In recent years, there has been a growing acceptance of the view that the goals of cancer treatment should include concerns about quality of life (QOL) as well as length of life. Patients with cancer experience a variety of symptoms due to their disease and its treatment, such as pain, fatigue, and nausea, that can have a significant negative impact on their well-being and functioning. The development of multidimensional self-report QOL instruments has allowed investigators to measure the adverse impact of disease and its treatment on well-being and functioning and evaluate the efficacy of interventions designed to prevent or treat these adverse effects. Findings from QOL research suggest that routine use of QOL instruments as part of clinical practice has the potential to improve the quality of care that patients receive as well as their health status. However, in addition to its many benefits, there are also many challenges to assessing quality of life in research and clinical practice.
Two features characterize most forms of QOL assessment currently used in oncology. First, it is generally recognized that quality of life is a multidimensional construct and is best measured using instruments that tap multiple domains of functioning and well-being.[1-3] Consistent with this view, most QOL instruments measure physical, social, and emotional aspects of functioning, as well as common symptoms of cancer and its treatment. Second, there is general agreement that quality of life is a subjective phenomenon and that patients are the best judges of their own quality of life.[1,2] Indeed, studies have shown that considerable disparities exist between concurrent ratings of quality of life made by patients and their physicians.[4,5] Accordingly, assessment of quality of life in oncology trials is typically performed using patient self-report questionnaires.
Two of the most widely used multidimensional QOL instruments in oncology are the General Version of the Functional Assessment of Cancer Therapy (FACT-G)  and the European Organization for Research and Treatment of Cancer Quality of Life Questionnaire (EORTC-QLQC30). (For a comprehensive list of these and other QOL scales discussed in this article, see the Appendix at the end of this supplement.)
The FACT-G (version 4) is a 27-item measure. For each item, respondents indicate on a 5-point rating scale (0 = not at all; 4 = very much) how true each statement (for example, "I have a lack of energy") has been for them during the past 7 days. The FACT-G yields a total score for overall quality of life as well as subscale scores for physical well-being, social/family well-being, emotional well-being, and functional well-being.
The EORTC QLQ-C30 is a 30-item measure. For each item, respondents indicate the rating that best applies to them. Seven items are rated yes or no for an unspecified time frame (eg, "Do you have any trouble taking a long walk?"); 21 items are rated on a 4-point scale (1 = not at all; 4 = very much) for the past week (eg, "Were you tired?"); and 2 items are rated on the 7-point scale (1 = very poor; 7 = excellent) for the past week (eg, "How would you rate your overall quality of life?"). The EORTC QLQ-C30 yields scores for five functional scales (physical, role cognitive, social, and emotional), three symptom scales (nausea, pain, and fatigue), and a global health and QOL scale. The measure also yields single-item ratings of additional symptoms commonly reported by cancer patients (dyspnea, appetite loss, sleep disturbance, constipation, and diarrhea) as well as the perceived financial impact of disease and its treatment.
Both the FACT-G and the EORTC QLQ-C30 have been shown to have adequate validity and reliability and to be able to distinguish patients according to their performance status.[6,7] A number of disease- specific modules (eg, breast, lung, and prostate) have been developed to supplement each of these core measures. These modules assess additional symptoms and QOL issues that are relatively specific to certain forms of cancer.
Linear Analogue Self-Assessment (LASA) scales are also widely used in QOL research in oncology. A LASA scale consists of a 100-millimeter line with descriptors at each end. Respondents mark their current status somewhere along the line, and then the distance in millimeters from the lower end point (0 point) is measured to obtain their scores. LASA scales have been developed to measure a variety of symptoms (eg, pain) and aspects of functioning (eg, physical activity), as well as overall quality of life.
These measures are popular, in part, because they are relatively easy and quick to administer. Moreover, there is evidence to suggest that many LASA scales compare favorably with more established QOL measures in terms of both validity and ability to detect changes over time. Although the use of LASA scales is appealing, caution is advised. Investigators need to determine whether the specific set of LASA scales to be administered has been validated for its intended use. In the absence of existing validity data, LASA scales should be used in combination with the more established FACT-G and EORTC-QLQ-C30 measures. (For a discussion of the use of these and other instruments in the assessment of anemia and anemia thearpy, see Dr. David Cella’s article in this supplement.)
Perhaps the most important methodologic issue to consider in evaluating QOL end points in an oncology clinical trial is the selection of appropriate outcome measures. In most instances, the use of a well-validated multidimensional self-report QOL instrument (eg, FACT-G, EORTC QLQ-C30) will meet this requirement. Depending on the nature of the trial, it may be necessary to supplement these core measures with additional measures that provide more information about those symptoms that are most relevant to the patient population under study. For example, the lung subscale for Functional Assessment of Cancer Therapy (FACT-L)  includes several items assessing respiratory difficulties. Likewise, in trials where relief of pain is a primary goal, it may be useful to collect additional information about the subjective experience of pain using a LASA scale or a measure such as the Brief Pain Inventory.
Number and Timing of Assessments
A second important issue to consider is the number and timing of QOL assessments. The desire to collect self-reported information at relatively brief intervals in order to increase the likelihood of detecting changes over time must be weighed against concerns about the burden to patients and the financial cost of conducting frequent assessments. Osoba has proposed a set of guidelines that may be useful in determining the timing of QOL assessments in oncology clinical trials. A baseline QOL assessment carried out before the initiation of treatment can be considered necessary for two reasons:
First, in randomized trials, the baseline assessment will indicate whether there are preexisting differences in quality of life between patients in the various treatment arms; if present, these differences would need to be adjusted for statistically in order to accurately determine treatment effects.
Second, the baseline assessment conducted prior to intervention provides an essential point of reference for identifying changes over time that may be attributable to the treatment under investigation.
In most instances, one or more on-treatment assessments are also necessary. As noted by Osoba, the frequency and timing of these assessments will depend on the research question(s) being asked. If, for example, the goal is to determine whether chemotherapy improves quality of life in patients experiencing disease-related symptoms (eg, pain), on-treatment assessments should be conducted just before the start of subsequent chemotherapy cycles to reduce the likelihood that results will reflect short-term treatment side effects. In instances where multiple chemotherapy cycles are being administered, the nature of the research question being asked and the financial costs of data collection will determine whether on-treatment assessments are conducted after each cycle or at less frequent intervals.
Finally, there is the issue of off-treatment assessmentsthose conducted following the completion or cessation of treatment. Once again, the nature of the research question and issues of cost will be the primary factors determining the number and timing of these assessments. In studies of patients with advanced disease (and a poor prognosis for survival), it may be both desirable and feasible to follow patients until disease progression occurs or even until death. Data collected during the off-treatment period would indicate if and for how long any of the observed on-treatment benefits to quality of life may have persisted.
Handling Missing Data
A third important methodologic issue to consider is the handling of missing data. This issue is of particular relevance to studies of quality of life end points. As Moinpour has noted, "In the very setting where quality of life questions are most compelling, they are the most difficult to evaluate because the missing data mechanism is often dependent on the very outcome being assessedthe health status and quality of life of the patient." That is, patients who are experiencing negative health outcomes, such as treatment toxicity or progressive disease, are also most likely to have missing QOL data. Under these circumstances, analyses based only on available (nonmissing) data may lead to erroneous conclusions. For example, if QOL data are missing on a consistent basis due to treatment toxicity, the analysis of only nonmissing data is likely to lead to an overestimate of the actual QOL benefits of the agent under study.
At present, there is no consensus on the optimal method for dealing with nonrandom missing QOL data in clinical trials. As a general strategy, Fairclough and colleagues suggest that two questions be considered in attempting to evaluate the impact of missing data. First, why are the data missing? If data are missing for reasons related to treatment toxicity or disease progression, then the missing data mechanism is "nonignorable" and statistical models appropriate for this situation should be explored. Second, how sensitive are the study results to different assumptions about the missing data mechanism? In the absence of a consensus on the "best" approach, sensitivity analyses are recommended to examine the effects of several different methods of handling missing data. Readers interested in learning about these methods may wish to consult a special issue of Statistics in Medicine (volume 17, numbers 5-7, 1998) devoted specifically to the topic of missing QOL data in oncology clinical trials.