The Oncologist, Vol. 13, No. 5, 596-607, May 2008; doi:10.1634/theoncologist.2007-0219 © 2008 AlphaMed Press
Health Economic Evaluation of Treating Anemia in Cancer Patients Receiving Chemotherapy: A Study in Belgian HospitalsaIMS Health, Brussels, Belgium; bGhent University, Ghent, Belgium Key Words. Anemia • Cancer • Chemotherapy • Economic • Erythropoiesis-stimulating agents • Transfusion Correspondence: Erik Spaepen, M.Sc., IMS Health, Rue de Crayer 6, 1000, Brussels, Belgium. Telephone: 32-2-627-3211; Fax: 32-2-627-3334; e-mail: ESpaepen{at}be.imshealth.com; Erik219{at}yahoo.com Received November 7, 2007; accepted for publication March 28, 2008. Disclosure: L.A. has performed contract work for Ghent University. N.D. has performed contract work for IMS Health. E.S. is an employee of IMS Health and has worked as a contractor for Amgen regarding erythropoietins. The study was performed by IMS for Amgen and funded by an educational grant from Amgen. The authors acknowledge that E.S. and N.D. received payment from Amgen for writing the article.
Background. Erythropoiesis-stimulating agents (ESAs) are used in chemotherapy-induced anemia (CIA) with the goal of improving quality of life and preventing RBC transfusions. This retrospective database study compared the three currently available ESAs, epoetin alfa (EPO-A), epoetin beta (EPO-B), and darbepoetin alfa (DARB), regarding costs and outcomes. Methods. Data were obtained from a Belgian longitudinal database, including medical and financial data on cancer patients receiving chemotherapy and ESAs, submitted by 46 Belgian hospitals. Propensity score matching was applied to correct for selection bias. The main effectiveness parameter was defined as transfusion- and anemia-readmission-free survival (TA-free survival) at 3 months. Costs were analyzed taking the health care payer perspective.
Results. Including 1,584 EPO-A, 380 EPO-B, and 429 DARB propensity-matched patients, TA-free survival rates were similar for the three groups (DARB, 84.37%; EPO-A, 84.60%; EPO-B, 84.94%). Overall inpatient costs were Conclusion. To our knowledge, this is the first real-life matched retrospective study comparing ESAs with regard to both costs and effects. For similar patient profiles, the patients in the DARB group consumed the smallest amounts of ESAs, with similar clinical outcomes. These data therefore suggest a greater efficiency of DARB in the treatment of CIA.
Anemia is a frequent complication of cancer and/or the treatment of cancer, with up to two in three cancer patients having a serum hemoglobin (Hb) level below the normal values at some point during their disease course [1]. The incidence and prevalence of anemia are especially high in patients receiving chemotherapy [1, 2]. The presence and severity of anemia vary as a function of cancer type, cancer stage, and chemotherapy type, duration, and intensity [2, 3]. Chemotherapy-induced anemia (CIA) is associated with a substantial burden in terms of direct medical costs, with inpatient services accounting for a substantial proportion of the high costs [4]. CIA, however, also has a financial impact, through indirect costs, on the patients and their caregivers, as well as society in general [5, 6]. Furthermore, CIA has a detrimental effect on the patient's quality of life (QoL) [7, 8]. Improvements in Hb levels have been linked to reductions in fatigue scores, as measured by instruments such as the Functional Assessment of Cancer Therapy–Fatigue scale, and significant improvements in psychological outcomes [9]. A meta-analysis of five randomized, controlled trials showed that, even after controlling for clinical and demographic variables, patients whose fatigue improved after treatment for anemia reported greater improvements in energy and overall health than patients who were treated for cancer alone [7, 10]. Different therapies have been used to treat anemia. RBC transfusion was one of the earliest and most widely used interventions for the management of CIA, but it often only leads to a temporary correction of Hb levels and, furthermore, exposes patients to the risk of dangerous blood-borne diseases [11]. Moreover, transfusions are resource intensive and their costs have increased over time [12, 13]. Despite this, RBC transfusion remains the recommended therapy for patients with Hb concentrations <8 g/dl, as a means of providing immediate, albeit temporary, symptomatic management [14]. Erythropoiesis-stimulating agents (ESAs) have been shown to increase Hb levels and reduce the need for RBC transfusions, and are extensively used in anemic patients [15, 16]. The first ESA approved for the management of CIA, recombinant human erythropoietin (rHuEPO), is efficacious and well-tolerated, but requires administration either once or three times a week. Darbepoetin alfa (DARB) has a longer serum half-life and a greater biological activity than the other currently available ESAs, and can be administered less frequently than rHuEPO [17–19]. For CIA, DARB was originally licensed for weekly administration at 2.25 µg/kg, but more recently, every-3-weeks administration of DARB 500 µg was approved in both the European Union and the U.S. [20–22]. Current guidelines recommend that treatment with ESAs be initiated in patients with an Hb level of 9–11 g/dl who are receiving chemotherapy, with the goal of improving QoL and preventing RBC transfusions. Target Hb concentrations of 12–13 g/dl are recommended; if no symptomatic improvement or increase in Hb is observed 4–6 weeks after initiation of ESA therapy, options include an increase in ESA dose, maintenance of ESA dose levels, or cessation of therapy (all possibly combined with transfusions). For patients achieving the target Hb concentration, individualized titration to the lowest effective maintenance dose is recommended [23, 24]. ESAs are perceived as expensive, so a number of health economic analyses have been undertaken to determine their health economic benefits objectively. The results of these analyses vary, depending on the methodology used to determine costs or cost-effectiveness [8, 25, 26], so a rigorous appraisal of these studies is therefore important. Recently, several abstracts and posters have suggested that standard epoetin regimens have greater health economic or outcome benefits than DARB [27–31]. The purpose of the current study was to compare the three currently available ESAs, epoetin alfa (EPO-A), epoetin beta (EPO-B), and DARB, with respect to direct costs and health outcomes in a real-life setting based on data from a large Belgian hospital database.
Data Source Data were obtained from the Minimum Basic Data Set (MBDS), which is also known as Minimale Klinische Gegevens/Résumé Clinique Minimum, to which Belgian hospitals have been required to register case-mix data for each admitted patient since 1991. It includes all in-hospital and day clinic admissions. In the publicly funded Belgian health care system, MBDS reporting is a legal obligation required to fulfill the reimbursement criteria established by the Belgian social security administration (the National Institute of Sickness and Disability Insurance [RIZIV-INAMI]) and to ensure adequate hospital funding by the Federal public health authorities. MBDS data are captured through trusted third parties, which ensures both patient and hospital anonymity. A patient retains a unique identification code through one calendar year, which allows anonymous analysis of the patient during a limited follow-up period. The 2003, 2004 (full years), and 2005 (first semester) MBDS data, the most recent database available at the time of this analysis, include case-mix data from 46 hospitals representing 16,141 beds and 2,467,698 patient stays. The MBDS contains daily basic inpatient data for each patient, including demographic information (e.g., gender and 10- to 15-year age category); data on all conditions diagnosed and inpatient procedures performed, both reported using the International Classification of Diseases, 9th Edition, Clinical Modification (ICD-9-CM) codes [32]; and the all patients refined-diagnosis-related group (APR-DRG) [33]. Only one APR-DRG is assigned per stay and is considered as the reason for hospitalization. The severity of the reason for hospitalization and mortality indices are also recorded. All pharmaceuticals administered to the patient, including the corresponding Anatomical Therapeutic Chemical (ATC) code [34], the pack used, the number of units administered, and the percentage of the cost that was to be reimbursed by the RIZIV-INAMI, were registered in the MBDS.
Patient Data Selection
Cost Calculation The costs of procedures performed during a given hospital stay were based on estimated procedure costs per APR-DRG derived from the Technical Cell (TC), a public technical institution that reports to the Belgian public health authorities and controls MBDS data [35]. The TC gathers data from the majority of hospital stays in Belgium (coverage rate, 92%) and supplies nationwide representative mean procedure costs per APR-DRG and severity index. To estimate the procedure costs in this sample, the estimated mean costs from the TC were linked to the patient data available in the MBDS database using the corresponding APR-DRG and severity index. As the most recent available data from the TC were from 2001, the Belgian annual health price index rates were used to extrapolate 2003, 2004, and 2005 costs from 2001 TC costs [36].
From the costs of procedures, the cost of blood transfusions was analyzed separately. Blood transfusions were detected by the ICD-9-CM procedure codes 99.03 (other transfusion of whole blood) and 99.04 (transfusion of packed cells). The number of transfusions was, by default, set at two units (mean number of transfused units) per admission with any of the relevant codes, with a cost of
Inpatient hospitalization costs were calculated by multiplying the length of stay (LOS) by the estimated daily inpatient hospitalization cost in Belgium. The daily cost was estimated using the average cost per day ( The overall cost per patient was obtained by adding the pharmaceutical product cost, estimated procedures cost, and inpatient hospital stay cost per admission together, then adding up all the stays per patient reported because of the longitudinal nature of the database. Because of the limitation of the observation periods of one calendar year, patients from 2003 and 2004 have a maximum observation period of 12 months, whereas patients analyzed in 2005 have a maximum observation period of only 6 months. To analyze the management of anemia from a health-economic perspective, all RIZIV-INAMI–reimbursed costs incurred by the patient were analyzed on a per-patient level. These analyses represent the overall inpatient costs incurred by an anemic patient with neoplastic disease receiving chemotherapy treatment. Hence, this total cost also includes all nonanemia-related costs during the considered observation period. To distinguish between anemia-related costs and other costs, for example, for chemotherapy, a separate cost analysis was performed on anemia-related costs. The following costs were considered as anemia-related: (a) any ESA used, (b) any blood transfusion performed, and (c) all costs incurred during an anemic admission (APR-DRG, 663). The cost of chemotherapy was excluded from all anemia-related costs, as well as the cost of hospital stays associated with the administration of chemotherapy. For the cost assessment, all costs starting from the index admission (i.e., the first date of ESA use) up to the end of the observation period were considered, in order to present a complete picture of all clinical costs.
Clinical Outcomes
Propensity Score Matching It is then possible to proceed in multiple ways. One option is to divide the total population (groups A and B) into quintiles based on the propensity scores. The analysis then takes into account these subgroups, which include patients from each treatment group with similar characteristics, because they share similar propensity scores. One such model is a hierarchical mixed-effects model, with random intercept effects for the propensity quintiles. If one study group (group A) contains many more patients than the other study group (group B), another way to proceed is to match patients from group A with multiple patients from group B on this propensity score. An example is a greedy 1-to-many (1–N) matching algorithm [41]: the patients from group A, the group with the smallest number of patients, is matched with multiple patients from group B (the group with more patients than group A), with propensity scores that are closest (absolute difference) to that of the group A patient. The algorithm is stated as "greedy," because once a match is created, it is never re-evaluated versus other potential patient matches [42]. The analysis then only takes into account these matched groups (one patient from A with multiple patients from B). This can also be performed with a hierarchical mixed-effects model as above. In the case of 1–1 matching, special matched-pairs analytical techniques can be performed [43]. The DARB patients (n = 539) and the EPO-A patients (n = 1,594) were matched using a greedy 1–N algorithm. In the first step, the two study groups were analyzed in a logistic regression model, where the probability of receiving DARB treatment was modeled using a set of 13 clinically important variables available in the database. The 13 variables used to calculate the propensity scores were: age >65 years, male gender, platinum therapy, i.v. iron use at index stay, RBC transfusion at index stay, severity index at index stay, lung cancer, breast/ovarian cancer, hematological cancer, other metastatic cancer, months on chemotherapy (index stay to end of treatment), hospital transfusion rate (proxy of hospital transfusion policy), and death within 1 month after the index stay. Patients were matched 1–N (one DARB patient with many EPO-A patients using a greedy matching algorithm). The algorithm used was customized to match as many as four EPO-A patients per DARB patient. In cases where the propensity score of the DARB patients was outside the range (minimum–maximum) of the EPO-A group, that is, no overlap in propensity scores, the DARB patients were not retained for analyses. Because the number of patients in the EPO-B group (n = 380) was of the same order of magnitude as the number of DARB patients (n = 539), these patients were matched using the quintiles method: in the first step, the two study groups were analyzed in a logistic regression model, where the probability of receiving DARB treatment over EPO-B was modeled using the same set of 13 variables as above. In the case where there was no overlap in propensity scores between the two groups, these nonoverlapping patients were not retained for the analyses. All patients were then allocated into quintiles of the propensity score. There were five quintile groups with similar baseline properties for DARB and EPO-B patients. To also distribute the EPO-A patients from the first 1–N match, the EPO-A patients were distributed into the same quintile to which their DARB match was allocated. Therefore, the five quintiles each contained patients from the three different study groups, with each quintile containing only patients with similar characteristics. This quintile grouping was then used in statistical analyses to correct for treatment selection bias.
Statistical Analyses Propensity score-corrected costs were analyzed in a hierarchical, mixed-effects model, with the propensity quintiles included as random effects (for treatment selection bias correction) and the three study groups (DARB, EPO-A, and EPO-B) included as fixed effects. The weighted, least-squares estimated mean and standard error of the mean (SEM) of the costs were calculated per study group for each cost of interest: total cost of pharmaceuticals, ESA cost, inpatient hospitalization cost, cost of procedures, and overall cost. Furthermore, from these models, estimates of the differences in costs among the three study groups were calculated. The clinical outcomes, ARFS and TFS and the composite outcome TA-free-survival, were analyzed by a stratified Cox regression model, which takes into account the censoring of events at the end of the patient observation periods. The Cox-estimated survival was stratified according to the propensity quintiles in order to correct for treatment selection bias; in this way, the hazard ratio of experiencing the event at any given point in time versus another study group could be calculated for each study group. Furthermore, the propensity-corrected, Cox-estimated survival curves were calculated and plotted for each study group. For each estimate, the mean, SEM, and 95% confidence interval (CI) are presented. For differences among study groups, p-values are presented as well.
Estimated ESA Treatment Duration
Propensity Score Matching In total, 2,513 patients fulfilling the aforementioned criteria were extracted from the database. At first, the pool of DARB patients (n = 539) was matched with the available EPO-A patients (n = 1,594) in a 1–N greedy matching algorithm. This methodology allowed the matching of 429 DARB patients with 1,584 EPO-A patients. The second step of the matching process involved matching the 429 matched DARB patients with the available EPO-B patients (n = 380). This algorithm made it possible to match the remaining 429 DARB patients with all 380 EPO-B patients available. Finally, to construct one matching frame for all analyses, the EPO-B matched patients were allocated to the same DARB–EPO-A quintile as their matched DARB counterparts. Table 1 shows the overall and matched patient numbers per study group and per propensity quintile.
Apparent and large differences were observed among the prematching patient characteristics in the three study groups (Table 2a). The variables with major intergroup differences were gender, platinum therapy, percentage of female cancers (breast and ovarian cancer), percentage of hematological cancers, and 1-month postindex mortality.
To check the validity of the matching process, the 13 variables used for the matching were analyzed with stratification on propensity quintiles (hierarchical mixed-effects model for continuous variables, stratified logistic regression model for dichotomous variables) [44]. The means and/or percentages were estimated for each study group. Table 2B shows the results of these analyses, indicating an improvement in the alignment of patient characteristics after matching.
Clinical Results
As for the global TA-free-survival, the separate components of its composite endpoint were also similar among the study groups. At 3 months, a period reflecting a common chemotherapy period and a time window often used in clinical trials of this type, 84.46% (95% CI, 79.29%–88.43%) of DARB, 84.86% (95% CI, 81.00%–87.99%) of EPO-A, and 85.51% (95% CI, 80.70%–89.19%) of EPO-B patients survived without additional RBC transfusion beyond 1 month after the index stay. Further, 89.16% (95% CI, 84.71%–92.38%) of DARB, 88.66% (95% CI, 85.18%–91.36%) of EPO-A, and 87.91% (95% CI, 83.31%–91.29%) of EPO-B patients survived without requiring readmission for anemia.
Cost Results
Anemia-associated costs mainly consisted of ESA costs, and were significantly lower in the DARB group ( 3,051 ± 218) than in the EPO-A ( 3,995 ± 144; p < .0001 versus DARB) and the EPO-B ( 3,752 ± 229; p = .0132 versus DARB) groups (Table 5A and B).
Estimated ESA Treatment Duration Propensity-score corrected days on ESA treatment were significantly lower in the DARB group (41.39 ± 2.33 days) than in the EPO-A (52.82 ± 1.28 days; p < .0001 versus DARB) and EPO-B (53.22 ± 2.47 days; p = .0004 versus DARB) groups (Table 6A and B).
ESA Differential Price Changes Over Time All the above-mentioned cost calculations applied cost per unit used as applicable to the period studied (2003 to mid-2005). In the interim, however, unit costs of ESAs could have changed substantially, and to a different extent for the three studied compounds. In order to assure validity of our outcomes in the light of the current Belgian environment, the ESA-related cost analyses were rerun applying 2006 prices (December 2006) for all ESAs at the individual pack-level to the raw data, in order to mimic the 2006 cost situation. Table 7 shows the total anemia-induced as well as ESA-specific cost results at 2006 prices. Although overall differences tend to become smaller, the differences in ESA costs remain significantly in favor of darbepoetin (p < .0001 for DARB versus EPO-A; p = .0496 for DARB versus EPO-B).
A retrospective analysis of a 2003, 2004, and 2005 Belgian hospital database to compare the costs and outcomes of patients treated with ESAs (EPO-A, EPO-B, or DARB) in Belgium was conducted. After adjusting for differences observed in the baseline characteristics of the patients in the different groups, patients treated with DARB experienced a significantly lower hospitalization cost over a 1-year period ( 16,949) than patients treated with EPO-A ( 19,472; p < .0001) or EPO-B ( 19,295; p = .008). Hospitalization costs were driven in all groups by pharmaceutical costs, representing 40%, 43%, and 46% of the total costs in the DARB, EPO-A, and EPO-B groups, respectively. Of these pharmaceutical costs, >30% were related to the ESA costs in each group. Clinical outcomes were similar after propensity score adjustment. The chosen methodology considered only clinical events registered 1 month after the index date. This approach was chosen because the full effect of the ESA treatment is not observed immediately after initiation [45], so any event occurring relatively soon after the index date is likely to be a result of the patient profile and severity of anemia at the index date, rather than a consequence of the ESA treatment. Indeed, events at the index admission were used in the propensity score model to assess the patient profile (e.g., blood transfusion or i.v. iron use at the index admission) instead.
The largest differences among the groups were observed in the anemia-related costs. The anemia-related costs were the lowest in the DARB group (
Lower costs in the DARB group cannot be explained by the difference in time to anemia readmission or by a longer TFS time in the DARB group, because the differences for these outcomes were small and not significant. Another explanation may be the need for a proportionally lower dosage in the DARB group. The current guidelines recommend dose reduction of the ESA to determine the lowest effective maintenance dose [23, 24], so lower ESA costs could be generated by a lower maintenance dose. However, the data available in the database (i.e., overall dose administered) did not allow the validation of this hypothesis. A further hypothesis is that the overall ESA treatment duration is shorter, with lower overall costs as a result. For Belgium, taking the official dosage recommendations as per label, the weekly doses for a 75 kg patient are 168.8 µg, 33,750 IU, and 33,750 IU for DARB, EPO-A, and EPO-B, respectively. The cost per unit in the database was assessed, and from there the theoretical weekly cost was calculated to be Previous studies in this area have shown conflicting results [25–31]. Some of these were presented as posters only, and it has been deemed impossible to assess the methodological quality of these studies [46]. However, several gaps were identified in the methods of the fully published studies [31, 46], including failure to appropriately describe the database used, use of differing observation periods [31], and failure to adjust for selection bias (for instance, different types of cancer in the different treatment groups) [47]. Performing a retrospective database analysis is a challenging exercise because many biases may be present. This concern was highlighted by the International Society of Pharmacoeconomic and Outcomes Research, which subsequently developed guidelines for such analyses [48]. For the current analysis, the criteria developed in these guidelines were followed as much as possible, and, by doing so, included careful description of the origin of the database, patient selection criteria, parameters investigated, costs included, and statistical analysis. Also, careful attention was paid to how the investigated parameters were adjusted to account for differences observed among the groups at baseline using a propensity score analysis. Some of the previously published analyses did not account for these differences [31]. These methodological differences may partly explain why the results obtained in the previously published analyses differ from the ones presented in this study. Limitations to the current analysis also exist. The database only includes longitudinal hospital data over a maximum period of 1 year; therefore, the assessment can be performed only on that year, preventing the analyses from taking into account the disease history of the patients, unless it happened to be mentioned in the database, or events that occurred in the subsequent year. This analysis, therefore, provides a one calendar year picture of hospital costs of these patients in a Belgian setting. Also, because the patients from 2005 had a shorter maximum observation period available in the database (only 6 months), the average costs were underestimated, because of the fact that some additional admissions were be observed. However, because the distributions of the study entry dates were similar in the three study groups, the relative observed differences should not be affected and may even be smaller than the true differences.
Based on the retrospective analysis of a longitudinal Belgian database, no difference in the clinical effectiveness of the three ESAs, as measured by survival without readmission for anemia or transfusion requirement, was observed. The anemia-related hospital costs, did, however, reveal significant differences, with the cost of patients treated with DARB being significantly lower than the cost of patients treated with EPO-A or EPO-B. Because this analysis takes into account baseline differences among the patients, it therefore provides valuable information on the real-life costs and effects associated with treatment of CIA by DARB as compared with EPO-A and EPO-B.
Conception/design: Erik Spaepen, Nadia Demarteau, Lieven Annemans Provision of study materials or patients: Erik Spaepen Collection/assembly of data: Erik Spaepen Data analysis and interpretation: Erik Spaepen, Nadia Demarteau, Simon Van Belle, Lieven Annemans Manuscript writing: Erik Spaepen, Nadia Demarteau, Simon Van Belle Final approval of manuscript: Erik Spaepen, Nadia Demarteau, Simon Van Belle, Lieven Annemans
Funding for this study was provided by Amgen. The authors also wish to acknowledge Helen Wilfehrt, Ph.D. (Amgen (Europe) GmbH, Zug, Switzerland) for providing valuable suggestions on grammar and syntax for correct usage of English and scientific language. E.S. is now affiliated with SBD Analytics, Bekkevoort, Belgium.
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||