Visualizing Cost-effectiveness Analysis

This issue of THE JOURNAL includes 2 cost-effectiveness analyses that compare cervical screening strategies.^{1 - 2} To many readers, these types of studies are both confusing and complex. In addition, because the results are based on mathematically modeled outcomes rather than "real" outcome data, some may not really trust the results. However, THE JOURNAL publishes these types of studies when they are of high quality according to the standards of the discipline and when they concern important health care questions that are difficult to address using other methods. In the case of cervical cancer screening, it is unlikely that clinical trials will be performed that could adequately compare all the possible variations of screening and treatment and follow-up cohorts of patients over their lifetimes to assess mortality and quality-of-life outcomes.

If a cost-effectiveness analysis is performed correctly, adequately tested, and presented well, many useful insights can be gained regarding the benefits and trade-offs between different interventions. Whether a particular intervention is "cost-effective," by whatever standard, may be a less important result than these insights, given that there is no official body in the United States that approves particular interventions for use based on cost-effectiveness analysis.

One helpful suggestion for clinicians evaluating these studies is to loosen focus on the specific numbers presented in the studies and concentrate attention on the graphic representation of the results, shown generically in Figure 1, specifically in both studies,^{1 - 2} and in many other published cost-effectiveness analyses in THE JOURNAL. Gains in life-years are plotted on the y axis, and total costs are plotted on the x axis. Each possible intervention strategy is represented by a point. Assuming that the analysis uses uniform and consistent methods and assumptions to calculate the gain in life-years and total costs of each strategy, such as those recommended by Gold et al,³ the graph can provide many useful insights. At the simplest level of analysis, the higher the point, the more effective the intervention, the more to the right, the more expensive. From the perspective of the prudent spender trying to purchase the most effective health care for a given amount of money, the strategies that form the solid line connecting the points lying left and upward are the economically rational subset of choices (points C, G, H, and I in Figure 1). Points lying beneath the line (points D, E, and F) represent strategies that are not as effective for any given amount of money as a point lying on the line—they are dominated strategies. The slope between any 2 points represents the incremental cost-effectiveness ratio (actually the inverse). The important perspective is that as the line gets flatter, the incremental cost-effectiveness ratio gets higher, representing diminishing returns of effectiveness per expenditure as the more effective strategies are used. This gives literal meaning to the term flat-of-the-curve medicine.⁴

The analysis itself cannot answer the question of which strategy is economically preferred, only which strategy is the most effective, in terms of life-years saved, for a given level of desired expenditure. Authors commonly discuss their findings in terms of comparisons to cost-effectiveness of other established interventions and conclude that a certain strategy is economically reasonable. This is often problematic due to differences in perspective, methods, and assumptions between studies. A critical examination of actual medical practice to observe where it appears to be limited or affected based on cost-effectiveness would probably reveal wildly inconsistent cost-effectiveness thresholds. Thus without an official role in setting practice guidelines, cost-effectiveness analysis in the United States can only provide a gentle appeal to the professional responsibility to practice fiscally efficient medicine.

The set of economically rational choices is an important result from a cost-effectiveness analysis, but it can be quite useful to notice other relationships between strategies. For instance, in Figure 1 the dominated strategies D and E lie well below the line. Barring great uncertainty and sensitivity in the assumptions that led to D and E having this position, the strategies represented by these points are unlikely to be good choices. Strategy F also is a dominated strategy, but it lies close to strategy G. Strategy H has a shallower slope between it and strategy G, indicating a higher incremental cost-effectiveness ratio, but it too lies close to G. Based on being close to nondominated strategy G, both strategies F and H might be considered reasonable alternatives with relatively insignificant economic consequences if there were other noneconomic reasons to prefer them, such as patient or physician acceptability, availability, or other factors.

Lines can be visualized between any 2 strategies, and the slope between them indicates the incremental cost-effectiveness between those 2 strategies. Although strategy B is a dominated strategy in Figure 1 and many analyses will not directly report any cost-effectiveness ratios involving strategy B, one may be particularly interested in the incremental cost-effectiveness ratio between B and G (if strategy C is not available, or to compare with another study that compares those 2 strategies). The relatively steep slope (dashed line Figure 1) indicates a cost-effectiveness ratio similar to but slightly higher than between A and C.

The ability to compare numerous different strategies simultaneously in one figure can offer other additional insights. Debate about cervical cancer screening has often centered on the superiority of one test over another at set screening intervals. However, both studies in this issue of THE JOURNAL directly compare many different screening intervals. The figures in both articles offer a way of examining this issue without undue clutter of actual numbers. Human papillomavirus (HPV) screening, either as a triage for atypical squamous cells of undetermined significance or in combination with Papanicolaou testing, may allow improved or equal efficacy with decreased frequency of screening, at intervals of up to every 3 years at costs no greater than traditional Papanicolaou testing at more frequent intervals. Thus, without resorting to the contortions needed to argue what a year of life is worth, HPV testing may offer a way to reduce the frequency of cervical screening while still maintaining current levels of effectiveness. Such a result seems plausible given the slow progression typical of cervical neoplasia and the spotty sensitivity of the Papanicolaou test, which has always required relatively frequent screening intervals to make up for its lack of sensitivity.⁵

A limitation of the analyses^{1 - 2} published in this issue, in addition to the need for assumptions and estimation of certain parameters, is that the models assume that the incidence and persistence of HPV infection is a random phenomenon solely determined by a few demographic characteristics. It is actually more likely that HPV infection is nonrandomly distributed in the population due to behavioral and possibly genetic determinants. Given this fact and the known causal relationship of HPV to cervical cancer, more efficient screening programs could be modeled that take advantage of this and stratify women into different screening regimens depending on their HPV status. Women without HPV infection could receive screening at intervals of several years, whereas women with HPV infection could receive more regular screenings. If medical practice guidelines are amenable to this kind of screening paradigm, similar to how colon cancer screening recommendations diverge subsequent to finding a high-risk adenomatous polyp,⁶ it is likely that both the cost and effectiveness of cervical cancer screening could be greatly improved.