NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Seidenfeld J, Aronson N, Piper M, et al. Uses of Epoetin for Anemia in Oncology. Rockville (MD): Agency for Healthcare Research and Quality (US); 2001 Jun. (Evidence Reports/Technology Assessments, No. 30.)

  • This publication is provided for historical reference only and the information may be out of date.

This publication is provided for historical reference only and the information may be out of date.

Cover of Uses of Epoetin for Anemia in Oncology

Uses of Epoetin for Anemia in Oncology.

Show details



This report is the product of a systematic literature review of controlled trials that compared outcomes of cancer-related anemia managed with and without the use of epoetin. Four groups of cancer patients are addressed: (1) patients with anemia resulting primarily from cancer therapy; (2) patients with anemia resulting primarily from their malignant disease and who may also be receiving cancer therapy; and (3) patients who are anemic as a result of bone marrow ablation prior to (3) allogeneic or (4) autologous stem-cell transplantation. Outcomes of interest include effects on transfusion, symptoms of anemia, and quality of life, as well as adverse events.

This systematic review is limited to controlled trials because many characteristics of cancer patients (e.g., disease stage, tumor progression, concurrent treatments) are likely to affect the outcomes of interest and thus confound the interpretation of the effects of epoetin treatment. Most of the trials included in this systematic review are randomized, but nonrandomized controlled trials are also included. Uncontrolled studies were excluded from this review.

The protocol for this review was prospectively designed to define study objectives, search strategy, patient populations of interest, study selection criteria and methods for determining study eligibility, outcomes of interest, data elements to be abstracted and methods for abstraction, and methods for study quality assessment.

To maximize the accuracy of study selection and data abstraction, two independent reviewers completed each step in this protocol. Disagreements were resolved by consensus of the two reviewers. The protocol provided that disagreements that could not be resolved by the two reviewers would be referred to another of the coauthors of this report for a third review and resolution by agreement of two of three reviewers. Data were abstracted directly into two separate electronic databases and the databases were compared electronically. Comparison of the two databases revealed some substantive differences but also included differences in spelling, capitalization, wording, spacing, and other nonsubstantive differences. It was not possible to quantify only the substantive differences; however, resolution by a third reviewer was seldom required.

This is a review of published evidence, which includes published abstracts from scientific meetings as well as accepted manuscripts in press at the time of this writing and for which permission was granted by the journal or authors for use of a prepublication copy. To supplement published abstracts, where available, we also obtained from the authors slides of presentations and manuscripts in preparation for journal submission. Abstracts and unpublished sources are identified as such in our review.

A supplementary meta-analysis accompanies this systematic review. We conducted a meta-analysis of the effect of epoetin on odds of transfusion in patients with anemia or at risk of anemia due primarily to cancer therapy. The selection of outcomes for meta-analysis is discussed in Chapter 3.

The development of the evidence report and supplementary analysis was subject to extensive expert review. A technical advisory group provided ongoing guidance on all phases of this project. In addition, a preliminary analysis of the evidence base for this report was reviewed by the Blue Cross and Blue Shield Association Medical Advisory Panel. The draft report was also reviewed by a panel of external reviewers that included experts and stakeholders.

The Technical Advisory Group included six members. James Wade, M.D., a medical oncologist, and James Armitage, M.D., a hematologist/medical oncologist, were appointed by the Technology Evaluation Center under the auspices of this task order. In addition, Charles Bennett, M.D., Ph.D., and Michael Gordon, M.D., both medical oncologists, were appointed by the American Society of Clinical Oncology (ASCO) and are members of the ASCO Health Services Research Committee. Alan Lichtin, M.D., a hematologist, and Steven Woolf, M.D., M.P.H., a systematic review methodologist, were appointed by the American Society of Hematology (ASH) and are members of the ASH Committee on Optimization of Hematologic Care (Dr. Bennett is also a member of this committee). A panel appointed jointly by these two organizations will utilize the data reported here as part of the evidence base for preparing guidelines on the use of erythropoietin in patients with cancer or MDSs.

Twenty external reviewers reviewed the study protocol and draft report, and revisions were made based on their comments. Eight reviewers were invited by the Technology Evaluation Center under the auspices of this task order for their expertise in medical oncology, hematology, transfusion medicine, quality of life, and systematic review methodology. One reviewer directs another Agency for Health Care Policy and Research (now the Agency for Healthcare Research and Quality, AHRQ) Evidence-based Practice Center and is a medical oncologist/hematologist. Ten of the external reviewers were appointed by professional organizations other than ASCO or ASH and by patient advocacy groups; these reviewers included clinical and research specialists involved in the treatment of cancer and/or management of cancer-related anemia and patient advocacy representatives.

One external reviewer was from the technical staff of Ortho Biotech, Inc., which markets epoetin alfa for the treatment of nonmyeloid cancer patients receiving concomitant chemotherapy and for the treatment of predialysis patients with progressive renal failure.

An early work product for this report, consisting of a preliminary analysis of evidence, was reviewed by the Blue Cross and Blue Shield Association Technology Evaluation Center Medical Advisory Panel. This interdisciplinary panel comprises experts in technology assessment methods and clinical research and also includes managed care physicians from Blue Cross and Blue Shield and Kaiser Permanente health plans. (Appendix A lists the members of the Technical Advisory Group, external expert reviewers, and the Blue Cross and Blue Shield Association Technology Evaluation Center Medical Advisory Panel.)

Search Strategy for the Identification of Articles

A comprehensive literature search was performed that attempted to identify all publications of relevant controlled trials (see the section Selection Criteria, Types of Studies). The search process began with the MEDLINE, CancerLit, and EMBASE databases. These online sources were searched for all articles published since 1985 that included at least one of the following textwords (tw) or MeSH® terms in their titles, their abstracts, or their keyword lists:

  • erythropoietin (MeSH®)
  • epoetin alfa (MeSH®)
  • erythropoietin (tw)
  • epoetin (tw)
  • Epogen (tw)
  • Procrit (tw)
  • Eprex (tw)
  • Marogen (tw)
  • Recormon (tw)
  • epo (tw)
  • Anemia/drug therapy (MeSH®; included all subheadings)
  • Anemia/therapy (MeSH®; included all subheadings)
  • Anemia/diet therapy (MeSH®; included all subheadings)

The search results were then limited to include only those articles that were indexed under the MeSH® terms "neoplasms" or "myelodysplastic syndromes" (including all subheadings) and that addressed studies on human subjects. The MEDLINE, CancerLit, and EMBASE databases were last searched in December 1998. Total retrieval through this date is 2,915 references.

To supplement the above strategy, issues of Current Contents on Diskette and issues of Medscape Oncology, an electronic medical journal (, were searched through October 30, 1999, to identify recently published articles that had not yet been indexed by the online databases. The resulting citations were compared with those in the primary bibliographic database to identify studies not cited in the MEDLINE, CancerLit, and EMBASE searches. We also searched abstracts presented at the 1999 meeting of the American Society of Clinical Oncology.

Additional bibliographic information and reprints of clinical studies were provided by Ortho Biotech, Inc., the pharmaceutical company that markets epoetin for use in oncology patients. Finally, all relevant review articles, editorials, and letters published in 1995 or later were retrieved. Reference lists from these articles were searched for studies not identified by the above methods.

A total of 28 additional published reports were identified by supplementary searches for a total retrieval of 2,943 references considered for this review.

Selection Criteria

Inclusion Criteria for Types of Studies

  1. The primary study selection criterion required that studies be designed as controlled trials comparing the outcomes of managing anemia with and without the use of epoetin in one of the three patient populations of interest.
    • In these trials, epoetin treatment (with transfusion used as necessary) was always compared with RBC transfusion alone. There were no trials that compared epoetin with any other alternative.
    • All randomized controlled trials relevant to the question and populations of interest were included in this systematic review.
    • Studies that used nonrandomized concurrent or historical controls were included if the reviewers could determine that patients included in the treatment and control groups were comparable. (See "Methods of the Review.")
    • Nonrandomized trials are identified as such in the tables and text, and were considered to be of lesser quality than randomized controlled trials.
  2. The minimum sample size for inclusion in this systematic review was at least 10 similarly treated evaluable patients in each arm, relevant stratum, or epoetin dose level, as applicable.

Exclusion Criteria for Types of Studies

  • All uncontrolled studies were excluded from this review.
  • Nonrandomized studies were excluded if sufficient information to determine comparability was not reported or if obvious selection bias was detected.
  • All otherwise eligible studies that did not meet the minimum sample size were excluded from this review.

Reports Published in Languages Other than English

Our literature search identified 70 titles and/or abstracts of reports that were published in languages other than English, including Japanese, Italian, French, Czech, Polish, German, and others. All 70 of these articles were retrieved in full copy. There were no controlled trials among them. In addition, for many of these, we also found papers published in English by some of the same authors, which confirmed our judgement that the criteria for inclusion in this systematic review were not met.

We then consulted with experts in the field, asking if they were aware of any controlled trials published only in languages other than English. We selected these experts based on their authorship of clinical studies on epoetin conducted in European, Latin American, or Asian countries but published in English-language journals. Additional contacts were selected based on their participation on international committees of either ASH or ASCO. None of the experts we consulted was aware of any controlled clinical trials conducted abroad and published only in a non-English language journal.

Types of Participants

All patients included in this systematic review had or were at risk of having cancer-related anemia. Grading systems for severity of anemia by Hb level were summarized in Table 1 in Chapter 1, Introduction. Malignancies were categorized as "solid tumors," "leukemia," "lymphoma," "multiple myeloma," or "myelodysplastic syndrome." Trials that compared outcomes of anemia management with or without epoetin were included if they enrolled patients of any age who had a malignancy and who had anemia from radiation therapy and/or chemotherapy for their disease; or were at risk for anemia from radiation therapy and/or chemotherapy for their disease, or had anemia caused by their underlying disease, or had been treated with a bone marrow ablative procedure and stem-cell transplant for their disease. In addition, specific criteria were used to classify and select studies relevant to each patient group.

Studies of Patients with Anemia Resulting Primarily from Cancer Therapy

We included trials that met either of the following two criteria:

  • Explicitly stated that the study objective was to investigate the effects of epoetin on treatment-related anemia; or
  • Enrolled only patients undergoing conventional-dose cancer therapy (nonmyeloablative chemotherapy and/or radiotherapy).

In some of these trials, patients had undergone prior therapy regimens and were already anemic; in others, patients had normal Hb levels prior to cancer therapy. Most trials defined an Hb cutoff value for enrollment; for the few that did not, cutoff values were inferred from the data provided. More specifically, these trials included:

  • Trials restricted to anemic patients enrolled patients with Hb values below a specified cutoff.
  • Three studies of anemic patients that also enrolled patients with Hb values above the specified cutoff, when they had an Hb decrease of a specified magnitude (1.5 g/dL) during a specified time period (4 to 8 weeks) prior to enrollment.
  • Trials including nonanemic patients that enrolled patients with Hb values above a specified cutoff.

In addition, when trials included patients with a prior history of radiation therapy and/or chemotherapy, we abstracted details of the prior therapy.

Trials were included whether or not they reported on ruling out any treatable causes of anemia prior to enrollment. When trials reported on ruling out treatable causes of anemia, we abstracted details of the treatable causes ruled out.

Studies of Patients with Anemia Primarily of Malignant Disease

We included trials that met both of the following two criteria:

  • Trials that enrolled anemic patients regardless of whether patients were receiving cancer therapy; and
  • Trials that ruled out one or more other treatable causes of anemia (e.g., iron, vitamin B12, and folate deficiencies; occult bleeding; hemolytic anemia) prior to patient enrollment.

Studies that enrolled patients with a prior history of radiation therapy and/or chemotherapy were included, and details of the prior therapy were abstracted.

We excluded trials that did not report ruling out at least one treatable cause of anemia prior to enrollment. We abstracted data on which of the following causes of anemia were ruled out: iron deficiency; vitamin B12 deficiency, folate deficiency, internal bleeding, hemolytic anemia; or whether all other causes of anemia were ruled out, without stating which specific causes were ruled out.

Studies of Patients with Anemia Resulting from Bone Marrow Ablation

We included trials that met all three of the following criteria:

  • Epoetin was administered beginning with stem-cell reinfusion and continuing through recovery of hematopoiesis; and
  • All patients received the same transplant procedure (i.e., autologous bone marrow, autologous peripheral blood stem cells, allogeneic bone marrow; or allogeneic peripheral blood stem cells); or outcomes were reported separately for subgroups defined by type and source of stem cells; and
  • Studies restricted entry to patients with malignant conditions; or reported outcomes separately for patients with malignant or nonmalignant conditions (e.g., aplastic anemia); or patients with nonmalignant diseases represented fewer than 10 percent of the patients in each study arm.

We excluded studies that reported only outcomes aggregated for patient groups that were mixed with respect to source and type of stem cells.

Studies of Groups of Cancer Patients Outside the Scope of this Systematic Review

Studies of the following uses of epoetin in cancer patients were outside the scope of this systematic review and were excluded:

  • Short-term preoperative treatment to correct anemia.
  • Short-term preoperative treatment to support collection of autologous blood prior to cancer surgery.
  • Mobilization of stem cells into the peripheral blood in preparation for harvesting stem cells for subsequent transplant.

Subpopulations of Interest

For patient Groups 1 (anemia primarily due to cancer therapy) and 2 (anemia due primarily to malignant disease), we sought and abstracted data that would permit analysis of outcomes in the subpopulations of interest listed below. Very few studies indicated that patients were randomized separately by subpopulation (stratification).

  • Type of malignancy: hematologic malignancies (leukemias, lymphomas, myeloma, etc.), tumors of solid organs and tissues (carcinomas, sarcomas, etc.).
  • Regimens: chemotherapy regimens with platinum, chemotherapy regimens without platinum, chemotherapy alone, radiation therapy alone, radiation therapy with chemotherapy.
  • Age: pediatric, adult, geriatric patients.
  • Transfusion history: prior transfusion, no prior transfusion.
  • Iron supplementation: iron supplement, no iron supplement.

We also sought and abstracted data on predictors of response to epoetin treatment (e.g., baseline serum epoetin or the ratio of observed to predicted epoetin level).

Types of Interventions

We required that studies be designed as controlled trials comparing the outcomes of managing anemia with and without the use of epoetin. Epoetin is intended to prevent the occurrence of anemia that is so severe that RBC transfusion becomes necessary. All trials that met the study selection criteria for this systematic review compared epoetin plus RBC transfusion as necessary to RBC alone. Red blood cell transfusion was initiated at a predefined Hb threshold (usually 7 to 9 g/dL) or at the discretion of the treating physician.

We identified four characteristics of epoetin administration that might affect treatment outcomes: route of administration, starting dose, class of dosing regimen, and duration of treatment as detailed below. Few studies that met the selection criteria for this systematic review directly compared the effectiveness of various doses, dosing regimens, treatment durations, or routes for administering epoetin. For this reason, we attempted indirect comparison of outcomes of the various characteristics of epoetin administration across the studies included in this systematic review.

  • Route of administration was subcutaneous or intravenous.
  • To facilitate comparisons among studies, all reported dosages were calculated as units per kilogram per week. All studies were classified within a range of starting doses. Where available, we report both starting dose and ending dose for each study.
  • Dosing regimens were classified as fixed dose with continuous treatment, decreasing dose, and increasing dose.
  • To permit comparison among studies, treatment duration was classified by ranges (e.g., <10 weeks, 12 to 16 weeks).

The three classes of dosing regimens were defined as follows:

  • Fixed and continuous dose. All patients treated with epoetin in a trial with a fixed and continuous regimen received the same dose throughout the study.
  • Decreasing dose. Studies that utilized a stop/start regimen temporarily discontinued epoetin when Hb levels rose above a predetermined threshold and then resumed treatment when Hb levels fell below a second threshold. Other trials decreased the epoetin dose only for responding patients, seeking to maintain Hb levels within a range specified in their protocol. The stop/start and decreasing dose regimens are considered together in this evidence report ("decreasing dose"), since either approach reduces the amount of epoetin given to responding patients.
  • Increasing dose. Patients who did not respond to the initial dose by a specified time had the dose increased, most often by a factor of two.

Types of Outcome Measures

Trials were included if they reported at least one of the following outcomes, each of which were compared and analyzed separately:

  • Hematologic outcomes
    - Change in Hb from baseline to final value after epoetin treatment
    - Number and/or percentage of patients responding to epoetin (each trial's definition of complete and partial response also was abstracted)
  • RBC transfusion outcomes
    - Number and/or percent of patients transfused
    - Number of RBC units transfused per patient (duration of the measurement period also was abstracted)
  • Recovery of hematopoiesis after stem-cell transplantation
    - RBC engraftment according to achievement of predefined Hb level without transfusion support
    - Days to RBC engraftment as defined by attainment of a predefined Hb level
    - Reticulocyte measures of engraftment
    - Platelet transfusions (for stem-cell transplant studies)
  • Anemia-associated symptoms (e.g., shortness of breath, dyspnea on exertion, angina, etc.)
  • Days in hospital
  • Functional status
  • Quality of life using any measure or instrument
  • Other outcomes
    - Regimen-related morbidity
    - Survival
    - Growth and development (pediatric)

Adverse Events

Data on adverse events were abstracted only from included studies that reported the percentage or numbers of patients experiencing specific epoetin-related adverse outcomes compared with a control group that did not receive epoetin treatment. Studies that only provided a general statement that epoetin was well tolerated were excluded from the analysis of adverse events.

The number of enrolled patients experiencing specified adverse events was abstracted exactly as reported by study authors. No attempt was made to stratify according to severity, since few studies presented information on severity. If studies simply reported the total number of patients experiencing epoetin-related adverse events, this was also abstracted.

Abstraction and analysis of data on adverse events present particular difficulties. The difficulties encountered in this project are representative of the general problem of the limitations of clinical trials as a source of data on adverse events. One well-recognized problem is that some adverse events may be so infrequent that clinical trials are not large enough to capture events that may be of concern when the treatment is used in the general population of patients. A second problem is inconsistency in which adverse events are reported and how they are measured. Efforts to improve standards in reporting of randomized trials have emphasized the need for more thorough and systematic reporting of the spectrum of effects for an intervention (McPeek, Gilbert, and Mosteller, 1980).

Randomized controlled trials are widely recognized as an incomplete source of data on adverse events because of the difficulty in capturing infrequent but serious events that may present a substantial risk when the intervention is used in large populations. However, analysis of uncontrolled patient series for epoetin-related adverse events would lack controls for several confounding factors, such as disease progression and cancer therapy, that independently result in similar adverse events. Presently there is a substantial history of experience in the ESRD population to characterize the serious adverse effects of epoetin, of which the risk of hypertension and other vascular effects are well recognized. Our objective was to estimate the frequency of occurrence in the oncology setting of the common adverse effects of epoetin.

Costs and Other Economic Outcomes

Our protocol required that data on financial costs were to be abstracted only from studies that reported on a group treated with epoetin and a control group managed without epoetin and met all study inclusion criteria for this systematic review. In addition, the studies were required to report costs for all patients treated (not just those who develop anemia) and capture (or attempt to capture) the full range of costs for patient care over the entire course of treatment (e.g., not just compare the drug costs for epoetin with the costs of the units of blood transfused).

No trials meeting the inclusion criteria for this systematic review reported cost data. As a result, our review of the literature on epoetin costs is limited to a summary of secondary analyses that are discussed in the introductory section of this systematic review.

Methods of the Review

Determining Study Eligibility

The following procedure was followed to systematically screen citations, select those to be retrieved, and identify those meeting the study inclusion/exclusion criteria. Initially, one of two independent reviewers evaluated each title and abstract against the study inclusion and exclusion criteria. Each reviewer was responsible for one-half of the retrieved citations. The reviewers sorted citations into three categories: "retrieve," "hold," and "uncertain." Full copies were obtained of all articles sorted into the "retrieve" category by either reviewer. Next, each reviewer evaluated all citations sorted into the "hold" category by the other reviewer. Full copies also were obtained of any reference initially sorted as "hold" which the second reviewer sorted as "retrieve." The two reviewers discussed all references sorted as "uncertain," with a bias toward being inclusive.

After articles were retrieved, each reviewer evaluated all articles against the inclusion and exclusion criteria. Articles were labeled as "included" or "excluded," and a code corresponding to the basis for exclusion was recorded in the bibliographic database. Table 5 lists the codes and reasons for exclusion. The results of the two independent reviews were compared, and any disagreements were resolved by consensus.

Table 5. Reasons for study exclusion -- oncology.


Table 5. Reasons for study exclusion -- oncology.

The resulting bibliography of included studies was circulated to the Technical Advisory Group and to the members of the ASH/ASCO Joint Guidelines Panel on Erythropoietin for review for possible omissions. No additional studies meeting the inclusion criteria were identified.

Data Abstraction

Two reviewers independently abstracted data from each eligible study, recording the information with electronic database software (Microsoft® Access 97). The data elements that were abstracted are listed in the data abstraction forms and defined in the accompanying table (Appendix C).

Data elements were grouped into the following broad categories: trial identifiers, study design and methods (including enrollment and withdrawal numbers), patient characteristics, outcomes, and predictors of response.

If an article did not report exact numerical values for one or more of these elements, the reviewers estimated them from figures if they were available in the published reports. After each reviewer completed data abstraction, databases were compared electronically and disagreements were resolved by consensus of the two reviewers, or by a third reviewer if required.

If p values were not reported, but sufficient information was provided, we calculated p values using chi-square analysis or Fisher's exact test (when one or more cells contained five or fewer data points).

Assessment of Comparability of Study Arms

Comparability of study arms was assessed for two purposes. Nonrandomized studies were excluded if the minimum set of data elements we required to determine comparability was not reported or if obvious selection bias was detected. For randomized trials, most of which were small, we sought to identify differences in study arms that might affect the interpretation of results or that might explain heterogeneity of results among studies. Any such findings are reported in the review of evidence.

Each trial was evaluated by two reviewers for comparability of study arms based on the data reported. We required a minimum set of specific elements to assess comparability of study arms. This set included patient age, tumor type, and baseline Hb value for studies of the anemia resulting from cancer therapy and the anemia caused by malignant disease; and patient age, tumor type, and information on the conditioning regimens used in each study arm for studies of anemia after stem-cell transplant. Other elements evaluated related to the severity of anemia, the severity of malignant disease, previous cancer therapy, or cancer treatment while on study.

Note that uncontrolled hypertension is a contraindication for use of epoetin alfa in the FDA-approved labeling for the drug. In addition, many of the trials that met inclusion criteria for this systematic review excluded patients with other comorbidities (e.g., cardiovascular disease, major organ dysfunction) that might affect either the decision to transfuse patients or the frequency of adverse events. Consequently, it was judged unlikely that imbalances between study arms with respect to comorbidities might affect the interpretation of results.

For studies that did not report statistical comparison among study arms, we compared the data elements reported. No studies lacking statistical comparison reported sufficient data to perform a statistical test; so we simply estimated equivalence from the raw numbers or percentages reported. If insufficient data were provided to assess overall comparability (e.g., insufficient elements reported, patient information not separated by study arm, or study published as an abstract only), this was noted.

The number of studies that reported each specific element also was compiled; we also noted which data elements were omitted from the studies for which there was inadequate information to assess overall balance between arms.

Classification of Alternative Hemoglobin Thresholds for Initiating Epoetin Treatment

Our systematic review protocol called for grouping studies by upper limit of Hb level at study entry, as reported in the patient eligibility criteria of each study. The thresholds of interest were: Hb >12 g/dL; Hb >10 and <12 g/dL; Hb <10 g/dL or requiring blood transfusions; and prevention studies. Prevention studies are those studies that enrolled patients whose Hb was above a minimum threshold but without an upper limit of Hb for enrollment. Upon examining the abstracted data, we found that the mean baseline Hb at study entry was generally lower than our initial classification would suggest. As a result, we revised our protocol and grouped studies by mean baseline Hb at study entry.

Our comparison of the studies selected for Part I of this systematic review (Chapter 3, Anemia Resulting Primarily From Cancer Therapy) illustrates why we found mean Hb level at entry to be more informative for grouping the included studies than Hb cutoffs for enrollment. Among these studies, the mean or median Hb level at entry was <10 g/dL in the preponderance of studies (9 of 10 trials, with 1,134 of 1,164 enrolled patients) that we had initially classified as setting upper limits for eligibility between 10 and 12 g/dL. Similarly, the mean or median Hb level was >10 and <12 g/dL in the majority of studies (5 of 6 trials, with 338 of 386 enrolled patients) that set an upper limit for eligibility >12 g/dL.

This systematic review classifies included studies into three categories defined by mean Hb at enrollment: Hb >12 g/dL; Hb >10 and <12 g/dL; and Hb <10 g/dL. We also abstracted and reported standard deviations, where reported. If the standard deviation was not reported, we abstracted the range, if reported. In the few studies where mean Hb was not reported, we used the median Hb. Where there was a discrepancy in mean Hb between the epoetin and the control arms, we classified the trial by the Hb level of the epoetin arm.

Quality Assessment for Sensitivity Analysis

The objective of quality assessment for this systematic review was to identify a group of "higher quality trials," for purposes of sensitivity analysis. Our meta-analysis includes a quantitative sensitivity analysis; and throughout this systematic review, we have included qualitative sensitivity analyses in our summaries of study conclusions. Our sensitivity analyses compare the results reported and conclusions reached from all included studies to results and conclusions drawn by examining the outcomes of only higher quality studies.

Sensitivity analysis based on study quality is useful because trials of lower quality generally overestimate the effectiveness of an intervention compared with higher quality trials. More than two decades ago, Chalmers showed that randomized studies report smaller treatment effects than nonrandomized studies (Chalmers, Smith, Blackburn, et al., 1981). Subsequently, many methodologists have attempted to identify the characteristics that define the quality of randomized trials and to test whether such characteristics have an effect on study results (Schulz, Chalmers, Hayes, et al., 1995).

Although many quality scales have been used to assess the quality of randomized controlled trials, there is a dearth of empirical evidence to validate such scales. Indeed, Juni and colleagues recently illustrated the hazards of using summary quality scores to select or pool studies for meta-analysis (Juni, Witschi, Bloch, et al., 1999). They identified 25 different quality scales, which they tested for a meta-analysis of 17 trials comparing low molecular weight and standard heparin. No significant association between summary quality scores and treatment effects was found; and the results of different quality scales yielded different conclusions as to which treatment was superior.

Although the use of quality summary scores is problematic, there are three domains of study quality that have been tested in empirical studies. These are concealment of treatment allocation during randomization, double-blinding, and handling of withdrawals and exclusions. Although there is evidence suggesting that these quality domains are associated with more valid estimates of treatment effects, all three domains have not been reported as significant in all studies (Juni, Witschi, Bloch, et al., 1999; Moher, Pham, Jones, et al., 1998; Mulrow and Oxman, 1997; Schulz, Chalmers, Hayes, et al., 1995). Moreover, assessment of study quality generally depends on information reported in journal articles, and the absence of such information may reflect incomplete reporting rather than flawed study design. This point is especially germane to studies published prior to the Consolidated Standards of Reporting Trials (CONSORT) statement, which was published in the Journal of the American Medical Association in 1996, to disseminate a standard for completeness of reporting in journal articles (Begg, Cho, Eastwood, et al., 1996).

In an editorial accompanying the Juni study, Berlin and Rennie (1999) suggest that, to be clinically relevant, quality assessment of trials should focus on key aspects of research design relative to the outcomes of interest. Thus, where an outcome requires subjective judgment, double-blinding may be of paramount importance but may matter less for outcomes where there is little discretion regarding assessment or interpretation. In this systematic review of use of epoetin in cancer-related anemia, we had substantial concerns about the impact of subjective judgments on treatment outcomes. The clinical outcomes of interest (transfusion, quality of life, fatigue and other anemia-related symptoms) can obviously be affected by physician behavior and patient perceptions.

For example, transfusion is the most commonly reported of these clinical outcomes. We anticipated that physicians might be more aggressive in transfusing patients they knew were in the control arm and less aggressive in transfusing those they knew were in the epoetin arm. One study included in this systematic review reported relevant data (Thatcher, De Campos, Bell, et al., 1999). In this unblinded study, there was variation among participating centers in the mean Hb at which first transfusion was initiated, and the range was markedly lower for the higher dose than the lower dose epoetin group (8.7 to 10.8 g/dL versus 8.4 to 12.9 g/dL). Many studies included in this systematic review had a transfusion trigger, i.e., a prespecified Hb level at or below which transfusion was initiated, intended to achieve consistency in transfusion practice. However, these studies rarely reported the mean Hb level at which transfusion actually occurred, so it is impossible to confirm whether transfusion practice was, in fact, consistent across study arms.

In developing our approach to sensitivity analysis, we put highest priority on comparing the outcomes of double-blinded randomized controlled trials to all other trials of weaker design with respect to minimizing the effects of subjective judgements on treatment outcome. While a high proportion of trials included in this systematic review were randomized, the preponderance of these were unblinded. We also assessed study quality with respect to the domains of allocation concealment and handling of exclusions and withdrawals. But in defining higher quality studies, we wanted to avoid criteria that might be overly sensitive to how a trial was reported, thus diminishing the pool of double-blinded randomized controlled trials for sensitivity analysis. For the domain of handling of exclusions and withdrawals, we set a stringent cutoff for the maximum number of patients that could be excluded from the analysis of results. Although we looked for, and often found, an explanation of exclusions and withdrawals, this was not required. As was the case in a prior evidence report for the AHRQ, information on concealment of allocation was reported infrequently (Aronson, Seidenfeld, Samson, et al., 1999) and was not required for our group of higher quality studies.

Criteria to Define Higher Quality Trials for the Sensitivity Analysis

Studies that met all three criteria below were included in the group of higher quality trials for purposes of sensitivity analysis:

  1. The study was a randomized controlled trial.
  2. The study was double blinded.
  3. At least one of the following conditions was true:
    1. Fewer than 10 percent of subjects within each study arm were excluded from the analysis and the percentage of subjects excluded from analysis in each arm was fewer than a 2:1 ratio; or fewer than 5 percent of subjects were excluded in each study arm; or
    2. If more than 10 percent of subjects were excluded from the analysis in any of the study arms, results were reported as an intention to treat analysis.

A study was classified as double blinded if stated as such in the publication without further description of the method of blinding and the study used a placebo. If a placebo was used, but there was no mention of double-blinding, the study was classified as single blinded. If a placebo was not used, or if there was no mention that a placebo was used, or if it was stated that the study was unblinded, the study was classified as unblinded.

"Excluded from the analysis" refers to all patients who were enrolled in the study but were not included in the analysis of results. Subjects excluded from the analysis are those not included in the results for any reason, including: not randomized, withdrawn after randomization, lost to followup, or with missing data. In our evidence tables, the number of excluded subjects for each study equals the number of enrolled patients minus the number of evaluable patients.

Quality Domains Assessed but Not Required

The following domains were assessed but not required for a study to be included in the group of higher quality trials for purposes of sensitivity analysis.

  1. Concealment of allocation
    • Was the initial allocation of patients to different treatment arms concealed from the subjects and investigators? (This could be achieved in a number of ways including a central randomization site or using opaque envelopes if one of the investigators is involved in randomization.)
  2. Explanation of reasons subjects were excluded from the analysis of results
    • Were the reasons that patients were excluded from the analysis of study results described in sufficient detail and separately for each treatment arm?

We found that, compared with studies of cancer chemotherapy, studies of epoetin provide less detail on patients who withdrew because they stopped chemotherapy. Epoetin studies generally report that the patient stopped chemotherapy without distinguishing among death, relapse, or refused further treatment. We answered yes to the above question whether or not details of chemotherapy were provided but also noted in our data abstraction which studies specified why chemotherapy was stopped.

Study Features to Control for Specific Confounders

The presence of the following study design features to control for confounders of treatment effect are relevant to the clinical setting of epoetin studies. These features were assessed but were not required for a study to be included in the group of higher quality trials for purposes of sensitivity analysis.

  1. Unbiased decisions to transfuse
    • Did studies that reported transfusion outcomes either specify an Hb level ("trigger") above which patients did not receive RBC transfusions and below which transfusions were always given; or
    • Did studies provide data to demonstrate that the mean or median Hb/Hct at transfusion was comparable for all study arms.
  2. Underestimation of epoetin effects because of presence of other causes of anemia
    • Were all of the following causes of anemia ruled out in patients who were anemic: iron deficiency, vitamin B12 deficiency, folate deficiency, occult bleeding, and hemolytic anemia?
    • For studies that examined the effect of epoetin on hematologic response, did the study either (a) verify the adequacy of iron status during the course of the study by measuring serum iron and transferrin saturation and report on the results of these tests and their implications for study outcomes, or (b) supplement patients with iron? In the latter event, supplementation of epoetin-treated patients alone was considered sufficient. Oral (as opposed to intravenous) iron supplementation is acceptable for cancer patients.
  3. Effects of patients' knowledge of Hb values on quality-of-life assessment
    • Were patients blinded as to their value of Hb or RBCs counts prior to administering the quality-of-life assessment instrument?

No study blinded patients to Hb values or red cell counts prior to administering a quality-of-life measurement instrument. Such blinding could be accomplished by having patients complete the questionnaires prior to informing them of current Hb levels. However, it is also possible that the patients' knowledge of prior Hb values might also influence response to the questionnaire.

Study Sponsorship

As a separate issue from assessment of study quality, the included trials were classified by source of research support. Using the acknowledgements of support or provision of study drug in published papers from each study and/or institutional affiliation of authors, the trials were categorized as having been funded by one of the following (See Appendix B):

  • Research grants from government or other nonprofit agencies only;
  • Pharmaceutical manufacturers only;
  • Pharmaceutical manufacturers and nonprofit agency research grants; or
  • No sponsorship reported.

We reviewed these results to see if any relationship between study sponsorship and study findings was apparent. The evidence does not suggest that the source of research support had a substantial impact on the outcomes reported from these studies (see Appendix B). However, the very small number of studies without support from pharmaceutical manufacturers limits the ability to detect a relationship between source of funding and outcomes.


We conducted a meta-analysis of the effect of epoetin on odds of transfusion in patients with anemia or at risk of anemia due primarily to cancer therapy. (For a discussion of the selection of the outcome for meta-analysis, see Chapter 3. We combined data from trials of epoetin therapy in patients with different types of cancer and receiving different types of treatment regimens. Because epoetin treatment affects anemia, a common consequence of cancer or cancer therapy, rather than cancer outcomes, pooling results on a variety of patient types provides outcomes that are generalizable across a broad spectrum of cancer patients.

Most meta-analyses are performed on a group of studies with a common endpoint. The assumption is often made that these studies all estimate the same parameter, such as an odds ratio, and the analysis is referred to as a fixed effects analysis. The opposite of a fixed effects model is a random effects model. The random effects model produces estimates that are more conservative than those produced by fixed effects models. The idea of a random effects model is that the parameter sampled from "mother nature" does not remain constant from study to study. Instead, it varies randomly, and is in fact a random variable sampled from some distribution. The problem then is to estimate the center of the distribution of the parameter of interest and the variance of the distribution. This methodology is especially appropriate for studies of epoetin therapy for anemia in cancer patients because of the differences in epoetin dose, patient Hb level at study entry, length of followup, and study quality in each study.

Random effects models differ from fixed effects models in that a measure, v, of the variation between studies is included in computation of the total uncertainty used to compute weights for each estimate. One conventional measure of v is
Image f3687_EQU001.jpg
where X2 is the usual chi-squared measure of heterogeneity for the m studies and where wj = 1/vj, and vj is the variance of the estimated odds ratio from study j. If the value of v is computed to be negative, it is usually set to zero. The random effects weighted mean odds ratio is just
Image f3687_EQU002.jpg
where αj is the estimated odds ratio from study j, and wj* = 1/[vj + v]. The variance of the weighted mean odds ratio in the random effects model is
Image f3687_EQU003.jpg

Since v is usually larger than zero, each wj* is usually larger than the corresponding fixed effects weight wj, and so the variance of the random effects weighted mean is usually larger than the variance of the fixed effects weighted mean. There are several methods for obtaining estimates of v, including some described by DerSimonian and Laird (1986) and Hedges and Olkin (1985). The method described by Hedges and Olkin (1985) is an empirical Bayes estimator and is the one used in this analysis. This particular estimator works well for as few as two studies, and if the studies are homogeneous, the estimates approach those of the fixed effects model. The calculations were carried out using the FAST*PRO software as described by Eddy and Hasselblad (1992). A test for heterogeneity was carried out according to DerSimonian and Laird (1986).

For some problems, one cannot assume that the effect of the treatment comes from the same population for each study. For example, the effect of treatment may actually depend on dose. These kinds of problems require the fitting of more complicated models known as mixed models, and several mixed model procedures have become available in the last several years. One of the earliest of these was the multiple logistic regression model with a random effects term. This model is part of the Egret software (Egret, 1996). The model can be written as
Image f3687_EQU004.jpg
where pik(x) is the probability of a transfusion event for the kth arm of the ith study, αj is a term for the log-odds of transfusion in the control group of the ith study, xij =1 if i equals j and xij equals 0 otherwise (i.e., xij matches the correct value of α for each study). β is the logistic regression coefficient for the effect of the epoetin dose in the ith study, xi, m+1 is the actual epoetin dose for the kth arm of the ith study, ϵ is a standard normal random variable, and α2 is the random effects variation. The terms are estimated using maximum likelihood methods.

Each study was treated exactly as it was designed; e.g., a three-arm study contributed three records. Each study had its own dummy variables to allow for study differences, and the effects of different doses were modeled using appropriate regression terms. Thus, the assumption of independence was not violated. This analysis is relatively complicated and is described in detail by Hasselblad (1998).


  • PubReader
  • Print View
  • Cite this Page

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...