- © 2007 by American Society of Clinical Oncology
Interpreting the Economic Literature in Oncology
- From the Christiana Care Helen F. Graham Cancer Center, Newark, DE; and Fox Chase Cancer Center, Philadelphia, PA
- Address reprint requests to Patrick A. Grusenmeyer, ScD, Helen F. Graham Cancer Center, 4701 Ogletown-Stanton Rd, Newark, DE 19711; e-mail: pgrusenmeyer{at}christianacare.org
Abstract
New treatment options provide hope for patients with localized and advanced cancer. However, these advances are associated with cost, both in terms of treatment-related expenditures and effects on quality of life. It is important that patients, physicians, insurers, and policymakers understand the relationship between costs and outcomes of new cancer treatments. Various methods of cost analysis can provide a structured manner to assess cost. Cost-effectiveness analysis (CEA) compares the cost of the intervention with the effect, resulting in a cost per effect (eg, cost per year of life gained) that can be compared across interventions. In this article, we review three recent CEAs in the oncology literature, including chemoprevention in breast cancer, adjuvant endocrine therapy in early-stage breast cancer, and salvage chemotherapy in advanced ovarian cancer. The important elements of CEA, including the recommendations of the US Public Health Service Panel on Cost Effectiveness in Health and Medicine as they relate to cancer treatments, are discussed. Many well-done CEAs in cancer treatment have been performed during the last decade. As with clinical trials, the rigor and methods of the analysis are critical to the reliability of the results. Therapies with high cost and small incremental improvement in survival and/or quality of life may find it difficult to meet the societal thresholds for what is considered cost effective. CEA is a method to assess the cost and effect of cancer treatments, providing important insights into the best use (ie, obtaining the most value for) of health care expenditures. As the literature indicates, one must be cognizant of the fact that there can be extraordinary costs associated with some newer cancer therapies that provide small incremental clinical benefit. Better understanding of the cancer economic literature can help lead to an informed dialogue on the health policy implications of resource allocation in cancer care.
INTRODUCTION
New treatment options are providing hope for patients with localized and advanced cancer. However, these advances are associated with cost, both in terms of treatment-related expenditures and effects on quality of life (QOL). It is important that patients, physicians, insurers, and policymakers understand the relationship between costs and outcomes of new cancer treatments.
This is particularly important as health care expenditures continue to increase. United States national health care expenditures were $1.9 trillion or $6,280 per person in 2004.1 Health care expenditures are projected to increase at an annual rate of 7.2% through 2015, reaching $4 trillion or 20% of gross domestic product by 2013.2-4 Recent innovations in cancer diagnosis and treatment are also associated with increased cost5,6 (Meropol et al7 and Dimasi et al8 in this issue). It is important to determine if these added resources are used in a manner that provides the most value for the expenditure.
There are many forms of economic analyses that can be used to determine the cost implications of new treatment strategies. Cost identification or cost minimization studies estimate and compare costs of different interventions. These analyses are important when alternative interventions provide similar results (ie, survival); consequently, the study seeks to identify the least costly option. Cost-effectiveness analysis (CEA) compares the cost of the intervention with the effect, resulting in a cost per effect (eg, cost per year of life gained) that can be compared across interventions. The addition of a QOL measure results in a cost-utility analysis providing the consideration of cost per quality-adjusted life years (QALY).
In 1996, the US Public Health Service Panel on Cost Effectiveness in Health and Medicine (USPCEHM) issued guidelines to help analysts design studies and users interpret results.9-11 It is important for readers to critically interpret CEAs, because although economic analyses are becoming increasingly common in the medical literature, they can be of varying methodologic quality.12 Oncology studies that evaluate potentially costly and toxic interventions can be particularly challenging to conduct; one systematic review of oncologic economic analyses found that many studies published from 1975 to 1997 did not incorporate important elements suggested by the USPCEHM.13 Given the growing body of literature in this area, the Evidence Based Medicine's Working group has published users' guides to help clinicians interpret the literature14 (Table 1).
Users' Guide for Economic Analyses of Clinical Practice
In this article, we first discuss three well-designed recent economic analyses of chemoprevention, adjuvant endocrine therapy for breast cancer, and salvage chemotherapy for advanced ovarian cancer. They were selected because they address important clinical questions using contemporary treatments and are examples of different analytic techniques. We review elements of CEA such as model design (including Markov models), time horizon, inclusion of costs, incremental cost-effectiveness ratios (ICERs), and reporting of sensitivity analyses used in these studies. These elements are then defined in greater detail in the remainder of the article.
CHEMOPREVENTION OF BREAST CANCER WITH TAMOXIFEN
Melnikow et al,16 using data from the National Surgical Adjuvant Breast and Bowel Project P-1 Breast Cancer Prevention trial, developed a Markov model to evaluate the effect of tamoxifen on mortality and assess cost effectiveness for a hypothetical cohort of 50-year-old women with a 1% to 5% risk of developing breast cancer. Markov models follow a cohort of hypothetical patients through various health states (eg, alive without cancer, alive with cancer, or dead). Incidence and mortality rates for breast and endometrial cancers were derived from the Surveillance, Epidemiology, and End Results survey. Relative risks of outcomes were derived from the National Surgical Adjuvant Breast and Bowel Project P-1 trial. Patients were observed for 50 years, or until death. Using a payer's perspective, costs were based on Medicare reimbursements.
Melnikow16 concluded that for women with a uterus and a projected 5-year breast cancer risk of 1.67%, tamoxifen provided an extremely limited improvement in survival, resulting in a cost-effectiveness ratio (CER) of $1,335,690 per year of life gained. The CER was highly sensitive to the cost of the drug, decreasing substantially to $414,916 and $123,780 if the cost of tamoxifen was reduced to health maintenance organization (HMO) or Canadian prices, respectively. For women at similar risk without a uterus, adding tamoxifen resulted in improved survival, due to the elimination of the risk of endometrial cancer from the tamoxifen and a CER of $177,116 per year of life gained. That ratio decreased substantially to $51,146 and $11,315 if the cost of tamoxifen was reduced using HMO or Canadian prices, respectively. Effectiveness and cost effectiveness of tamoxifen improve for women at increased risk of breast cancer, and consequently Melnikow et al recommend tamoxifen for women at higher 5-year risk (> 3%) if the price of tamoxifen approximated Canadian prices. QOL was not considered in this study, but may be important in this analysis. Given that tamoxifen prevents breast cancers that are less likely to be fatal, its adverse effects may lower the QOL of women taking the drug, but would improve QOL in women in which it prevented a breast cancer.
ADJUVANT ENDOCRINE THERAPY IN EARLY-STAGE BREAST CANCER
Hillner17 used data from the Arimidex, Tamoxifen Alone or in Combination (ATAC) study to measure the cost effectiveness of treatment with anastrazole compared with tamoxifen for early-stage breast cancer. ATAC found a disease-free survival advantage favoring the anastrazole arm, and also noted that unlike tamoxifen, which had a higher risk of venous thromboembolism and uterine bleeding, anastrazole was associated with a higher risk of bone mineral density loss and hip fracture.
Hillner17 used a Markov model that observes women who are well and receiving adjuvant hormonal therapy through different health states. They can either remain well or develop adverse effects (vaginal bleeding, venous thromboembolism, or hip fracture). They continue to receive therapy for 5 years unless they develop an adverse effect, recurrent breast cancer, or contralateral breast cancer while receiving therapy. Patients are observed for up to 20 years or until death. The model uses transition probabilities for progression, toxicity, and recurrence from ATAC, as well as published reports. Costs of treatment for toxicity and recurrence are based on expert opinion and published reports. Given that there was no QOL assessment in ATAC, utilities were based on published reports.
This model demonstrates the importance of long-term follow-up, particularly in assessing curative therapies in low-mortality disease. It found that the incremental costs decreased at longer periods of follow-up, reflecting the long-term benefits of anastrazole. For example, at 4 years of follow-up, anastrazole was associated with relative high ICER (the cost of the incremental benefit) of $167,500 per disease-free year; at 20 years of follow-up, the ICER in the anastrazole arm decreased to $16,700 per disease-free year. The model estimated a survival benefit based on the reduction in systemic recurrence and estimated that the ICER at 12 years was approximately $100,000 per life year, whereas the ICER at 20 years was $40,600 per life year. The study found the short-term QOL benefit favoring anastrazole due to its lower risk of vaginal bleeding and thromboembolism was offset by longer-term increase in hip fractures. Because this model uses data directly derived from one clinical trial, it does not explore other possible alternatives for this clinical scenario, such as letrozole or exemestane.
SALVAGE CHEMOTHERAPY FOR ADVANCED OVARIAN CANCER
Rocconi et al18 evaluated the role of salvage therapy for platinum-resistant ovarian cancer. Unlike Hillner's17 analysis, this study combines data across multiple clinical trials to construct a model to compare possible treatment strategies for the same disease. Their model compares a base case of best supportive care (BSC) alone versus second-line single-agent and doublet regimens as well as third-line monotherapy.
This model incorporates commonly used regimens for second- and third-line therapy. Patients receiving BSC alone are expected to live 3 months. Patients receiving second-line monotherapy with liposomal doxorubicin have a median progression-free survival (PFS) of 4.1 months, whereas patients receiving combination therapy with gemcitabine and cisplatin have PFS of 6 months. After treatment with second-line chemotherapy, patients are estimated to survive an additional 2 months with BSC. Patients who proceed to third-line therapy are treated with topotecan and have an estimated PFS of 3 months with an additional 1 month with BSC. The study was performed from the third-party payer perspective using 2004 dollars.
The authors present the incremental cost per life year gained by increasingly aggressive regimens. They rank the treatment options in order of increasing effectiveness (BSC v second line monotherapy v second-line combination v third-line following second-line combination). They concluded that second-line monotherapy provides a 2-month survival advantage over BSC at $64,000 per year of life gained. However, the additional 3 months gained by second-line combination therapy compared with monotherapy comes at $302,000 per life year gained, whereas the additional 2 months gained by third-line therapy compared with second-line combination therapy comes at a cost of $303,000. In their one-way sensitivity analyses, they found that second-line monotherapy must provide a overall survival (OS) of 7 months or cost less than $16,500 to be cost effective; second-line combination therapy must provide an OS of 20 months or cost less than $25,000 to be cost effective as defined in this article (less than $50,000 per life year saved).19,20 It would be important to know if the results of their sensitivity analyses were significantly different under higher thresholds. In addition, the study compares a limited number of potential regimens for advanced ovarian cancer. However, the authors note that these regimens are among the most commonly used, and given the results of the sensitivity analyses, it is unlikely that other agents would have significantly improved the OS or cost profile that would make combination second-line therapy cost effective. However, it is possible that other salvage regimens may be slightly more effective and come at higher treatment-related costs than those used in this article; it would be important to know if sensitivity analyses find them to be cost effective under higher thresholds.
CEA
These three articles were chosen to demonstrate challenges of designing and interpreting CEAs in oncology. The remainder of the article is devoted to a discussion of the important elements of CEAs to aid readers in interpreting other economic analyses in the oncology literature.
DESIGN OF STUDIES
Studies are often based on clinical trials that directly compare results (eg, survival or cancers prevented) of two or more treatment arms. However, studies may not report all outcomes that are necessary components of economic analyses. For example, long-term survival may not be available in a study of adjuvant chemotherapy in stage II colon cancer, which reports 5-year survival data. Other studies compare outcomes of treatments of varying efficacy that are not directly compared in clinical trials. In both situations mathematical models such as Markov models may be used to extrapolate clinical outcomes and estimate costs.
Markov Models
Markov models follow a cohort of hypothetical patients through various health states (eg, alive without cancer, alive with cancer, or dead; Fig 1). Markov models provide a framework to model recurring events and to extend forecasts and assessments further out in time (for example, to encompass the patient's lifetime). In a Markov model, uncertain events are modeled as transitions during specified intervals (cycles) between mutually exclusive health states. The hypothetical patients proceed through the predetermined intervals of evaluation based on the clinical scenario. For example, patients with early-stage breast cancer may be evaluated at yearly intervals, whereas patients with advanced lung cancer may be evaluated at monthly intervals. A Markov model allocates and reallocates members of a population into health states according to defined transition probabilities. These transition probabilities are determined by the investigator and derived from the literature, ongoing studies, or expert opinion.21 The model tracks and aggregates the dimensions of interest (cost, survival, QALYs, and so on) for each health state so that a comparison among the choices can be made to optimize the dimension of interest. Analysts can create Markov models based on individual clinical trials or combine data across studies (see Sources of Effectiveness Data).
Example of the Markov model. Patients can transition between health states based on transition probabilities derived from the literature or expert opinion. Ovals signify discrete health states. P1, probability of initially developing cancer or probability of cancer recurrence; P2, probability of cure; P3, probability of dying without cancer; P4, probability of dying with cancer. Curved arrows indicate that patients can stay in that health state for more than one stage.
Both Melnikow et al16 and Hillner17 used a Markov model. Hillner's study used a Markov model with seven health states: well and receiving adjuvant therapy, systemic breast cancer recurrence, local recurrence or contralateral breast cancer, adjuvant therapy halted, vaginal bleeding or venenous thromboembolism, hip fracture, and death.
The transition probabilities (the chance of moving from one health state to another) were developed primarily from ATAC. The model tracked two hypothetical cohorts of women: one receiving anastrozole and one receiving tamoxifen.
Choice of Comparators
Economic analyses should include relevant treatment strategies, including maintaining the current standard of care for that clinical scenario. For example, in studies such as Hillner's17 of adjuvant hormonal therapy in hormone receptor–positive breast cancer, tamoxifen is a necessary option, whereas chemoprevention studies, such as the study by Melnikow et al,16 may include “do nothing” as a comparator. Studies of salvage chemotherapy, such as the study by Rocconi et al18 of platinum-resistant ovarian cancer, may include BSC as appropriate treatment option.
ELEMENTS OF CEAs
Perspective
CEAs can be performed from different perspectives that reflect the different costs and consequences. For example, studies from a private health insurer's perspective might emphasize direct medical costs, including physician fees, hospitalization costs, and medication expenditures. However, they may not take into account indirect medical costs, such as a patient's travel time and the value of missed work. The USPCEHM recommends that all analyses be performed from a societal perspective, which evaluates costs and consequences to the society as a whole. In addition, using a societal perspective facilitates cross-study comparison of results, rather than comparison of studies from different perspectives.11 The national perspective of the study should also be stated clearly, given that variations in resource use and drug pricing may vary widely among countries (see Costs). For example, the three studies profiled in this article were written from the perspective of a US third-party payer, but as the study by Melnikow et al,16 shows, using Canadian drug prices can alter the CER.
Time Horizon
The follow-up period for the study should be stated clearly, and be long enough to measure the relevant outcome. These results may need to be extrapolated from other clinical trials or population-based registries such as Surveillance, Epidemiology, and End Results. Studies that do not include adequate follow-up may not include long-term risks or benefits. For example, if a study of adjuvant therapy uses five-year survival data to measure effectiveness, it may miss potential long-term toxicity and subsequently overstate the benefit of treatment. The article by Hillner17 shows the importance of a sufficient horizon since the long-term toxicities of anastrazole (hip fracture) would not be reflected in a study with shorter follow-up.
CERs
Results are presented by cost per unit effect. CER can be calculated using the following equation: (cost regimen 2 − cost regimen 1)/(outcome regimen 2 − outcome regimen 1), here regimen 2 is the new regimen and regimen 1 is the comparator, and the effectiveness of the therapies may be adjusted for QOL.
ICER
The USPCEHM recommends that all studies report ICERs, rather than average CERs. The regimens are ranked in order of increasing cost or effectiveness, and the ICER is calculated by comparing each regimen to the one immediately preceding it. Treatments that are less effective and more costly are considered dominated and should not be included in the calculations.9
For example, the study by Rocconi et al 18 of advanced ovarian cancer shows that second-line monotherapy comes at a reasonable cost per life year saved compared with BSC. The next most effective option (second-line combination therapy) provides an additional 2 months survival compared with second-line monotherapy. These 2 months come at a substantial incremental cost of more than $300,000 per life year gained compared with second-line monotherapy. Publishing incremental CERs is important because they allow readers to understand the cost of obtaining each additional unit of survival.
Costs
Quantifying costs can be challenging. Cost data are not often collected alongside a clinical trial and must be estimated as part of the model. The increase in costs associated with a new regimen (numerator) should be opportunity cost, which can be thought of as the value of the funds if they were used as the next best alternative use. Unfortunately, given the pricing complexities of the health care market, it may be difficult to quantitate true opportunity costs, given that payors and providers may negotiate different rates for the same service or medication. Some commonly used methods of adjustment include using ratios of cost to charges or third-party payer reimbursements as a proxy for costs.10 The choice of which costs to include depends on the perspective of the study. For example, studies from a hospital perspective would include the cost of providing inpatient care. Studies from a managed care organizations would also include cost for home care and outpatient medications as well. However, studies from a societal perspective would also include the value of lost work for both the patient and caregiver. In addition, because medical expenditures may vary widely among countries, the article should state clearly the source of costs of drugs and other resources.
Authors should state their cost assumptions explicitly, so that readers can determine if these assumptions reflect their patients' costs. Because chemotherapy costs represent a substantial portion of costs for most cancer patients, the article should state the source of costs, drug dose, and assumptions of patients' body-surface area, as well as whether the model includes costs of supportive medications and treatment for complications. The study by Rocconi et al18 on salvage chemotherapy uses average wholesale costs for liposomal doxorubicin, gemcitabine and cisplatin for a 5-ft, 6-in.-tall woman weighing 65 kg, but states that it excludes costs related to chemotherapy complications.
Outcomes
The difference in outcomes (denominator) of the CER refers to the additional benefit gained from the intervention. Two important components of measuring outcomes include the source of the effectiveness data and the role of QOL adjustment.
Source of Effectiveness Data
The quality of the efficacy data is one of the most important components of a CEA.10 Sources may include primary data collection efforts, or secondary data sources such as published clinical trials. For example, Hillner's17 study of adjuvant endocrine therapy uses data from ATAC, whereas the study by Rocconi et al 18 of salvage chemotherapy for ovarian cancer compiles data across several clinical trials. The sources should be clearly referenced in the analysis; readers should determine if the source is relevant to their patient population. For example, clinical trial participants may be younger, healthier, or more motivated than the typical patient,22 and sensitivity analyses may be helpful in determining if these results are applicable to another clinical population.
QOL Adjustment
Most cancer treatments are associated with adverse effects, which can vary from mild to life threatening. In addition, the underlying malignancy can impair a patient's QOL, particularly in those with advanced cancer. QALY, rather than life years alone, are used to quantitate QOL in economic analyses. They are incorporated as preference weights or utilities, which can vary from one (perfect health) to zero (dead). Cost utility analysis incorporates the QOL into the CEA. The USPCHEM recommends using QALYs as the measure of health effect.11
Collecting data on QOL can be challenging, and unfortunately, many clinical trials do not incorporate or report a QOL component. Another option is to refer to the literature for measurements of these preference weights. The Institute for Clinical Research and Health Policy Studies at Tufts University23 maintains a comprehensive listing of published preference weights for different health states; examples of cancer-related preference weights are listed in Table 2. The lack of QOL adjustments in two of the three studies examined in this article indicates the difficulty in collecting and applying QOL data in economic analyses. Hillner17 adjusted for QOL using the Tufts (formerly Harvard) registry. As noted previously, QOL adjustments could play a role in the assessment of tamoxifen in breast cancer prevention.
Selected Preference Weights for Cancer Related Conditions
Sensitivity Analyses
The CER is calculated using a mathematical model requiring the estimation and/or extension of different input parameters such as outcomes (survival), transition probabilities, and costs. These input parameters are developed from clinical trials results, expert panels, and cost schedules. Each of these parameter estimates, for different reasons, might have a degree of uncertainty, leaving some question about whether these are the best estimates. However, the selection of an optimal decision according to the model depends heavily on these values. Consequently, once an optimal decision is determined, it is appropriate to assess the degree to which the selection is sensitive to changes or different estimates in these parameters. A formal sensitivity analysis is conducted to determine how variations in model inputs change the results of the CEA.25
One-way sensitivity analyses vary one variable at a time over a range of plausible values to determine if the changes affect the outcome (the CERs) appreciably. For example, the investigator may vary the measure of cost between 50% and 150% of the baseline to determine if changes in pricing change the CERs dramatically. Multivariate sensitivity analyses adjust two or more variables simultaneously to determine if certain combinations of changes to the input parameters (eg, making one regimen more costly and less effective) alter the results. For example, the study by Rocconi et al 18 varied the efficacy (OS) and cost of therapy of individual regimens separately to determine the minimum survival or the highest cost a regimen could have and still be cost effective. Hillner's17 study varied inputs such as the anastrazole-induced reduction in systemic recurrence, cost of anastrazole, and risk of non–breast cancer-related pain to see how these changes affected the final CERs. If the CERs remain similar despite rigorous sensitivity analysis, the results may be considered robust; however, if small changes in one or more variables significantly change the CERs, then the results can be considered sensitive to variations in those parameters.
Discounting
Individuals prefer benefits in the present versus potential rewards in the future. The value of future costs and benefits is less than that in the present. Therefore, USPCEHM recommends that future costs and benefits be discounted between 3% and 5% annually.10 In the study by Melnikow et al,16 future costs were discounted 3% annually throughout the model. This is especially important in an analysis with a long time horizon, as in breast cancer prevention.
Funding Source
The funding source for the CEA (federal or state government, foundation, pharmaceutical industry, or device manufacturers) should be stated clearly in the article. Studies of the role of sponsorship on economic analyses of oncology medications have found that industry-sponsored reports are more likely to show favorable outcomes than those funded by nonprofit sources.26,27 The authors of the three articles profiled in this review all identified funding sources for their research. This is commonly required for most medical journal publications.
DISCUSSION
The alternative interventions can be compared with each other to determine which strategies are dominated (less effective and more costly) and therefore should not be implemented. Results can also be compared with previous CEAs performed for other treatments as part of a league table that lists models in order of cost per unit of effect. Although it may be difficult to compare CEAs developed using different methodologies directly, league tables allow comparison to other accepted interventions that have already been adopted. Examples of other CEAs are listed in Table 3.
League Table of Selected Interventions
Despite their recommendations to use CEA, the Panel on Cost Effectiveness in Health and Medicine recognized its limited use by stating that “there is little indication…that cost effectiveness analysis contributes systematically to resource allocation decisions in United States Medicine.”25 Private insurers use CEA analyses to varying degrees when making decisions regarding coverage35; however, managed care plans and HMOs have been reluctant to use CEA explicitly, due in part to the perception as rationing. In its place, managed care plans have used increasingly aggressive methods to control costs, including stringent utilization review and disease management programs.36
Other countries with national health insurance may use CEA either implicitly or explicitly. For example, the United Kingdom's National Institute for Clinical Excellence makes recommendations regarding adoption of new devices and drugs based on economic analyses and clinical information.37 Although the National Institute for Clinical Excellence has received criticism for its “one size fits all” policies, some experts advocate for a politically independent agency in the United States that integrates economic and clinical effectiveness data.37-39
As health care costs continue to increase, payers in both the public and private sector may be forced to contain costs, and economic analyses may be helpful in guiding decision making for cancer treatments. Clinicians can inform the debate by understanding the possibilities and limitations of economic analyses and guide payers and policymakers in making informed decisions that best allocate resources so that the promises of new cancer treatments are available to as many patients as possible.
The goal of economic analysis in health care is to ensure that the money spent will provide the most value for the dollar. Health care is a value because it allows the consumer to increase the length and improve his or her QOL.39 “As we get older and richer, which is more valuable: a third car, another television set, more clothing - or an extra year of life?”40
The collection of QOL and cost data within or alongside clinical trials would aid in the development of CEAs and cost utility analyses. Although QOL analysis is being incorporated into more clinical trials, the time has come for the incorporation of cost data so that policy makers, aided by the advice of health care providers, can make more informed decisions.
AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST
The authors indicated no potential conflicts of interest.
AUTHOR CONTRIBUTIONS
Conception and design: Patrick A. Grusenmeyer, Yu-Ning Wong
Administrative support: Patrick A. Grusenmeyer, Yu-Ning Wong
Collection and assembly of data: Patrick A. Grusenmeyer, Yu-Ning Wong
Data analysis and interpretation: Patrick A. Grusenmeyer, Yu-Ning Wong
Manuscript writing: Patrick A. Grusenmeyer, Yu-Ning Wong
Final approval of manuscript: Patrick A. Grusenmeyer, Yu-Ning Wong
Footnotes
-
Supported by Grant No. R25CA057708 from the National Institutes of Health (Y.-N.W.).
Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.
- Received September 5, 2006.
- Accepted October 16, 2006.
