(Stroke. 2004;35:1497.)
© 2004 American Heart Association, Inc.
Comments, Opinions, and Reviews |
Editorial Comment—What Can Models Teach Us About Stroke Treatment?
Sorting Out the Missing Bits
Center for Clinical Health Policy Research, Duke University Medical Center, Durham, North Carolina
The report in this issue of Stroke on the cost-effectiveness of recombinant tissue plasminogen activator (rt-PA)1 suggests – with notable uncertainty – that, when compared with standard acute stroke care, thrombolytic therapy provides, on average, greater health benefits (in terms of average quality adjusted life years [QALYs]) at a reasonable average medical cost. This is not itself a unique finding; other similar efforts suggest – with notably greater certainty – that rt-PA is a good value for money.2–4 What have we learned from these modeling exercises? Is it true, as the authors propose, that we need another trial of rt-PA? Before considering the answer, it is useful to briefly review a few basics of disease models – what they are and what they can teach us.
Disease models are mathematical representations of a clinical condition, its development, and its outcome, and models are often used to evaluate the impact of potential diagnostic or therapeutic strategies. These models are usually implemented in computer code, ranging from something as simple as a spreadsheet formula to sophisticated clinical event simulations. Sandercock and colleagues1 frame the question of rt-PA use as a decision tree, with a Markov model as a "calculation engine." The inputs to this model are epidemiological, clinical trial, and health economic data (specifically applicable to the context of the UK National Health Service [NHS]); the outputs are the month-to-month proportion of individuals in various stroke-relevant health states and their health costs.
When properly constructed, outputs of disease models can be used in a cost-effectiveness analysis. Cost-effectiveness analysis is a well accepted approach to formulating public resource allocation decisions. The incremental cost-effectiveness ratio, ie, the extra resource inputs required divided by the extra benefits achieved, is a standard metric that policy makers use to quantify value for money. Under specific assumptions, a society that allocates resources based on incremental cost-effectiveness ratios will optimize "social welfare."5 Whether one accepts the theory or, more to the point, accepts the applicability of the theory to the real world, incremental cost-effectiveness ratios have an intuitive appeal. When calculated using standard methods,6 they provide a compact way of assessing value and making allocation decisions in the face of limited resources.
The current report represents an application of modeling in clinical medicine that is unique to a limited number of countries – a guide to public policy making. The UK is notable for having a formal organizational structure, the NHS Health Technology Assessment Programme, that is responsible for analyses intended to inform the NHS’s funding decisions. Other applications of models include providing insights into disease/treatment dynamics and assisting in clinical research design.7,8
With this information, let us turn to the question presented in the current paper: Is it cost-effective to use rt-PA in the context of the UK NHS? Available effectiveness data has not been collected in the UK setting, the limited trial-based information about resource use is not UK-based, and long-term outcomes (in the UK or elsewhere) have not been studied. The latter point is vital because both health benefits and costs accrue over a lifetime; failure to account for these long-term effects leads to a skewed assessment of the potential costs and benefits of an acute stroke treatment. While a single data source cannot provide a satisfactory answer, a model can offer guidance. Assuming that effectiveness data can be extrapolated to the UK, and given UK-specific cost estimates, one can turn the crank of the calculation engine and estimate the incremental cost-effectiveness of rt-PA for the NHS. Moreover, one can account for uncertainty in the input estimates by performing a so-called Monte Carlo analysis, ie, repeating the analysis many times, each time randomly drawing from likelihood functions of relative risk reduction for poor outcomes, derived from empirical source data. The uncertainty in inputs (in this case, treatment efficacy) is reflected in variations in the outputs.
Based on trial evidence, the NHS model indicates that there is approximately a 3 out of 4 chance that tPA will provide a net benefit over one year, or over a lifetime. For the lifetime outcome, ie, the outcome relevant to policy analysis,9 rt-PA is likely to not merely be "cost-effective" but actually cost-saving. While one may quibble with some assumptions, input values, or details of the analysis, the results are relatively robust, and are consistent with prior analyses. So, why do the authors suggest that the solution is another trial of rt-PA? And while we are on that point, why, nearly a decade after the NINDS trial of rt-PA,10 is there continued controversy about whether clinicians should use this therapy and whether policy makers should make systematic efforts to facilitate its use?11
If the model were a perfect representation of reality, then, as theory indicates, rt-PA is probably a good use of public resources and – the authors’ claims of inadequate certainty notwithstanding – a society that selects this and similar therapies will tend to optimize the health of its population in light of available resources. However, the current model, like all models, is of necessity an imperfect representation of reality. The NHS model illustrates that even a relatively sophisticated mathematical simulation may not capture the full range of decisional issues with complete fidelity. It is acknowledged that the current model does not capture the full array of resource costs associated with modifying the health care system to accommodate universal use of rt-PA. Another notably missing feature is an accounting of the "disutilities" such as regret following a treatment-associated complication. These missing bits may be key to what makes treatment acceptable or not.
This is not an indictment of modeling. On the contrary, the discord between a model and current opinion and behavior provides insight, and, in turn, grist for further discussion and – yes – research. But the crucial research agenda is not a more precise estimate of rt-PA efficacy. Rather, it is a better understanding of the economic costs (from the perspective of the various payers) and the decisional conflict engendered by this treatment. Moreover, since the case for rt-PA rests largely with the long-term implications of observed short-term improvements, it is essential to determine whether the benefits of rt-PA persist; are patients who improve following rt-PA more likely to subsequently deteriorate than patients with comparable levels of poststroke disability?
When it comes to decision-making, no analysis (and, for that matter, no clinical trial) can provide an answer reflexively. Evidence and analysis is never entirely definitive. Models will never replace human judgment. However, they can, with proper formulation and interpretation, be extremely useful guides to the complex decisions we face in clinical practice and policy.
 |
References |
|---|
 Top References |
|---|
1. Sandercock P, Berge E, Dennis M, Forbes J, Hand P, Kwan J, Lewis S, Lindley R, Neilson A, Wardlaw J. Cost-effectiveness of thrombolysis with recombinant tissue plasminogen activator for acute ischemic stroke assessed by a model based on UK NHS costs. Stroke. 2004; 35: 1490–1497.
2. Fagan S, Morgenstern L, Pettita A, Ward RE, Tilley BC, Marler JR, Levine SR, Broderick JP, Kwiatkowski TG, Frankel M, Brott TG, Walker MD. Cost-effectiveness of tissue plasminogen activator for acute ischaemic stroke. Neurology. 1998; 50: 883–890.
3. Chambers M, Koch P, Hutton J. Development of a decision-analytic model of stroke care in the United States and Europe. Value Health. 2002; 5: 85–97.
4. Samsa GP, Reutter RA, Parmigiani G, Ancukiewicz M, Abrahamse P, Lipscomb J, Matchar DB. Performing cost-effectiveness analysis by integrating randomized trial data with a comprehensive decision model: application to treatment of acute ischemic stroke. J Clin Epidemiol. 1999; 52: 259–271.[CrossRef][Medline] [Order article via Infotrieve]
5. Garber AM, Phelps CE. Economic foundations of cost-effectiveness analysis. National Bureau of Economic Research; 1995.
6. Gold MR, Siegel JE, Russell LB, eds. Cost-Effectiveness in Health and Medicine. New York, NY: Oxford University Press; 1996.
7. Matchar DB, Samsa GP, Matthews JR, Ancukiewicz M, Parmigiani G, Hasselblad V, Wolf PA, D’Agostino RB, Lipscomb J. The Stroke Prevention Policy Model: linking evidence and clinical decisions. Ann Intern Med. 1997; 127: 704–711.
8. Samsa GP, Matchar DB. Have randomized trials of neuroprotective drugs been underpowered? An illustration of three statistical principles. Stroke. 2001; 32: 669–674.
9. Torrance GW, Siegel JE, Luce BR. Framing and designing the cost-effectiveness analysis. In: Gold MR, Siegel JE, Russell LB, eds. Cost-Effectiveness in Health and Medicine. New York, NY: Oxford University Press; 1996: 68.
10. National Institute of Neurological Disorders and Stroke (NINDS). Tissue plasminogen activator for acute ischemic stroke. N Engl J Med. 1995; 333: 1581–1587.
11. Hoffman JR. Tissue plasminogen activator (tPA) for acute ischaemic stroke: why so much has been made of so little. Med J Aust. 2003; 179: 333–334.[Medline] [Order article via Infotrieve]
Related Article:
Stroke 2004 35: 1490-1497.
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||