The Cost-Effectiveness of Primary Stroke Centers for Acute Stroke Care
Background and Purpose—Primary stroke centers (PSC) have demonstrated improved survival in patients with acute ischemic stroke (AIS). The objective of this study was to evaluate the cost-effectiveness of treating AIS patients in a PSC compared with a nonPSC hospital setting.
Methods—We developed a decision analytic model to project the lifetime outcomes and costs of 2 hypothetical cohorts of 75 AIS patients. Clinical data were derived from a recent observational study comparing PSC- and nonPSC-admitted patients, clinical trials, longitudinal cohort studies, and health state preference studies. Cost data were based on Medicare reimbursement and other published sources. We used a healthcare payer perspective, and the primary outcomes were incremental life expectancy, quality-adjusted life years, and healthcare costs. We performed sensitivity and scenario analyses to evaluate uncertainty in the results.
Results—Admission to a PSC resulted in a gain of 0.22 years of life (95% credible range [CR], 0.12–0.33) and 0.15 quality-adjusted life years (95% CR, 0.08–0.23) per patient, at a cost of $3600 (95% CR, $2400–$5000) per patient, compared with admission to a nonPSC hospital. The incremental cost/quality-adjusted life year gained was $24 000, and all probabilistic simulation results were below the $100 000/quality-adjusted life year threshold. In scenario analyses accounting for as few as 7 and as many as 500 AIS patients/year per PSC, cost-effectiveness improved as the number of AIS patients admitted per year increased.
Conclusions—Our study indicates that care at a PSC for patients with AIS is cost-effective and improves outcomes across a wide range of possible scenarios.
Approximately 795 000 new or recurrent stroke cases and 133 000 stroke deaths occur each year in the United States, at an estimated domestic cost of $74 billion.1 Stroke mortality is expected to increase because of aging of the population and the impacts of changing sociodemographic factors.2
One widely supported measure for improving stroke care is the establishment of primary stroke centers (PSC).3,4 The components of a PSC include: specialized stroke units, stroke treatment protocols, and specially trained staff.3 The benefits of a PSC include: reduced times to physician contact and brain imaging, increased recombinant tissue-type plasminogen activator (rtPA) use, increased admission rates for stroke victims, and the potential involvement of a neurologist in stroke treatment.5 However, establishing a PSC can be challenging, in part because of lack of financial resources.6
Since the Brain Attack Coalition first published recommendations to establish PSCs in 2000, approximately 14% of acute care hospitals in the US achieved PSC designation.3 Measures of the impacts of PSC certification have been limited by a lack of precertification hospital data.7 However, a recent, large, observational study found that acute ischemic stroke (AIS) patients admitted to stroke centers adhering to Brain Attack Coalition guidelines had significantly lower mortality and were more likely to receive thrombolytic therapy compared with patients admitted to nonPSC hospitals.8 Although this study provided critical data on the comparative clinical effectiveness of PSCs, no cost-effectiveness analyses of PSCs have been reported.9 The objective of this study was to estimate the long-term clinical and economic impacts of treating AIS patients in a PSC compared with a nonPSC hospital setting.
We developed a decision-analytic model to evaluate the long-term incremental cost-effectiveness of treating AIS patients in a PSC compared with a nonPSC hospital setting. The analysis was conducted from a healthcare payer perspective using a lifetime horizon to enable assessment of life expectancy. All costs were inflated to 2010 dollars using the Medical Consumer Price Index, and long-term costs and outcomes were discounted at 3% per year. The model was programmed in Microsoft Excel.
The average age and size of the hypothetical cohort was based on the population reported by Xian et al.8 In this observational study, 15 297 AIS patients were admitted to 104 designated stroke centers in New York State over 2 years, and adjustments were made to account for diversity in other patient characteristics between the PSC and nonPSC groups. From these figures, we estimated 75 AIS patients were treated per year at each PSC, which we used in our base case analysis.
The model compares 2 simulated, equivalent cohorts of AIS patients: 1 cohort admitted to a PSC, and the other admitted to a nonPSC hospital setting. We created a decision tree to model short-term (1-year) treatment and outcomes (Figure 1). A proportion of patients could receive rtPA (online-only Supplemental Appendix). In the base case analysis, we assumed treated patients arrived within 3 hours postAIS onset; in a scenario analysis, we assessed a 4.5-hour rtPA therapeutic window.10,11
Patients who survived the short-term model were classified as nondisabled or disabled. Using the NINDS trial modified Rankin scale (mRS) outcomes (ranging from 0 [nonsymptomatic] to 5 [severe disability]),12 we grouped mRS scores 0 to 1 as nondisabled and mRS 2 to 5 as disabled, and derived group-weighted parameters from mRS-specific ones. We compared this approach with using individual mRS health states and found that the methods gave near-equivalent results; thus, our simplified approach captured the dissimilarities among mRS groups.
Following the short-term model, nondisabled and disabled patients entered a long-term (post-1-year) Markov model to simulate clinical outcomes such as recurrent stroke, transition to disability, and death over the patients' remaining years of life (Figure 2). In the long-term model, patients remained in health states experienced at the end of the short-term model until either nonstroke death or a stroke recurrence resulted in the same or worse health state or stroke death. The long-term model simulated 40 years, by which time all simulated patients died. The derivation of model parameters (Table 1) is described in the online-only Supplemental Appendix.
We calculated life years, quality-adjusted life years (QALY), and lifetime direct medical costs in PSC and nonPSC settings. The incremental cost-effectiveness ratio (ICER) was calculated as the difference in costs divided by the difference in QALYs.
We performed a series of scenario analyses to evaluate assumptions in our model.
A Cochrane Collaboration meta-analysis of stroke unit studies concluded that treatment at a formal stroke unit was associated with a 17% reduction in the odds of death (odds ratio [OR], 0.83; 95% CI, 0.71–0.96) following AIS.13 We used this estimate and doubled the base case stroke unit costs to assess the outcomes of PSCs that included more specialized stroke unit care.
We modeled different-sized cohorts to assess the effect of various admission rates on model outcomes, including identifying a threshold cohort size that resulted in a $100 000/QALY ICER, and up to 500 AIS patients per year to represent outcomes at a large PSC.
We modeled 30-day race/ethnicity-stratified PSC mortality reductions to estimate the impact of a PSC for different racial/ethnic populations.8
We explored larger increases of rtPA treatment at PSCs after a review of the literature on PSCs identified up to 15% greater use of rtPA compared with nonPSCs.
We evaluated a scenario in which rtPA could be given within 4.5 hours of AIS onset to model the extended therapeutic window recently recommended by the American Heart Association/American Stroke Association.11 We used estimates of nondisabled status, symptomatic intracranial hemorrhage, and death from the European Cooperative Acute Stroke Study III (ECASS III) for patients arriving within 3 to 4.5 hours postAIS onset.10 We assumed 10% to 30% relative increases in the proportion of patients receiving rtPA.
We performed a threshold analysis to determine the total cost of a PSC that would result in a $100 000/QALY ICER.
We modeled the additional annual cost for an on-call neurologist for weekends only, evenings and weekends, and 24 hours/day for 7 days/week. We assumed a 24-hour day, $400 per diem; a weekend was 2 days, and an evening was 12 hours. We made no adjustments to existing PSC costs.
We performed sensitivity analyses to examine the influence of uncertainties in the model inputs and to judge the robustness of the findings. Single-variable (1-way) sensitivity analyses were performed with the value of each input varied over ranges shown in Table 1. Clinical parameter ranges were obtained from reported (when available) and data-derived CIs. We explored a conservative range for increased risk of mortality postAIS. We derived utility ranges from previous studies.14,15 PSC-specific costs were varied by ±50% to reflect substantial variability in these estimates; all other costs were varied by ±20%.
We also performed probabilistic sensitivity analysis using Monte Carlo simulation. All model parameters were jointly varied over 10 000 simulations using specified previous distributions that approximated their ranges, enabling the calculation of 95% credible ranges (CR) for each incremental outcome (in Bayesian statistics, a credible range is a posterior probability interval, analogous, but not equivalent, to a CI in frequentist statistics). Probabilistic results were plotted on a cost-effectiveness plane, and a cost-effectiveness acceptability curve was generated to demonstrate the probability of a PSC being cost-effective versus a nonPSC at various willingness-to-pay thresholds.
Admission to a PSC resulted in a gain of 0.22 years of life (95% CR, 0.12–0.33) and 0.15 QALYs (95% CR, 0.08–0.23) per patient, at an increased cost of $3621 (95% CR, $2405–$4939), compared with admission to a nonPSC. The incremental cost-effectiveness ratio for AIS care in a PSC setting compared with a nonPSC setting was $23 990/QALY (95% CR, $18 386–$34 126; Table 2).
Scenario analysis results are presented in Table 3. When we modeled the Cochrane Review's mortality impact of a stroke unit, the difference in mortality was greater between the 2 settings and the model resulted in greater incremental life expectancy for the PSC population. Incremental cost also increased because of the increase in stroke unit cost and the increased lifetime health costs of a larger population of stroke survivors; however, the cost per QALY was lower relative to the base case.
Patient cohort size was negatively correlated with incremental cost. As greater numbers of patients entered the model, PSC-associated costs were distributed among a wider population, leading to lower incremental costs per patient and lower ICERs. The ICER surpassed the $100 000 willingness-to-pay threshold with 7 or fewer patients/PSC per year. A PSC that treated 500 AIS patients per year resulted in an ICER of $16 589.
Race/ethnicity-stratified estimates of PSC mortality reduction showed that cost and life expectancy were positively correlated with increased PSC mortality reduction, whereas the ICER was negatively correlated.
When we modeled higher increases in rtPA treatment at a PSC compared with a nonPSC, life years and QALYs increased, whereas cost decreased, because rtPA treatment led to a higher proportion of nondisabled patients who incurred fewer costs over time.
Using an extended therapeutic window of 4.5 hours, both PSC and nonPSC settings exhibited gains in life years and QALYs at lower cost as the percentage of patients receiving rtPA increased.
The addition of an on-call neurologist at a PSC increased the incremental cost per patient and the ICER; however, even with a 24 hour/day, 7 day/week on-call neurologist, the results were less than $50 000/QALY.
One-Way Sensitivity Analyses
The results of the 1-way sensitivity analyses are shown in Figure 3A–C. Incremental cost and QALYs were primarily influenced by the mortality reduction associated with PSC admission. Second, incremental QALYs were sensitive to health state utilities and stroke recurrence, and costs were sensitive to health state and PSC-associated costs. The ICER was most sensitive to the utility of being disabled, the PSC mortality reduction, long-term annual costs for disabled patients, and PSC staff costs.
Probabilistic Sensitivity Analysis
All simulations were cost-effective at a $100 000/QALY threshold; 99% of simulations resulted in ICERs less than $50 000/QALY. See online-only Supplemental Figure S1 for the cost-effectiveness acceptability curve.
We evaluated the long-term clinical, quality of life, and economic outcomes associated with admission to a PSC versus a nonPSC setting using a disease-based, decision-analytic model. Our analysis supports PSCs increasing average life expectancy and QALYs in a cost-effective manner. On average, AIS patients treated at a PSC live approximately 82 days longer than do patients treated at a nonPSC, at an additional cost of approximately $3600 per patient. Various sensitivity and scenario analyses demonstrated that these results were robust despite uncertainty in several model parameters.
For a PSC that treats 75 AIS patients per year, approximately 16.8 life years or 11.3 QALYs may be gained per annual cohort. A PSC that treats 500 AIS patients per year could gain as much as 112 life years and 75 QALYs per cohort. Because the cost per patient of a PSC decreases as the number of patients increases, higher volume centers appear to provide greater economic value. However, even a PSC that admits only 8 patients per year was cost-effective at a $100 000/QALY willingness-to-pay threshold. A telemedicine hub and spoke model, with transfer of appropriate patients to a PSC, may better serve hospitals expected to serve fewer patients.16,17
The results are primarily driven by the mortality reduction associated with treatment in a PSC. Although greater survival portends greater total cost over patients' remaining lifetimes in both hospital settings, the incremental cost per QALY of a PSC decreases as survival increases. This result indicates that additional efforts to reduce mortality via PSC may be cost-effective.
Although there is variability in the level of resources required to run a PSC, the results of various sensitivity and scenario analyses indicate that even with substantial increases in cost, such as adding a 24-hour on-call neurologist, a PSC provides good value for the money. However, although some PSCs may have staff neurologists and neurological on-call services, such PSCs would not be representative of the ones in the study by Xian and colleagues.
A widely accepted body of literature supports the need to increase rtPA treatment rates for eligible patients,5,10,14,18 and this ability is often cited as a strength of PSCs. Scenario analyses, in which we increased the probability of rtPA treatment at a PSC, showed that greater rtPA use correlates with greater quality-adjusted life expectancy and lower cost. The results of this analysis support additional efforts to increase rtPA treatment opportunities for appropriate patients in both settings.
When we modeled a 4.5-hour therapeutic window for rtPA, we calculated increases in life years and QALYs at reduced cost in both hospital settings. However, the extended therapeutic window had little impact on the incremental differences between a PSC and nonPSC. If nonPSCs are found to be more reluctant to administer rtPA in this extended window, the benefits of a PSC versus a nonPSC may become more evident.
Comparison With Previous Studies
Our systematic literature review did not identify any previous decision-analytic modeling studies comparing PSCs with nonPSCs, in the US or abroad.
We used the comparative effectiveness data from Xian and colleagues, in which an instrumental variable approach was used to adjust for potential confounding arising from differences in characteristics of patients presenting to a PSC versus a nonPSC setting.8 Residual confounding in this study may have led to an overestimate of the PSC-associated mortality reduction in AIS patients. However, the findings of our analysis were robust over a range (1.5%–4.4%) of mortality reduction estimates.
Although we did include symptomatic intracranial hemorrhage for AIS patients postadmission, our model did not account for patients admitted with hemorrhagic stroke (approximately 19% of stroke victims in New York State during the Xian et al study period). Hemorrhagic stroke is associated with higher mortality compared with AIS, and stroke centers are more likely to admit hemorrhagic stroke patients than are nonstroke centers.8,19,20 Therefore, our results are specific to AIS patients only.
We assumed that PSCs already had neuroimaging capabilities, and a 24-hour call team was available for treatment of all arrivals, criteria typically used to designate PSCs. The cost to establish a PSC at hospitals without these capabilities may be significantly higher, and our results do not apply to such institutions.
The literature is inconsistent in defining the increase in annual mortality following stroke, particularly in disabled patients. However, we used conservative estimates of this increase, and indeed, higher disabled patient annual mortality tended to lower lifetime costs because of the shorter life expectancy. In addition, our model does not account for the declining trend in stroke recurrence rates caused by improved risk factor management.
We assumed the effect of rtPA administration was independent of hospital setting. The safety and effectiveness of rtPA in a PSC setting may be greater than in a nonPSC setting, although we are not aware of studies that compared outcomes in these settings. Nonetheless, we found that PSC-mediated increases in rtPA use in patients arriving between 0 to 3 hours or between 3 to 4.5 hours resulted in cost-effective improvements in patient outcomes.
Treatment of AIS patients in a PSC appears to be cost-effective, and efforts should be made to increase opportunities for patients to reach these facilities instead of nonPSC hospitals. Our findings support existing recommendations for establishing PSCs in more locations throughout the country. A telemedicine hub and spoke model that transfers appropriate patients to a hospital with a PSC may better serve hospitals expected to serve fewer than approximately 10 AIS patients per year. Future research based on data from the growing number of PSCs will be useful for refining these results.
Sources of Funding
This study was supported by Genentech, Inc., and NIH grant 1R01 HL096944 (Levine).
The study was sponsored by Genentech, Inc. The authors had complete control over the analysis and interpretation of the data and content of the manuscript. Mr Guzauskas has served as a consultant for Genentech, Inc. Ms Villa is employed by Genentech, Inc. Dr Levine has served on an advisory board and as a consultant for Genentech, Inc (honorarium donated to stroke research), is an associated editor for MEDLINK, has served as an expert witness in medical-legal cases on acute stroke, and receives grant support from the National Institutes of Health. Dr Veenstra has served as a consultant for Genentech, Inc.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.111.648238/-/DC1.
- Received December 15, 2011.
- Accepted February 3, 2012.
- © 2012 American Heart Association, Inc.
- Lloyd-Jones D,
- Adams RJ,
- Brown TM,
- Carnethon M,
- Dai S,
- De Simone G,
- et al
- Elkins JS,
- Johnston SC
- Alberts MJ
The Joint Commission. Advanced Certification for Primary Stroke Centers. http://www.jointcommission.org/certification/primary_stroke_centers.aspx. Accessed April 14, 2011.
- Fonarow GC,
- Gregory T,
- Driskill M,
- Stewart MD,
- Beam C,
- Butler J,
- et al
- Del Zoppo GJ,
- Saver JL,
- Jauch EC,
- Adams HP
- Wilson JT,
- Hareendran A,
- Hendry A,
- Potter J,
- Bone I,
- Muir KW
Stroke Unit Trialists' Collaboration. Organised inpatient (stroke unit) care for stroke. Cochrane Database Syst Rev. 2007;4:CD000197.
- Fagan SC,
- Morgenstern LB,
- Petitta A,
- Ward RE,
- Tilley BC,
- Marler JR,
- et al
- Samsa GP,
- Reutter RA,
- Parmigiani G,
- Ancukiewicz M,
- Abrahamse P,
- Lipscomb J,
- et al
- Pervez MA,
- Silva G,
- Masrur S,
- Betensky RA,
- Furie KL,
- Hidalgo R,
- et al
- Nelson RE,
- Saltzman GM,
- Skalabrin EJ,
- Demaerschalk BM,
- Majersik JJ
- Gropen TI,
- Gagliano PJ,
- Blake CA,
- Sacco RL,
- Kwiatkowski T,
- Richmond NJ,
- et al
- Stradling D,
- Yu W,
- Langdorf ML,
- Tsai F,
- Kostanian V,
- Hasso AN,
- et al