© 2003 by Oxford University Press
Journal of the National Cancer Institute Monographs, No. 31, 102-110,
2003
© 2003 Oxford University Press
ARTICLE |
Chapter 15: Public Health Policy and Cost-Effectiveness Analysis
Correspondence to: Sue J. Goldie, M.D., M.P.H., Harvard Center for Risk Analysis, Department of Health Policy and Management, Harvard School of Public Health, 718 Huntington Ave., 2nd floor, Boston, MA 02115-5924 (e-mail: sgoldie{at}hsph.harvard.edu).
 |
ABSTRACT |
|---|
 TopNotes Abstract IntroductionDecision Science and Cost...Mathematical ModelsEvaluating UncertaintyPrevious Cervical Cancer ModelsThe Efficiency CurvePriority Areas for Future...References |
|---|
Recent scientific advances are providing an opportunity to revisit strategies for cervical cancer prevention. How to invest health resources wisely, such that public health benefits are maximized—and opportunity costs are minimized—is a critical question in the setting of enhanced cytologic screening methods, human papillomavirus DNA testing, and vaccine development. Developing sound clinical guidelines and public health policy will require careful consideration of the incremental benefits, harms, and costs associated with new interventions compared with existing interventions, at both an individual and a population level. In addition to an intervention’s effectiveness, public health decision making requires the consideration of its feasibility, sustainability, and affordability. No clinical trial or single cohort study will be able to simultaneously consider all of these components. Cost-effectiveness analysis and disease-simulation modeling, capitalizing on data from multiple sources, can serve as a valuable tool to extend the time horizon of clinical trials, to evaluate more strategies than possible in a single clinical trial, and to assess the relative costs and benefits of alternative policies to reduce mortality from cervical cancer.
|
Introduction |
|---|
|
TopNotesAbstractIntroductionDecision Science and Cost...Mathematical ModelsEvaluating UncertaintyPrevious Cervical Cancer ModelsThe Efficiency CurvePriority Areas for Future...References |
|---|
Despite substantial progress in our understanding of cervical carcinogenesis and the causal role of oncogenic human papillomavirus (HPV), cervical cancer continues to be a leading cause of cancer death among women worldwide (1–7). There are important differences in the most relevant policy questions related to cervical cancer control for developed and developing countries. In countries with organized screening programs, there has been a marked reduction in the incidence of invasive cancer; however, screening and treatment have not been equally accessible to all groups of women (8). Effective and cost-effective public health strategies for increasing screening coverage are needed. From a health economic and policy perspective, among the most pressing concerns are the escalating costs associated with current screening practices. For example, in the United States, more than 6 billion dollars is spent each year on the evaluation and management of low-grade lesions, the majority of which would regress without intervention (9). Rigorous cost-effectiveness analyses are needed to evaluate the best ways to integrate new cytologic methods and HPV DNA testing into existing screening programs.
In countries classified as low-income economies (the gross national income per capita was less than or equal to $755 in 2000) (10), the most pertinent challenge is how to implement a sustainable screening program in the setting of competing health priorities and limited resources. Data from studies conducted in the last several years support a potential role for simple visual screening methods and HPV DNA testing followed by treatment, with and without colposcopic triage (11–13). However, the long-term effectiveness of these strategies is still uncertain. In some countries classified as middle-income economies (the gross national income per capita was more than $755 but less than $9266 in 2000), cytology screening is available to a limited extent but has not achieved reductions in cervical cancer. In these settings, one of the most pressing questions is how to improve the quality of cytologic screening programs and how to more accurately target previously unscreened women over the age of 30 years, those at the highest risk of cervical cancer. Finally, from a global public health perspective and particularly relevant in resource-poor countries, a challenge of the highest priority is how to ensure the investment of adequate resources to develop a type-specific HPV vaccine.
Public health decision making requires consideration of an intervention’s real-world "effectiveness," as opposed to a singular focus on "efficacy," and the likelihood that it will be sustainable. However, evaluating the effectiveness of any large-scale public health intervention is difficult because there are multiple elements. The optimal cervical cancer screening policy requires the consideration of screening test performance, alternative options to evaluate and manage abnormal results, and the effectiveness of different treatment options for precancerous lesions. A model of the underlying natural history of disease must be sophisticated enough to allow for a reasonable representation of heterogeneity of risk in the population and must be able to incorporate information on accessibility, compliance, and feasibility in the targeted setting. A single clinical trial or longitudinal cohort study will not be able to consider all of these components and to assess all of the possible strategies for all possible populations. Moreover, clinical trials do not generally include all of the necessary cost and quality-of-life data needed to allow the results to be readily understood in terms of policy. These facts, together with the need for decision making in the setting of incomplete information, make the use of decision-analytic models an attractive tool for public health.
|
DECISION SCIENCE AND COST-EFFECTIVENESS ANALYSIS |
|---|
|
TopNotesAbstractIntroductionDecision Science and Cost...Mathematical ModelsEvaluating UncertaintyPrevious Cervical Cancer ModelsThe Efficiency CurvePriority Areas for Future...References |
|---|
Decision science incorporates elements of economics, statistics, and psychology and provides an explicit, quantitative, and systematic approach to decision making under uncertainty (14). The discipline encompasses a collection of quantitative methods that have been developed to guide the management of complex problems that require simultaneous consideration of several competing choices, alternative perspectives, and inevitable trade-offs. The components of a decision analysis are formally synthesized and structured over time using a model. While different types of models may be chosen to accommodate the complexity of the decision, all rely on the use of quantitative mathematical analysis to compare the "expected value" (i.e., expected consequences) of different alternatives.
The basic principle of a decision-analytic approach is that all of the consequences of decisions (e.g., individual clinical outcomes, population-based outcomes, and costs) should be identified, measured, and valued. When a decision analysis formally compares the relationship between the health and economic consequences associated with different public health care interventions, it is considered to be a cost-effectiveness analysis. The application of economics to public policy does not necessarily mean that less money should be spent, but rather that the use of resources might be more efficient (15). Different types of economic evaluations are commonly confused. For example, there are distinct differences between cost-minimization analysis (how much money can be saved?) and cost-effectiveness analysis (how much health improvement can be gained, dollar for dollar?) (16). The results of a cost-effectiveness analysis are summarized using an incremental cost-effectiveness ratio. In this ratio, all of the health outcomes associated with a particular strategy (compared with an alternative) are included in the denominator, and all of the costs or changes in resource use with a particular strategy (compared with an alternative) are included in the numerator. This type of analysis defines the "opportunity cost" of choosing one clinical or public health approach over another.
Cost-effectiveness ratios should be reported as dollar per unit of effectiveness, stating the year of the costs, for example, $25 000 per year of life saved (YLS) (2002 dollars). Cost-effectiveness analyses are always incremental, with the ratios comparing each intervention with the next most effective alternative. This means that the costs and clinical benefits associated with the intervention of interest should be compared not only with existing practice but also with all other reasonable options. When all of the possible alternatives are not included, there is a risk of coming to an incorrect conclusion that an intervention is cost-effective, but only because it was compared with a cost-ineffective alternative. For example, if cervical cancer screening with annual Pap smears is evaluated compared with doing nothing (no screening), then the cost-effectiveness ratio is less than $50 000 per YLS. If one correctly includes the option of an every-2-year screening, the cost-effectiveness ratio of an annual Pap screening will exceed several hundred thousand dollars per YLS. Similarly, a cost-effectiveness analysis of type-specific HPV vaccination will need to consider the incremental costs and benefits associated with using the vaccine in comparison to current screening practices.
To improve the quality and comparability of cost-effectiveness analyses, the U.S. Public Health Service Panel on Cost-Effectiveness in Health and Medicine recommended that, for the analyses intended to inform resource allocation decisions, a standard set of methodologic practices be used to facilitate comparison across different studies and types of interventions (15,16). The methodologic rigor of cost-effectiveness analyses conducted in the next decade will ideally reflect these and other published recommendations (16–21).
|
MATHEMATICAL MODELS |
|---|
|
TopNotesAbstractIntroductionDecision Science and Cost...Mathematical ModelsEvaluating UncertaintyPrevious Cervical Cancer ModelsThe Efficiency CurvePriority Areas for Future...References |
|---|
A decision-analytic approach relies on the use of a mathematical model to formally structure the components of the decision over time. Models combine information about the natural history of disease and the performance of screening and treatment with other relevant demographic and epidemiologic characteristics of the target population. Clinical and economic outcomes can be extrapolated beyond the time horizon of a single clinical study. In addition to relating clinical and biologic information, they provide insight into the relative importance of different components of the screening process and investigate how cost-effectiveness ratios will change if values of key parameters are changed (20–22). Mathematical models have been used routinely to guide important policy decisions in the areas affecting public health, ranging from environmental regulation of pesticides to assessing priorities for the development of new vaccines. Models have also been used to influence the coverage decisions by public and private insurers and clinical practice guidelines in areas such as cardiovascular disease prevention, human immunodeficiency virus (HIV) treatment, and cancer screening (16,21–25).
Models may be classified in several ways. For example, they may be classified according to the structure that they employ to account for events that occur over time (e.g., decision trees and state transition models) (16). Decision trees work well in relatively straightforward clinical problems with short-term outcomes and a fixed time horizon but are less useful for diseases in which clinical events occur repeatedly or over a long period of time. To evaluate policy options for population-based screening and to examine outcomes over an extended period, more complex models, such as Markov models, are required.
A Markov model is composed of a set of mutually exclusive and collectively exhaustive health states (Fig. 1
). Each person in the model can reside in only one health state at any point in time, and all persons residing in a particular health state are indistinguishable from one another (12,26). Transitions occur from one state to another at defined recurring intervals (Markov cycle) of equal length (e.g., monthly or yearly) according to a set of transition probabilities. These probabilities can be made dependent on population characteristics, such as age, sex, and chronic disease, by specifying the probabilities as functions of these characteristics and may be constant or time dependent. An important assumption inherent in a Markov model framework is that the transition probabilities depend only on the current health state and not on previous health states (26).
|
A state-transition framework is employed in which members of a population are allocated and subsequently reallocated into different health states over time. Values are assigned to each health state to reflect the cost and utility of spending one Markov cycle in that state. The contribution of these values to population outcomes (e.g., life expectancy, quality-adjusted life expectancy, and lifetime costs) depends on the length of time spent in each state. By synthesizing data on costs, effects, and quality of life, a Markov model permits one to compare the outcomes associated with different clinical strategies.
The two most commonly used methods of evaluating a Markov model are cohort simulation and Monte Carlo simulation (first order) (22,26). Fig. 2 shows a simple schematic of a cohort simulation. Patients enter the model as a cohort and transit through the model simultaneously. Their distribution into the initial health states may vary, depending on the objective of the analysis. For example, in a Markov model developed to simulate HPV screening in women over the age of 30 years, one might start an analysis with an entire cohort of healthy 13-year-old girls and apply the intervention at a later age. On the other hand, one might want to start the model with a cohort of 30-year-old women and distribute the women on entry into the model into different health states according to the age-specific prevalence of the disease. For each model cycle, the fraction of the cohort initially in each health state is divided among all health states according to a set of defined transition probabilities. In each subsequent cycle, there will be a new distribution of the cohort among the various health states. The calculation of a Markov process yields the average amount of time spent in each health state. If the only attribute of interest is the duration of survival, then the average time spent in the individual states is added together to estimate the expected survival. If both attributes of quality of life and length of life are used, then the quality-adjusted time spent in each health state is added together to estimate the quality-adjusted survival time. The model outcomes that are easily generated with a cohort simulation include age-related prevalence and cumulative incidence of HPV, low- and high-grade squamous intraepithelial lesions (LSILs and HSILs, respectively), and invasive cancer.
|
Figure 3
shows a simple schematic of a Monte Carlo simulation (first order). The patients are randomly selected from a hypothetic cohort and enter the model one at a time. The characteristics (e.g., age, sexual history, and parity) of each person are randomly drawn from distributions derived from data. Using simulation, the model tracks the women individually, one after the other, from entry into the model until death. By examining the clinical course of a disease represented by the particular pathway an individual woman took through the health states before dying, the model can generate a survival time (or quality-adjusted survival time) for that individual. By running large numbers of simulated cases (e.g., 1 000 000), a distribution of survival values is obtained. If the sample is large enough, the mean value of this distribution will approximate the value estimated by cohort simulation. The variance of these values reflects the variability of the patient-level life spans rather than the variability of the population-level life expectancy. For a Monte Carlo simulation to be reliable, it is important to select a cohort size that is large enough that the variability in the estimate of the sample means is small compared with the differences of interest between strategies (22).
|
Because of the Markov assumption that transition probabilities depend only on the current health state and not on previous health states, to make future events depend on clinical history, one needs to expand the number of health states so that each state represents a unique health state history. For example, a woman with previously treated HSILs—even with subsequent normal cytology—would not return to a health state that contains women who had never had an abnormal cytology result. Because each health-state definition must describe all of the relevant current and past clinical information, the number of health states can grow prohibitively large, increasing the number of parameters to be estimated and requiring memory capacity beyond most available software (22). The major advantage of a Monte Carlo simulation is that it only requires that health states describe the current clinical situation, because each individual’s history is tracked specifically for that individual as she transitions through a model. Although the required sample size for a Monte Carlo simulation will depend on the magnitude of the transition probabilities, the differences in values between health states, and the anticipated effect sizes between strategies, often it is quite large (e.g., several hundred thousand to one million). Therefore, a disadvantage of this type of simulation is that it generally takes much longer to run the analysis, since each individual transits through the model one at a time.
Markov models are state-transition models that are "closed" in that no one enters or exits the cohort at any time during the simulation. State-transition models can also be dynamic in that they allow people to enter or exit the model over time. Whereas closed models are useful for comparing the outcomes associated with different health interventions, dynamic models are useful for evaluating the nature of epidemics or disease trends over time (22). For example, to assess the impact of HPV vaccination of males and females on transmission of disease, one might use an epidemic model (27). In an epidemic model, difference equations are used, in which the numbers of susceptible, immune, and infected persons in a population are modified each time period according to an equation that relates the change in the number of persons in each category to the number of persons in each category in the preceding time period as well as to variables that may be modified by intervention, such as contact rates and infectivity rates.
|
EVALUATING UNCERTAINTY |
|---|
|
TopNotesAbstractIntroductionDecision Science and Cost...Mathematical ModelsEvaluating UncertaintyPrevious Cervical Cancer ModelsThe Efficiency CurvePriority Areas for Future...References |
|---|
A critical part of any decision analysis is to evaluate the uncertainty in the model structure, parameter estimates, and assumptions (16,21,22). Weinstein et al. (21) provide a comprehensive discussion of several concepts related to the validation of decision-analytic models. Verification of a model is defined as demonstrating that the output of the model is consistent with the known facts about the disease. This process may include simple technical tests (often referred to as "debugging") and more complex calibration exercises in which statistical methods are used to maximize the likelihood function for observed data given a set of parameter values (21,22). Predictive validity refers to whether a model produces outputs that are consistent with observations that are independent of data used as inputs to the model.
Statistical issues in cost-effectiveness studies are different from those that arise in experiments or other data analyses. Rather than testing hypotheses using traditional statistical significance as a criterion, model-based evaluation studies aim to portray the scope and nature of the uncertainties that surround the estimates of costs, benefits, and cost-effectiveness ratios that they produce. The most common way to evaluate the stability of the conclusions of an analysis over a range of parameter estimates and structural assumptions is to conduct sensitivity analyses. The range of values used for a sensitivity analysis can be based on a statistical variation for point estimates or on probability distributions. Alternatively, expert opinion can be used to evaluate the range of values. To maintain the model’s transparency, an analyst should disclose the rationale for his or her choice of values. In a sensitivity analysis, some critical component in the calculation is varied over a plausible range, and the cost-effectiveness ratio is recalculated. The resulting difference in the ratio provides some indication of how sensitive the results might be to a change in that parameter. If the results are insensitive to a reasonable variation in a parameter, then the analyst can be relatively sure that the conclusions are insensitive to the baseline assumptions about that parameter. If the major results do change, sensitivity analysis may help to identify where better data are needed.
Another type of sensitivity analysis is one that adopts probabilistic simulation approaches (e.g., second-order Monte Carlo methods) (28). These methods assume some knowledge of the distributions surrounding a model’s input parameters and take random draws from these distributions to calculate outcome variables. Probabilistic sensitivity analysis can be performed using a Markov model analyzed as a cohort simulation and, using recently developed techniques, can be performed on a Markov model analyzed as a first-order Monte Carlo simulation (22,29).
|
PREVIOUS CERVICAL CANCER MODELS |
|---|
|
TopNotesAbstractIntroductionDecision Science and Cost...Mathematical ModelsEvaluating UncertaintyPrevious Cervical Cancer ModelsThe Efficiency CurvePriority Areas for Future...References |
|---|
Previous cervical cancer models have provided important groundwork for thinking about the data requirements necessary to evaluate new strategies for cervical cancer screening, such as HPV testing and the potential benefits of HPV vaccination (30–50). In general, prior analyses have shown that the cost-effectiveness of screening is sensitive to the age at which screening is started, generally indicating that, if resources are limited, it is more cost-effective to focus screening efforts on women older than 25 years of age. The cost-effectiveness analyses that have evaluated newer technologies have found that more sensitive screening tests are likely to be cost-effective compared with conventional Pap smears only if they allow for a longer screening interval, are associated with better specificity, or are used in conjunction with less aggressive and costly treatment of low-grade abnormalities (48). Recently, models have begun to incorporate state-of-the-art epidemiologic information on the natural history of HPV, a necessary prerequisite for assessing the potential benefits, harms, and costs of integrating HPV testing into cervical cancer screening programs.
|
THE EFFICIENCY CURVE |
|---|
|
TopNotesAbstractIntroductionDecision Science and Cost...Mathematical ModelsEvaluating UncertaintyPrevious Cervical Cancer ModelsThe Efficiency CurvePriority Areas for Future...References |
|---|
The results of a cost-effectiveness analysis are typically presented in the format of an "efficiency curve" (Fig. 4
). This curve represents the relationship between the incremental costs of one strategy (compared with the next best alternative) and the incremental benefits provided by that strategy (compared with the next best alternative). Fig. 4 shows the lifetime costs and clinical benefits (expressed here as the reduction in the lifetime risk of cancer) of different screening strategies performed at different screening intervals. From the perspective of one trying to purchase the most effective intervention for a fixed amount of resources, the strategies that form the solid line connecting the points are the economically efficient subset of choices (51). Strategies lying on this efficiency curve dominate those lying to the right of the curve because they are more effective and either cost less or have a more attractive cost-effectiveness ratio than the next best strategy. The relationship between the incremental benefits and the costs is formally described using an incremental cost-effectiveness ratio. The incremental cost-effectiveness ratio associated with moving from one screening strategy to a more costly alternative is represented by the difference in cost divided by the difference in clinical benefits (e.g., gains in life expectancy or quality-adjusted life expectancy and reduction in cancer risk) associated with the two strategies. The slope between the two strategies is steeper when the net gain in life expectancy or lifetime cancer risk reduction is greater.
|
Clustered in the lower-left corner of the efficiency curve in Fig. 4
are once in a lifetime screening strategies using a one-visit visual inspection or a two-visit HPV DNA testing. These are the least costly screening options reported in a previous cost-effectiveness analysis of South Africa (42). The twice and three times in a lifetime screening strategies are also located in the lower left but are positioned slightly higher up the curve—in this region of the curve, the slope is steep, demonstrating rapid escalating clinical benefits for minimal costs. The incremental cost-effectiveness ratios associated with these screening strategies reflect this relationship—for example, single lifetime screening ranges from cost saving to $50 per YLS compared with the next best strategy. In contrast, note the position of several previously reported screening strategies in the United States in the upper-right corner of the efficiency curve in Fig. 4—in this region of the curve, the slope is quite flat. The graph shows that, although reduction in cancer incidence is similar with every 1-, 2-, and 3-year screening interval, the total lifetime costs increase dramatically with more frequent screening. The incremental cost-effectiveness ratios associated with the most frequent screening strategies reflect this relationship—for example, a strategy of liquid cytology with either repeat cytology or reflex HPV testing of atypical cells of uncertain significance (ASCUS) results performed every 1 year compared with this same strategy conducted every 2 years and has an incremental cost-effectiveness ratio of more than $500 000 per YLS (46). This unattractive ratio reflects the dramatic increase in the costs associated with increasing screening frequency from every 2 years to every 1 year, for an additional gain of only 4 hours of average life expectancy. Every 3-year screening using liquid-based cytology and HPV triage for ASCUS results has a cost-effectiveness ratio of $60 000 per YLS compared with this same strategy conducted every 4 years. In the United States, we are faced with the challenge of how to ensure that the huge price we pay for achieving a few additional hours of life-expectancy gain does not represent the opportunity cost of reducing disparities in screening that would provide far greater public health benefits. In other words, as new technology becomes available, how do we keep from indiscriminately moving to the right on the "flat of the curve"—expending more and more resources with rapidly diminishing clinical gains? Although there is no consensus on a "threshold cost-effectiveness ratio" (i.e., the acceptable cost per year of life gained), incremental cost-effectiveness ratios for particular settings are often placed in context by interventions that are widely mandated in that same region of interest. For example, an annual screening mammography for women 55–65 years of age in the United States costs between $32 000 and $120 000 per YLS.
In developing countries, the relevant public health challenge is how to get on the curve at all—rather than "doing nothing" or "indefinitely waiting for more information," how do we implement a screening program that would include at least one, two, or three screens per lifetime? For countries with cost-effectiveness thresholds below $2 or $3 per YLS, a single lifetime screening test may be the only feasible strategy. To place these much lower incremental cost-effectiveness ratios into perspective, the cost per YLS in developing countries for childhood immunization programs is $12–$17 and for acquired immunodeficiency syndrome prevention programs is $3–$5 (42).
|
PRIORITY AREAS FOR FUTURE COST-EFFECTIVENESS ANALYSIS |
|---|
|
TopNotesAbstractIntroductionDecision Science and Cost...Mathematical ModelsEvaluating UncertaintyPrevious Cervical Cancer ModelsThe Efficiency CurvePriority Areas for Future...References |
|---|
As models of the natural history of HPV and cervical cancer evolve in sophistication, the focus of future cost-effectiveness analyses should be driven by the most pressing policy questions facing different regions of the world. Three general priority areas include 1) the optimal use of new technology in countries with established screening programs, 2) the feasible alternatives to conventional cytology in developing countries, and 3) communicating results of cost-effectiveness analyses.
Optimal Use of New Technology in Countries With Established Cytology Screening
As new screening options become available, it will be important to formally evaluate their incremental costs and benefits compared with existing practice, to identify potential harms that may be associated with their use, to critically assess the need for more information before widespread adoption, and to comparatively evaluate specific strategies that may incorporate new technology in different ways.
What are the most important issues that future cost-effectiveness analyses should address? 1) Assessing HPV DNA testing (with or without cytology) as a primary screening test—the incremental benefits from newer tests need to be defined in explicit comparison with existing alternatives, the subgroups of women should be identified for whom alternatives to cytology may be cost-effective (e.g., women >35 years of age), and new strategies should be considered that capitalize on the negative predictive value of repeated negative results on cytology and undetectable HPV; 2) the formal assessment of the health and economic consequences associated with different strategies to follow-up women who are HPV DNA positive; 3) the reassessment of when to start screening (e.g., 3 years after sexual initiation versus age-based cutoff), the consideration of different age-specific management approaches of cytologic abnormalities (e.g., ignoring LSILs), and the quantification of the potential risks of using HPV testing in primary screening strategies for younger women; 4) better methods to elicit and measure the preferences of individual women and the quality-of-life impact of newer technologies (52); and 5) cost-effectiveness analyses of anal cytology screening in gay men that extend previously done work (53,54) to incorporate the most recent natural history data, the direct and indirect costs of the provider training, and men’s screening attitudes and preferences.
Will modeling be necessary to evaluate new cervical cancer screening technologies? Although randomized controlled trials provide the most valid estimate of cancer screening efficacy in that they adjust for both the known and the unknown confounding variables, they pose several economic and practical difficulties. First, the magnitude of the typical population health gain achieved by a cancer screening program is small because only a small proportion of the population is at risk of the disease and is able to realize a screening benefit (55). Furthermore, several decades of observation might be necessary from the time of initiating a program to when an effect on cancer incidence would be measurable. Second, a randomized controlled trial of different screening strategies is not always considered to be an acceptable alternative if the screening already is widely accepted as a standard of care. Therefore, in the case of cervical cancer screening, it is unlikely that clinical trials will be performed that could adequately compare all of the possible variations of screening and treatment and follow-up cohorts of patients for the decades required to assess long-term mortality and quality-of-life outcomes, and disease simulation modeling will undoubtedly be necessary.
Are there enough natural history data to be confident of the results of cost-effectiveness analysis? A valid concern about the use of models is that the data used are generally from diverse sources and are subject to varying degrees of bias because of confounding variables, patient selection, unidentified sources of heterogeneity, or method of analysis (22). On the other hand, complete information on all transition probabilities required for a natural history model of cancer is never available. For example, to conduct a study to assess the monthly probability of progression from HSILs to invasive cancer would be unethical. However, information is generally available on the incidence of diagnosed invasive cancer in an unscreened or partially screened population using cancer registries. In such a situation, after incorporating the data that are directly available, one may derive the unobserved progression rate by calibrating the model to the observed incidence and stage distribution of diagnosed invasive cancer (22).
We can be considerably confident about the consistent findings from cost-effectiveness analyses regarding screening frequency with conventional cytology and the use of HPV testing for triage of ASCUS. However, analyses of HPV DNA testing for primary screening must be scrutinized carefully because there is still a good deal of uncertainty with respect to the natural history of HPV in women of different ages. The most important natural history data required for future models of cervical cancer screening include the age-related distribution of HPV and the time-dependent transition probabilities conditional on detectable HPV. It will be important for epidemiologists, decision analysts, and statisticians to collaborate closely in the next several years to develop new methods of estimating transition probabilities governing HPV acquisition, persistence, and clearance for use in mathematical models and then to test these models for predictive validity.
Do models match, given their differences in assumptions, such as the question of whether HPVs and LSILs are discrete steps? Model corroboration, also referred to as convergent validity, refers to the ability of independently developed models to produce similar results (21). Despite the differences in structural assumptions, several published models provide reasonably consistent projections of cancer incidence and mortality in unscreened and screened populations. Thus, for the major policy questions that have been on the table for more than a decade, they are fairly consistent. However, for analyses of primary screening with HPV DNA testing, independent models may differ considerably because of the following factors: 1) the different definitions of HPV infection and inconsistent methods to distinguish transient from persistent HPV infection; 2) the assumptions of homogeneity versus heterogeneity in health states reflecting HPV infection, LSILs, and HSILs (56); 3) the variable inclusion of time dependence in transition probabilities governing the incidence of squamous intraepithelial lesions that are conditional on HPV; and 4) the structural assumptions made with respect to modeling LSILs and HPV infection as a single health state versus two separate health states. It can be challenging to choose the specific health states reflecting precursor lesions, because data on their natural history are limited and are often subject to bias. However, often distinct health states are required, even if the prognosis associated with each are identical, to permit the evaluation of different screening and diagnostic algorithms applied to patients residing in each of them. For example, a model that does not distinguish between HPV infection and LSILs as distinct health states may be less able to accurately represent the different costs or clinical practice patterns that may be associated with each. Thus, the choice of health states in a model must reflect the consideration of natural history as well as real-world screening, management, and treatment interventions.
Evaluation of Alternatives to Conventional Cytology in Low-Income Developing Countries
Relatively recent analyses have provided quantitative insight into the relative importance of different components of a screening program in developing countries. There is a pressing need for additional region-specific analyses that consider alternatives to conventional cytology screening programs.
What are the most important screening issues that future cost-effectiveness analyses should address in resource-poor settings? Screening alternatives, such as visual inspection and HPV DNA testing, each require different types of resources, and the relative availability of these in different settings will impact the most feasible and cost-effective alternative for specific countries. As additional studies are conducted to establish the efficacy of cryotherapy when performed without colposcopy in low-resource settings, these data will need to be integrated into existing models. In addition, cost-effectiveness analyses in developing countries must include costs that are often overlooked in analyses in developed countries with established health care infrastructure. For example, it will be important to include the initiation costs of new screening programs, the ongoing costs of training and the supervision of providers, and the maintenance costs in the context of competing health priorities—many of these will likely be region specific. Accurately estimating the costs and valuation of women’s time in nonmonetary communities will be difficult. Because of the interactions between HPV and HIV, the pattern and stage of the HIV epidemic must also be included in these analyses.
Is it necessary to develop a country-specific model for every potential setting in which a preventive cervical cancer effort might be made? A basic natural history model (reflecting the underlying biology of disease) should be broadly transportable across settings, provided that the model is flexible enough to accommodate epidemiologic factors, such as different HPV types, mean age of sexual debut, frequency of risk factors, and competing morbidities. However, the intervention component of a model is less transportable because of different baseline levels of infrastructure, economic situations, and clinical practice patterns. We are currently working on a global policy model for developing countries to conduct analyses for several representative areas of the world. By grouping together different areas of the world on the basis of epidemiology, infrastructure, economic profile, and accessible resources, the model will be capable of comparing the costs and benefits of different cervical cancer screening strategies given knowledge of some of the basic parameters about the population. This model will serve to assist decision makers such as ministries of health in assessing the best possible prevention strategy given their regional circumstances.
How likely are cost-effectiveness results to change by population over time? Optimal strategies will likely change over time as new biologic information becomes available, as screening penetration and clinical practice change, and as new technology becomes available. For example, in developing countries where fewer than 5% of the women have ever been screened, the initiation of screening will uncover mainly prevalent disease, but the long-term cost-effectiveness will be driven mainly by the incidence of disease. An advantage of decision-analytic models is their ability to incorporate new information as it becomes available and to continuously reassess the optimal set of strategies.
What is the cost-effectiveness of an HPV type-specific vaccine compared with screening? With the promise of a type-specific HPV vaccine on the horizon (57), a new option to reduce the burden of cervical cancer will need to be entered in an already complex equation. In addition to the multiple conventional factors considered in cervical cancer screening models, information on vaccine efficacy, program accessibility, compliance, and feasibility will be important. To evaluate the cost-effectiveness of a type-specific HPV vaccine, a more detailed model will be required—one that is capable of simulating the entire natural history of cervical disease conditional on HPV type (58–60). There have been very few models calibrated simultaneously to the age-specific prevalence of HPV, LSIL, HSIL, and cervical cancer and to the distribution of HPV types across the spectrum of cervical disease. A new generation of models that incorporate this level of sophistication will allow for analyses that provide insight into the implications of using different surrogate markers in clinical vaccine trials and will be able to forecast the impact of vaccination on HPV, LSIL, and HSIL. The latter will be critical for understanding the benefits and costs of introducing vaccination into a partially screened population. Finally, an additional important priority is to develop methods to link HPV transmission and decision-analytic models. These methods will facilitate the assessment of the population impact of type-specific HPV vaccines for young males and females.
Presenting and Communicating Cost-Effectiveness Results
A third general priority relates to the methodologic rigor of cost-effectiveness analyses, the transparency of the assumptions made, and the translation and communication of the results to different target audiences (61).
How can analyses be made useful and accessible to planners in many places? Communicating the results of a decision-analytic study or a complex cost-effectiveness analysis is the responsibility of the analyst. First, the users of these models need to be able to feel comfortable that they meet well-established standards of quality and methodologic rigor. Analysts should compare their results with those of others and make the differences and similarities transparent to the reader. When possible, analysts should work together to resolve discrepancies. There should be the use of a consistent standard of evidence on the clinical effectiveness of different screening strategies. Second, analyses need to be presented in multiple formats (e.g., journal articles and monographs) for different target audiences. Journals are encouraged to post appendices on websites so that detailed descriptions of assumptions, methods, and any additional analyses are available. Third, to be responsive to a variety of stakeholders, analyses should adopt multiple perspectives. All analyses intended to inform policy should include an analysis from a societal perspective in which all of the costs and all of the benefits are included, regardless to whom they accrue (15). However, costs are frequently not "spent" and "saved" by the same party—for example, a governmental payer is less likely to be interested in patient time costs than in direct short-term medical costs. In addition, often managed-care organizations, state-health clinics, and other provider organizations require cost-effectiveness information over a shorter time horizon (e.g., 2- or 5-year time horizon) (56).
Scientific advances that contribute to our understanding of the natural history of HPV and cervical cancer will undoubtedly continue into the next decade. Disease-simulation modeling, capitalizing on data from multiple sources, can serve as a valuable tool to extend the time horizon of clinical trials, to evaluate more strategies than possible in a single clinical trial, and to assess the relative costs and benefits of alternative policies to reduce the mortality from cervical cancer. Cost-effectiveness analyses, when conducted rigorously and explained clearly, can illuminate the trade-offs with different policy alternatives, can take into account regional priorities, can provide qualitative and quantitative insight into the relative importance of different parameters, and can help to both prioritize and guide the design of future clinical studies. There are strengths and limitations of both experimental and modeling approaches; however, when used together, these methods can provide formidable assistance to public health decision makers faced with challenging policy choices.
|
NOTES |
|---|
Dr. Goldie’s research is supported in part by Public Health Service grant R01CA93435-01 from the National Cancer Institute, National Institutes of Health, Department of Health and Human Services; and by EngenderHealth, a member of the Alliance for Cervical Cancer Prevention, supported by the Bill and Melinda Gates Foundation.
|
REFERENCES |
|---|
|
TopNotesAbstractIntroductionDecision Science and Cost...Mathematical ModelsEvaluating UncertaintyPrevious Cervical Cancer ModelsThe Efficiency CurvePriority Areas for Future...References |
|---|
1 International Agency for Research on Cancer (IARC). IARC monographs on the evaluation of carcinogenic risks to humans. Human papillomaviruses. Vol 64. Lyon (France): IARC; 1995.
2 Bosch FX, Lorincz A, Munoz N, Meijer CC, Shah KV. The causal relationship between human papillomavirus and cervical cancer. J Clin Pathol 2002;55:244–65.
3 Franco EL, Rohan TE, Villa LL. Epidemiologic evidence and human papillomavirus infection as a necessary cause of cervical cancer. J Natl Cancer Inst 1999;91:506–11.
4 Gustafsson L, Ponten J, Bergstrom R, Adami HO. International incidence rates of invasive cervical cancer before cytological screening. Int J Cancer 1997;71:159–65.[CrossRef][ISI][Medline]
5 Sankaranarayanan R, Black RJ, Swaminathan R, Parkin DM. An overview of cancer survival in developing countries. IARC Sci Publ 1998;145:135–73
6 Pisani P, Parkin DM, Bray F, Ferlay J. Estimates of the worldwide mortality from 25 cancers in 1990. Int J Cancer 1999;83:18–29.[ISI][Medline]
7 Ponten J, Adami HO, Bergstrom R, Dillner J, Friberg LG, Gustafsson L, et al. Strategies for global control of cervical cancer. Int J Cancer 1995;60:1–26.[ISI][Medline]
8 Miller BA, Kolonel LN, Bernstein L, Young, Jr., JL, Swanson GM, West D, et al. Racial/ethnic patterns of cancer in the United States 1988–1992. National Cancer Institute, NIH Publication 96-4104. Bethesda (MD): National Institutes of Health; 1996.
9 Follen M, Richards-Kortum R. Emerging technologies and cervical cancer. J Natl Cancer Inst 2000;92:363–5.
10 World Development Indicators. The World Bank. Washington (DC): 2002 International Bank for Reconstruction and Development/The World Bank; 2002. Also available at www.worldbank.org/data. Last accessed December 23, 2002.
11 Sankaranarayanan R, Wesley R, Somanathan T, Dhakad N, Shyamalakumary B, Amm A, et al. Visual inspection of the uterine cervix after the application of acetic acid in the detection of cervical carcinoma and its precursors. Cancer 1998;83:2150–6.[CrossRef][ISI][Medline]
12 Visual inspection with acetic acid for cervical-cancer screening: test qualities in a primary-care setting. University of Zimbabwe/JHPIEGO Cervical Cancer Project. Lancet 1999;353:869–73.[CrossRef][ISI][Medline]
13 Denny L, Kuhn L, Pollack A, Wainwright H, Wright TC Jr. Evaluation of alternative methods of cervical cancer screening for resource-poor settings. Cancer 2000;89:826–33.[CrossRef][ISI][Medline]
14 Clinical decision analysis. 1976. In: Weinstein MC, Fineberg HV, editors. Clinical decision analysis. Philadelphia (PA): Saunders; 1980.
15 Russell LB, Gold MR, Siegel JE, Daniels N, Weinstein MC. The role of cost-effectiveness analysis in health and medicine. JAMA 1997;276:1172–7.
16 Gold MR, Siegel JE, Russell LB, Weinstein MC, editors. Cost-effectiveness in health and medicine. New York (NY): Oxford University Press; 1996.
17 Weinstein MC, Siegel JE, Gold MR, Kamlet MS, Russel LB. Recommendations of the Panel on Cost-Effectiveness in Health and Medicine. JAMA 1996;276:1253–8.[Abstract]
18 Drummond MF, O’Brien B, Stoddart GL, Torrance GW, editors. Methods for the economic evaluation of health care programs. New York (NY): Oxford University Press; 1997.
19 Muening P, editor. Designing and conducting cost-effectiveness analyses in medicine and health care. San Francisco (CA): John Wiley & Sons; 2002.
20 Halpern MT, Luce BR, Brown RE, Geneste B. Health and economic outcomes modeling practices: a suggested framework. Value Health 1998;2:131–47.
21 Weinstein MC, Toy EL, Sandberg EA, Neuman PJ, Evans JS, Kuntz KM, et al. Modeling for health care and other policy decisions: uses, roles and validity. Value Health 2001;4:348–61.[CrossRef][Medline]
22 Kuntz KM, Weinstein MC. Modeling in economic evaluation. In: Drummond M, McGuire A, editors. Economic evaluation in health care. New York (NY): Oxford University Press; 2001.
23 Frazier AL, Colditz GA, Fuchs CS, Kuntz KM. Cost-effectiveness of screening for colorectal cancer in the general population. JAMA 2000;284:1954–61.
24 Hagen MD, Garber AM, Goldie, SJ, Lafata JE, Mandelblatt J, Meltzer D, et al. Does cost-effectiveness analysis make a difference? Lessons from Pap smears. Med Decis Making 2001;21:307–23.[CrossRef][ISI][Medline]
25 Freedberg KA, Losina E, Weinstein MC, Paltiel AD, Cohen CJ, Seage GR, et al. The cost effectiveness of combination antiretroviral therapy for HIV disease. N Engl J Med 2001;344:824–31.
26 Sonnenberg FA, Beck JR. Markov models in medical decision making: a practical guide. Med Decis Making 1993;13:322–38.
27 Anderson RM, Garnett GP. Mathematical models of the transmission and control of sexually transmitted diseases. Sex Transm Dis 2000;27:636–43.[ISI][Medline]
28 Doubilet P, Begg CB, Weinstein MC, Braun P, McNeil BJ. Probabilistic sensitivity analysis using Monte Carlo simulation: a practical approach. Med Decis Making 1985;5:157–77.
29 Halpern E, Weinstein MC, Hunink MG, Gazelle GS. Representing both first and second order uncertainties by Monte Carlo simulation for groups of patients. Med Decis Making 2000;20:314–22.
30 Eddy DM. Screening for cervical cancer. Ann Intern Med 1990;113:214–26.
31 Fahs MC, Mandelblatt J, Schechter C, Muller C. Cost effectiveness of cervical cancer screening for the elderly. Ann Intern Med 1992;117:520–7.
32 Gustafsson L, Adami HO. Optimization of cervical cancer screening. Cancer Causes Control 1992;3:125–36.[CrossRef][ISI][Medline]
33 Sherlaw-Johnson C, Gallivan S, Jenkins D. Evaluating cervical cancer screening programmes for developing countries. Int J Cancer 1997;72:210–6.[CrossRef][ISI][Medline]
34 Jenkins D, Sherlaw-Johnson C, Gallivan S. Can papilloma virus testing be used to improve cervical cancer screening? Int J Cancer 1996;65:768–73.[CrossRef][ISI][Medline]
35 van Ballegooijen M, van den Akker-van Marle ME, Warmerdam PG, Meijer CJ, Walboomers JM, Habbema JD. Present evidence on the value of HPV testing for cervical cancer screening: a model-based exploration of the (cost-) effectiveness. Br J Cancer 1997;76:651–7.[ISI][Medline]
36 Cuzick J, Sasieni P, Davies P, Adams J, Normand C, Frater A, et al. systematic review of the role of human papillomavirus testing within a cervical screening programme. Health Technol Assess 1999;3(14):1–196.[Medline]
37 Goldie SJ, Weinstein MC, Kuntz KM, Freedberg KA. The costs, clinical benefits, and cost-effectiveness of screening for cervical cancer in HIV-infected women. Ann Intern Med 1999;130:97–107.
38 Myers ER, McCrory DC, Nanda K, Bastian L, Matchar D. Mathematical model for the natural history of human papillomavirus infection and cervical carcinogenesis. Am J Epidemiol 2000;151:1158–71.
39 McCrory D, Mather D, Bastain L, Datta S, Hasselblad V, Hickey J, et al. Evaluation of cervical cytology. Evidence report/technology assessment No. 5 (Prepared by Duke University under contract No. 290-97-0014). AHCPR Publ No. 99-E010. Rockville (MD): Agency for Health Care Policy and Research. February; 1999.
40 Brown AD, Garber AM. Cost-effectiveness of 3 methods to enhance the sensitivity of Papanicolaou testing. JAMA 1999;281:347–53.
41 Goldie SJ, Freedberg KA, Weinstein MC, Wright TC, Kuntz KM. Cost effectiveness of human papillomavirus testing to augment cervical cancer screening in HIV-infected women. Am J Med 2001;111:140–9.[CrossRef][ISI][Medline]
42 Goldie SJ, Kuhn L, Denny L, Pollack A, Wright TC. Policy analysis of cervical cancer screening strategies in low-resource settings: clinical benefits and cost-effectiveness. JAMA 2001;285:3107–15.
43 van den Akker-van Marle ME, van Ballegooijen M, van Oortmarssen GJ, Boer R, Habbema JD. Cost-effectiveness of cervical cancer screening: comparison of screening policies. J Natl Cancer Inst 2002;94:193–204.
44 Mandelblatt J, Lawrence W, Gaffikin L, Ringers P, Blumenthal P. Costs and benefits of different strategies to screen for cervical cancer in less-developed countries. J Natl Cancer Inst 2002;94:1469–83.
45 Hutchinson ML, Berger BM, Farber FL. Clinical and cost implications of new technologies for cervical cancer screening: the impact of test sensitivity. Am J Manag Care 2000;6:766–80.[ISI][Medline]
46 Kim J, Wright TC, Goldie SJ. Cost-effectiveness of alternative triage strategies for atypical squamous cells of undetermined significance. JAMA 2002;287:2382–90.
47 Mandelblatt JS, Lawrence WF, Wamack SM, Jacobson D, Yi B, Hwang YT, et al. Benefits and costs of using HPV testing to screen for cervical cancer. JAMA 2002;287:2372–81.
48 Myers ER, McCrory DC, Subramanian S, McCall N, Nanda K, Datta S, et al. Setting the target for a better cervical screening test: characteristics of a cost-effective test for cervical neoplasia screening. Obstet Gynecol 2000;96(5 Pt 1):645–52.
49 Montz FJ, Farber FL, Bristow RE, Cornelison T. Impact of increasing Papanicolaou test sensitivity and compliance: a modeled cost and outcomes analysis. Obstet Gynecol 2001;97:781–8.
50 Taylor LA, Sorensen SV, Ray NF, Halpern MT, Harper DM. Cost-effectiveness of the conventional Papanicolaou test with a new adjunct to cytological screening for squamous cell carcinoma of the uterine cervix and its precursors. Arch Fam Med 2000;9:713–21.
51 Mark D. Visualizing cost-effectiveness analysis. JAMA 2002;287:2428–30.
52 Neumann PJ, Goldie SJ, Weinstein MC. Preference-based measures in economic evaluation in health care. Ann Rev Public Health 2000;21:587–611.[CrossRef][ISI][Medline]
53 Goldie SJ, Kuntz KM, Weinstein MC, Freedberg KA, Welton M, Palefsky JM. The clinical-effectiveness and cost-effectiveness of screening for anal squamous intraepithelial lesions in homosexual and bisexual HIV-positive men. JAMA 1999;281:1822–9.
54 Goldie SJ, Kuntz KM, Weinstein MC, Freedberg KA, Palefsky JM. Cost-effectiveness of screening for HPV-induced anal squamous intraepithelial lesions in homosexual men. Am J Med 2000;108:634–41.[CrossRef][ISI][Medline]
55 Wright JC, Weinstein MC. Gains in life expectancy from medical interventions—standardizing data on outcomes. N Engl J Med 1998;339:380–6.
56 Kuntz KM, Goldie SJ. Assessing the sensitivity of decision-analytic results to unobserved markers of risk: defining the effects of heterogeneity bias. Med Decis Making 2002;22:218–27.[Abstract]
57 Koutsky LA, Ault KA, Wheeler CM, Brown DR, Barr E, Alvarez FB, et al. A controlled trial of a human papillomavirus type 16 vaccine. N Engl J Med 2002;347:1645–51.
58 Goldie SJ, Grima D, Kohli M, Wright TC, Weinstein MW, Franco E. The potential impact of a human papillomavirus 16/18 vaccine. In: Proceedings of the 2002 20th International Papillomavirus Conference. Paris (France); 2002. p. 243.
59 Sanders GD, Taira AV. Cost-effectiveness of a potential vaccine for human papillomavirus. Emer Infect Dis 2003;9:37–48.[ISI][Medline]
60 Myers E. Cervical cancer screening in an era of HPV vaccination. In: Proceedings of the 2002 20th International Papillomavirus Conference. Future direction in HPV disease prevention. Paris (France); 2002. p. 33.
61 Siegel JE, Weinstein MC, Russel LB, Gold MR. Recommendations for reporting cost-effectiveness analyses. JAMA 1996;276:1339–41.[Abstract]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  M. Vera-Llonch, E. Massarotti, F. Wolfe, N. Shadick, R. Westhovens, O. Sofrygin, R. Maclean, Y. Yuan, and G. Oster Cost-effectiveness of abatacept in patients with moderately to severely active rheumatoid arthritis and inadequate response to methotrexate Rheumatology, April 1, 2008; 47(4): 535 - 541. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||