|Home | About | Journals | Submit | Contact Us | Français|
Disease simulation models of the health and economic consequences of different prevention and treatment strategies can guide policy decisions about cancer control. However, models that also consider health disparities can identify strategies that improve both population health and its equitable distribution.
We devised a typology of cancer disparities that considers types of inequalities among black, white, and Hispanic populations across different cancers and characteristics important for near-term policy discussions. We illustrated the typology in the specific example of cervical cancer using an existing disease simulation model calibrated to clinical, epidemiological, and cost data for the United States. We calculated average reduction in cancer incidence overall and for black, white, and Hispanic women under five different prevention strategies (Strategies A1, A2, A3, B, and C) and estimated average costs and life expectancy per woman, and the cost-effectiveness ratio for each strategy.
Strategies that may provide greater aggregate health benefit than existing options may also exacerbate disparities. Combining human papillomavirus vaccination (Strategy A2) with current cervical cancer screening patterns (Strategy A1) resulted in an average reduction of 69% in cancer incidence overall but a 71.6% reduction for white women, 68.3% for black women, and 63.9% for Hispanic women. Other strategies targeting risk-based screening to racial and ethnic minorities reduced disparities among racial subgroups and resulted in more equitable distribution of benefits among subgroups (reduction in cervical cancer incidence, white vs Hispanic women, 69.7% vs 70.1%). Strategies that employ targeted risk-based screening and new screening algorithms, with or without vaccination (Strategies B and C), provide excellent value. The most effective strategy (Strategy C) had a cost-effectiveness ratio of $28200 per year of life saved when compared with the same strategy without vaccination.
We identify screening strategies for cervical cancer that provide greater aggregate health benefit than existing options, offer excellent cost-effectiveness, and have the biggest positive impact in worst-off groups. The typology proposed here may also be useful in research and policy decisions when trade-offs between fairness and cost-effectiveness are unavoidable.
Disease simulation models can be used to select both medically and economically effective cancer prevention strategies. However, all segments of the population may not benefit equally because of disparities in access to preventive care.
A typology of cancer disparities among black, white, and Hispanic populations was applied to the example of cervical cancer using a disease simulation model calibrated to clinical, epidemiological, and cost data for the United States.
Explicitly modeling disparities between subgroups identified prevention strategies that could reduce the risk of cervical cancer overall as well as make access to prevention more equitable. In addition, when combined with vaccination, the proposed changes to screening were more effective and less costly than current screening patterns.
Although trade-offs between fairness and cost-effectiveness are unavoidable, explicit modeling of inequalities can improve access to health care for all segments of the population.
The model was applied to a simplified example of cervical cancer prevention that included only a select number of strategies. Systemic disparities and differences attributable to choice can be difficult to model because they may be hard to distinguish in some instances. The incremental costs of implementing targeted programs to reduce disparities were not included.
From the Editors
Disease simulation models can guide policy decisions about cancer control by comparatively assessing the health and economic consequences of different prevention and treatment strategies. Health policy and cancer control policy specifically have two fundamental goals, improving population health and distributing that health fairly. These goals are often in tension, because measures to improve one may not be optimal for and may conflict with the other. Increased emphasis on the second goal is reflected by the focus on disparities of various US health agencies, including the National Cancer Institute (1–6). Our objective is to facilitate policy efforts to pursue both improvement and fair distribution. To do so, we evaluate cancer interventions to reveal the impact on both population health and on relevant health disparities and to identify strategies compatible with both.
The evaluation of different cancer interventions is particularly applicable to disease modeling. Predicting the population-level impact of a prevention program for cervical cancer is complex, because the time course from infection to cancer spans multiple decades, primary and secondary prevention efforts focus on different components of the disease spectrum, and randomized controlled trials are not feasible or ethical (7,8). Computer-based disease simulation modeling can provide a formal framework to synthesize data in an internally consistent and epidemiologically plausible way, extrapolate intermediate outcomes beyond the time horizon of individual studies, and compare multiple strategies targeting different points in the disease course.
Dramatic advances in the cancer field provide an opportunity to enhance the disease models of the 1990s. Increased knowledge about the natural history of disease has been accompanied by advances in modeling methodology, programming efficiency, and the ability to harness enormous computer power for data analysis and simulation. Contemporary models not only reflect the average course of disease in the population but also can capture important heterogeneities (eg, biological differences, disease determinants, risk factors).
Disease-specific cancer models are often used to generate the comparative health outcomes (ie, comparative effectiveness) associated with different prevention strategies. When this type of decision analysis also includes a comparison of economic consequences, it is considered to be a cost-effectiveness analysis, in which the underlying principle guiding the valuation of resources is opportunity cost, reflecting competing societal demands for limited resources (9). Traditionally, cost-effectiveness analyses maximized the health of a population subject to a resource constraint and did not explicitly address the distribution of benefits (10); this “neutral stance” has been the focus of ethical criticism because it fails to consider equity or fairness issues.
Because many cancer simulation models now incorporate subgroup heterogeneities, producing estimates of both aggregated and subgroup-specific outcomes is feasible. The ability to reflect differences between racial and ethnic subgroups prompts consideration of how we might more formally include distributional equity (ie, fairness in the distribution and magnitude of health benefits provided by interventions) in cost-effectiveness analyses; however, to do so, careful thought is required about which factors that contribute to inequalities are of potential ethical relevance. If we are to encourage decision analysts to consider the distribution of important health outcomes among subgroups—and the policies to potentially alter them—then we need to be clear about the definition of “disparity.” This is not straightforward because different definitions are used by various health and research agencies (1–4,6).
Several US agencies have developed definitions of health disparities (Table 1), but they vary in how they differentiate inequalities from disparities (1–4,6). For example, the definition by the Agency for Healthcare Research and Quality (AHRQ) treats any inequality greater than 10% of the reference group as a health disparity between population groups, even if the inequality is attributable to a genetic vulnerability or an informed personal choice (1). In our view (11,12) and that of others (6,13), only inequalities that arise from certain problematic differences in the distribution of the socially controllable factors influencing health should count as health disparities; those resulting from genetic or other biological variation or from informed voluntary choices are inequalities.
The AHRQ definition has the advantage that it is purely arithmetic and therefore transparent. It also has the disadvantage of ignoring the fact that many people reasonably want to distinguish and pay special attention to those inequalities that are unfair or unjust. Although we recognize that there may be some disagreements about what counts as unjust, even an egalitarian view of distributive justice (14) would consider differences arising from informed religious choices as fair and not as a disparity. A second feature of some US agency definitions is that they focus on health inequalities among certain social groups, without specifying why those inequalities should count as ethically problematic disparities. In some cases, the groups are the ones that clearly suffer from past or present discrimination, so the definition may rest on the assumption that those groups typically suffer from ethically problematic distributions of the socially controllable factors that affect health. Nevertheless, these approaches do not specify exactly what explains why the health inequalities between these social groups are unjust. If we want to have clear models of the impact of strategies on health for different groups, we must clearly define the mechanisms that underlie their health differences.
One way to defend current agency definitions would be to claim that the term disparity does not carry with it the implication that the inequality is ethically problematic (unjust or unfair), but then we need another notion (perhaps “inequity”) that carries this implication for a subgroup of disparities that we have extra reason for reducing. We, however, treat the terms inequity and disparity as equivalent and reserve for inequality the more general notion of difference, without any morally problematic implication. We note, of course, that we have social obligations to meet health needs, whether or not they are the result of an ethically problematic distribution of the socially controllable factors affecting health. In Daniels’ (12) account of justice and health, such general obligations are rooted in our obligation to protect the range of opportunities open to people by keeping them as close as we can to functioning normally. People might reasonably think, however, that we have additional reason to redress health disparities because they result from some form of social injustice in socially controllable factors (12).
Definitions matter to model-based analyses intended to inform policy. If we can focus on the causes of differences that are ethically problematic and so make sure we are quantifying disparities, then analyses that seek to provide information on both average outcomes and the distribution of outcomes will be informative. If we simply identify differences between groups, we will not be able to distinguish strategies that reduce disparities from strategies that simply alter selected inequalities. Our premise, based on the ethical arguments above, is that we have additional reason to reduce those differences that are disparities when faced with policy alternatives.
We propose a typology of cancer disparities that accommodates the variation in types of inequalities across different cancers and allows for an explicit discourse about their ethical importance. The typology differentiates between inequalities in health status caused by differences in genetic or other biological factors (or by differences in informed behaviors or preferences) and those that result from ethically unacceptable differences in access to services or in the distribution of socially controllable factors (eg, poverty, socioeconomic status, insurance). We first discuss this typology in the context of breast, colorectal, and cervical cancer. We then demonstrate its application using an existing model that captures inequalities in cervical cancer incidence, stage distribution, and mortality among racial subgroups. We conduct an analysis that compares the health and economic outcomes of different prevention strategies and present the results using aggregated population-based measures in a conventional format, followed by an augmented format that presents outcomes across racial and ethnic groups. In this context, we evaluate the total population-level benefits, the distribution of benefits, and the projected widening or narrowing of disparities associated with several strategies.
In the typology of cancer disparities proposed here, an individual’s initial cancer risk evolves from early in life (through biological and/or genetic factors), throughout the life span (during which he/she may acquire risk factors), and then through a period in which access to cancer screening or other intervention may modify the risk (Figure 1). Following a diagnosis of cancer, the mortality rate and risk of recurrence (ie, long-term survival) are modified by the access, acceptability, quality, and ongoing availability of health care (Figure 1). Although certain factors (eg, genetic risk) are not modifiable, others are, at least in theory. Social determinants may affect cancer incidence through direct or indirect influence on environmental exposures and risk factors; socially controllable factors may also affect the likelihood of cancer incidence, mortality, and long-term survival through their influence on access to health care (prevention and treatment) and the quality of that care (Figure 1).
Our typology assumes that an inequality in a health-related outcome counts as a disparity only if it is the result of an ethically problematic distribution of socially controllable factors affecting cancer-related negative health consequences (eg, incidence, stage distribution, mortality) (11). Inequalities in negative health outcomes from genetic or other biological variation, and as a result of informed voluntary choices, may indeed be the focus of health services efforts but are not disparities.
Although we note the general potential relationships between social determinants and cancer risks and outcomes (Figure 1), we recognize that for each specific cancer, selected factors will be more or less salient. Specific cancers are associated with inequalities in different outcomes (eg, incidence, stage distribution, mortality); the reasons for the inequalities, when known, can help us to appropriately identify those that are disparities.
In addition, mapping specific cancers within our typology framework (Figure 1) can help identify interventions that would be required to narrow those inequalities considered disparities. For each category of social determinants and indicators of health-care accessibility, we indicate identifiable characteristics important for near-term policy discussions (Figure 1, right three columns). These include 1) whether the social determinant or health access indicator is generally included in disease-specific cancer models, 2) whether the influential factors contributing to the inequality are modifiable or tractable, and 3) based on the first two criteria, what the level of priority should be for inclusion in near-term policy options. In general, we assume that access and care factors are more tractable than social determinants for near-term policy choices, which should be considered in model-based analyses. This difference in tractability is analytically and ethically relevant because our obligation to improve a situation depends on our ability to actually do so.
We applied our typology in three different cancers, cervical (Figure 2), breast (Figure 3), and colorectal (Figure 4), to illustrate how the association (magnitude and direction of influence) between social determinants and outcomes differs and to help identify disparities that would be early priorities for policy intervention.
Cervical cancer is caused by infection with oncogenic types of human papillomavirus (HPV). Although the age-related prevalence of type-specific HPV varies, in the United States, differences in observed incidence rates between racial subgroups (eg, black vs Hispanic women) are predominantly a function of access to population-based screening and follow-up (eg, treatment of screen-detected precancerous abnormalities) (1,4,15–20). Screening for cervical cancer detects cervical abnormalities before they become cancerous, so it reduces both the incidence and mortality of cancer. Differences in cervical cancer outcomes are therefore largely attributable to differences in cervical cancer screening participation. Although self-reports of recent cervical cancer screening by black women are similar to those in white women, Hispanic women are much less likely to have been screened in the last 3 years. In addition, compared with white women, a greater proportion of black and Hispanic women report having never received a Pap test (1,4,15–20). Differences in ever screening between race and ethnic subgroups are particularly important in light of findings that approximately half of cervical cancers are among women who have not been screened in the past 3 years or ever. In black women, mortality is two to three times higher, and 5-year survival is lower compared with white women. This is attributable to multiple factors including diagnosis at later stages, higher rates of loss to follow-up, and a lower probability of receiving appropriate treatment. Because 80% of cervical cancer deaths occur in women of low socioeconomic status, other social determinants that can be overrepresented in some racial subgroups (eg, income level or insurance status, knowledge of and access to cervical cancer screening, distribution of health services within a region) also influence access and quality of care (1,4,15–20).
Cervical cancer incidence, mortality, and recurrence would all be considered disparities because they are secondary (at least in part) to socially controllable factors (Figure 2). Given that cervical cancer screening and treatment differences are generally included in cervical cancer disease models and are tractable, we would consider them to be near-term policy targets for priority attention. This implies that in a cost-effectiveness analysis, a policy analyst might include strategies that target racial and ethnic minorities, as well as women of low socioeconomic status who lack insurance for enhanced access to screening, improved treatment access, and quality of care.
In contrast to cervical cancer screening, breast cancer screening reduces mortality rather than incidence, because detection of cancer in earlier stages allows for more successful treatment than cancer detected in later stages. The incidence of breast cancer in the United States is higher among white women than black women and is also higher among women of higher socioeconomic status. This increased incidence is above the increased detection of cancer attributable to higher rates of screening (1,4,15–17,21–23). In contrast to the inequality in incidence of cervical cancer, differences in breast cancer incidence across races tend to be attributable to factors that are less socially controllable and thus would not necessarily be considered a disparity (Figure 3).
Despite the higher incidence of breast cancer in white women, black women are most likely to die of the disease (1,4,15–17,21–23). Differences in screening rates among racial groups are attenuated when controlling for insurance; regardless of race, women with no insurance are less likely to get a mammogram than women with insurance, and women of lower socioeconomic status have worse 5-year survival rates than those of higher socioeconomic status (16,21–23). Black women are less likely than women of all other races to get appropriate treatment, which contributes to the higher mortality rate (16,21–23). We consider mortality and survival differences in black vs white women a disparity because they are secondary (at least in part) to socially controllable factors (Figure 3).
Given that breast cancer screening and treatment differences are generally included in breast cancer disease models and are tractable, we consider these disparities deserving of priority attention. Alternatively, modifying social determinants that contribute to lower socioeconomic status and poverty would likely reduce disparities in the long term but are less tractable in the short term. For example, the disproportionate number of minorities in low socioeconomic status categories is itself the result of discriminatory practices outside the health system (jobs, education, residence), but the latter is less tractable in the short term. The implication would be that a policy analyst might include in cost-effectiveness analyses strategies that target screening of women lacking insurance and strategies that target improved treatment access or quality of care for black and Hispanic women, as well as women of low socioeconomic status.
Colorectal cancer screening can detect precancerous polyps, which can be removed before they progress to cancer, and can detect colorectal cancer at earlier stages when treatment is associated with higher survival rates. Colorectal cancer incidence and mortality are higher in black men than white men, and these inequalities have widened during 1990–2010 (1,4,15–17,24–26). Although biological differences are unknown, environmental and behavioral factors do differ (eg, consumption of fruits and vegetables, physical exercise, multivitamins, smoking) (15,26). There are differences in race-specific screening rates; for blacks these differences can be largely explained by low socioeconomic status, whereas Hispanics have lower screening rates at all income levels (1,15,26). Differences in mortality are also attributable to differences in access to treatment. Individuals with lower socioeconomic status are less likely than those of higher status to be treated according to standard recommendations, and for every stage, blacks have a worse 5-year survival rate than any other group (1,17,24). The distribution of risk factors, incidence, mortality, and recurrence would be considered a disparity (Figure 4).
The inequalities observed in colorectal cancer outcomes vary in their tractability. For example, modifying the consumption of fruits and vegetables requires changes in well-entrenched dietary patterns that are supported (or caused) by reduced access to low-cost fruits and vegetables in urban communities. In contrast, some treatment differences—also influenced by socially controllable factors—may be more tractable. For example, strategies that target improved treatment access or quality of care for black men, and improved access to screening for individuals of lower socioeconomic status, would be reasonable to consider in a policy or cost-effectiveness analysis.
We synthesized epidemiological, clinical, and economic data from the United States (27–31) using a previously described empirically calibrated individual-based Monte Carlo simulation model of HPV infection and cervical cancer (32–36). The natural history of disease in an individual was characterized as a sequence of monthly transitions between mutually exclusive health states that distinguished HPV infection with type 16, type 18, other high-risk types, and other low-risk types; cervical intraepithelial neoplasia (CIN) grade 1 (CIN1) and CIN grades 2 and 3 (CIN2 and CIN3); and local, regional, and distant cervical cancer. Individual girls enter the model at age 9 years, before sexual debut and free of HPV infection, and transition between health states throughout their lifetimes according to transition probabilities (ie, model input parameters) that depended on HPV type, age, history of previous HPV infection, type-specific natural immunity, previously treated CIN, and screening pattern. Our model accommodates complex cervical cancer prevention strategies and tracks each woman’s history (such as vaccination, screening, treatment, and past abnormalities) and resource use; a running tally was maintained for all events, length of time spent in each health state, and the cost and quality of life associated with each health state. Racial subgroup heterogeneities such as competing mortality risk and screening coverage were included (37,38). Summary statistics for each individual were compiled to compute population measures such as average lifetime costs and quality-adjusted life expectancy.
With the introduction of HPV DNA testing for screening and prophylactic vaccines against the two most common cancer-causing HPV types, 16 and 18, there are opportunities to improve both outcomes and efficiencies of current cervical cancer programs in the United States (8,39–42). In earlier work, we used a simulation model that incorporated the natural history of cervical carcinogenesis, multiple HPV types, risk factors, and comorbidities to compare population-level health and economic outcomes for different strategies and reported aggregated average results for the entire modeled population (32–36). For this work, however, we disaggregated outcomes by racial and ethnic subgroups, which differ in their socioeconomic, demographic, and cervical cancer risk profile, as well as in their access to preventive and quality treatment health services.
We compared the health and economic consequences of three population-based screening strategies (strategies A1, A2, and A3). With the exception of women older than 30 years in strategy A3, all strategies assumed screening with cytology (Pap smears) and HPV triage (cytology followed by HPV DNA testing for equivocal results), a fairly standard screening protocol (Table 2). All three strategies assumed continued racial and ethnic screening patterns. Strategy A2 combined adolescent HPV16 and HPV18 vaccination (33% coverage) with screening, and strategy A3 combined adolescent HPV16 and HPV18 vaccination (33% coverage) with screening using HPV DNA testing as a primary test in women older than 30 years, reserving cytology for women older than 30 years who have high-risk HPV-positive results.
Conventional cost-effectiveness calculations (Table 2) using aggregated population-based measures (eg, mean reduction in cancer risk, average lifetime costs, average life expectancy) showed that depending on the strategy, the average cancer reduction ranged from 60% to 74.5%. Combined vaccination and screening (strategy A2) had a cost-effectiveness ratio (cost per year of life saved [YLS]) of less than $50000. The cost for strategy A3, combined vaccination with the new screening algorithm in women older than 30 years, was more than $100000 per YLS. Although there is no consensus on a strict benchmark ratio below which a strategy would be considered cost-effective, a common heuristic is $50000–$75000 per YLS (43–45); following this practice would imply that strategy A2 provides “good value for money.” However, the conventional analysis provides no information on the distribution of benefits.
The augmented analysis (Table 2) included model-based outcomes for three subgroups, including the race-specific reduction in lifetime risk of cancer, and the absolute difference in the mean reduction in cancer compared with the best-off subgroup (white women). Despite overall reduction in cancer risk, benefits were unequally distributed, and disparities were wider. For example, the overall reduction in cancer incidence for strategy A1 (current screening patterns) was 60%, but reductions ranged from 54.8% for Hispanic women to 62.5% for white women (Table 2, Figure 5, A). The gains in overall average US cervical cancer reduction by adding vaccination to screening represent an increased reduction in lifetime risk of cancer of approximately 9% (from 60% to 69%) (Figure 5, A). With respect to the distribution of outcomes across subgroups, disparities were widest for Hispanic women (increase from 54.8% to 63.9% in cancer incidence reduction), followed by black women (increase from 60.1% to 68.3%), compared to white women (increase from 62.5% to 71.6% in cancer incidence reduction) (Figure 5, A). Without altering race-specific screening patterns, and only adding HPV vaccination, disparities persisted despite the overall gain in clinical benefits for the average population and for each individual subgroup.
The results for both average and distributional outcomes, coupled with the typology that identifies cervical cancer incidence and mortality as true disparities, motivated us to examine if additional strategies would result in disparity reductions among groups. We assessed two new strategies we refer to as targeted screening protocols (strategies B and C) (Table 3, Figure 5, B). These strategies assumed initial targeting of racial and ethnic minorities to bring screening participation up to the mean population level, with subsequent frequency dependent on the clinical risk profiles of individual women (eg, history of prior abnormal cervical screening tests).
The use of the more sensitive HPV DNA test at a lower frequency provided the same protection as traditional cytology used more frequently, as evidenced by the average reduction in cancer incidence increasing from 60% with current screening patterns (strategy A) to nearly 70% with the new screening algorithm (strategy B) (Table 3). Note that although all subgroups experience benefits, the magnitude of cancer risk reduction is greatest in the worst-off populations (eg, from 54.8% for Hispanic women at current screening patterns to 70% reduction with targeted screening compared with a smaller increase for white women, from 62.5% to 69.7%), and disparities are dramatically reduced (Figure 5, B).
The expanded cost-effectiveness analysis, which made outcomes across racial and ethnic groups explicit, found that both new strategies (B and C) provided health gains for all subgroups, reduced disparities, and were more effective and less costly than original strategies that widened disparities (Table 3). Without vaccination, targeted screening to racial and ethnic minorities, with subsequent risk-based screening, resulted in a reduction in cancer incidence of nearly 70% compared with 60% with current screening patterns. Targeted screening also had lower lifetime costs (an average of $1115 compared with $1415 for current screening patterns), resulting in cost savings of approximately $300 per woman over her lifetime. The most effective strategy, of vaccination at coverage rates of 33% and targeted screening with new screening algorithms, had a cost-effectiveness ratio of $28200 per YLS when compared with the same strategy without vaccination. Further, disparities were essentially eradicated for Hispanic women (achieving 77% reduction in cancer incidence, the same as the average reduction) and were reduced by more than 50% for black women (difference in cancer incidence reduction compared with white women falling from 2.4% with current screening patterns to 0.9% with vaccination and targeted screening); these racial differences demonstrate that targeted screening was effective at nearly eliminating disparities for Hispanic women, but black women would also require improved access and quality care once diagnosed with cancer.
Explicitly modeling disparities between subgroups according to our proposed typology of cancer disparities allowed us to identify points of overlap among effectiveness, equity, and efficiency in the context of different options for cervical cancer prevention. Using a previously published cervical cancer model (32–36), we showed that several new strategies would reduce the lifetime risk of cervical cancer averaged across all subgroups; however, although some strategies widened disparities between groups, others dramatically reduced them. Furthermore, we identified strategies associated with both favorable average and distributional outcomes. Adding vaccination to current screening patterns or implementing a new screening algorithm and targeting risk-based screening resulted in reduced cancer incidence across all groups; the targeted risk-based screening strategies, with or without vaccination, resulted in the greatest reduction in disparities for black and Hispanic women, as compared with white women. The proposed changes to screening (new screening algorithm and targeted risk-based screening) were more effective and less costly than current screening patterns; strategies that combined these screening changes with vaccination were more effective and more cost-effective than the status quo. These points of convergence are “win-win” in the sense that they have the biggest positive impact in worst-off groups as well as on population health overall. Our claim is that such win-win strategies are most desirable from the perspective of both goals of health policy, population health improvement, and health equity.
We also demonstrate a format for reporting results that allows a clear and explicit portrayal of the distributional effects of different strategies. Although our typology points us in the right direction in terms of identifying potential targets, it cannot help us describe or quantify the trade-offs between strategies that improve overall population health but widen disparities and those that may be marginally less effective or less cost-effective but reduce disparities. Presenting the results of both clinical and cost-effectiveness analyses in the format suggested here can make these trade-offs transparent. We believe that this will promote a more specific dialogue in those cases for which trade-offs between fairness and cost-effectiveness are unavoidable.
To illustrate this point, consider Mechanic’s (46) claim that policies that improve aggregate health, even when increasing health disparities, are morally acceptable provided that they also improve the absolute level of health for the less-than-equal groups. This claim ignores the fact that there may be a range of strategies that could produce different absolute levels of improvement for less-than-equal groups with or without accompanying changes in the aggregate health improvement. Yet, policy makers must choose among these alternatives even though they have different implications for health equity. One reason for the vagueness of Mechanic’s (46) claim is that we do not have good (ie, quantified) examples of how different intervention strategies may affect the health levels of worse-off groups relative to aggregate population health. Our discussion of cervical cancer suggests that quantified illustrations of these trade-offs and similar examples can be used to make clearer arguments about the ethics of following one intervention strategy rather than another. By providing quantified illustrations, modeling gets us beyond purely hypothetical cases, even if some hypotheticals are embedded in the modeling alternatives. Thus, modeling allows for more rigorous discussion of policies and their actual consequences.
We emphasize that our suggested typology and stylized example represent a preliminary exploration of how we might begin to better align the methods we use in model-based cost-effectiveness analysis with our policy goals of improved population health and fair distribution of that health. Accordingly, there are several limitations that should be noted. First, we present a simplified example of cervical cancer prevention in which we purposefully restrict our analysis to a select group of strategies. We chose to do so in part because the model is already well described (32–36) but also to ensure that the most important point of this article does not get lost in the context of a complex analysis. Thus, the results of this particular example are not intended to inform current policy but rather to illustrate the potential advantage of including the distribution of outcomes in model-based analyses of cancer control. It is important to realize that policies may unintentionally widen disparities as well as be targeted to reduce disparities. The assessment of both of these possibilities is critical to good health policy.
A second limitation is that although our typology assumes that disparities are those inequalities that result from an ethically problematic distribution of socially controllable factors, cancer outcomes that differ as a result of informed voluntary choices can be challenging to disentangle. For example, we view most smoking as a socially controllable factor amenable to policy decisions rather than an informed voluntary behavior because it is primarily the result of behaviors in adolescents who then become addicted to nicotine.
Third, our judgments about relative tractability are admittedly uncertain. For example, some treatment or screening interventions that have poor uptake by certain groups may seem to be tractable; whereas we ought to be able to improve access to them in theory, the obstacles to doing so may be difficult in practice. Even so, favorable results that assume tractability may indeed motivate more intensive efforts to remove barriers to implementation.
Fourth, we did not include all possible strategies that might indeed be tractable; for example, we did not model the costs and benefits of improved access to high-quality cervical cancer treatment. However, in one of our examples of targeted screening in which disparities were essentially eradicated for Hispanic women but only reduced by 50% for black women, the favorable economic profile of this strategy coupled with the explicit and persistent racial disparity could motivate targeted investments to improve high-quality stage-specific cancer treatment for black women. Similarly, as better data become available, strategies that address factors such as race-specific health-seeking behavior, differential access to care based on location (eg, rural vs urban), and the relative influence of socioeconomic factors such as insurance status would allow for refinement of this framework.
Fifth, we did not include specific incremental programmatic costs that might be required for implementation of targeted programs to reduce disparities, which are high-priority empirical data that will be needed in future analyses. Sixth, for this initial work, we have focused on health-care interventions, avoiding the somewhat less tractable social determinants of health. However, if we develop a better understanding of how to quantify trade-offs involving interventions for which we have a clearer idea of the mechanisms involved and which are tractable features of health policy, we will be in a better position to address how to include interventions that modify social determinants of cancer. The extension of this work clearly requires a robust research agenda.
In line with a point made by Neumann and Weinstein (47) about the use of cost-effectiveness analysis in policy decisions, we cannot avoid difficult trade-offs by simply disregarding the costs and benefits of our choices. Furthermore, the socioeconomic and demographic changes we expect in coming years and the potential for widening disparities in the United States (48), coupled with anticipated changes in how health care is delivered and financed, support the critical importance of reliable information about the distributional effects (as well as average costs and benefits) of interventions to prevent and treat cancer.
In conclusion, this preliminary work represents a potential framework from which we can consider how to prioritize interventions that reduce inequalities that are most unjust while also considering their tractability. Use of decision analytic methods to explicitly quantify and examine both the population benefits and impact on disparities associated with different strategies allows us to identify those policy choices that might contribute to both equity and efficiency. More importantly, in the majority of cases in which trade-offs between equity and efficiency are inevitable, illuminating the nature and magnitude of these trade-offs will reveal influential data gaps, promote critical dialogue and debate, broaden the conversation about what we value as a society, and move us one step closer to having those values reflected in our policies.
Agency for Healthcare Research and Quality (R01-HS015570-02 to S.J.G. and N.D.); National Cancer Institute at the National Institutes of Health (R01-CA093435 to S.J.G.).
The study sponsor(s) had no role in the design of the study; the collection, analysis, and interpretation of the data; the writing of the article; or the decision to submit the article for publication.
The authors gratefully acknowledge the contributions of the decision science and ethics working group on disparities, supported by the Center for Health Decision Science in the Harvard School of Public Health and the Harvard Global Heath Institute.