differences between mastectomy and breast conservation therapy), and thus, as with treatment trials, it is important to confirm QOL study results. Only the regular and systematic measurement of QOL in all phase III treatment trials will permit clinicians to determine which therapies have the best survival, response, and QOL outcomes.
Quality of life endpoints are particularly critical for the evaluation of therapies in advanced metastatic disease (eg, non-small-cell lung cancer), where survival gains of treatment are modest and treatment toxicity is considerable. However, a major limitation of QOL evaluation in lung cancer trials has been the high rate of missing QOL data due to death and deterioration [22,23]. There is keen interest in developing statistical methods for dealing with missing QOL data, typically from the sickest patients, in longitudinal studies .
There are other methodological problems that need to be addressed prior to the clinical application of QOL outcomes study results. Several popular generic and cancer-specific tools are now being widely used in clinical and research studies . Although statistical comparison tests can have adequate power to detect changes between groups of patients, the clinical meaning of individual changes in scores in the clinical setting is uncertain. For example, what does a 2-point change in physical functioning on a particular QOL instrument mean in an individual patient when assessed at two different points in time? That is to say, what is the clinical significance for the individual patient, and could this merely represent an error in measurement? Furthermore, will two different instruments that measure physical functioning show the same direction of change? These questions are not currently answerable for most of the commonly used QOL instruments. As noted by Guyatt et al, instruments that are reliable and valid in the cross-sectional evaluation of a sample may not be responsive to change over time .
Another parallel problem is the lack of head-to-head comparisons of different instruments simultaneously in the same patient population. This type of information would be useful to examine the relative strengths and limitations of some of the leading QOL tools, as well as to develop some form of common scoring system across several measures. This last question is being addressed by Cella in an ongoing research program (personal communication).
An example of the value of such comparisons comes from a recent study that evaluated long-term survivors of early-stage prostate cancer. Litwin et al  found that a generic health-related QOL tool did not discriminate between healthy controls and patients who received either surgery or radiation therapy. Two cancer-specific measures did a bit better, but it was only a very specific symptom measure that distinguished between the patients and normal healthy men . Thus, the sensitivity and specificity of psychometrically validated QOL tools may not be robust enough to detect subtle differences in functioning among patient groups. Furthermore, these tools have not been tested for monitoring the QOL of individual cancer patients in the clinical setting.