Cancer produces multiple distressing symptoms that affect functioning and quality of life, especially for patients with advanced disease. Cancer therapy can significantly reduce disease-related symptoms or prolong the time to symptom onset or worsening; it also can damage or functionally interfere with normal tissues, thus producing new symptoms or exacerbating existing symptoms. Even targeted therapies, whether used alone or in combination with traditional chemotherapy or radiotherapy, produce symptoms that can be dose limiting. Conversely, effective control of treatment-related symptoms may improve patient health, minimize toxicities that impair function, and increase adherence to curative treatments—not only maintaining or even enhancing health-related quality of life (HRQOL), but also potentially increasing survival.1 Thus, alleviating or delaying the onset of symptoms caused by cancer and cancer therapy is of paramount importance.
Reporting the symptom status of patients in oncology clinical trials provides critical information about the effects of treatment and can be an important factor in planning treatment. Symptom reporting can also provide critical information for pharmaceutical companies in determining the relative benefit, safety, and dosage parameters of new agents and for those making regulatory and funding decisions that influence the economic impact of cancer on patients, the health care system, and society. The importance of developing a pathway for assessing tumor-related signs and symptoms within the drug-approval process has been recognized,2 and there is increasing advocacy for the inclusion of symptom measures in cancer clinical trials.3, 4 Consensus on how to assess the status of multiple symptoms as clinical trial outcomes is lacking, however.
To begin to establish such a consensus, an independent working group (Assessing the Symptoms of Cancer Using Patient-Reported Outcomes [ASCPRO]) of academic researchers, patient advocates, pharmaceutical industry representatives, representatives from the US National Cancer Institute, and participant observers from the US Food and Drug Administration (FDA) was established in 2006 to review the current use of symptom measures as clinical trial and clinical research outcomes and to make recommendations to facilitate the implementation of symptom assessment.5 To date, ASCPRO has published a background article5 and recommendations on the measurement of cancer-related fatigue in clinical trials.6
ASCPRO commissioned a Multisymptom Task Force to address methods for representing changes in multiple symptoms that may co-occur during a trial of an anticancer therapy, including the utility of a composite endpoint. Although differences in symptom benefit provide meaningful information for evaluating treatments, defining and measuring a composite multisymptom endpoint is conceptually and methodologically challenging. These definitional and methodological issues were the focus of the discussion and recommendations of the Multisymptom Task Force presented in this report.
A symptom is defined as patient-observed, subjective evidence of disease or physical disturbance.7 The severity of a patient's symptoms (whether caused by disease or by treatment toxicity) and the impact of those symptoms upon normal functioning constitute symptom burden. Symptom burden is an important component of HRQOL, an inclusive concept comprising many domains. Within HRQOL, symptom burden is most closely aligned with the biologic and physiologic changes associated with disease and treatment.5 Because a symptom can only be known through the patient's subjective report, it is by definition a patient-reported outcome (PRO). Because symptoms like pain can broadly affect other aspects of HRQOL, some have identified them as causal variables rather than indicator variables.8
SYMPTOMS AS CLINICAL OUTCOMES
The primary function of symptom measurement in a clinical study is to inform patients and clinicians about the symptomatic impact of a treatment. Several trial scenarios are possible, including but not limited to: 1) the treatment reduces disease-related symptoms; 2) the treatment delays the onset of disease-related symptoms; 3) the treatment itself produces symptoms that need to be considered in the overall evaluation and use of the product; and 4) 2 therapies are equally effective in treating disease, but 1 is less toxic. Early evidence of multiple treatment-related symptoms can serve as a warning sign to drug developers that adherence to treatment may be compromised.
The analysis of differences in symptom severity has played a role in numerous studies comparing active treatment arms or comparing treatments versus placebo.9, 10 For example, studies of renal cell carcinoma11, 12 have demonstrated that selected, targeted therapies were associated with less-severe symptoms than cytokine-based therapies. Other studies have demonstrated that 1 arm of combination therapies13 and some surgical approaches14 was associated with less symptom burden. In some cases, symptomatic improvement is noted even when standard outcomes (survival, progression-free survival) do not improve.15
In 2009, the FDA issued Guidance for Industry. Patient-Reported Outcome Measures: Use in Medical Product Development to Support Labeling Claims,16 which supports the use of patient report in trials in which symptoms are an outcome: This guidance advises using a PRO instrument when measuring a concept that is known best by the patient or that is measured best from the patient's point of view, such as symptoms, signs, or aspects of functioning related directly to disease status. Often, PRO measures represent disease effects on health and functioning from the perspective of the patient. The guidance emphasizes rigorous evaluation of the concepts to be measured and a clear specification of the statistical endpoint model to be used in analyzing trial data (Fig. 1).16
The inclusion of a symptom benefit in an FDA-approved label is an important option for the pharmaceutical industry, for both anticancer and supportive-care agents. Symptom benefit is specifically endorsed in the FDA's 2007 Guidance for Industry. Clinical Trial Endpoints for the Approval of Cancer Drugs and Biologics,17 although the FDA requires a clear distinction between improvement in tumor symptoms and lack of drug toxicity if symptom assessments are intended to serve as primary endpoints to support drug approval. Thus, it is possible for symptom measures to support approval of an anticancer treatment, but the regulatory hurdles are high. Furthermore, recommended approaches for assessing specific symptoms by patient report in clinical trials have been lacking, and barriers to the use of symptom measurement and other PROs in clinical trials have persisted.18 This has resulted in a decrease in the number of applications to the FDA that include symptom assessment.19 However, even if a symptom benefit is not included in the label, symptom assessment in a trial provides critical information for patients and the health care team about what to expect from treatment and which treatment-related supportive care measures may be needed, and it also may provide critical information for advisory boards that approve drugs and evaluate reimbursement and quality-of-care issues.19, 20
The Multisymptom Task Force developed the following recommendations addressing issues related to assessment of the overall symptom status of patients with cancer during clinical trials and afterward (for a summary of these recommendations, see Table 1).
Table 1. Summary of Issues and Related Multisymptom Task Force Recommendations
Task Force Recommendation
Need for a definition of a multisymptom (composite) outcome
• A definition for a multisymptom outcome is proposed
Difficulties inherent in attributing symptom source (disease, treatment)
• Symptoms may be produced by disease, treatment, both, or neither
• Determine the set of symptoms reported by patients in the target population before intervention
• Comorbid conditions, previous treatment, and supportive care can be confounding factors
• Track and compare symptom severity over the course of a trial to clarify the effects and relative benefits of a treatment or treatments
Numerous validated instruments are available
• Can existing tools can be modified to suit a particular study?
• Multiple symptoms can be measured with existing instruments as long as they are well validated, have good psychometric properties, and are sensitive to change
• Do new scales need to be developed?
• Existing symptom scales can be enriched to make them appropriate for a target population, and new symptom items can be added to existing scales as long as the addition of these new items follows the standards for development of other self-report measures, as described herein
• The conceptual or measurement model developed for a study will influence the choice of which scale(s) to use and which symptoms on those scales to include in the analysis
Many symptoms that could be measured
• The degree to which these symptoms are present and change over time can differ dramatically, depending on the type and stage of cancer or the specific treatment
• Symptoms to be measured should adequately characterize the patient's symptom experience, in accordance with documented evidence of relevance to the patient, clinical or preclinical evidence, and the clinical trial objectives
• Focusing on change in a symptom cluster overall may risk losing sight of critical changes in individual symptoms that are important for targeted patient populations
• Certain symptoms are common and distressing across cancers (eg, fatigue, pain, poor appetite, nausea, and often distress and depression) and should be included in any symptom assessment
Symptoms may cluster together, related to various diseases and treatments
• Co-occurring symptom clusters vary over time, depending on patient factors, analytic techniques, and interactions between disease and treatment
• Using cluster-based scoring as a basis for representing multiple symptoms in clinical trials or for deriving any multisymptom outcome may be premature
Necessity of including all relevant symptoms
• Some symptoms are only now emerging or may be unique to a disease or treatment
• Symptoms identified by patients cannot be ignored
• Clinicians do not always recognize relevant symptoms or accurately assess symptom severity
• Evidence of patient relevance can be obtained through qualitative interviews with individuals or focus groups, prioritization by patients using questionnaires, and cognitive debriefing of existing scales
• Not all symptom items in an existing scale apply to a targeted patient population
• Administer all items on an assessment tool, even those that do not directly apply to the target population
• Incorporate additional single-symptom scales to target an expected symptom if relief of such symptom may be an outcome of the treatment being studied
Need for a well designed clinical endpoint model
• Clinical research and clinical trials designed to evaluate symptoms as outcomes have a much different set of measurement requirements than the management of individual patients
• The conceptual framework of a clinical trial should guide each stage of its design: choice of symptoms to be measured, instrument(s) to use, new symptom items, if any, and structure of the analysis plan
• Prespecify intent: efficacy, treatment tolerability, palliation, or a combination of effects
How might a composite symptom score be operationalized?
Possible ways forward
• Use a mean score of the most frequent and bothersome symptoms for the target population, and specify a level of reduction in this mean that could be justified as clinically important (this helps to define a symptomatic responder)
• Use statistical approaches to determine and weight the relative contribution of each symptom to the overall endpoint
• Evaluate potential aggregate composite scores from large available data sets and cross-compare psychometric properties, interpretability, and stability; develop responder definitions
• Perform sensitivity analysis to determine whether multisymptom endpoint results are consistent with the individual components of the composite endpoint
1. Defining a Multisymptom Outcome
The FDA guidance16 states that a single-item PRO instrument is insufficient for capturing a general concept that includes multiple items or domains, such as physical functioning, or disorders defined by clusters of specific symptoms and signs. A PRO instrument consisting of a single item about improved or worsened symptoms probably would be uninformative about the effects of a treatment on each specific symptom or sign and would be inadequate as an endpoint to support labeling claims, although it could be helpful for calibrating a composite score.
At the same time, the guidance recognizes potential limitations of composite endpoints used in labeling claims that also were acknowledged by the Multisymptom Task Force as applicable to multisymptom outcomes. These include the possibility of large variations in component symptoms that would be obscured by an overall component score, and the inclusion of individual symptoms in the component that were not prespecified in the protocol. The guidance stipulates that the symptoms that drove the change in the composite score should be identified in the label and that individual symptoms should be tested only when a statistically significant treatment benefit is demonstrated by the composite score. In short, all symptoms that are included in a composite endpoint should have similar clinical importance and relevance to the patient, and no single symptom should exclusively influence the results.17
Addressing the relative merits of dealing with composite symptom domains versus individual symptoms first requires a clear working definition of a multisymptom outcome. The Multisymptom Task Force recommends the following:
A multisymptom outcome is a composite endpoint that reflects the severity and impact of symptoms reported as relevant by patients having a particular disease or undergoing a particular treatment. The Multisymptom Task Force further proposes that the measurement of symptomatic change, as a subset of HRQOL, may be a sufficient outcome in many clinical trials, giving health providers, patients, and regulators enough information to make decisions about whether to use, approve, or reimburse a treatment.21
2. Attributing Symptom Source
Explicit in the FDA guidance is that, if symptomatic benefit is to be claimed for an oncology drug in regulatory review, then the symptoms must be those of the disease and not the toxicities of therapy. Determining the causes of symptoms is elusive, however. Symptoms can be produced by disease, treatment, both, or neither, and attempts at symptom attribution have been unreliable. For example, a study of clinician attribution of adverse events in a large number of clinical trials indicated that half of the events attributed to study treatment were recorded for patients who received placebo.22
For several reasons, it is practically impossible to identify a singular source of individual symptoms in most situations. First, complex underlying chemical and biologic mechanisms are involved in both illness and treatment. Second, various factors, such as comorbid conditions and the effects of supportive care (eg, analgesics for pain, bronchodilators for dyspnea), can confound a straightforward interpretation of a treatment's impact on symptom severity. Third, patients in an oncology trial may have received cancer treatment before enrollment. Such confounding factors also affect more accepted oncology endpoints, such as overall or progression-free survival. Nonetheless, to assess only disease-related symptoms when a mix of disease-related and treatment-related symptoms are to be expected does not reflect biologic reality and denies important information about differences among treatments.
Whether or not exact attribution is possible, symptom assessment nonetheless permits informative comparisons of symptom burden from therapies. For example, a treatment that reduces the mass of a lung lesion may be associated not only with a reduction in both shortness of breath and pain but also with improved sleep. If 2 treatments reduced tumor-related symptoms to the same extent, then a significant difference in symptoms related to 1 treatment versus another would provide added benefit to the patient. This information may have a substantial impact on patient-clinician choice of therapy even if it does not meet the requirements for inclusion in labeling information.
Determining symptom attribution in all but the most obvious of circumstances (eg, treatment-induced mucositis, rash, vomiting) is likely unachievable. A more informative approach is to determine the set of symptoms reported by patients in the target population before the intervention and then to report changes in symptom status during and at the end of the study. Comparisons of symptom severity over the course of a trial can provide clear indication of the effects and relative benefits of the treatment or treatments.
3. Determining How Multiple Symptoms Should be Measured
A primary issue for symptom researchers and clinicians is whether existing multisymptom assessment tools can be used to measure any multisymptom outcome, or whether a new measure must be developed and validated for each application or trial. In a systematic review of cancer symptom assessment instruments, Kirkova et al23 identified 21 instruments, each measuring more than 5 symptoms, as being appropriate for clinical use in the assessment of multiple cancer-related symptoms. Some of these measures were groups of symptom items that were subsets of broader HRQOL assessment instruments.24, 25 Other measures asked patients to rate the severity of multiple symptoms and other dimensions of the symptom experience, including interference, distress, and frequency.26-28 Since the Kirkova et al review was published, additional symptom rating scales have been developed or adapted from existing scales,29-32 and expert opinion about modifying existing PRO measurements has been summarized.33
The Multisymptom Task Force recommends that multiple symptoms can be measured with existing, well validated scales (ie, those that have good psychometric properties and sensitivity to change). Even scales that do not specifically adhere to the FDA guidance for labeling claims produce important data for evaluating treatment outcomes as long as they have been developed according to accepted validation criteria. The conceptual or measurement model developed for a study will influence the choice of which scale(s) to use and which symptoms on those scales to include in the analysis. Most well validated multisymptom scales have a history of performance in clinical research that can guide the scale selection process.23 Conversely, new multisymptom measures require a great deal of time and expense to develop and have no history of effective usage.
4. Determining Which Symptoms to Measure
Converging evidence from both expert opinion and patient report has identified a subset of symptoms that are common across cancers, that increase with disease progression and shortened survival, and that are associated with significant distress for patients. Clinicians have identified fatigue, pain, poor appetite, and nausea as sentinel symptoms,34, 35 and they often include distress and depression.36 Patients with advanced cancer universally rate fatigue/tiredness as the most severe symptom,27, 37, 38 with pain, disturbed sleep, emotional distress (distress/anxiety/worrying), and poor appetite among the top symptoms.39-41 One study examined symptom ratings of cancer patients from several countries and observed little difference in the ordering of symptoms by severity (generally: fatigue, disturbed sleep, distress, pain, and lack of appetite), suggesting similarity across national, cultural, and linguistic boundaries.42
The consistency of these findings across studies suggests that some symptoms are common enough across various cancers to be considered as the basis for a multisymptom outcome. Nonetheless, the degree to which these symptoms are present and change over time can differ dramatically, depending on the type of cancer or treatment. For example, patients with lung cancer are likely to experience significant pain, dyspnea, and coughing early on, and their symptoms may improve with treatment; conversely, patients with renal cell cancer can be fairly asymptomatic until the metastasis worsens, so that prolonging time at their baseline score is advantageous. Similarly, sore throat is likely to be reported during radiotherapy but not chemotherapy.29 The expectation of symptom improvement or worsening over time will be very different for these populations, and symptom analyses must take into account the type and stage of cancer and the expected toxicity of the therapy to be studied. The growing literature on longitudinal assessment of cancer symptoms can help set these expectations.43
Investigators have recently noted that certain subsets of symptoms (for example, nausea and vomiting) are highly likely to be rated similarly by patients. Various analytic techniques, including factor analysis, cluster analysis, and multidimensional scaling, have been used to identify these subgroups of symptoms, often called symptom clusters.27, 44, 45 How symptoms cluster together depends on disparate factors, such as the particular symptoms rated,46 the analytic techniques used to derive the clusters,47 and even the age and sex of the patient.48 Clusters are likely to change over the trajectory of treatment because of interactions between disease-related and treatment-related symptoms.43, 49 Some analyses, such as investigations of sickness behavior clusters, suggest that the symptoms forming a cluster have a similar biologic basis.49, 50
Nonetheless, focusing on change in a symptom cluster overall risks losing sight of critical changes in individual symptoms that are important for targeted patient populations.51 Specific symptoms may change in different directions in response to an intervention (eg, treatment may cause an increase in fatigue, whereas pain is alleviated). Other methods of choosing critical symptoms may be more appropriate, depending on the goal of the clinical trial and the specific patient cohort. For example, in patients with advanced solid tumors, the co-occurrence of particular symptoms is common enough to produce reliable indexes that aggregate relevant symptom reports, and a significant reduction in a composite of such symptoms or a significant prolongation of the time until they appear seems clearly beneficial. In a recent trial, myelofibrosis-specific symptoms were identified and used to develop a composite score and define a symptomatic responder.52, 53
The Multisymptom Task Force recommends that the symptoms to be measured should adequately characterize the patient's symptom experience, in accordance with documented evidence of relevance to the patient, clinical or preclinical evidence, and the clinical trial objectives. In some cases, it may be necessary to supplement a multisymptom measure with 1 or more focused symptom measures. The Multisymptom Task Force believes that it is premature to consider cluster-based scoring as a basis for representing multiple symptoms in clinical trials or for deriving any multisymptom outcome. An index of summative impact, however, can be and has been successfully used.52
5. Incorporating Symptoms Specific to the Target Population
How to deal with emerging or unique symptoms related to disease or treatment presents a measurement dilemma. Should an entirely new scale be developed each time a new symptom emerges, or is it possible to append new symptoms to existing well validated scales? If the latter is justified, then significant efficiencies in both time to deployment and cost of scale development are achieved.
Obtaining patient input is especially important to ensure that relevant symptoms that are not recognized by clinicians are nonetheless included. Evidence of patient relevance can be obtained through a variety of methods, such as qualitative interviews with individuals or focus groups or prioritization by patients using questionnaires and/or cognitive debriefing of existing scales. Cognitive debriefing allows patients in the target population to rank the relevance of existing scale items, suggest new symptoms that the existing scale does not capture, and rate the ease of use and clarity of the instrument being validated.
The Multisymptom Task Force recommends that new symptom items can be added to existing scales as long as the addition of these new items follows the standards for development of other self-report measures, as described herein. In the absence of patient input supporting or confirming the additional measurement precision, using multiple and convergent approaches to establishing item relevance should be considered.
Although many existing scales have symptoms that are commonly reported by patients with cancer, not all items may apply to the targeted patient population. Nonetheless, the Multisymptom Task Force recommends administration of all items, including both items from the original scale and any new items. This approach augments previously derived information about the validity and utility of the parent scale while allowing a subset of the most relevant symptoms for a given cohort to be analyzed separately to verify their significance for the target population. Additional symptom scales focused on a single symptom may be required if there is reason to expect that relief of that symptom may be an outcome of the treatment being studied.
6. Developing a Symptom Endpoint Model
The selection of symptom assessment methods for clinical research and clinical trials designed to evaluate symptoms as outcomes has a much different set of requirements than the selection of symptom measures for the management of individual patients. Intended labeling claims, the basis for the FDA guidance,16 provide an excellent framework for thinking about assessment design at both the qualitative research stage and the clinical trial stage.
The symptom-related questions to be posed will dictate components of the endpoint design, including symptom assessment during the trial. At the qualitative research stage, hypothesized treatment effects should guide patient selection and interview criteria. For example, for reduction of existing symptoms, the qualitative assessment should focus on symptoms present at the outset of therapy; conversely, for delayed worsening, the qualitative assessment must be highly sensitive to the timing of new or worsening symptoms. The pattern of symptoms occurring over the course of disease or treatment, as described by patients in qualitative interviews, can be especially informative for identifying relevant symptom items and guiding the frequency of symptom assessments in a clinical trial. For example, when symptom effects occur rapidly, repeated use of a limited list of single-item symptoms is probably the only practical choice; when symptom effects occur gradually, assessment will be less frequent and thus, if warranted psychometrically, more items can be administered. The complexity of the concept being measured also affects the number of items. Ways of comparing these trajectories, such as measuring the area under the curve, have been used in the regulatory approval process for analgesics and may apply here.54
The Multisymptom Task Force recommends that a trial's conceptual framework guide each stage of its design, including the choice of which symptoms should be measured and with what instrument(s), whether new symptom items need to be developed, and the structure of the analysis plan. Investigators should prespecify whether they are trying to measure efficacy, treatment tolerability, palliation, or a combination of effects; the study design and analysis methods should be sufficient to evaluate these prespecified objectives.
7. Developing a Composite Symptom Score
A composite symptom score could be developed to represent variations in symptom burden during a clinical trial or to operationalize a multisymptom outcome as defined above. For example, studies have demonstrated that fatigue, pain, dyspnea, and coughing are highly prevalent in and distressing for patients with lung cancer, especially those who have advanced disease.29, 43, 55-58 How might a composite measure of these 4 symptoms be constructed, considering that composite endpoints are difficult to interpret if the improvement is not consistent among the components of the composite? That is, if only fatigue improved, then it would not be appropriate to conclude a benefit in the overall composite score.
The Multisymptom Task Force considers that a variety of methods are appropriate for deriving a composite score to combine ratings of multiple symptoms. A simple mean score of the most frequent and bothersome symptoms for the target population could be used, and a level of reduction in this mean that would be justified as clinically important could be specified. More complex statistical approaches also may be used. For example, using multivariate regression, it is possible to determine the relative contribution of the rankings of each symptom to another clinical dimension, such as progression of disease, or to other patient-reported measures, such as satisfaction with treatment, symptom interference or function, or global well being. Data-reduction techniques, such as principal component analysis or cluster or factor analysis to examine the underlying dimensionality of multiple symptoms, also may be considered. These methods could be used to weight the relative contribution of each symptom to the composite score. Another approach involves defining indicator variables for each symptom or collection of symptoms that would count the number of issues reported by the patient or record the presence of a severe symptomatic problem on the basis of a review of multiple symptoms.59 Patient testing and interviews also could be used to determine the relative importance of each of these symptoms to patients in the group targeted for the trial. A simple arithmetic mean of the ratings of all relevant items, together with what constitutes a meaningful reduction in this mean, may be an adequate composite measure.
Composite scores aggregated from multiple sources, including existing clinical trial data sets from the pharmaceutical industry and National Cancer Institute collaborative groups, could be cross-compared for interpretability by patients and by those who make judgments about treatment benefit and also could be investigated for their psychometric properties, effects on potential trial sample size, and performance across time (stability). These analyses also may be helpful in developing potential responder definitions. For a discussion of the advantages and disadvantages of composite scores, see the Initiative on Methods, Measurement, and Pain Assessment in Clinical Trials (IMMPACT) recommendations for using multiple endpoints in pain clinical trials.60 If a composite score is used as a multisymptom endpoint in a trial, then sensitivity analyses should be performed to determine whether the results are consistent with the individual components of the composite endpoint.
In conclusion, many issues involved in symptom assessment for cancer clinical trials have yet to be resolved, thus forming a rich agenda for future research. Nonetheless, we now know that patient report reliably represents patients' symptoms and meets the standards of assay sensitivity that are expected of clinical assessments and laboratory tests used as trial endpoints.
Members of the Multisymptom Task Force agreed that the assessment of changes in the status of multiple symptoms as measured by patient report represents a set of patient-perspective outcome measures that should be included in all cancer treatment trials, regardless of whether or not a labeling claim based on a symptom reduction benefit is being sought from a regulatory agency. A multisymptom outcome, if conceptually justified by the trial design and intent of the therapy, is critical for the appraisal of new and existing cancer agents, because it may provide outcomes data sufficient to make decisions about the value of a therapy or to allow judgment about the relative value of 1 therapy versus another.
Patient-based symptom data add critical information to the drug development and approval process, even if they do not meet the high regulatory bar for inclusion in the product label. For instance, a symptom measure may provide a patient's perspective on the benefits of delayed disease progression or on a symptom related to drug toxicity, or it may demonstrate how a patient's symptoms improve in response to palliative care. Whether or not symptom ratings are used to inform labeling claims, symptom data contribute critical information that informs the evaluation of treatment benefit and risk for consideration by prescribers, patients, and policy makers. Quality-assurance and comparative-effectiveness research increasingly demand the assessment of symptom status as a representation of the patient's experiences in a clinical trial or clinical encounter.21 We currently have the measurement tools and analytic techniques required to deploy a multisymptom outcome, and relatively straightforward methods for developing additional symptom items as needed are available. Future research, much of which could use existing databases, should examine the potential interpretability and utility of such a composite measure.