Disease Management & Health Outcomes

, Volume 11, Issue 11, pp 709–721

Health Outcomes Assessment in Cancer

Current Measurement Strategies and Recommendations for Improvement


    • Center for Outcomes and Effectiveness ResearchAgency for Healthcare Research and Quality
  • Carolyn M. Clancy
    • Center for Outcomes and Effectiveness ResearchAgency for Healthcare Research and Quality
Review Article

DOI: 10.2165/00115677-200311110-00003

Cite this article as:
Lawrence, W.F. & Clancy, C.M. Dis-Manage-Health-Outcomes (2003) 11: 709. doi:10.2165/00115677-200311110-00003


Measuring the outcomes of cancer care has become increasingly important both in clinical practice and in health policy. Responsiveness to patient-centered needs, preferences, and outcomes is one of the hallmarks of quality healthcare.

Health-related quality of life (HR-QOL) measures can be considered within a framework based upon: (i) whether the measure is a generic instrument applicable across a wide range of health conditions, or whether it is specific to cancer or a specific cancer site; (ii) whether it measures a single domain of health or multiple domains; and (iii) whether or not the measure is preference based. Judicious selection of a set of instruments from within different areas of this framework can provide a detailed description of relevant aspects of a patient’s health for a wide variety of research and clinical needs.

Current health outcomes research is focused not only on the development of improved measures of health, but also on how to expand the use of these measures from research settings into clinical practice and health policy in ways to improve the process and outcomes of cancer care. Shared decision-making tools incorporating HR-QOL data can assist patients in clarifying decision alternatives for difficult cancer treatment decisions. Observational studies of HR-QOL of cancer patients can help patients better understand potential outcomes of their choices. HR-QOL measures are being used in quality of care initiatives.

Cancer care is composed of a spectrum of services, ranging from prevention and early detection, through to diagnosis and treatment, as well as end-of-life care. As the importance of the patient’s perspective has become more clearly recognized, health outcomes measures have become more widely used and can contribute to improved care across the spectrum of cancer services. While further research needs to focus on developing better measures of health, it is equally imperative that future research focus on methods to incorporate health outcomes measurement into practice in ways to improve clinical practice, health policy, and ultimately to improve the outcomes of care of patients with cancer.

Cancer care is fraught with difficult decisions. Advances in treatments and palliation have led patients to confront many decisions between care alternatives without a clear answer on which choice is best for them. The clinical literature exemplifies the choices that patients must make. For example, two recent studies[1,2] confirmed that women with early-stage breast cancer treated with lumpectomy and radiation therapy had similar overall survival rates to women treated with a radical mastectomy. However, women with the more invasive surgery may be less likely to experience a recurrence. Thus, a woman deciding on treatment for her early-stage cancer must decide between the adverse impacts of more invasive up-front therapy and the possibility of future recurrence. Likewise, recent research suggests that radical prostatectomy for men with localized prostate cancer, compared with expectant management, reduced disease-specific mortality, but did not significantly impact on overall mortality[3] or overall quality of life (QOL).[4] Thus, a man with localized prostate cancer needs to estimate the benefit of an invasive therapeutic procedure and predictable adverse effects[5] and a possible increased probability of mortality. These trial results, and a lack of demonstrated mortality benefit for prostate cancer screening,[6] may need to be considered in a man’s decision of whether or not to undergo prostate cancer screening.

Health-related quality of life (HR-QOL) measurement must reflect the patient’s subjective experiences with health.[7] An interest in measuring HR-QOL and other patient-centered outcomes has increased in research settings. This interest is evidenced by the growth in numbers of cancer-related articles dealing with HR-QOL, and by the increasing number of cooperative oncology groups that have QOL committees[810] and that are incorporating patient-reported outcomes into clinical trials.[11]

Measuring the outcomes of care has also become increasingly important in clinical practice and in health policy.[1214] The evidence that links the process of care with the outcomes or end results is an integral component of quality of care. The Institute of Medicine in the US has defined quality of care as “the degree to which health services for individuals and populations increase the likelihood of desired health outcomes and are consistent with current professional knowledge”.[15] Thus, understanding these outcomes can help to determine which services are worth providing, where more evidence is needed to demonstrate that an intervention is effective, and where clinicians and health systems have opportunities for improvement.[12] Responsiveness to patient-centered needs, preferences, and outcomes is one of the six hallmarks of quality of care.[16] This quality criterion recognizes that care providers need to be responsive to the patient, and that the care provided needs to address outcomes that are important to the patient. As new funding in clinical research returns new choices in diagnostic and therapeutic interventions, and as the population ages, the need to understand the impact of care processes on the end results of care will increase.

This article focuses on the science of measuring patient-centered outcomes in the cancer field. The term outcomes assessment can be broadly interpreted.[14] The goal of this paper is to provide an overview of methods for the measurement of health and well-being from a patient’s perspective. To accomplish this goal, we provide a conceptual framework for understanding measures, give examples within this framework of measures used in cancer outcomes assessment, and finally make recommendations for measurement strategies and future research. Furthering the understanding of a patient’s perspectives of HR-QOL is essential to improving healthcare decision making and health outcomes across the spectrum of cancer care.

1. A Framework for Health Outcomes Measurement

The primary goal of the healthcare system is to maintain and improve health. The WHO defines health as “a state of complete physical, mental, and social well-being and not merely the absence of disease or infirmity”.[17] Other sources, such as Healthy People 2010 in the US[18] and the Ottawa Charter for Health Promotion (adopted at the First International Conference on Health Promotion held in November 1986, Ottawa, Ontario, Canada),[19] also define health as a positive construct, or the ability to function across a variety of health domains, and not just the lack of a negative, e.g. the absence of specific conditions such as cancer. Particularly since cancer has increasingly shifted from being considered an acute condition rather than a chronic one,[20] more emphasis in the caring of patients with cancer needs to be placed on maintaining health as an outcome rather than solely trying to reduce tumor burden.

Health can be considered as having two main attributes: longevity and HR-QOL.[21] Healthcare interventions may involve trade-offs between these two attributes, and individuals may be willing to trade longevity to maintain HR-QOL.[22,23] Longevity (e.g. as measured by 5-year survival rates) and clinical outcomes (e.g. tumor progression) have long been considered in outcomes of cancer care. In 1948, Karnofsky and colleagues[24] examined the impact of therapy on not only clinical outcomes but also on patients’ abilities to perform their usual activities. Since this time, the measurement of functional status and HR-QOL in cancer care has expanded tremendously both in scope and in complexity. Several researchers[25,26] have proposed a hierarchical framework of outcomes, starting with biologic and physiologic measures as the most basic. These outcomes influence specific symptoms, which may influence a person’s function in one or several domains considered to be important to HR-QOL. Table I lists some of the domains commonly thought to be important to HR-QOL; functioning in these domains influences overall HR-QOL. The list in table I is by no means universally considered comprehensive; for example, some investigators propose spirituality as a domain of HR-QOL.[27] We use the terms health status and functional status to refer to a person’s level of ability or functioning on the various domains of health, and HR-QOL to include these terms as well as judgments of preference for states of health.
Table I

Examples of domains of health-related quality of life

At each level of the hierarchical framework, external factors and a person’s intrinsic characteristics can influence the relationship between one level and the next. For example, chemotherapy-induced nausea may be lessened with medication, improving a person’s physical functioning. Social support from a family member may allow a person with low physical function to perform his or her usual daily activities, improving overall quality of life. As we move further up the hierarchy of outcomes from the biologic and physical variables, it may be more difficult to establish a direct relationship between an intervention and a health outcome, yet we measure something that becomes more directly relevant to the patient. While in the past research and clinical practice has focused on measurement of biologic and physiologic variables,[7] investigators have developed valid and reliable tools that incorporate a patient’s experience as a critical dependent variable.

1.1 Measuring Health Outcomes

Within this framework, a wide variety of health outcomes measures are available that address different types of patient-reported outcomes across the spectrum of cancer care, ranging from prevention to care at the end of life. To better understand the measures and their uses, it is helpful to have a taxonomic classification of the measures themselves. Working knowledge of this taxonomy helps to clarify the rationale behind many choices in measuring outcomes in the extensive literature on cancer and HR-QOL. The goal of this section is to familiarize the reader with some of the common health outcomes measures used in the cancer field, and gain insights as to why certain measures might be chosen in certain situations.

Health outcomes measures in cancer can be considered along three dimensions, which we use to organize our discussion of the measures. The first dimension is the intended population in which the instrument is used: is the measure specific to a particular disease or condition, or is the measure a generic instrument designed for people with a wide range of health conditions? Second, instruments can be divided into whether they measure one domain of health, or whether they cover multiple domains. Finally, instruments can be divided by whether they are preference-based or not preference-based measures; i.e. whether or not the instrument reflects preference for, or the overall value of, a particular state of health.[28] For example, preference-based measures could compare the desirability of health states involving primarily physical function limitations to health states involving primarily emotional function limitations.

These dimensions form a grid into which measures can be classified. Some of the boxes in the grid do not have any associated instruments. For example, since preference-based measures incorporate a valuation of overall health, there are currently no single-domain preference-based measures of health. The various measures that fit within this framework are discussed in sections 1.2, 1.3 and 1.4 to give the reader a sample of the type of quality of life outcome measures available and potential strategies for using the measures. Table II shows the instruments discussed in this paper within the context of this classification framework.
Table II

Generic or disease-specific instruments used to determine health-related quality of life across multiple or single domains

1.2 Generic Versus Disease-Specific Measures

This dimension focuses on the generalizability of instruments for health conditions. Generic measures are usable across a broad range of conditions, allowing comparability of a specific patient group to people with other health conditions. A measure more specific to a particular disease can focus more on concerns associated with that disease, which may allow greater sensitivity to changes in health due to the disease at the cost of reduced comparability across conditions.

1.2.1 Generic Instruments

Generic measures are designed to measure HR-QOL in people with a wide variety of health conditions. These measures have usually been validated in large samples and in people with widely varying health status. The use of these measures in wide-ranging samples provides the advantage of allowing a comparison of patients with cancer with the population as a whole or with patient populations with other conditions. Since these measures are not designed to focus on any specific disease process, they tend to take a broad approach to including different domains of HR-QOL. In addition, since they are not focused on any particular disease, they can be used across the entire spectrum of cancer care, from prevention and early detection through to treatment, survivorship, and end-of-life care. This section presents several of the more popular instruments that have been used in cancer or cancer-related settings, along with the domains of health they measure. While the surveys vary widely in format, length, and approach, they have large overlaps in the concepts that they measure.

Two of the more popular generic HR-QOL instruments currently in use are modified from instruments originating from the RAND Health Insurance Experiment[55] and the Medical Outcomes Study.[21] These are the Short-Form (SF)-36 Health Survey[21,56,57] and SF-12 Health Survey,[30] which are 36-item and 12-item general health surveys, respectively. The SF-36 can be scored on eight domains of health:[21] physical function, role limitations due to physical function, general health perceptions, bodily pain, energy and vitality, social function, mental (emotional) health, and role limitations due to emotional function. Each domain is scored from 0–100, where 100 is the best level of ability measured for that domain. The survey can also be scored using two summary scales: the Physical Component Summary (PCS), representing physical function, and Mental Component Summary (MCS) scales,[29] representing emotional function. The PCS and MCS scores have been scaled so that the mean of both scales in a general population is 50, with a standard deviation of 10,[58] providing an automatic comparison to the sample the instrument is used in. The SF-12 is a 12-question subset of the SF-36, with questions representing each of the 8 scales, and scored using similar PCS and MCS summary scales.[30] These summary scales of the shorter survey account for over 90% of the variance of the PCS and MCS scales of the longer SF-36. Both of these surveys are reliable and have been well validated in general populations and in samples with a variety of health conditions. Over 100 English-language publications report the results of using the SF-36 for a wide variety of cancers in clinical settings, including, for example, breast cancer,[59,60] prostate cancer,[40,41,61] sarcomas,[62] and Medicare managed-care beneficiaries with cancer.[63] These surveys can be administered by interviewer, by telephone, or by written survey.

The Sickness Impact Profile,[31,32] an older generic HR-QOL measure, is still commonly used in many indications, including cancer.[6467] The survey is a 136-item yes/no format questionnaire that is based upon a conceptual model of sickness as manifested in changes in behavior.[68] The survey has 12 dimensions representing ambulation, mobility, body care and movement, communication, alertness behavior, emotional behavior, social interaction, sleep and rest, eating, work, home management, and recreation and pastimes. This survey has the advantage of being extensively validated and of representing a wide range of health domains; its primary disadvantage is that it is lengthy.

The Nottingham Health Profile[33,34] measures health within six content areas: energy, emotional reactions, social isolation, pain, sleep and physical mobility. A second part of this survey asks patients how their health affects life activities, although currently this portion is not frequently used. This survey has been used in patients with a variety of cancer types, for example, lung,[6971] breast,[72,73] and pancreatic cancer.[74] While this instrument has been validated and been shown to be responsive to conditions such as cancer, it may be less sensitive than other instruments to changes in health for those with more minor functional impairments.[75]

The Dartmouth Primary Care Cooperative Information Project (COOP) chart system[35,36] is a generic HR-QOL measure specifically developed for use in an outpatient clinical setting. This survey has not been used as extensively for cancer as some of the other generic instruments, but has been used to study outpatients in cancer clinics,[76,77] inpatients receiving chemotherapy[78] and patients with head and neck cancer.[79] The COOP survey was designed for interviewer or self-administration, and is composed of single item scales measuring nine domains: physical fitness, feelings, daily activities, social activities, change in health, overall health, social support, pain, and overall quality of life. Each scale has a simple figure illustrating the state of health. In addition to being simple and fast to administer, this survey has a version that is specific to adolescents.[80]

1.2.2 Generic Cancer Instruments

Cancer-specific measures can be subdivided into whether they are designed to be used for people with any cancer (generic cancer instruments), or whether they are intended specifically for use with people with specific types of cancer (site-specific cancer instruments). Generic cancer instruments frequently represent a compromise between the generic measures and the site-specific measures, in that they tend to take a broad approach to measuring domains of HR-QOL, but they also have questions dealing specifically with concerns of cancer patients. These instruments allow comparability between samples with different primary cancer sites, but may be more sensitive to cancer specific issues than generic measures.

The European Organization for Research and Treatment of Cancer (EORTC) Quality of Life Questionnaire (QLQ-C30) [37,38] is the most commonly studied generic cancer instrument reported in the medical literature.[81] Similar to the generic health instruments, this instrument is a brief (30 questions) questionnaire measuring health on multiple domains, including physical function, role function, cognitive function, emotional function, social function, and global quality of life. In addition to these scales, however, the instrument adds symptom scales and items (e.g. fatigue, pain, nausea and vomiting) particularly relevant to cancer patients. A recent randomized trial in patients undergoing palliative chemotherapy[77] found that administering the EORTC QLQ-C30 at outpatient visits helps to increase physician recognition and discussion of patient health problems.

The Functional Assessment of Cancer Therapy — General (FACT-G),[39] one of the Functional Assessment of Chronic Illness Therapy surveys, is another commonly used multi-domain instrument designed for use in cancer patients. This questionnaire measures physical function, social well-being, emotional well-being and functional well-being. Similar to the EORTC QLQ-C30, the FACT-G survey can be used for interviewer administration, telephone administration or as a written, self-administered questionnaire, and the survey is available in numerous languages.

1.2.3 Site-Specific Cancer Instruments

Site-specific instruments have the advantage of being able to focus on specific symptoms and concerns related to a specific type of cancer; because of this focus, these surveys may be more sensitive to change in health status of patients with cancer than generic instruments. In general, there have been two approaches to developing site-specific instruments, a modular approach and a separate survey approach. Both the EORTC QLQ-C30 and the FACT-G are designed to be modular surveys, which include the general cancer instrument, and optionally additional site-specific modules. These modules have questions particularly relevant to patients with a specific primary cancer, including breast cancer,[82,83] lung cancer,[84,85] prostate cancer,[86,87] and a variety of other primary sites.

Other site-specific surveys are designed solely for one cancer type. For example, the University of California Los Angeles Prostate Cancer Index[4042] contains the SF-36 and items measured on six additional scales related to prostate cancer symptoms including urinary function, urinary bother, bowel function, bowel bother, sexual function and sexual bother. The additional prostate cancer-specific scales make the instrument more responsive to different cancer-related states of health than the generic measure alone.[41] The Breast Cancer Chemotherapy Questionnaire[43] covers some more general issues such as emotional function and fatigue, but also covers specific issues such as consequences of hair loss due to chemotherapy.

1.3 Domain-Specific Versus Multi-Domain Measures

The HR-QOL instruments discussed in section 1.2, whether scored with a single score or with multiple scale scores, represent multiple domains of HR-QOL. However, sometimes a focus on a specific domain of health, rather than a specific disease, is desired when measuring outcomes. For example, in the National Surgical Adjuvant Breast and Bowel Project P-1 breast cancer prevention study[88] (a randomized trial of tamoxifen for the prevention of breast cancer in high-risk women), investigators measured QOL as an endpoint in addition to breast cancer development. Since women started the trial without breast cancer, investigators used a generic HR-QOL measure, the SF-36. To augment the patients’ responses to specific concerns, investigators supplemented the generic measure with the Center for Epidemiologic Studies — Depression scale[52,53] (a measure focusing on the degree of depressed mood), the Medical Outcomes Study Sexuality Scale[54] (to examine sexual function), and a symptom checklist. A variety of generic and disease-specific instruments are available to measure domains such as emotional, sexual function, cognitive function, sleep, pain, and intervention- or cancer-related symptoms.

1.4 Preference-Based Versus Non-Preference-Based Measures

Most of the measures discussed in section 1.2 describe a person’s health based on specific domains, whereas preference-based measures provide a summary valuation of the desirability of a state of health.[8991] While some non-preference-based measures have summary scores produced by summing scores for different domain-based component scores, preference-based measures differ by explicitly linking the score to a judgment of value for the state of health.[28] This valuation allows for a comparison of widely disparate health states; for example, a preference for moderate congestive heart failure, with moderate limitations on physical function due to cardiac disease, could be compared with that of moderate depression, in which physical function may not be limited but emotional function is compromised. A respondent could provide a summary statement of desirability for each state, and different respondents may have a higher preference for different states, depending on their individual values for different domains of health. A recent Institute of Medicine report in the US on summary measures of health[92] emphasizes the need for a summary valuation of the complex, multidimensional entity that comprises health, and that adequately reflects the tradeoffs between the dimensions.

Health utilities are commonly used preference measures that are considered to be ratio scaled and are bounded by 0 (representing death) and 1 (representing excellent health). Health utilities can be combined with measures of longevity to form a composite measure of length of life and quality of life measured in quality-adjusted life years (QALYs). QALYs represent an important outcome measure in decision and cost-effectiveness analyses in healthcare;[28] a recent systematic review[93] identified 40 cost-effectiveness analyses using QALYs as an outcome metric in the field of oncology.

Two main approaches to measuring health utilities include direct and indirect assessments. In a direct assessment, the respondent is asked to provide their own value for a state of health, using the anchor points of death and excellent health. The most common ways of directly assessing preferences include rating scale techniques,[94] the time trade-off (TTO),[89] and the standard gamble (SG).[94,95] The rating scale assessment is the simplest assessment. While exact techniques vary, in general, respondents are asked to rate their health on a scale from 0 (representing death or the worst state of health imaginable) to 100 (representing full health). With regards to the TTO, the respondent must consider a hypothetical trade-off of length of life to have excellent health instead of his or her current state of health, while with the SG the respondent must choose between having his or her current health for certain, or a gamble involving a chance of getting excellent health but with a risk of death. While there is currently a debate about whether the rating scale represents a true health utility,[96] all three of these measures have been used as health utilities for the purpose of measuring HR-QOL and longevity outcomes in QALYs.[97]

In contrast to the direct assessments, indirect utility measures reflect an average group preference rather than the respondent’s own preference for the respondent’s state of health. These measures are typically termed ‘utility indices’, or multidimensional descriptive HR-QOL survey instruments which produce a summary score representing an average utility. The mechanisms for creating the scoring functions vary for utility indices. However, these indices all consist of a descriptive health survey on which the respondent describes his or her state of health on multiple domains. The survey is then scored to produce a summary measure that represents an average preference score of the population that the scoring algorithm was derived in. For example, the commonly used scoring algorithm for the EuroQoL EQ-5D survey[98] would produce a score that represents an average TTO value of a sample group in England from which the scoring algorithm was created.

The three most commonly used generic utility indices are the EuroQoL EQ-5D,[45] the Health Utility Index (HUI),[44] and the Quality of Well-Being Index (QWB).[46,47] These three instruments are based upon brief multi-domain health surveys (e.g. the EQ-5D assesses five domains including: mobility; self-care; ability to perform usual activity; pain and discomfort; and anxiety and depression). All three instruments have been used in cancer-related settings for adults. One study has examined the QWB in childhood cancer,[99] and the HUI has been more extensively evaluated in the published literature for measuring preferences for children with cancer.[100107] A new utility index, the SF-6D,[48] based upon the questions from the SF-36 survey, has recently been published. Since it can be calculated for any person taking the SF-36, it will undoubtedly become a popular instrument. Cancer-specific utility indices[49,50] and prostate cancer-specific utility indices[51,108] have been published.

2. Strategies for Measurement

The most important tasks in deciding upon a measurement strategy are to consider (i) why the measurement is being done and (ii) what aspects of health outcomes are important to measure.

These questions are not always straightforward to answer. While it is important to measure the end results of care in research and clinical settings, how will the results be used? For example, if deciding whether or not a specific intervention improves patients’ overall health, a preference measure or a broad-ranging multidimensional measure of health may be most appropriate. However, one of the cancer-specific symptom indices may provide the information most directly linked to a palliative therapeutic intervention for individual patients. If we wish to compare burden of illness of one disease with another, then we need to use a generic instrument that is applicable to a patient with both diseases. However, a disease-specific instrument may provide more detailed information and may be more responsive to change for patients with a specific cancer. If we are concerned about whether an intervention improves one domain, e.g. mental health of cancer survivors, then we might choose domain-specific instruments that provide detailed measurements of the domain of interest.

Table III provides a brief outline of some general issues to be considered when selecting specific instruments for HR-QOL measurement. Extensive recommendations have been developed for use within research settings.[109] Frequently, there is fundamental tension between comprehensiveness, responsiveness, and burden to the respondent. Instruments that require the patient to report about his or her health on a number of domains, and to provide in-depth detail within these domains, increases the length of the survey and the time necessary to complete the survey. This issue of respondent burden may be of particular importance in conditions such as cancer, which may impact on health in multiple domains. Prioritizing the areas of health measured and pilot-testing instruments within specific populations can help to ensure that a survey is comprehensive and responsive while minimizing respondent burden.
Table III

General issues to be considered when selecting health outcomes instruments

Within the constraints of respondent burden, using combinations of measures across the grid discussed in section 1.1 can be a helpful approach to developing a measurement strategy for a specific situation. After considering what the most important dimensions would be to measure, the decision-maker might choose disease-specific and/or domain-specific measures to ensure maximal responsiveness in these dimensions. A generic health instrument, or generic cancer instrument, which measures a broad range of health dimensions using relatively few questions might be added to ensure that the measurement strategy is not missing unexpected effects in other areas of health. If the decision-maker specifically needs to know if overall health is better or worse, a preference-based measure could provide the valuation, whereas the multidimensional generic measures may provide a description of health function explaining why overall health is better or worse. Choosing different categories of instruments can be synergistic for an overall measurement strategy, although care must be taken to avoid the redundancy of questions and needlessly increasing respondent burden.

3. Future Directions in Health Outcomes Measurement

3.1 Novel Psychometric Methods

HR-QOL conceptualization and measurement have become increasingly important topics in the medical literature. An evaluation of articles solely dealing with the development or assessment of HR-QOL instruments found almost 4000 articles.[81] New measures are continuing to be developed and older measures continue to be refined. From a methodological point of view, however, one of the most exciting issues in outcomes measurement is the application of newer psychometric approaches to instrument development and measurement. Item response theory, which has been extensively used in the educational field, is now starting to be applied to measurement of health outcomes.[110114] This analytic approach simultaneously evaluates an individual’s level of function on a specific dimension (e.g. physical function) from his or her responses to survey questions as well as the level of difficulty of the questions (e.g. endorsing “I can walk a mile” would be more difficult than “I can walk a block”).

The item response theory approach to instrument development, compared with classical test theory, has several advantages if the underlying assumptions are satisfied and if sufficient data are available to generate the necessary models. While the advantages are briefly addressed in this article, the reader is referred to the references cited in this section[110114] for a more extensive discussion.

Established levels of difficulty for individual survey items allow the possibility of computerized adaptive testing.[110] Current fixed instrument approaches require that a person respond to all items, regardless of their level of ability on a particular dimension; this approach can be redundant for low ability questions asked of high-ability respondents and vice-versa. For example, if someone reports that they can walk a mile, they do not need to be asked if they can walk a block, yet in standard paper-based assessments both questions would need to be asked. Computerized adaptive testing would allow storage of numerous questions representing a wide range of abilities for a particular dimension of HR-QOL. As the respondent answers each item, a new estimate of the person’s level of ability is calculated and the computer selects the next question that would best further refine the estimate. Using a computerized adaptive testing approach, several questions can be as accurate as longer static questionnaires in determining the level of function for a particular domain. Measuring health status across multiple domains of health is frequently important to understanding the health status of patients with cancer. Computerized adaptive testing could dramatically reduce respondent burden while maintaining or even improving responsiveness of the instrument, particularly in these settings where information on multiple domains of health is necessary.

Scale equating is also an exciting application of item response theory.[112] The wealth of instruments available to measure single or multiple domains of HR-QOL makes comparisons across different populations potentially difficult. How do we compare the physical function, for example, of a patient who has taken the FACT-G with one who has taken the SF-36? The levels of difficulty of individual items within an instrument, as determined by various item response theory approaches, allow the estimation of a person’s function on a scale of the trait that is independent of the original questionnaire. Therefore, the levels of function gleaned from different questionnaires could be projected onto the same scale, and people completing different instruments measuring the same domains of health could be compared. Investigators have also examined the use of item response theory methods to equate instruments across countries and cultures.[115,116] These effects may allow for improved generalizability of outcomes findings across populations in the future.

4. From Measurement to Action

The measurement of HR-QOL outcomes in cancer has become pervasive in cancer research, yet is reasonably new to routine clinical care. However, work is starting to turn to how outcomes measurement can influence cancer care and health outcomes. Previous conceptual work has considered the link between outcomes research and its impact on policy, practice and outcomes. Stryer and colleagues[117] describe this link using a research pyramid. The first level at the base of this pyramid is research that impacts on other research findings, e.g. research that identifies problems, generates hypotheses or develops new tools. Level two represents research that impacts on a policy or healthcare program. Level three represents research that is linked to a documented change in what clinicians or patients actually do in practice. Finally, a level four change at the top of the pyramid links research to a change in health outcomes. While much of the work in outcomes research in cancer represents level one evidence, impacts at higher levels are starting to be demonstrated. This section discusses methods in which outcomes research in cancer care is having, or can be having, higher-level impacts on policy, practice and outcomes.

4.1 Informed Choices

The choices that patients face in cancer care can be difficult, and often involve trade-offs between current morbidities and risks of future adverse outcomes. Patients are being encouraged to think about the potential outcomes of their treatment decisions, e.g. the US National Breast Cancer Coalition Fund’s Guide to Quality Breast Cancer Care,[118] a guide that advises patients with breast cancer to think about their choices and potential outcomes within the framework of their own values and preferences. Clinicians and researchers have become increasingly interested in methods to improve patients’ understanding of the choices they face and the quality of the decisions they make regarding their care.[119] This research has, in part, led to the development of decision aids that assist the patient in clarifying the decision alternatives and provide information on the options and outcomes relevant to the patient.[120] Such decision support tools can improve knowledge of intervention alternatives and outcomes,[119,121] yet more outcomes information is needed so that patients can be informed about possible outcomes of their treatment choices.

There are several possible sources of outcomes information to inform future patients in their decision-making process. First, outcomes data from large clinical trials can be an excellent source of information. For example, in a recent trial, tamoxifen was found to reduce breast cancer incidence in high-risk women.[88] Health outcomes data collected as part of this trial suggest that, overall physical and emotional function were similar between women who were taking tamoxifen and those who were not taking tamoxifen, however women taking tamoxifen were more likely to experience some symptoms such as vasomotor symptoms.[122] These clinical trial outcomes are important sources of outcomes information, but can have issues of generalizability if the person faced with a decision is not similar to those in the study cohort.

Observational data can also provide outcomes information with which to inform patients about the potential outcomes of their choices. For example, the Prostate Cancer Outcomes Study, a longitudinal cohort study, identified almost 3500 men with prostate cancer from the Surveillance, Epidemiology, and End Results Program[5] national cancer registry. This observational study has been used to examine health outcomes issues, such as HR-QOL after either radical prostatectomy or radiation therapy[123] and androgen deprivation,[124] as well as clinical practice issues, such as understanding the determinants of treatment choice for clinically localized prostate cancer.[125] These, and other observational studies, can help to understand outcomes for a wide variety of patients, including those who may not be eligible or who would choose not to participate in a clinical trial.

To help understand the outcomes of treatment choices, in the US a national outcomes database has been proposed,[126] including the development of a registry of patients with a particular condition (e.g. prostate cancer) and tracking the treatment choices and health outcomes of these patients. Outcomes experienced by these patients could be used to inform similar patients (e.g. of similar age, with similar tumor stage and grade, and/or with similar comorbid illnesses) about the possible range of outcomes associated with treatment alternatives. The ideal goal would be for all patients to contribute outcomes information to assist future patients with decision making through understanding the range and likelihood of possible outcomes. Currently, the Joseph H. Kanter Family Foundation (Falls Church, Virginia, USA), which aims to help patients and providers to make more informed healthcare decisions, and the Agency for Healthcare Research and Quality in Rockville, Maryland, USA, are exploring the mechanisms for implementing a national outcomes database for common health conditions including common cancers.[126]

HR-QOL is increasingly being measured in clinical settings, yet for these measures to have an impact on practice they need to be routinely incorporated. Measuring HR-QOL in the clinical setting may improve physician-patient communication[77] and facilitate decision making, although the impact of measurement on health outcomes remains uncertain and in need of further research. Particularly in busy outpatient settings, a number of barriers exist to the rapid collection and feedback of HR-QOL surveys, but advances in information technology may help to overcome these barriers. User-friendly computerized HR-QOL assessments appear to be acceptable to patients,[127,128] and can provide rapid assessment and feedback to the clinician and patient. Computerized administration will become especially valuable as reliable and valid computerized adaptive testing-based instruments are further developed and disseminated, allowing for the collection of a broad range of HR-QOL information with minimal respondent burden, which is especially important in cancer care settings.

Further research is needed to determine the most effective ways to improve shared decision making between providers and patients, to help inform patients of their alternatives, and to help patients understand the possible outcomes of their choices. Investigators are starting to explore the relationship between decision making of cancer patients and their health outcomes,[129,130] an important area of future research.

4.2 Improving Practice and Systems

In addition to using outcome measures to help patients understand the potential outcomes of their own healthcare choices, outcomes measures are being incorporated into quality of care initiatives. To the extent that quality care can improve or maintain HR-QOL, the outcomes of patients with chronic conditions such as cancer should reflect the quality of care they receive within health systems such as clinics, hospitals, or health plans. The Centers for Medicare and Medicaid Services (CMS; Baltimore, Maryland, USA) have a major focus on improving the quality of care of Medicare beneficiaries. The Medicare Health Outcomes Survey (HOS)[131,132] represents a major CMS initiative using a measurement of health status as a quality of care measure. The HOS measures the 2-year change in health status of Medicare managed care beneficiaries using the SF-36 survey, and will publicly release case-mix adjusted changes in health status at the plan level.[131] Cross-sectional data from the HOS show variations in health status at the plan level,[131] although longitudinal data have not yet been published. Further work is needed to link these outcome measures to the optimal care processes in order to improve quality of care.

In order to link these outcome measures to improvements in quality of cancer care, the measurement of outcomes must focus on measures that are consistent across different sites and systems for providing cancer care, and on the measurement of domains of health status that are relevant to patients and to healthcare systems. The US National Cancer Institute’s Cancer Outcomes Measurement Working Group,[133] a group of 35 scientists largely from government and academia, is currently working on developing a core group of outcomes measures for evaluating cancer care across the entire continuum, including prevention, early detection, diagnosis, treatment, and end-of-life care. This group is evaluating potential outcomes measures based upon validity, reliability, sensitivity, acceptability, feasibility in application and importance to decision makers.[133] The goal of this initiative is to use level one data as a basis for an initiative which can facilitate the production of higher level evidence on the pyramid of research findings to help decision making at the practice and policy level.

5. Conclusions

As the importance of the patient’s perspective has been increasingly recognized, patient-reported health outcomes have become increasingly used and important outcome measures applicable across the entire spectrum of cancer care, from prevention through to end-of-life care. HR-QOL information can be helpful at several levels. HR-QOL data may assist patients in understanding the potential consequences of their care choices, and reporting their own health on standardized measures may help enhance physician-patient communication. These measures may help clinicians to recognize the impact of cancer and cancer therapies on their patients’ daily functioning. At the level of healthcare systems, HR-QOL can be used as outcomes measures to better understand what care processes favorably influence health outcomes. HR-QOL measures continue to represent important outcomes in cancer research to better understand the impact of novel cancer therapies on quality of life in addition to survival.

Research needs to continue into improving our ability to measure health outcomes in ways that are meaningful to patients with cancer. Of similar importance is the need for research into ways to seamlessly integrate HR-QOL measurement into routine practice, and into optimal methods for measurement to have an impact on the quality and outcomes of care. The use of measures assessing health outcomes is starting to expand from clinical research studies that are aimed at finding more effective interventions to also include work on improving the quality of cancer care by guiding improvements in clinical practice and in health policy. As patients face difficult decisions about their cancer care, a detailed understanding of the patients’ perspectives on HR-QOL and of the outcomes of cancer care can help patients make the best decisions regarding their own care.


Funding for this manuscript was provided by the Agency for Healthcare Research and Quality, in Rockville, Maryland, USA. The authors have provided no information on conflicts of interest directly relevant to the content of this review.

Copyright information

© Adis Data Information BV 2003