The most crude measure of health outcome is vital status-alive or dead. For many diseases, such as cancer, curative treatment is not always possible. Still, the outcomes of noncurative care may be very different. For instance, therapeutic strategies may be associated with similar survival but different toxicities; alternatively, one therapy may yield better survival but more severe side effects, while another treatment may offer poorer survival but better quality of life during the patient's remaining months or years. Thus, decisions about alternative therapies are often based on quality of life considerations, in addition to the likelihood of survival [1,2]. Even when cures are possible, the costs of treatment may exceed an individual's or society's willingness to pay.
Outcomes research is the study of the net effects of the health care process on the health and well-being of individuals and populations. As such, it encompasses a wide breadth of issues, including research on practice patterns, effectiveness, appropriateness of care, and measurement of patient preferences and health status. Thus, outcomes research can best be seen as a discipline that endeavors to ask which treatment works best, under which circumstances, for which individuals, and at what cost.
The growing interest in the costs and quality of health care during the past decade has contributed to the dramatic growth of the field of outcomes research . Clinicians, consumers, employers, insurance companies, and government agencies are increasingly looking to the results of outcomes research to assist them in determining how to get the most value for their health care efforts. For comprehensive reviews of outcomes research, the interested reader is referred to several excellent sources [4-7].
This paper concentrates on the measurement of health status and preferences for oncology-related health outcomes, broadly referred to as "quality of life" (QOL). Following a summary of the historical evolution of QOL measurement in cancer research, we will de scribe the tools commonly used in the measurement of quality of life and the major methodological issues that investigators or consumers of QOL research should consider. Measurement of patient and population preferences for health outcomes and incorporation of these preferences into economic evaluations of cancer interventions are also highlighted. We conclude with recommendations for future cancer outcomes research.
Evolution of Cancer Outcome Measures
The most general disease outcome measure is death. For cancer, 5-year survival or the interval of disease-free survival has customarily been used to evaluate the success of treatment. Clinical events, such as severity of illness, tumor response, or stage shifts, have served as intermediate measures of outcome, principally because they are believed to be associated with differences in survival. For example, in the landmark Health Insurance Plan (HIP) mammography screening trial, downstaging of breast cancer was initially used to assess effectiveness [8,9]. Proxy measures of intermediate clinical outcomes have included events such as number of consultations, days in the intensive care unit, and need for blood transfusions .
While quality of life has been an implied medical outcome since the time of Hippocrates [3,11], the landmark paper by Karnofsky marked the first explicit effort to systematically assess the impact of cancer treatment on the patient's quality, and not quantity, of life. The Karnofsky performance index uses a scale from 0% to 100% for physician rating of functional status, with 100% reflecting ability to carry on normal activities and 0% representing death. The index was originally used in conjunction with subjective symptoms and tumor response to evaluate the use of nitrogen mustards as palliative cancer therapy . Although the Karnofsky scale was a seminal contribution, it does not meet today's standards for validation and demonstration of instrument reliability (see "Measurement and Analysis Issues" section) .
The next major tools designed to assess the impact of cancer therapy on quality of life were not developed until the early 1980s. In the prototype, the Spitzer Quality of Life Index, QOL continued to be physician-rated . Thus, physician or researcher assessment of symptoms, toxicities, and/or quality of life remained the standard in cancer outcomes research for almost 4 decades.
Increasing consumerism and patient participation in health care decisions, occurring in parallel with the growth in interest in outcomes of care in the late 1970s and 1980s, set the stage for the development of patient-based measures of general and cancer-related quality of life [3,14]. For example, in 1976, Ware and colleagues  presented results validating a patient self-reported measure of general health status from the Rand Health Insurance Experiment. Other patient-rated general health QOL measures developed in this period include the Sickness Impact Profile (SIP) [16,17], Psychosocial Adjustment to Illness Scale (PAIS) [18,19], and Nottingham Health Profile .
In parallel with the development of these patient-rated general health QOL measures, patient assessment tools for measuring the quality of life of cancer patients began to appear during the past 20 years; preliminary incorporation of such measures into cooperative group, randomized, controlled trials occurred in the late 1980s [3,21]. Examples of the instruments that have been used (and will be discussed below) include the LASA-P , Padilla Quality of Life Index , EORTC-QOL [24,25], CARES [26,27], FLIC , and, most recently, the Functional Assessment of Cancer Therapy (FACT) . Cancer site-specific tools have also been developed in this same time period.
Despite this explosion of QOL measures, incorporation of QOL outcomes into medical research has been slow, and, when used, outcomes have often been poorly measured. For instance, in a recent review of QOL measurement across a variety of medical conditions, Gill and Feinstein  found that 159 different measures of quality of life were used in the 75 articles reviewed. Despite the large number of measures, fewer than half defined the target domains, only 17% included a patient rating of quality of life, and just 9% elicited patients' preferences for health outcomes. In phase III breast cancer randomized controlled trials, as another example, only 4% of trials published from 1985 to 1989 and 6% published from 1990 to 1994 included any QOL assessment; only one study included a measure of patient preference .