Published by Oxford University Press 2007.
Challenges to Use of Health-Related Quality of Life for Food and Drug Administration Approval of Anticancer Products
Affiliations of authors: Office of Oncology Drug Products (EPR, RP), Study Endpoints and Label Development Team, Office of New Drugs (JAS, DLK, LBB), and Office of Biostatistics (RS), Center for Drug Evaluation and Research, Food and Drug Administration, Silver Spring, MD
Correspondence to: Laurie B. Burke, RPh, MPH, Study Endpoints and Label Development Team, Office of New Drugs, Center for Drug Evaluation and Research, Food and Drug Administration, Bldg 22, 10903 New Hampshire Ave, Silver Spring, MD 20993 (e-mail: laurie.burke{at}fda.hhs.gov).
 |
ABSTRACT |
|---|
 Top Abstract IntroductionLack of RandomizationLack of BlindingMissing DataMultiplicityIntrinsic MeaningNotesReferences |
|---|
The U.S. Food and Drug Administration (FDA) approves labeling claims of drug efficacy based on substantial evidence of clinical benefit demonstrated in adequate and well-controlled investigations. Patient-reported outcomes (PROs) may support marketing claims of clinical benefit, either alone or with other study endpoints. Health-related quality of life (HRQL) is a PRO that comprehensively measures patients' reported health status. We present an overview of why HRQL-based efficacy claims have not to date been accepted by the FDA for inclusion in anticancer product labels. Persistent challenges to allowance of such claims include shortcomings in randomization and blinding of clinical trials, missing data, statistical multiplicity, and unclear intrinsic meaning of selected HRQL findings.
|
INTRODUCTION |
|---|
|
TopAbstractIntroductionLack of RandomizationLack of BlindingMissing DataMultiplicityIntrinsic MeaningNotesReferences |
|---|
The Food and Drug Administration (FDA) approves labeling claims of clinical benefit following from treatment with drugs and biologics on the basis of substantial evidence demonstrated in adequate and well-controlled investigations. Section 505(d) of the Act establishes substantial evidence as the evidence standard for making conclusions that a drug will have a claimed effect. Reports of adequate and well-controlled investigations are to provide the basis for determining whether there is substantial evidence to support claims of effectiveness for new drugs. These requirements are specified in the Federal Food, Drug, and Cosmetic Act (the Act). Following from the Act, the Code of Federal Regulations (CFR) further clarifies acceptable measurement standards for performance of clinical trials intended to support product labeling claims. 21 CFR 314.126 describes characteristics of an adequate and well-controlled investigation. The guidance for industry entitled Providing Clinical Evidence of Effectiveness for Human Drug and Biological Products reviews the nature and quantity of evidence necessary to meet FDA's substantial evidence standard [http://www.fda.gov/cder/guidance/1397fnl.pdf].
Clinical benefit can be measured either directly or indirectly. Direct measures assess effect of treatment on patient function, disease-related symptoms, or survival. Such measures generally are sufficient on their own to support drug approvals. Indirect measures or surrogates of benefit require in addition an understanding of both disease pathophysiology and the treatment's effect on a disease process. While few valid surrogates of treatment benefit exist (1), examples of surrogate markers that have been accepted by FDA for documenting efficacy of anticancer drugs include response rates (measuring radiographically observed tumor size reduction) and time to progression (measuring delays in tumor growth).
Patient-reported outcomes (PROs) are direct measures of clinical benefit that assess any aspect of health status from the patient's perspective and without interpretation by health care providers or others. These measures provide information on disease symptoms and effects of medical therapy that can only be known to the patient, e.g., pain intensity. PROs are measurable concepts of a patient's health status that range from relief of a single symptom to complex collections of subcategories referred to as "domains." PRO instruments, including questionnaire items, instructions, and guidelines for scoring and interpretation, are used to measure PRO concepts.
In 2006, the FDA published draft guidance to inform and request comment from medical product developers, clinicians, and researchers regarding FDA's current views on development and implementation of PRO-based measures that may potentially support treatment benefit claims in product labeling (2). This draft guidance advocates clear, prospective delineation of measurement concepts underlying health labeling claims and outlines principles for development, validation, implementation, and interpretation of reliable and accurate PRO measures. Unlike regulations, regulatory guidances provide FDA's perspective on a given topic rather than mandatory rules. Reasonable deviations to approaches outlined in the guidance are acceptable with adequate documented justification. Furthermore, the draft PRO guidance will be finalized only after comments submitted following its initial publication are reviewed and suggested revisions are considered.
PROs are not synonymous with either quality of life (QOL) or health-related quality of life (HRQL). QOL represents level of satisfaction with a person's life situation, including economic security; religious and political freedom; safety, number, and character of relationships; quality of education; and health care. Because many elements of QOL do not relate directly either to disease states or therapy, QOL is not an appropriate outcome for evaluating treatment benefit. Consequently, FDA does not accept inclusion of QOL claims in product labeling.
By contrast, HRQL refers only to those aspects of QOL that do involve a person's health status. Conceptually, HRQL represents a summation of how patients feel or function relative to their health and treatment. HRQL is a complex, multidomain concept that measures at a minimum physical, psychologic (including emotional and cognitive), and social functioning domains. Instruments to measure HRQL collect information on the presence, frequency, and intensity of symptoms, feelings, or behaviors in each of the above domains (3). HRQL instruments used in cancer treatment trials typically incorporate an additional disease-specific domain that focuses on symptoms attributable to the cancer being evaluated.
Improvement in HRQL has become a commonly sought treatment claim by drug manufacturers and a prominent endpoint in cooperative group trials. HRQL assessments may be incorporated into clinical trials for multiple purposes. Examples include 1) when experimental therapy is predicted to improve survival or disease-free survival yet cause potentially greater toxicity; 2) when survival and progression-free survival are presumed similar but one arm is predicted to be less toxic; and 3) when marginal differences in survival or progression-free survival are expected but differences in HRQL between treatment arms could influence treatment recommendations (4).
Since most conventional cancer therapies are associated with significant toxicity, therapeutic selection must entail consideration of both a treatment's potential risks and benefits (5). In this sense, HRQL measures are intuitively appealing since they seek to integrate potential benefits with adverse effects of anticancer treatment. Correspondingly, clinical studies of oncology products have incorporated HRQL assessments with increasing frequency over the past decade (6–12). However, the FDA has not yet used HRQL data as the primary evidence to support an anticancer product approval. The reasons for this dichotomy are addressed below.
Evidence supporting marketing claims of anticancer products follows from adequate and well-controlled investigations. By regulation, such studies should address explicitly and minimize known, potentially confounding sources of bias. HRQL data submissions thus far have faced challenges to acceptance by FDA due to their susceptibility to multiple known sources of bias. These are summarized in Table 1 and discussed below.
|
Lack of Randomization |
|---|
|
TopAbstractIntroductionLack of RandomizationLack of BlindingMissing DataMultiplicityIntrinsic MeaningNotesReferences |
|---|
Single-arm clinical trials compare outcomes in enrolled patients with those from historic controls. However, assessment of treatment effect in single-arm trials may be confounded by variation in a disease's natural history within the enrolled population; patients enrolled in the trial may not be representative of the larger population of patients with a particular diagnosis. In addition, comparison to historic controls is complicated by subsequent evolution in treatment, diagnostic testing, referral patterns, and supportive care. Thus single-arm trials are prone to confounding by selection bias. This type of bias can be addressed most effectively by conducting randomized clinical studies, preferably with stratification, if appropriate. Randomization tends to balance treatment arms for both known (stratified) and unknown prognostic factors.
|
|
Lack of Blinding |
|---|
|
TopAbstractIntroductionLack of RandomizationLack of BlindingMissing DataMultiplicityIntrinsic MeaningNotesReferences |
|---|
In unblinded trials, patients and investigators may be aware of treatment assignment. Patient knowledge of assigned treatment could influence their responses to questions. Reporting bias follows from subjective, unblinded measurement of outcome and may confound results of unblinded trials measuring HRQL. Ideally, trials intended to support HRQL labeling claims would be randomized, double blind, and placebo controlled. In practice, however, blinding in anticancer product trials is frequently difficult either to achieve or to maintain. Different administration schedules and routes of treatment routinely make blinding either impractical or unethical. Unblinding may also occur during conduct of a blinded trial due to different drug toxicities that may be recognized by patients and/or health care providers.
|
Missing Data |
|---|
|
TopAbstractIntroductionLack of RandomizationLack of BlindingMissing DataMultiplicityIntrinsic MeaningNotesReferences |
|---|
HRQL results from anticancer trials often have a significant proportion of data that is either incomplete or missing. Because instruments measuring HRQL are self-administered, the time and effort required to complete these questionnaires may present a significant burden to patients enrolled in oncology clinical trials. Trial patients may be either unable due to disease status or uninterested in spending the time, typically 15 minutes or so, required to complete each HRQL assessment. Alternatively, data collection ceases completely following a patient's death or decision to withdraw due either to toxicity or absence of treatment effect.
Large amounts of missing data increase the risk of informative censoring. When this occurs, responses are absent for systematic reasons that lead to biased conclusions. For example, patients on an experimental treatment arm may drop out of a clinical study due either to toxicity or lack of efficacy. Subsequent positive results might be due simply to absence of data from dissatisfied patients who left the study early. In practice, even a small amount of missing data may prevent accurate and unbiased HRQL assessment of treatment.
Compensatory adjustments for missing data have been proposed, including MCAR (missing completely at random) and MAR (missing at random) analyses. MCAR assumes that the probability of an observation being missing is independent of the observation's value. MAR assumes that the probability of a missing observation is independent of the observation's value, provided that additional factors associated with missing values are considered in the analysis. Effect of the factor(s) leading to missing data is almost never known and must be assumed to formulate a statistical model. These assumptions are generally invalid in the case of missing PRO data. Although use of these methods may partially compensate for missing data, the only reliable way to generate unbiased, robust HRQL results is to minimize the proportion of missing data (13).
PRO instruments completed by proxy (where others respond for the patient from the patient's perspective) have also been proposed as a method to reduce missing data and ensure greater compliance with HRQL data collection. However, assessments completed by proxies are not equivalent to those generated by patients themselves unless parallel responses by patient and proxy can be demonstrated. The FDA has generally not considered data obtained from a proxy to be appropriate for product labeling because validity of the proxy response is difficult to demonstrate (14).
|
Multiplicity |
|---|
|
TopAbstractIntroductionLack of RandomizationLack of BlindingMissing DataMultiplicityIntrinsic MeaningNotesReferences |
|---|
Multiplicity refers to the elevated risk of falsely concluding that a difference between groups has occurred when multiple hypotheses, usually unplanned, are performed. This is a form of analysis bias. If multiple hypotheses are tested in an uncontrolled manner, one or more tests may generate by chance alone an erroneous but "statistically significant" result. The likelihood of obtaining a statistically significant, false-positive result rises with the number of hypotheses tested. Regardless of trial objectives, prospective identification of hypotheses and careful statistical planning decrease the risk of false-positive conclusions. A prospective statistical plan should outline every primary and secondary endpoint, order of testing for all endpoints, and allocation of type I error rate or "alpha" (the probability of making a false-positive assumption) to each hypothesis being tested.
Although many anticancer product trials incorporate HRQL instruments, few such trials incorporate a statistical analysis plan that fully addresses the evaluation of HRQL measures and their evaluation in relationship to other endpoints, such as by hierarchic testing. Statistical plans typically omit discussion of hypotheses and rationale underlying HRQL measures. By nature, HRQL assessments integrate numerous questions into either multiple scores or a single summary score that incorporates results from multiple domains. Thus, a large number of possible statistical comparisons among scores are generated, and the risk of multiplicity accounting for any observed statistically significant results remains prominent.
Observed HRQL effects may also result from timing and frequency of assessment. Toxicity of therapy can markedly affect data obtained from HRQL measurement. Timing of toxicities versus administration of the HRQL tool must be carefully considered. HRQL should not be used as a substitute for measurement of therapeutic toxicities but can offer complementary information.
|
Intrinsic Meaning |
|---|
|
TopAbstractIntroductionLack of RandomizationLack of BlindingMissing DataMultiplicityIntrinsic MeaningNotesReferences |
|---|
Interpretation of HRQL data must consider whether all relevant information has been captured by the HRQL instrument, whether specific findings within a set of HRQL testing results are consistent, and whether claimed statistical significance of a given finding has actual clinical relevance to the population studied. As an illustration, HRQL assessments may be inconsistent when different instruments are used to test the same patient populations. Holzner et al. (15) compared results from the two HRQL instruments most frequently used in cancer trials in four different cancer patient groups. In each one of the four paired comparisons, the instruments generated disparate results in one of four HRQL domains. This occurred despite the fact that these two multidomain instruments were designed to measure similar HRQL concepts. This raises concern about the extent to which such instruments generate a valid assessment of concepts to be measured.
Consistency of endpoint results bolsters robustness of findings in clinical trials. One might expect that if HRQL treatment benefit has occurred, then a degree of internal consistency between HRQL domains should exist. Either all subsidiary domains should be positively affected by treatment or alternatively some domains should improve while others remain stable. Yet treatments typically do not favorably affect all components of HRQL to a measurable and meaningful extent. If worsening is observed in some domain(s), conclusions regarding improvement in overall HRQL are not valid.
In addition to internal consistency of HRQL domains, robustness of an HRQL claim is expected to be corroborated by consistent findings associated with other conventional endpoints such as overall survival, time to progression, or tumor response rates. Alternatively, HRQL results may corroborate data on adverse events. As noted above, hierarchic testing of such endpoints in a prospectively defined statistical plan is vital to control the risk of claiming false-positive findings.
Results of HRQL should provide convincing evidence to clinicians and patients of a meaningful treatment effect in a risk–benefit analysis. Small, inconsistent differences may be statistically positive. However, if the effect is small in magnitude and accompanied by considerable toxicity, then evidence of clinical benefit falls short (5).
In conclusion, we have described challenges that impede incorporation of HRQL data from clinical trials into anticancer product labels. Measurement principles described here apply to all endpoints supporting efficacy claims of anticancer products. Sponsors must strive to minimize measurement error and biases to describe accurately the benefits and toxic effects of medical products for which marketing authorization will be sought. Nonrandomized trials, lack of blinding, missing data, statistical multiplicity, and unclear intrinsic meaning of results present fundamental challenges to incorporation into product labeling of claims based on HRQL data. Sponsors who desire specific HRQL claims should discuss with the FDA in advance their plans for use of HRQL data to support efficacy claims of anticancer products. Such discussions should cover instrument(s) proposed for use, hypotheses to be tested, and the statistical analysis plan.
|
NOTES |
|---|
This paper is part of a special edition summarizing the 2005 NCI-CCOP "Assessing Quality of Life (QOL) in Cancer Symptom Management Trials Workshop."
Opinions expressed in this paper are the professional views of the authors and do not reflect the official position of the U.S. Food and Drug Administration.
Present address: GlaxoSmithKline, Upper Providence, PA (E. P. Rock).
Present address: Mapi Values Ltd, Cheshire, United Kingdom (J. A. Scott).
|
REFERENCES |
|---|
|
TopAbstractIntroductionLack of RandomizationLack of BlindingMissing DataMultiplicityIntrinsic MeaningNotesReferences |
|---|
(1) Fleming TR. Surrogate endpoints and FDA's accelerated approval process. Health Aff (Millwood) (2005) 24:67–78.
(2) Center for Drug Evaluation and Research, U.S. Food and Drug Administration. Draft guidance for industry: patient-reported outcome (PRO) measures: use in medical product development to support labeling claims. Available at: www.fda.gov/cder/guidance/5460dft.doc. [Last accessed: September 27, 2007.].
(3) Velikova G, Start D, Selby P. Quality of life instruments in oncology. Eur J Cancer (1999) 35:1571–80.[CrossRef][Web of Science][Medline]
(4) Paskett ED, Schrag D, Komblith A, Lamont EB, Weeks JC, et al. Cancer and leukemia group B cancer control and health outcomes committee: origins and accomplishments. Clin Cancer Res (2006) 12(Suppl):3601s–5s.
(5) Weng CSW. Where is health-related quality-of-life assessment in oncology clinical study heading? J Biopharm Stat (2004) 14:1–3.[Medline]
(6) Dooms CA, Pat KE, Vansteenkiste JF. The effect of chemotherapy on symptom control and quality of life in patients with advanced non-small cell lung cancer. Expert Rev Anticancer Ther (2006) 6:531–44.[CrossRef][Web of Science][Medline]
(7) Hassan I, Cima RC, Sloan JA. Assessment of quality of life outcomes in the treatment of advanced colorectal malignancies. Gastroenterol Clin North Am (2006) 35:53–64.[CrossRef][Web of Science][Medline]
(8) Chandu A, Smith AC, Rogers SN. Health-related quality of life in oral cancer: a review. J Oral Maxillofac Surg (2006) 64:495–502.[CrossRef][Web of Science][Medline]
(9) Blazeby JM, Avery K, Sprangers M, Pikhart H, Fayers P, Donovan J. Health-related quality of life measurement in randomized clinical trials in surgical oncology. J Clin Oncol (2006) 24:3178–86.
(10) Efficace F, Bottomley A, Osoba D, Gotay C, Flechtner H, D'haese S, et al. Beyond the development of health-related quality-of-life (HRQOL) measures: a checklist for evaluating HRQOL outcomes in cancer clinical trials—does HRQOL evaluation in prostate cancer research inform clinical decision making? J Clin Oncol (2003) 21:3502–11.
(11) Bottomley A, Efficace F, Thomas R, Vanvoorden V, Ahmedzai SH. Health-related quality of life in non-small-cell lung cancer: methodologic issues in randomized controlled trials. J Clin Oncol (2003) 21:2982–92.
(12) Buchanan DR, O'Mara AM, Kelaghan JW, Minasian LM. Quality-of-life assessment in the symptom management trials of the National Cancer Institute—supported community clinical oncology program. J Clin Oncol (2005) 23:591–8.
(13) Sridhara R, Chen G, Chi GYH, Griebel DJ. Evaluation of health-related quality-of-life measures in oncology drug product applications: issues and concerns. J Biopharm Stat (2004) 14:23–30.[CrossRef][Medline]
(14) Sprangers MA, Aaronson NK. The role of health care providers and significant others in evaluating the quality of life in patients with chronic disease: a review. J Clin Epidemiol (1992) 45:743–60.[CrossRef][Web of Science][Medline]
(15) Holzner B, Kemmler G, Sperner-Unterweger B, Kopp M, Dunser M, Margreiter R, et al. Quality of life measurement in oncology—a matter of the assessment instrument? Eur J Cancer (2001) 37:2349–56.[CrossRef][Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  R. Sridhara, J. R. Johnson, R. Justice, P. Keegan, A. Chakravarty, and R. Pazdur Review of Oncology and Hematology Drug Product Approvals at the US Food and Drug Administration Between July 2005 and December 2007 J Natl Cancer Inst, January 29, 2010; (2010): djp515v1 - djp515. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 P. A. Ganz and P. J. Goodwin Health-Related Quality of Life Measurement in Symptom Management Trials J Natl Cancer Inst Monographs, October 1, 2007; 2007(37): 47 - 52. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||