Taking account of future technology in cost effectiveness analysis

Introduction

Cost effectiveness analysis provides a tool for evaluating allocation of resources by characterising different healthcare interventions in terms of the extra cost per added unit of health benefit (box 1).1 Such analyses are being used increasingly to set national and international health priorities. The UK's National Institute for Clinical Excellence, for example, is charged with guiding decisions on use of new and existing technologies in the NHS, based in part on cost effectiveness considerations.2 3 In recent years innovation in healthcare technology has occurred at an unprecedented pace for some problems, with new options rapidly supplanting existing interventions.4 We explore how cost effectiveness analysis could be extended to reflect evolving technologies, and how accounting explicitly for future treatment prospects might affect a typical analysis, using treatment for hepatitis C virus (HCV) infection as an example.

Treatment for hepatitis C virus infection

An estimated 2.7 million Americans and 6.7 million Europeans have chronic HCV infection and are at risk of cirrhosis, end stage liver disease, and liver cancer.7^–9 Treatment decisions are complicated by variable progression rates and require difficult trade-offs between costs, side effects, and uncertain clinical benefits.10^–12 Various treatments have emerged in recent years, and a series of decision analyses have examined their cost effectiveness (table 1). Analyses typically have found that each new treatment has an attractive cost effectiveness ratio compared with many common interventions. Most ratios, in fact, have fallen below $10 000 (£5607, €8218)/QALY compared with the next most effective option. Given the rapid evolution of treatments for HCV infection, this example offers an illustration of how anticipated technological changes can be used to enrich conventional cost effectiveness analyses.

Modelling the natural course of infection

We developed a Markov model (box 2) of the natural course of HCV infection (figure), incorporating assumptions similar to those used previously.12^–13 The model includes stages of fibrosis14 leading to clinical cirrhosis and its complications (such as decompensated liver disease and primary liver cancer), and the possibility of liver transplantation. Details on the model and data sources are on bmj.com.

Model of natural course of hepatitis C virus infection

Credit: RUSSEL KIGHTLEY/SPL

Retrospective analysis

We begin by travelling back in time to 1996, to revisit the decision problem confronting a patient with chronic HCV infection considering interferon monotherapy, the only approved treatment at that time. Adopting the conventional assumption that the “no treatment” strategy excludes the downstream potential for improved therapies, we calculated the incremental cost effectiveness ratio of monotherapy as $8700/QALY gained compared with no treatment (see bmj.com for details of the model). With the benefit of hindsight, however, we might ask whether no treatment was the only relevant comparator. In other words, should the alternatives to immediate treatment also have included deferred treatment?

A previous study considered one important element of this question, investigating watchful waiting with periodic liver biopsy versus immediate empirical treatment for chronic HCV infection.17 The authors found that immediate treatment was cost effective compared with biopsy management. We focus on an additional facet of this question—waiting and watching for technological innovation. For a patient in 1996, how would anticipation of imminent advances in treatment have changed the decision problem?

Perfect foresight scenario

As a starting point, consider a perfect foresight scenario, in which a person must decide between immediate treatment, deferred treatment, or no treatment but we assume perfect knowledge about the timing and nature of a future improved treatment. Specifically, we assume combination therapy with interferon alfa and ribavirin would be available within three years and provide sustained viral clearance in 42% of treated patients versus 11% with monotherapy. To bias towards immediate monotherapy, we assume combination therapy will be more costly ($11 800 v $2500), have more severe side effects (yielding an average loss equivalent to 27 healthy days v 18), and that monotherapy will not reduce the effect of subsequent retreatment with combination therapy, which would also be provided after three years to people who had not responded to monotherapy.

The results provide a sharp contrast to our naïve analysis using the conventional comparison with no treatment. With perfect knowledge of a future more effective regimen, a strategy of waiting three years for the new treatment would have lower costs and greater benefits than immediate treatment - that is, would “dominate” immediate treatment in the cost effectiveness idiom (table 2).

With the key temporal component of this example, an important factor is the degree to which people “discount” the value of future consequences. Standard practice in cost effectiveness analysis applies annual discount rates of 3-5%, and the first set of results in table 2 reflects a 3% discount rate. Delaying treatment would defer costs and side effects but would also allow the disease to progress. If future consequences are discounted, the relative costs of immediate treatment are higher because of the timing; without discounting, costs are similar. For health outcomes, discounting reduces the advantages of immediate therapy because the benefits of treatment relate largely to averting future disease outcomes (discounting therefore compresses incremental gains from earlier treatment) and the negative effects of treatment (toxicities and adverse events) count less for deferred treatment when discounted but count equally for immediate and deferred therapy without discounting. The net effect of these factors is that immediate therapy looks more attractive without discounting; it nevertheless retains a much less favourable cost effectiveness ratio compared with the conventional analysis (table 2).

These results depend on several other key variables, including rates of progression of disease, costs and adverse outcomes associated with different regimens, and timing of new treatment options. Since this case is intended to illustrate a more general methodological question, we present just one example of the sensitivity of the perfect foresight results to important assumptions in the model. When we vary the year in which an improved treatment arrives, immediate treatment offers lower benefits than deferred treatment (at higher costs, making immediate treatment a “dominated” strategy) even if improvements are up to five years away. Moving from a one year to a five year delay, the difference in overall benefits between immediate and deferred treatment ranges from −0.03 to −0.002 QALYs—that is, losses equivalent to 110 healthy days if a new treatment would arrive in one year or one healthy day if the treatment were five years away.

Relaxing assumption of perfect foresight

The above analysis is based on an unrealistic assumption of perfect knowledge of future treatments. However, uncertainty about emerging treatments can be incorporated in a decision analysis framework in the same way that other uncertain outcomes, such as developing cirrhosis, are captured. Define p as the probability of a new treatment being available in three years (with the specifications of combination therapy). In the deferred treatment strategy, we assume that individuals receive combination therapy in year 3 with probability p, or will fall back on monotherapy if no new treatment emerges that year, with probability (1-p). In the immediate treatment strategy, individuals receive monotherapy now, but may (with probability p) have access to combination therapy in year 3, in the event of non-response or relapse. Allowing for this uncertainty, immediate treatment remains dominated by deferred treatment provided that p > 50%. The probability of a new treatment must be 30% or lower for the incremental cost effectiveness of immediate monotherapy to fall below $50 000/QALY (table 3).

We have examined different scenarios regarding the timing and likelihood of better treatment separately, but we may also combine these two dimensions. For example, imagine that the cumulative probability of an improved treatment rises linearly over time so that there is a 10% chance of the new treatment emerging in one year, 20% in two years, and so on, up to 50% in five years. In this scenario, the incremental costs of immediate treatment compared with deferred treatment would be $800, with incremental benefits of 0.007 QALYs (less than 3 days), implying an incremental cost effectiveness ratio of more than $115 000/QALY—that is, more than a 13-fold increase over the assumption of no potential improvements in treatment.

Conclusion

Our analysis shows that taking account of possible future advances in treatment can change conclusions about the cost effectiveness of current interventions for conditions where technology is evolving rapidly. We have presented a simple model of HCV infection that does not do justice to its complex natural course or the nuances of treatment decisions because we wanted to illustrate a more general problem. Certain key features of the problem make consideration of evolving technologies important to economic evaluations, including progression over a relatively long period, uncertain efficacy and toxicity of current treatment, and steady scientific progress toward new treatments. Other conditions that share one or more of these features include chronic lymphocytic leukaemia, cystic fibrosis, primary pulmonary hypertension, and Parkinson's disease. Our approach will be less relevant for conditions that require urgent treatment (for example, acute renal failure) or have current treatments that are very effective and well tolerated (for example, acute cardiac ischaemia).

Technological change is one of many factors that can influence decisions regarding optimal timing of treatment. Incorporating the interactions between these various factors will enrich the clinical relevance of the analysis and enhance its utility in decision making. Quantifying uncertainties regarding technological innovation is methodologically challenging. Although prospective assessment of probabilities that specific improvements will occur is difficult, however, the implicit alternative is to assign them all probabilities of 0. At a minimum, plausible predictions of changes in costs can be made in some cases based on the limited patent life of licensed pharmaceuticals. More sophisticated analyses are also possible incorporating additional dimensions of uncertainty such as response rates, costs, or stepped improvements over time. These complexities would be straightforward extensions of the conceptual logic presented here.

Development of sound clinical guidelines, public health policy, and investments for new technology will require careful consideration of the incremental benefits, harms, and costs associated with new interventions—including those not yet discovered—compared with existing ones. Even at this early stage, we encourage analysts to model explicitly the full spectrum of alternative options, so that decision makers have estimates of their costs, benefits, and risks (including associated uncertainties) when faced with difficult choices about imperfect treatment. Minimally, economic evaluations should make explicit, and justify, the assumptions that are otherwise implicit, including no development of alternative technologies and no change in costs for existing technologies.