Cost-Effectiveness of Thrombolysis With Recombinant Tissue Plasminogen Activator for Acute Ischemic Stroke Assessed by a Model Based on UK NHS Costs
Background and Purpose— Thrombolytic therapy is licensed for use in highly selected patients with acute ischemic stroke. We aimed to model the health economic impact of limited use of thrombolytic therapy and to assess whether it was likely to be cost-effective when used more widely in the UK National Health Service (NHS).
Methods— The authors formed a discussion panel to develop the decision-analysis model of acute stroke care. It consisted of Markov state-transition processes, with probabilities of different health states determined by certain key variables. The range of estimates of efficacy of recombinant tissue plasminogen activator (rt-PA) was taken from an update to a Cochrane systematic review of randomized trials of thrombolysis. Data on outcome after stroke were taken from our hospital-based stroke register, supplemented by data derived from relevant literature sources.
Results— The model suggested that compared with standard care, if eligible patients were treated with rt-PA up to 6 hours, there was a 78% probability of a gain in quality-adjusted survival during the first year, at a cost of £13 581 per quality-adjusted life-year (QALY) gained. Over a lifetime, rt-PA was associated with cost-savings of £96 565 per QALY. However, the estimates were imprecise and highly susceptible to the assumptions used in the economic model; under several plausible assumptions, rt-PA was much less cost-effective than standard care, and under others, a great deal more cost-effective.
Conclusions— The estimates of effectiveness and cost-effectiveness were imprecise. Although the benefits appeared promising, the data did not support the widespread use of thrombolytic therapy outside the terms of the current restricted license in routine clinical practice in the NHS. There is a case for new large-scale randomized trials comparing thrombolytic therapy with control up to 6 hours to determine more precisely the effects of rt-PA on short-term and long-term survival and its cost-effectiveness when used in a wider range of patients.
Thrombolytic therapy with recombinant tissue plasminogen activator (rt-PA) was first licensed for the treatment of acute ischemic stroke in the USA in 1996, and soon afterward in Canada and Germany. A conditional license has been granted in many European countries. The published economic evaluations of rt-PA for acute ischemic stroke have had significant limitations, because they were based on the North American1 or Australian2 health care systems, or based their efficacy estimates on a limited subset of the randomized trial evidence,1–3 or performed only limited sensitivity analyses or modeled only very limited use within the first 3 hours.1–3 To decide whether UK National Health Service (NHS) should implement thrombolytic treatment, and if so, how widely, we undertook an economic analysis constructed from the perspective of the NHS commissioned by the NHS Health Technology Assessment Programme.4 This assessment sought to explore a range of scenarios encompassing limited use under the current restrictive license for patients presenting within 3 hours and wider use up to 6 hours after onset.
See Editorial Comment, page 1497
Subjects and Methods
From the perspective of the UK NHS, is thrombolytic treatment for acute ischemic stroke (compared with standard care) cost-effective as judged by the incremental cost per quality-adjusted life-year (QALY) gained?
The perspective was a broad health care and personal social services perspective. We included the direct costs of hospital stay, rehabilitation, and long-term care. We did not include assessments of any indirect economic costs, such as loss of work-related earnings, or of the capital and revenue costs of developing services for patients with acute stroke to the point at which acute stroke care was delivered across the whole NHS to the standard required.
Assessment of Alternatives to Thrombolytic Treatment
It is difficult to define, in economic terms, a standard package of general care for patients with acute stroke (even more so to define one for patients treated with thrombolysis). We have therefore assumed that the alternative treatments being compared are “standard care” and “standard care plus thrombolysis.”
Form of Evaluation
We have adopted a cost-utility approach, assessing health gains in QALYs. We have modeled costs and effectiveness over the short-term (1 year) and the long-term (lifetime).
Steps to Improve Generalizability of Results
The patients included in the trials of thrombolysis were highly selected and were largely recruited from non-UK centers. So, to produce results that were more relevant to the NHS, we undertook a modeling approach, applying data on efficacy from the trials to a population of stroke patients treated within the NHS (or similar publicly funded health service).
Choice of Measure of Benefit
The use of QALYs as the measure of benefit enabled us to encompass the utility values that stroke patients assign to the different health states after stroke (ie, death, survival in a dependent state, or survival in an independent state).5,6
The authors formed a discussion panel to construct a decision-analysis model of the pathways that acute stroke patients follow after being admitted to hospital.4 The model was constructed by discussion among the reviewers, analysis of our own stroke registry (Lothian Stroke Register) data, and review of the literature. The model was entered into a software package (Data 3.5 software; TreeAge Software Inc) and is shown in Figure 1. We defined 5 groups of patients (see Figure 1 legend for definitions). Table 1 lists all of the base-case values (with plausible ranges) used in the model and the sources of the estimates. These include estimates of treatment effect for rt-PA to those that formed the basis for its current approval for use in clinical practice within 3 hours. To predict the health and economic outcomes of rt-PA after the first year, we used a Markov modeling approach.7–9 The Markov model used age-specific mortality, risk of recurrent stroke, and stroke-specific case-fatality to estimate the probabilities of being dead, dependent, and independent at the beginning of each year. The Markov process was run repeatedly in 1-year cycles until the end of the cohort lifetime, and totals were computed for the accumulated health outcomes and costs.
Assumptions About Cost of Implementing rt-PA
We sought to assess the typical additional costs of implementing rt-PA treatment in a “typical” district general hospital. However, we were unable to define a nationally agreed level of resource use required to deliver thrombolysis for acute stroke or to obtain reliable measures of the variation in the current level (and cost) of acute stroke care in UK hospitals. We therefore sought to identify, in a qualitative way, the specific extra resources we considered necessary to deliver thrombolysis (in the context of a randomized controlled trial) in our own hospital (Table 2). However, the resources currently allocated to acute stroke care vary greatly between centers across the UK, so any quantitative estimates of the extra implementation costs derived from these local data could not be reliably extrapolated to other hospitals in the UK.
Adjustment for Timing of Costs and Benefits
We accounted for the longer time horizon over which costs and health benefits may accrue by discounting outcomes and cost at an annual rate of 6%.
We performed a number of 1-way sensitivity analyses and threshold analyses to explore the impact of varying key parameters in the model: rt-PA efficacy (we used a range that encompassed larger benefits expected when used in a highly selected population within 3 hours and expected smaller benefits when used in a wider variety up to 6 hours); system efficiency (this ranged from the small proportion currently treated to a scenario with greatly increased efficiency leading to a high proportion treated); utility values; costs of rt-PA treatment; length of hospital stay; and unit cost per inpatient day. We also performed a multiway first-order Monte Carlo simulation to determine how likely certain levels of cost-effectiveness were when we simultaneously incorporated all ranges of values for variables listed in Table 1.
Cost-Effectiveness at 12 Months
Table 3 presents the costs and outcomes at 12 months per 100 patients treated with rt-PA. The base-case analysis assumed that only 5.3% of the patients admitted to hospital were eligible for rt-PA treatment, showed that treatment with rt-PA costs an additional £11 001, and resulted in a QALY gain of 0.81 per 100 patients treated. This gives a marginal cost-effectiveness ratio for rt-PA treatment of £13 581 per QALY gained. The multiway Monte Carlo simulation showed that the 5th and 95th percentiles for the increase in costs at 12 months were −£44 065 and £47 095, respectively, and that the corresponding percentiles for the impact on health outcomes were −0.4020 and 1.8259 QALYs, respectively. The analysis also showed that there was 85.5% probability of an increase in QALYs with rt-PA treatment. If we assumed that rt-PA increased QALYs, the 5th and 95th percentiles for the incremental cost-effectiveness ratio for this group were −£81 680 (cost-savings) and £142 505 (additional costs) per QALY gained. The lower sections of the Tables summarize the sensitivity analyses.
Cost-Effectiveness at End of the Cohort Lifetime
Costs accrue in the short-term (eg, initial acute care), whereas survival gains accumulate over a far longer period, and analyses performed at 12 months therefore underestimate expected yield (eg, in terms of QALYs gained) relative to costs. Nursing home/long-term care is higher for a stroke survivor with high levels of disability, and hence the cost-savings associated with reduced disability, are large and only partly offset by the costs associated with increased survival. The base-case analysis showed that over the cohort lifetime, giving rt-PA then became the dominant strategy (Table 4). Treatment with rt-PA was more effective (gain in QALYs of 3.63 per 100 patients treated), less expensive than standard treatment (cost savings of £350 532), and resulted in a reduced cost of £96 565 per QALY gained. The multiway Monte Carlo simulation showed that there was a 76.6% probability of increased QALYs. If we assume that rt-PA increases QALYs, the 5th and 95th percentiles for the incremental cost-effectiveness ratios for this group were −£908 153 (net savings) and −£37 858 (net savings) per QALY gained.
The lower halves of Tables 3 and 4⇑ summarize the sensitivity analyses. The impact of assuming the most optimistic estimate of rt-PA efficacy was to increase the number of QALYs gained from 3.63 to 19.41, reduce the costs from £350 532 to £267 713 per 100 patients treated, and change the marginal cost-effectiveness ratio from £96 565 to £13 793 saved per QALY gained. When we assumed the least favorable estimate of rt-PA effectiveness, rt-PA resulted in a loss of 13.21 QALYs, and the incremental cost-effectiveness ratio could not be calculated. Detailed sensitivity analyses are presented in the full report.4
Our analyses, based on an up-to-date estimate of the effectiveness of rt-PA and modeled on the NHS, suggests that rt-PA might well be cost-effective. In the base-case analysis, treatment with rt-PA was associated with an additional cost of £13 581 per QALY gained during the first 12 months after treatment. This estimate was considerably higher than the published estimates for treatment with rt-PA for myocardial infarction,10,11 but it was still well within the range of cost-effectiveness for health care interventions offered within the NHS.12 Donaldson has recently highlighted a limitation of such cost-effectiveness analyses; that is, if a new treatment requires more resources, misuse of cost-effectiveness ratios may lead to inefficient treatments being adopted.13 When the model was run to the end of the cohort lifetime, there appeared to be a substantial cost savings of £96 565 per QALY gained. The short-term and long-term cost-effectiveness estimates were very imprecise. At 12 months, the 5th and 95th percentiles for the impact on costs ranged from a cost saving of £44 065 to an extra cost of £47 095. There was therefore considerable uncertainty about the exact size of the incremental cost-effectiveness ratio for rt-PA in acute stroke. The cost-effectiveness estimates were sensitive to rt-PA efficacy and costs of rt-PA. Other parameters thought to be important, such as “system efficiency” and patient values, did not have any significant impact on the incremental cost-effectiveness ratio.
Summary of Previous Work
Our results are not as optimistic as earlier estimates. From the perspective of the North American health care system (which included nursing home costs), for every 1000 patients treated, rt-PA increased hospitalization costs by $1.7 million but decreased rehabilitation costs by $1.4 million and nursing home costs by $4.8 million.1 Multiway sensitivity analyses indicated a >90% probability of cost-savings. The study had some limitations for current health care planning outside the USA: the estimate of efficacy was based on a single trial;14 costs were based on the US health care system; the possibility that treatment might increase case fatality was not modeled; and the estimate of the gain in QALYs was very imprecise (and included the possibility of almost no benefit). A further study, commissioned by a pharmaceutical company (but conducted by an independent economist), concluded that the savings related to disability and long-term care considerably outweighed any potential extra costs of acute therapy, given a broad cost perspective and a time horizon of ≥2 years.3 However, the authors also pointed out that the fixed costs of developing and maintaining a capability to diagnose acute stroke and provide early thrombolysis would need to be taken into account in a more comprehensive analysis. Furthermore, any downstream savings attributed to the avoidance of social care costs associated with disability are unlikely to be very convincing to budget holders focused on hospital and drug cost alone.3
Why Might Our Results Be Different?
As expected, the cost-effectiveness estimate at 12 months was heavily influenced by the source of the data in the model.12 This may invalidate the comparison between our study and previous studies of cost-effectiveness of rt-PA in stroke13 and may explain the different short-term results. In contrast to earlier studies, we found that the cost-savings were not realized within the first 1 to 2 years after treatment. One likely explanation is that the other studies were based on more optimistic estimates of rt-PA effectiveness, from the NINDS trial alone, in which treatment was given within 3 hours14 or just 3 of the major rt-PA trials;14–16 however, our sensitivity analyses did include a value for the effectiveness of rt-PA comparable to that seen in NINDS. Earlier studies also used more favorable values for patients’ preferences.17,18 We based our estimates of the effectiveness of rt-PA on the results of a systematic review of all the available evidence from randomized controlled trials of rt-PA to date. Furthermore, we used a more conservative estimate of the patient valuation of the dependent state, which, as it turned out, was close to the estimate derived from a recent systematic review of patient utilities after stroke.6
Generalizability of These Results
Another uncertainty relates to the generalizability of the findings.12 It is likely that both resource use (eg, length of stay) and the valuation of resources (eg, mean unit cost per inpatient day) will vary considerably within the NHS. Hence, we used national official figures to “average out” local differences in unit costs,19 and we believe that the resources used by patients registered in the Lothian Stroke Register are reasonably representative of the resources used by stroke patients admitted to other UK hospitals. Our analysis did not include the costs of implementing rt-PA in NHS hospitals. We assumed that there were no capacity constraints in the health care system and that there were no extra costs associated with giving rt-PA to more patients. For example, we assumed that all admissions were “equal,” regardless of when they occur; that CT scanning equipment was always readily available, and that the correct number and mix of health care professionals and hospital beds were always in place. We sought to assess the additional costs of developing stroke services to deliver rt-PA treatment by identifying the specific service components we considered likely to be required to deliver thrombolysis in our hospital, over and above those required for “standard” acute stroke care (Table 2). However, we were unable to find a nationally agreed level of resource use required to deliver thrombolysis for acute stroke and no reliable measures of the variation in the current level (and cost) of acute stroke care in the NHS.
Implications for Practice
Our economic model, constructed from the perspective of the NHS, suggests that thrombolysis for acute stroke holds the promise, under favorable assumptions, of being cost-effective in terms of QALYs gained, particularly when the longer-term cost and health outcomes were considered.
However, the range of possible incremental cost-effectiveness ratios was considerable, and the conclusions from the economic modeling were very sensitive to the economic assumptions made and a number of parameters, including the effectiveness of rt-PA (detailed in Table 1). The less favorable estimates indicated that rt-PA could be either marginally cost-effective or harmful (ie, standard therapy was the preferred option).
The primary analyses suggested cost-effectiveness or even cost-savings. However, in view of the lack of precision of the estimates and lack of data on the cost of “rolling out” the treatment to the many centers that do not currently have the resources to give rt-PA, we were unable to model the cost of widespread use of rt-PA for stroke in the UK. However, these data do not preclude the use of rt-PA for the treatment of patients who meet the stringent conditions of the present product license (in the small number of appropriately staffed and equipped centers).
Implications for Research
The cost-effectiveness of rt-PA could not be assessed reliably because of the imprecise estimates of its efficacy. Large-scale randomized trials would be needed to provide sufficiently precise estimates.
If trials established reliably that thrombolysis was effective, then better estimates of the costs of implementing thrombolysis for acute stroke in the NHS will be needed. A more “dynamic system approach” to explore the relationships between different system components and their impact on patient treatment strategies would be informative.
Because the cost-effectiveness estimates were very sensitive to a relatively small set of parameters, future research could focus on the relationship between thrombolytic therapy, resource consequences, and health effects. More data are needed on the effect of the level of disability at 6 months after stroke on subsequent survival, recurrence, and eligibility for re-treatment with thrombolysis.
Peter Sandercock is a Coprincipal Investigator for the Third International Stroke Trial (IST-3) of thrombolysis in acute stroke. The start-up phase of trial (2000-2002) was funded by a grant from the Stroke Association. The expansion phase of the trial (2003-2005) is funded by a grant from the Health Foundation. Other support has consisted of a donation of trial drug and placebo from Boehringer Ingelheim for the first 300 patients to be included in the start-up phase, and help from DESACC Ltd with image analysis aspects of the study. He is coordinating editor of the Cochrane Stroke Review Group. He is a member of the NHS Health Technology Assessment Commissioning Board. Richard Lindley is a Coprincipal Investigator of the Third International Stroke Trial (IST-3) of thrombolysis in acute stroke. He was editor of Stroke Matters, a journal published with an unrestricted educational grant from Boehringer Ingelheim, the manufacturers of rt-PA, until 2002. He received a grant of £5000 from Boehringer Ingelheim for a qualitative study of ethical aspects of thrombolytic therapy.Joanna Wardlaw is the contact reviewer for the Cochrane systematic review of thrombolytic treatment for acute stroke. She is a member of the IST-3 Management Committee and leads a collaborative neuroradiological group assessing the scans of patients entered in the trial. She is director of the SHEFC Brain Imaging Research Centre for Scotland. This is located within the Division of Clinical Neurosciences at the University of Edinburgh. The Centre houses a research magnetic resonance scanner which was funded by the UK Research Councils Joint Research Equipment Initiative, supplemented by grants and donations from various other sources. Boehringer Ingelheim was one of the commercial enterprises which gave a grant towards the purchase of the scanner. She has also attended meetings held by Boehringer Ingelheim to discuss the licencing of rt-PA. She attended as an unpaid independent external adviser but was refunded her travel expenses. She is on the CT review committee for ECASS 3, a trial of rt-PA funded and run by Boehringer Ingelheim for which she will receive payment for her time.Martin Dennis is a member of the IST-3 Management Committee and is participating as a collaborator in IST-3. He is responsible for running stroke services at the Western General Hospital. This has been supported by Boehringer Ingelheim who supplied £800 of patient leaflets. He was founding president of the British Association of Stroke Physicians. Boehringer donated £4000 towards the cost of establishing the Association.Eivind Berge is the national coordinator of IST-3 Norway. Stephanie Lewis is the trial statistician for IST-3. Peter Hand and Joseph Kwan Have entered patients in IST-3. All of the above authors have at some time received lecture fees or travel expenses to attend conferences from a variety of pharmaceutical companies, including Boehringer Ingelheim. None of the authors have a contractual consultancy arrangement with any pharmaceutical company. Furthermore, none of the authors knowingly has any financial interest in or holds any stock in any of the companies whose products are mentioned in this report.
This article is a short summary of a section of a monograph published by the NHS Health Technology Assessment Programme. We would like to thank all the people who have helped with this review, in particular: Brenda Thomas, who designed and performed many of the complex electronic bibliography searches required and was a coauthor of the HTA monograph; the members of the Cochrane Stroke Group Editorial Base staff (Alison McInnes and Hazel Fraser); the Stroke Group Handsearching and translating volunteers whose efforts have helped ensure the Stroke Groups’ Specialised Register of Trials was as comprehensive as possible, and who translated reports of nonEnglish language studies; the patients who were included in the Lothian Stroke Register; the many Medical, Nursing, Administrative, Computing, and Statistical staff who have helped with collection, management, and analysis of the data; Janelle Seymour (Health Services Research Unit, Aberdeen), who made a substantial contribution to the component of the model which relates level of disability, Health Related Quality of Life and Utility values in stroke patients. The study was funded by a grant from the NHS Health Technology Assessment (HTA) Programme (Grant 98/02/02). The opinions and views expressed do not necessarily reflect those of the NHS Executive.
Please see Appendix I (available online at http://stroke.ahajournals.org) for a complete list of Conflicts of Interest.
- Received May 15, 2003.
- Revision received January 7, 2004.
- Accepted February 16, 2004.
Fagan S, Morgenstern L, Pettita A, Ward R, Tilley B, Marler JR, Levine SR, Broderick JP, Kwiatkowski TG, Frankel M, Brott TG, Walker MD. Cost-effectiveness of tissue plasminogen activator for acute ischemic stroke. Neurology. 1998; 50: 883–890.
Dewey H. Cost of acute stroke: is rt-PA cost-effective? Paper presented at: 6th International Symposium on Stroke and Thrombolytic Therapy; 2000; Hamilton Island, Australia.
Sandercock P, Berge E, Dennis M, Forbes J, Hand P, Kwan J, Lewis S, Lindley R, Neilson A, Thomas B, Wardlaw J. A systematic review of the effectiveness, cost-effectiveness and barriers to implementation of thrombolytic and neuroprotective therapy for acute ischaemic stroke in the NHS. Health Technology Assessment. 2002; 6: 1–112.
Dorman P, Dennis MS, Sandercock P. Are the modified “simple questions” a valid and reliable measure of health related quality of life after stroke? J Neurol Neurosurg Psychiatry. 2000; 69: 487–493.
Post PN, Stiggelbout A, Wakker P. Utility of health states after stroke. A systematic review of the literature. Stroke. 2001; 32: 1425–1429.
Sonnenberg FA, Beck JR. Markov models in medical decision making; a practical guide. Med Decis Making. 1993; 13: 322–328.
Bellsey J, Sheldon AM. Healthcare applications of decision analytic modelling. Health Econ Prev Care. 2000; 1: 37–43.
Milne R, Nuitjen M, Weinstein M, Zhou X. Decision analytic modelling in the evaluation of health technologies: a consensus statement. Pharmacoeconomics. 2000; 12: 443–444.
Vermeer F, Simoons ML, de Zwan C. Cost benefit analysis of early thrombolytic treatment with intracoronary streptokinase. Br Heart J. 1988; 59: 527–534.
Simoons ML, Vos J, Martens LL. Cost-utility analysis of thrombolytic therapy. Eur Heart J. 1991; 12: 694–699.
Donaldson C, Currie C, Mitton, C. Cost-effectiveness analysis in health care: contraindications. BMJ. 2002; 235: 891–894.
Hacke W, Kaste M, Fieschi C, von Kummer R, Davalos A, Meier D, Larrue V, Bluhmki E, Davis S, Donnan G, Schneider D, Diez-Tejedor E, Trouillas P. Randomised double-blind placebo-controlled trial of thrombolytic therapy with intravenous alteplase in acute ischaemic stroke (ECASS II). Lancet. 1998; 352: 1245–1251.
Solomon NA, Glick HA, Russo CJ, Lee J, Schulman KA. Patient preferences for stroke outcomes. Stroke. 1994; 25: 1721–1725.
National Health Service in Scotland, Information and Statistics Division, Common Services Agency, Edinburgh. Scottish Health Service Costs. Year ended 31st March 1999. Edinburgh; 1999.
Wardlaw JM, Lewis SC, Dennis MS, Counsell C, McDowall M. Is visible infarction on computed tomography associated with an adverse prognosis in acute ischaemic stroke? Stroke. 1998; 29: 1319.
Hankey GJ, Jamrozik K, Broadhurst RJ, Forbes S, Burvill PW, Anderson CS, Stewart-Wynne EG. Five-year survival after first-ever stroke and related prognostic factors in the Perth community stroke study. Stroke. 2000; 31: 2080–2086.
Wardlaw JM, del Zoppo G, Yamaguchi T, Berge E. Thrombolysis for acute ischaemic stroke (Cochrane Review). In: The Cochrane Library, Issue 3, 2003. Oxford: Update Software.
Scottish Health Service Costs 1998–99 for Western General Hospital. Edinburgh; 1999.
Forbes JF, Dennis MS. Costs and health outcomes of stroke patients. Final Project Report submitted to the Health Service and Public Health Research Committee. Edinburgh: Chief Scientist Office, Scottish Home and Health Department; 1995.
British National Formulary (BNF40), September 2000.
Chambers M. The MEDTAP model. Cerebrovasc Dis. 1998; 8 (suppl 4): 50. Abstract.
Benade M, Warlow C. Costs and benefits of carotid endarterectomy and associated preoperative arterial imaging: a systematic review of health economic literature. Stroke. 2003; 33: 629–638.