You are on page 1of 9

Vaccine 24S3 (2006) S3/155–S3/163

Chapter 18: Public health policy for cervical cancer prevention: The role of decision science, economic evaluation, and mathematical modeling
Sue J. Goldie a,∗ , Jeremy D. Goldhaber-Fiebert b , Geoffrey P. Garnett c
a

Department of Health Policy and Management, Harvard School of Public Health, 718 Huntington Avenue, 2nd Floor, Boston, MA 02115, USA b Harvard University, Cambridge, MA, USA c Department of Infectious Disease Epidemiology, Imperial College London, London, UK Received 4 April 2006; accepted 15 May 2006

Abstract Several factors are changing the landscape of cervical cancer control, including a better understanding of the natural history of human papillomavirus (HPV), reliable assays for detecting high-risk HPV infections, and a soon to be available HPV-16/18 vaccine. There are important differences in the relevant policy questions for different settings. By synthesizing and integrating the best available data, the use of modeling in a decision analytic framework can identify those factors most likely to influence outcomes, can guide the design of future clinical studies and operational research, can provide insight into the cost-effectiveness of different strategies, and can assist in early decision-making when considered with criteria such as equity, public preferences, and political and cultural constraints. © 2006 Elsevier Ltd. All rights reserved.
Keywords: Decision science; Economic evaluation; Mathematical modeling; HPV

1. Introduction When making decisions we can either base choices on evidence and the explicit comparison of alternative strategies or we can use the intuitive opinions of experts. Decision science provides a quantitative framework for evaluating available evidence and uses explicit models to synthesize data, acknowledge uncertainty, and extrapolate from specific studies [1]. Several innovations are changing the landscape for cervical cancer prevention and control, with important differences in the most relevant policy questions for different settings. In countries with existing screening programs, pressing questions center around the optimal use of HPV-DNA testing, reducing disparities, and defining synergies between screening and vaccination. Important questions in low-resource settings include how to implement and sustain screening strategies with fewer technical and infrastructure requirements

Corresponding author. Tel.: +1 617 432 2010; fax: +1 617 432 0190. E-mail address: sue goldie@harvard.edu (S.J. Goldie).

than required by conventional cytology-based programs, how to target the appropriate age groups for screening, and how to overcome the logistical barriers associated with delivering a three-dose vaccine during early adolescence. Key questions include: what is the coverage that could realistically be achieved with screening between ages 35 and 40 versus vaccination at ages 10–12? Will both boys and girls need to be vaccinated? Is there a combined screening/vaccination strategy that could be cost-effective or will decision makers need to choose between the two? No single empirical study can evaluate all possible strategies to inform these complex policy questions. By integrating the best biologic, epidemiologic, economic, and behavioral data, the use of modeling in a decision analytic framework can assist in early decision-making, can highlight where better data are needed and identify those factors most likely to influence outcomes, can inform clinical study design and guide the conduct of operational research, and can provide insight into the potential cost-effectiveness of different strategies. Here we review the elements of cost-effectiveness analysis and identify the methodological issues most rele-

0264-410X/$ – see front matter © 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.vaccine.2006.05.112

such as lifeexpectancy. The choice of interventions included in a costeffectiveness analysis may vary between settings. as the ratios evaluate the costs and benefits of each strategy relative to the next most effective strategy [3]. Although such clinical outcomes are easily understood. compared to an alternative use of those resources. Decision analysis and cost-effectiveness analysis Decision science offers an explicit. and trade-offs. the relevant baseline comparator for the question of “What will an HPV16/18 vaccine add to our current cervical cancer program?” is current screening practice [10]. and is recommended for studies that inform the broad allocation of healthcare resources. When the baseline comparator is specified as the “null” (i. inclusion of boys. and systematic approach to decision-making under uncertainty [1]. no intervention). and the political and cultural consequences of decisions [2]. A decision analysis is considered a cost-effectiveness analysis when it compares the relative health and economic consequences associated with different interventions. which reflects competing societal demands for resources. dollar for dollar. A societal perspective incorporates all costs and all health effects regardless of who incurs the costs and who obtains the benefits. measured. all health outcomes (compared to an alternative) are included in the denominator and all costs or changes in resource use (compared to an alternative) are included in the numerator. such as affordability. An example might be the inclusion of several vaccination strategies that differ according to mode of operational delivery. The underlying principle guiding the valuation of resources is opportunity cost. equity.2. Interventions 2. The implication is that resources should be used as efficiently as possible in order to maximize the health benefits to the population. In order to compare ratios across different interventions and diseases. we are asking how much health improvement can be gained. calculated by subtracting the age at which death occurs from life expectancy at that age). Framework and perspective The choice of perspective depends on the analytic goal and context of the decision. interventions to reduce cervical cancer mortality include primary prevention (vaccination). These dimensions are not mutually exclusive. 2. years of life lost (a measure of the impact of an adverse health event. such as cases of cancer prevented. as is often the case in analyses considering multiple new interventions. Other perspectives. Several published guidelines advocate standard methods and assumptions to improve the quality and comparability of costeffectiveness analyses [2–9]. and intensity of recruitment.e. the incremental cost-effectiveness ratio that is calculated for the first intervention is sometimes referred to as an average cost-effectiveness ratio. Finally. There are many additional criteria considered in priority setting. district facility. 2. and dictates which costs and which health benefits to count and how to value them. / Vaccine 24S3 (2006) S3/155–S3/163 vant to upcoming policy questions in the cervical cancer field. perspective. and valued while also considering the uncertainty that exists about the outcomes at the time decisions are made. it may be useful to assess alternative strategies relative to the current standard of care. such as the payer Interventions can be categorized along many dimensions [6.7]. measures to improve quality of care.. For example. public preferences. The premise of a decision analytic approach is that all consequences of decisions should be identified. For example.3. In analyses intended for local decision-making. or referral hospital) or be population-based (organized immunization or screening programs). For example. secondary prevention (screening). and vaccination followed by a single lifetime screening between ages 35 and 40. Goldie et al. quantitative. .S3/156 S. Results of cost-effectiveness analyses are summarized using a cost-effectiveness ratio. a disadvantage is that results can only be compared to studies using the same outcomes. Cost-effectiveness analysis takes a utilitarian approach to healthcare policy since it explores the maximization of aggregate population health. may be more appropriate for specific local decisions.J. in the US. The choice of the baseline comparator may also vary depending on both the setting and the objective of the analysis. and therapy (treatment for invasive cancer). Health outcomes Health units for measuring the population impact can be disease-specific clinical outcomes. screening three times per lifetime. age of vaccination. Interventions may also include policy-related activities such as legislation (mandatory vaccination). In this ratio. different perspectives. In a cost-effectiveness analysis. reasonable alternatives might include vaccination alone. interventions differ in terms of their complexity and the intensity of demand they place on the health system. 2. the denominator must be expressed in a common metric. Costeffectiveness analyses are always comparative. Interventions can include the provision of personal health services (services at the clinic or school level. and economic incentives such as taxes or subsidies. and several core principles are described briefly below. This means that the costs and clinical benefits associated with the intervention of interest should be compared to all other reasonable options. and interventions might include combinations of features from different categories. A collection of quantitative methods is used to guide the management of complex problems that involve competing choices.1. in a low-income country annual Pap screening is not a realistic option – rather.

but also for the downstream events that follow. 2. measurement. the appropriate rate to use. immunization against HPV-16/18 could prevent cervical cancer decades later and can appear less cost-effective if the health gain is heavily discounted during that interval. For example. preventing cervical cancer saves the lives of relatively young women who have crucial roles in caring for dependents and maintaining the stability of the household and the larger community. travel distance required for screening.5–7]. The striking difference in the types of costs between locations illustrates the potential importance of the local organization of healthcare. particularly with respect to temporary health states experienced by women participating in screening programs (e. There is debate over how well health preferences are currently measured. 2. Although there is consensus about the need for discounting costs. Non-health outcomes Non-health benefits not captured in traditional costeffectiveness analyses range from a reduction in impoverishment from improving earning capacity.g.5. 2.. patient time is conventionally valued using wage rates. For analyses conducted from a societal perspective.3. diagnosis of HPV infection) and with diagnosis of sexually transmitted infections. 1 shows an example of screening cost components for selected countries in which many of these details were included. and economic conditions. Goldie et al. or reducing catastrophic out-of-pocket payments. specimen transport. The implication of this is clearly shown in Fig. child or dependent care. further tests and treatment). The fundamental distinction among various qualityadjusted measures is whether they describe a person’s state of health (i. costs incurred at the administrative levels rather than the point of care delivery). and proportion of the population in rural areas are all rarely formally considered but may be very influential on intervention costs. (3) patient time costs (e. more research is needed in this area.. where there is the added potential for social stigma [11]. Although there is no consensus on how to measure and quantify non-health benefits for inclusion in a cost-effectiveness analysis. calculated as the number of years of life saved adjusted for the quality of life during those years).7. additionally. especially in poor countries: details such as specimen transport.g..g.. which depicts the differential impact of discounting on screening as opposed to vaccination. For example. 2. false positive results.g. Cost units Costs should be presented in currency units that remove price inflation. laboratory tests. but may differ considerably between interventions. calculated as the future years of disabilityfree life that are lost as the result of premature deaths or cases of disability occurring in a particular year) [2. While the former may reflect under. The identification. and whether the rate should be constant [3... Key cost categories include: (1) direct healthcare costs (e. (4) programmatic costs (e. Discounting Future costs and benefits should be discounted to their present values to reflect inherent uncertainty about the future and preferences for timing of consumption. road network density. 2. and highlights the importance of including non-medical costs. These differences need to be considered in country-specific analyses. at a 3% discount rate. patient transportation costs. or disability-adjusted life years (a unit for measuring the health lost because of a particular disease. and therefore account for differences in price levels across countries. there is controversy about the discounting of benefits.2. they should at least be identified.4.1. work-up of abnormal cytology. For example.g. screening test. Preference measures reflect the fact that people with similar symptoms or levels of function may value that level of health differently..or overvaluation of the local currency they represent what is actually paid for locally-produced inputs [5].S. Finally. in contrast.5. available infrastructure. for analyses intended to inform resource allocation and compare studies from multiple countries. Discounting may undervalue interventions for which the benefits appear long after the costs have been paid. a cost of US$ 1 next year would be equivalent to US$ 0. . For example. there are obvious data limitations. attempt to reflect what the local currency is worth in purchasing power.g.13]. and valuation of these resources are not always straightforward. comparing resources used for HPV vaccination combined with screening to those used with vaccination alone). but this method underestimates the value of women’s time for those who either work in the informal sector or take care of the household and dependents [12].5. time spent by family for caregiving). costs should be expressed as US dollars or international dollars. to improving the probability of children remaining in school [2]. Cost outcomes The numerator of the cost-effectiveness ratio represents the difference in resources used when implementing one strategy compared to those used in the next best strategy (e. subsequent healthcare visits for treatment. Prices in local currency can be converted to US dollars using exchange rates or to international dollars using purchasing-power parity rates. not only for the intervention itself. utility) [3].97 today.8]. clinic visit.6–8.e. the time spent by the patient receiving care). The exchange rate for domestic currency into international dollars is the amount of domestic currency required to purchase the same quantity of goods and services as US$ 1 could purchase in the US [3.. / Vaccine 24S3 (2006) S3/155–S3/163 S3/157 quality-adjusted life years (a unit for measuring the health gain associated with a clinical or public health intervention.e. health status) or ascertain a value for a state of health (i. Fig. Programmatic costs are rarely included in analyses because they are difficult to estimate.5.J. costs must reflect resource use. Purchasing-power parity rates. (2) direct non-healthcare costs (e.

reduction in lifetime risk of cancer on the right vertical axis) of different screening strategies performed at different screening intervals are shown. An additional important issue relates to how the rate of change in costs compares with the change in benefits as vaccination coverage increases or “scales-up”. 3. such as increased costs of distribution and staffing in remote locations. Strategies lying on the efficiency curve dominate those lying to the right of the curve because they are more effective.5. supplies (medium blue). Although certain outcomes cannot be captured in a costeffectiveness ratio with current methods. Analysts may also choose to present cost-effectiveness results in other formats. 3. these might include intermediate outcomes (e. The scale and consequences of a screening or vaccination strategy that is subject to limited monetary resources can be represented by expressing the results as the cost and effects of applying the intervention to a target population of 1 million. as shown in Fig. analysts must begin .S3/158 S. For example. for example. / Vaccine 24S3 (2006) S3/155–S3/163 Fig.. the perperson cost of delivering an intervention is reduced as coverage is increased) with. and equipment and laboratory (light blue). The functional relationship between cost and scale is frequently ignored but is worthy of further empirical study [6. The cost-effectiveness ratios associated with these screening strategies reflect this relationship in that they are two. Other cost issues Certain costs (aside from vaccine price) that are of particular relevance to vaccination strategies include costs of delivery strategies.3. bulk manufacturing and supply chain management. twice. while the colored regions within bars represent categories of costs.7. Direct medical costs are subdivided into those attributable to staff (dark blue). 1. Scale-up refers to the changes in an intervention’s effectiveness and costs as coverage is expanded to larger percentages of the eligible population. and the costs of achieving incremental increases in coverage rates.g. comprehensiveness of coverage by the primary clinic.J.to five-fold higher.. and/or difficulty in travel.14]. which signifies rapidly escalating clinical benefits for only modest increases in costs. There are no empirical data on any of these for a threedose vaccine targeting adolescents. When data were unavailable to disaggregate the direct medical costs. cases of cancer prevented) and discounted and undiscounted results. economies of scale might be achieved (i. wages. For some costs. Reporting the results of a cost-effectiveness analysis Cost-effectiveness results are often displayed in the format of an efficiency curve. and either cost less or have a more attractive cost-effectiveness ratio than the next best strategy. Some strategies lie near the efficiency curve. costs attributable to vaccine wastage. where slight changes in direct medical costs or test performance would make them equally cost-effective as the strategies lying on the efficiency curve. and three times per lifetime dominate other strategies [12]. and receiving care (yellow). Other costs include transportation costs (red) and the time associated with women traveling. However. there may also be diseconomies of scale. The slope between two strategies is steeper when the net gain in health per unit cost is greater. The two and three times in a lifetime screening strategies are positioned higher on the curve—in this region of the curve the slope is steep. The considerable variation in patient time and transport costs reflects differences in rural population and density. In this example. strategies of screening with HPV-DNA testing once. The height of each bar represents the total costs associated with the initial screening visit for HPV-DNA testing and cytology. Goldie et al.e. Screening cost components shown for selected countries for two cervical cancer strategies expressed in 2000 international dollars. where the lifetime costs and clinical benefits (discounted life expectancy on the left vertical axis. An advantage of using a tabular format is that a range of outcomes that different decision makers find helpful may be presented for each strategy. waiting. we placed these into a single category (green). 2.

S. In contrast.15]. respectively [6. using the per capita GDP as a rough indicator of monetary resource constraints is not unreasonable. with the benefits realized decades later.J. 4. distributional impacts and equity considerations. A commonly-cited rule of thumb is based on a report by the Commission on Macroeconomics and Health (CMH). Panel A: the cost-effectiveness ratio for a range of discount rates. The impact of discount rate on the cost-effectiveness ratio (CER) associated with a vaccination and screening program.g.. / Vaccine 24S3 (2006) S3/155–S3/163 S3/159 Fig. provided there is acknowledgement that costeffectiveness is only one relevant input for policy decisions.5–3 times the per capita GDP) allows loose classification of . Because of the interaction between costeffectiveness. Therefore. both the costs and benefits associated with screening every 5 years between ages 35 and 55 occur in approximately the same time frame. These include identifying non-health benefits and the extent to which interventions differentially place demands on the health-system capacity. 0. below which an intervention would be considered cost-effective. disease burden. On the single dimension of cost-effectiveness. to qualitatively identify results that could influence the comparative attractiveness of different interventions. Panel B: the timing of costs (C) and benefits (B). the costeffectiveness ratio alone is an inadequate guide to priority setting: additional criteria such as affordability. and when comparable methods are used to conduct analyses. the vast majority of costs are up front. there is no universal criterion that defines a threshold cost-effectiveness ratio. capacity to deliver interventions. and available funds. and public preferences can often be more influential [2]. 2. having some threshold range (e. Interpreting the results of a cost-effectiveness analysis While interventions that improve health at a cost should ideally be compared with other interventions that compete for the same resources. and subsequent rec- ommendations that interventions are “very cost-effective” and “cost-effective” if they have cost-effectiveness ratios less than the per capita GDP or three times the per capita GDP. Goldie et al. For a vaccination strategy.

and probability of immunity following HPV infection.J. some of the dominated alternatives have been left out of the illustration. there may indeed be a lower “threshold ratio” required to compete for scarce vaccination resources if the ratios associated with existing vaccines (e. state transition models). Use of models Models can be classified along several dimensions: according to the structure used for events that occur over time (e... and evaluate uncertainty and variability. Models utilize a wide range of approaches to calibrate to data. deterministic. In state-transition models used to assess cervical cancer prevention strategies. in a transmission model. Goldie et al. Three-visit strategy refers to Pap testing followed by colposcopy-directed biopsy. where the state of the system at the next time-step depends upon the current system and terms reflecting movement between categories over the entire interval. While in a disease simulation model.S3/160 S. allow healthy debate and discourse over alternative model structures. and assumptions. An example of an efficiency curve derived from an analysis of potential screening strategies with costs derived for South Africa and including different number of lifetime screens (1 time at age 35. the probability of HPV acquisition is permitted to depend on age and a general sexual risk category.16. Diagrams are often developed to highlight the most influential model parameters and. interventions into broad categories of very cost-effective. how the cost-effectiveness of an HPV vaccine compares with the cost-effectiveness of other well-accepted vaccinations will be critical.g. such as the type used to assess the cost-effectiveness of screening. Difference equations represent discrete jumps in time. differential equation models. decision trees. HPV type. 3. all models involve trade-offs. For an HPV-16/18 vaccination strategy in particular. the requirement to consider the cost/benefit profile relative to other investments in other disease areas is more prominent. 40. by making assumptions explicit. / Vaccine 24S3 (2006) S3/155–S3/163 Fig. in addition to comparing screening and vaccination approaches within the cervical cancer paradigm. which are the most cost-effective. estimate uncertain or unknown parameters. In contrast. and treatment when diagnosed with pre-invasive or invasive lesions. which depends upon the mathematical framework. 5. sexual risk. When cost-effectiveness analysis is used to inform “allocative efficiency”. important factors influencing the probabilities governing movement between heath states include age.g. Both models of “transmission dynamics of infection” and “disease simulation models” categorize individuals into mutually-exclusive health states. It also allows us to conduct uncertainty analyses that identify the probability that a cost-effectiveness ratio is below some threshold. and 45): the lifetime costs (horizontal axis) and clinical benefits (discounted life expectancy on the left vertical axis. consequently. exposure to screening. meaning that given several competing options for cervical cancer screening. reduction in lifetime risk of cervical cancer on the right vertical axis) of different screening strategies. and whether they reflect the transmission dynamics of infection [3. full details can be found in Ref. parameters. In the past. though complex models of this variety are often analyzed using discrete approximations.g. whether they are open or closed and according to the nature of the target population (e.. As a general rule. including more detail means that more parameter values are required and. Examples of two model schematics are shown in Fig. 2 times at ages 35 and 40. the probability of an individual acquiring an infection is modeled as dependent on the detailed sexual contact patterns of that individual and the distribution of the infection within the population at that . by the method of calculation (e. and potentially cost-effective. represent systems in continuous time. Two-visit strategy refers to HPV testing followed by treatment if HPV-positive. 4. Therefore. a longitudinal cohort model. One-visit strategy refers to testing with either HPV or VIA. not cost-effective.18]. the model becomes more complicated and challenging to analyze [16. either ordinary or partial. followed by immediate treatment if screen positive. childhood immunization) being used are much lower than the GDP rule of thumb cited above.g. stochastic).17]. For simplicity.. Movement between health states can be described by a system of equations. most cost-effectiveness analyses for cervical cancer screening have concentrated on questions of “technical efficiency”. or 3 times at ages 35. cross-sectional population model). For example. [12].

allowing people to enter or exit over time. In model (b) women who acquire infection can resolve. can provide estimates of the reduction in 16/18-associated cervical cancer but are more limited in terms of exploring detailed questions around the natural history. smoking.J. It is important to note that these figures represent stylized examples and are not intended to be comprehensive representations of any particular model. simplifying assumptions are unavoidable. 4. although these models may also be open.). can include considerable detail on different screening and treatment strategies.” in that no one enters or exits the cohort at any time during the simulation. models that describe the transmission dynamics of infection are necessary to estimate the impact of herd immunity. Even for HPV-16. / Vaccine 24S3 (2006) S3/155–S3/163 S3/161 Fig. Since so many variables are unknown. Most published cervical cancer screening models are state-transition models that are “closed. individual-based risk factors such as parity. Many of the parameters required for transmission models are largely unknown for HPV types other than HPV-16 [22]. Goldie et al. A static state-transition model is well suited to simulate chronic disease with multiple stages over relatively long time periods. and resource use associated with non-16/18 types. Because interventions that affect the individual have consequences for the risks of infection in the rest of the population. and are efficient in terms of computational intensity. women have acquired immunity and that women always progress to cervical intraepithelial neoplasia (CIN)-3 via CIN-2.23. and can accommodate the detailed cost measures associated with each health state [16. remain infected. “open” population-based models and first-order Monte Carlo simulation allow for more flexibility in capturing multiple dimensions of heterogeneity (e. Such depictions make explicit model assumptions. Transmission models that only include HPV16.24]. As models evolve in complexity. The squares describe mutually-exclusive categories and the arrows indicate the transitions between these categories described by the model. Advantages of closed models are that they are transparent. in model (a) it is assumed that when infection is treated or resolves. the figures illustrate one aspect of the model – the description of the natural history – but the actual models can contain further complexities through stratification of the model according to other variables such as age. Calibration techniques that fit model output to epidemiologic data on age-related prevalence of type-specific HPV.g. neither of which may be valid. explore the relative value of vaccinating boys in addition to girls. which can then be debated. However. the data are limited for transmission rates in men and women and for detailed sexual behavior stratified by variables such as gender. In both cases. or only HPV-16 and -18. etc.. or can develop chronic infection but can progress straight to more severe neoplasia without progressing through earlier stages.S. thereby reflecting both variability and uncertainty and permitting the risk of future events to depend on one or more prior events [25]. age-related incidence of cervical intraepithelial neoplasia (CIN) and inva- . screening history. and explore the impact of sexual mixing patterns on the projected age-specific HPV prevalence following vaccination. can easily accommodate probabilistic uncertainty analysis. type-specific HPV. etc. For example. Flow diagrams representing a simple transmission model (a) and a disease simulation model of HPV infection and progression to cervical cancer (b). specific time [19–21]. age. the requirement for parameter values quickly multiplies and input values will almost certainly never be available for all parameters. and other risks. outcome.

Merck and Co. Statistical issues in cost-effectiveness studies are different from those that arise in experiments or other data analyses. 1996.J. and cost-effectiveness ratios. NY: Oxford University Press. Fineberg HV. UK: Oxford University Press and the World Bank. Evans DB. 2003. Goldie et al. This process will include simple tests as well as more complex calibration exercises. p.. Breman JG. MA. Oxford textbook of public health.18]. Goldie is supported in part by the US National Cancer Insitute (grant #R01 CA093435) and The Bill and Melinda Gates Foundation (grant #30505). Oxford. limited set of parameters. Decision science and cost-effectiveness analyses offer important tools that can help to address key cervical cancer policy questions relating to screening and vaccination. BMJ 2005. Goldhaber-Fiebert is supported in part by a National Science Foundation Graduate Research Fellowship. Conclusion Evaluating the effectiveness of a public health prevention program is complex. all parameters should be varied simultaneously. Siegel JE. In: Detels RG. Oxford. Validation of a model requires a demonstration that the output of the model is consistent with known facts about the disease. et al. 2006. Grima D. NY: Oxford University Press. Disease Control Priorities Project. [8] Jamison DT. 2006. Adam T. Boston.uk. Edejer TT.and middle-income countries. editors.331(7525):1137–40. Cost-effectiveness analysis for priority setting. UK: Oxford University Press. 271–85. et al. Research Grants (GlaxoSmithKline) Acknowledgments Dr. In a sensitivity analysis. Jha P. New York. Gordillo-Tobar A. 4. Guide to the Methods of Technology Appraisal. whereas to understand the uncertainty. Sex Transm Infect 2006. We gratefully acknowledge the valuable contributions of Steven Sweet and Meredith Holtan of the Harvard School of Public Health. editors. model-based evaluation studies aim to portray the scope and nature of the uncertainties that surround the estimates of costs. Mr. Mills A. Levin C. Wardle J.nice. Probabilistic sensitivity analysis can be performed using a Markov model analyzed as a cohort simulation. http://www. PA: Saunders. N Engl J Med 2005. 2002. An important caveat is that data for estimating probability distributions around mean parameter estimates are seldom available in low.who. McEwen J. Stoddart GL. DCPP Working Papers. [13] Jamison DT. The range of values used for a sensitivity analysis can be based on statistical variation for point estimates or on probability distributions. Bethesda. editors. Weinstein MC. 1976. and the distribution of HPV types within CIN and cancer. 2006. Russell LB.353(20):2158–68. [4] Drummond MF. [7] Disease Control Priorities Project. Torrance GW. benefits.96(8):604–15. and using recently-developed techniques. Predictive validity refers to whether a model produces outputs that are consistent with observations independent of data used to calibrate the model. Lim SS. April 21.org/main/Home. particularly when the course from infection to disease spans multiple decades. 1980 ed. parameter estimates. 1997. References [1] Weinstein MC. some critical component in the calculation is varied over a plausible range and the cost-effectiveness ratio is recalculated. In: Jamison DT. . [2] Musgrove P. Methods to assess the costs and health effects of interventions for improving health in developing countries. April 21. The resulting difference in the ratio provides some indication of how sensitive the results might be to a change in that parameter. editors. Kohli M. [12] Goldie SJ. Fogarty International Center. Sanofi-Pasteur). when new activities are building upon existing interventions. Tanaka H. Social and psychological impact of HPV testing in cervical screening: a qualitative study. Wright TC. Goldhaber-Fiebert JD. Cost-effectiveness analysis: concepts and applications. in which statistical methods are used to maximize the likelihood function for observed data given a set of parameter values. Cost-effectiveness of cervicalcancer screening in five developing countries. Probabilistic sampling assumes some knowledge of the distributions surrounding a model’s input parameters and takes random draws from these distributions to calculate outcome variables. Fox-Rushby J. [11] McCaffery K. Mahe C. p.27]. Projected clinical benefits and cost-effectiveness of a human papillomavirus 16/18 vaccine. April 21.html. are increasingly being used [25]. [5] Evans DB. 2nd ed. O’Brien B. 903–19. [6] WHO (World Health Organization): WHO-CHOICE: World Health Organization statistical information system: CHOICE (CHOosing Interventions that are Cost Effective). Nazroo J.int/whosis/en/. Weinstein MC. Gaffikin L. and assumptions [3. Musgrove P. Beaglehole R. Claeson M.S3/162 S.dcp2. [3] Gold MR.org. National Institutes of Health. for their outstanding technical assistance. 2006. Disease Control Priorities in Developing Countires. Clinical decision analysis.. Alleyne G. [10] Goldie SJ. MD.16. and when resource constraints limit the range of reasonable choices. Philadelphia. Jamison JS. http:// www. New York. Bosch FX. The most common way to evaluate the stability of conclusions over a range of parameter estimates and structural assumptions is to conduct sensitivity analyses. 4th ed. Inc. Waller J. Disclosed potential conflicts of interest GPG: Consultant (GlaxoSmithKline.82(2):169–74. [9] NICE (National Institute for Clinical Excellence). Measham AR. Costeffectiveness in Health and Medicine. DCPP Working Paper No. The process described above focuses on discerning the influence of a particular. A critical part of any decision analysis is to evaluate the uncertainty in the model structure. Discounting. can be performed on a Markov model analyzed as a firstorder Monte Carlo simulation [26. Sanofi-Pasteur MSD. http:// www. editors. J Natl Cancer Inst 2004. Methods for the economic evaluation of health care programs. Rather than testing hypotheses using traditional statistical significance as a criterion. 6. / Vaccine 24S3 (2006) S3/155–S3/163 sive cancer.

Calibration of a natural history model of cervical cancer using longitudinal primary data [Abstract]. Modelling in economic evaluation: an unavoidable fact of life. Sandberg EA.20(3):314– 22. [17] Buxton MJ. et al. Economic evaluation in health care. [15] Investing in Health for Economic Development. Goldhaber-Fiebert JD. p. Modeling the sexual transmissibility of human papillomavirus infection using stochastic computer simulation and empirical data from a cohort study of young women in Montreal Canada. 2001. Beck JR. [24] Sonnenberg FA. In: Drummond MF. Med Decis Making 1993. [26] Halpern EF. Tellier PP. J Natl Cancer Inst Monogr 2003. Kuntz KM. [21] Garnett GP.191(Suppl 1):S97–S106. McGuire A. 2001. et al. Neumann PJ. S3/163 [22] Burchell AN. Prince RL. Merging Theory with Practice. Mahmud SM. Representing both first. Hunink MG. . / Vaccine 24S3 (2006) S3/155–S3/163 [14] Johns B. Murray CJ. Chapter 15: public health policy and cost-effectiveness analysis. Int J Technol Assess Health Care 2002. [16] Kuntz KM. Van Hout BA. roles. Mathematical models of the transmission and control of sexually transmitted diseases. Probabilistic uncertainty analysis and stochastic league tables. Evans DB.31:102–10. [25] Kim JJ. [18] Weinstein MC. Hanley J. In: Proceedings of the 22nd international papillomavirus conference and clinical workshop. Health Policy Plan 2005. Evans JS. Modeling in economic evaluation. Garnett GP. editors. Mahmud S. New York. Sheldon TA. NY: Oxford University Press.163(6):534–43. Report of the Commission on Macroeconomics and Health.20(1):1–13. Waddell HC. [19] Anderson RM. [20] Garnett GP. Modeling for health care and other policy decisions: uses.S. Torres TT. Uncertainty in cost-effectiveness analysis. Health Econ 1997.18(1):112–9. Costs of scaling up health interventions: a systematic review.and second-order uncertainties by Monte Carlo simulation for groups of patients. Weinstein MC. Switzerland: WHO (World Health Organization). [23] Goldie SJ. Geneva. 141–71. Role of herd immunity in determining the effect of vaccines against sexually transmitted disease. Toy EL. Villa LL.J. Public health paradoxes and the epidemiological impact of an HPV vaccine. Markov models in medical decision making: a practical guide. Value Health 2001. Szucs T.4(5):348–61. J Infect Dis 2005. Richardson H.27(10):636–43.19(1–2):101–11.13(4):322– 38.6(3):217–27. Drummond MF. [27] Baltussen RM. et al. Med Decis Making 2000. Am J Epidemiol 2006. 2005. Hutubessy RC. Weinstein MC. Sex Transm Dis 2000. Kuntz KM. Trottier H. Goldie SJ. Goldie et al. J Clin Virol 2000. and validity. Gazelle GS.