Cardiovascular Perspectives The Limits of Cost-Effectiveness Analysis

The competitive worldwide economic environment and ever-increasing costs of health care have created a setting in which understanding costs and making sure that we achieve good value in health care are paramount. One approach to seeking value is through the use of cost-effectiveness analysis. Although this science is now several decades old, it has been refined over the last several years, with increasingly sophisticated statistical and standardized methods.1,2 Is costeffectiveness analysis useful? Does it help in medical decision making and in allocation of scarce resources? In the accompanying article, “Cost, Effectiveness, and CostEffectiveness” Diamond and Kaul3 argue that costeffectiveness analysis is not a useful approach. Although we agree with many of the points that Diamond and Kaul raise, we do not agree with their conclusion. Cost-effectiveness analysis involves an assessment of both cost and effectiveness. The distribution of each needs to be understood. A cost-effectiveness analysis is only as valid as its underlying measures of effectiveness and cost, a discussion that is beyond the scope of this article. However, the methods to make these assessments vary considerably. There are standards for cost-effectiveness, but at times, perfectly adhering to these standards is not realistic, and compromises are often made that may be entirely scientifically legitimate.4 Cost-effectiveness is, by nature, incremental. Thus, it is necessary to look at the added costs compared with a control group. Selection of the appropriate control group is a challenge itself. At times, the appropriate control is placebo, and at other times, it is active therapy; the appropriate control is dependent on the clinical question being asked. However, when cost-effectiveness analysis is conducted using data from a clinical trial, the selection of the control group will not be a decision that the analyst can affect (at least after the trial has been completed). When additional costs and incremental measures of effectiveness of a new form of therapy are available, along with description of the distribution of each, an incremental costeffectiveness ratio (ICER) may be measured (along with its own distribution).5 An ICER is a ratio of the incremental cost of the new therapy divided by the incremental measure of benefit. When the measure of benefit is expressed in lifeyears or quality-adjusted life-years, the ICER will be measured in cost per life-year or quality-adjusted life-year gained. The ICER should not be viewed only as a single number, as there is uncertainty to both the measure of the cost and the measure of effectiveness. The first level of uncertainty is based on chance or sampling error alone. This may best be considered when patient-level data are available. The distribution of an ICER based on sampling error of the numerator and denominator is somewhat complicated, as the 95% confidence interval of a ratio is not easily defined. Nonetheless, there are approaches to understanding the distribution of the ICER. A popular one is to examine the confidence intervals of cost and effectiveness by sampling from the distribution of each, an approach called bootstrap analysis. By sampling both the cost and effectiveness distributions concurrently, multiple estimates of the ICER can be made.2 These may then be displayed in a plane (Figure 1) in which each point is an estimate of the ICER. In quadrant A, the new therapy is more effective but less costly than the previous standard. In quadrant B, the new therapy dominates the standard, being more effective and less expensive, whereas in quadrant D, the new therapy is dominated by the standard, being less effective and more expensive. The individual ICER estimates can also be used to construct a cost-effectiveness acceptability curve (Figure 2), in which a range of willingness-to-pay thresholds is displayed on the x axis and the y axis indicates the probability that the new therapy has an ICER below this threshold. Thus, the threshold value of the ICER can be varied, and the posterior probability of the ICER meeting that threshold may be noted, a fundamentally Bayesian approach. As the threshold value rises, the probability of the ICER being judged cost-effective increases. Cost-effectiveness analysis will almost always include a series of assumptions, as it is generally not possible to measure everything necessary for a comprehensive analysis. In addition, even when measurements are available, they may not adequately represent values appropriate for the analysis at hand. Thus, cost-effectiveness analyses generally include sensitivity analyses in which the input variables for assessing both cost and effectiveness are varied. Cost-effectiveness acceptability curves may then be created for each value, or probabilistic Bayesian analyses may be performed, in which all variables affecting cost and effectiveness are varied.6 In the approach outlined above, there is no mention of some threshold below which an ICER must be for a new