Long-Term Costs of Stroke Using 10-Year Longitudinal Data From the North East Melbourne Stroke Incidence Study
Background and Purpose—Stroke is costly, although little is known about the long-term costs of survivors of stroke. In previous cost-of-illness studies, lifetime costs have been modeled based on estimates to 5 years after stroke. Building on previous work from the North East Melbourne Stroke Incidence Study (NEMESIS), we aimed to describe resource use at 10 years and recalculate the lifetime societal costs of ischemic and hemorrhagic (intracerebral hemorrhage) stroke.
Methods—Ten-year patient-level resource use data were obtained and updated prices and population demographic statistics for 2010 were applied to our cost-of-illness models. We incorporated incidence data from a larger study region of NEMESIS than that used in the previous model and new 10-year survival and recurrent stroke rates. One-way sensitivity and probabilistic multivariable uncertainty analyses were undertaken.
Results—For ischemic stroke, the overall average annual direct costs at 10 years (US dollars [USD] 5207) were comparable to those for survivors between 3 and 5 years (USD5438). However, the contribution of some costs varied (eg, medications contributed 13% at 5 years and 20% at 10 years). For intracerebral hemorrhage, annual direct costs were considerably (24%) greater at 10 years than estimated using 3 to 5 year data. Greater average lifetime costs per case were found using the updated models (ischemic stroke: previous model USD51806 and current USD68 769; intracerebral hemorrhage: previous model USD43 786 and current USD54 956 per case). Following sensitivity and multivariable uncertainty analyses, the findings were robust.
Conclusions—Costs to 10 years after stroke have not previously been reported. Our findings demonstrate the importance of estimating resource use over longer periods for forecasting lifetime estimates.
Similar to other countries, stroke is an important cause of disease burden and health expenditure in Australia.1 The quality-adjusted life years lost per case is ≈5.09 for ischemic stroke (IS) and 6.17 for intracerebral hemorrhage (ICH).2 Building on the first Australian cost-of-illness study and using 5-year outcome data,3,4 Cadilhac et al5 found that the societal lifetime costs per first-ever (incident) case in 2004 were US dollars (USD) 47 354 for IS and USD39 628 for ICH (Australian dollar [AUD]-purchasing power parity 2004 value is 1.367 or AUD64 733 for IS and AUD54 172 for ICH).6 Generally, few cost estimates are based on long-term resource use observations and fewer from comprehensive community-based population studies.
The results and methods of previous cost-of-illness studies in stroke vary.7,8 Because the costs of stroke peak within the first year, but decline over time, it is crucial to account for long-term resource use to not overestimate lifetime costs.9 Furthermore, the direct costs of informal care and indirect costs of productivity losses after stroke are often omitted, despite these costs being substantial.7,9,10 In a recent review, Luengo-Fernandez et al7 identified only 5 cost-of-illness studies for stroke that were based on population-based cohorts and only 2 studies11,12 that included long-term assessment of resource use, beyond 1 year after stroke. In January 2014, we conducted a literature review in MEDLINE and EconLit using systematic review methodology and found 2 additional, population-based cost-of-illness studies5,13 that included long-term resource use data. However, we did not identify any study that included resource data beyond 5 years after stroke. Building on previous work from the North East Melbourne Stroke Incidence Study (NEMESIS),5 we aimed to describe resource use at 10 years and recalculate the lifetime societal costs of IS and ICH for Australia.
The Model of Resource Utilization Costs and Outcomes for Stroke (MORUCOS) has been developed and used for estimating the lifetime costs of stroke in Australia and is primarily based on the NEMESIS study.14 This model was applied to different stroke subtypes by Dewey et al3 for the reference year 1997 and updated for 2004 by Cadilhac et al.5 NEMESIS was conducted in 2 phases. First, a pilot study was conducted in Melbourne between May 1996 and April 1997 in an 8 postcode region (population 133 816).3 In a second phase, conducted between May 1997 and April 1999 the study region was expanded to 22 postcodes (306 631 residents).5,15
MORUCOS incorporates an incidence-based, patient-level costing approach from a societal perspective and accounts for all stroke-related costs: direct costs of treatment, informal care and out-of-pocket expenses, and indirect costs of productivity losses. Among participants who provided written informed consent to be followed-up and were known to be alive, interviews of resource use were conducted at 3 months, 6 months, and 12 months after the index stroke, as well as at 3, 4, and 5 years. All participants who were still alive at 10 years after stroke were approached to participate in the 10-year interviews. In the surveys of resource use and costs, only those resources that could be directly attributed to the stroke event were included. That is, if a person was in a nursing home and had a stroke and then returned to the nursing home, we did not include the costs of nursing home care for that person. (Methods in the online-only Data Supplement.)
In MORUCOS, population demography estimates are linked with epidemiological data and resource use data obtained from NEMESIS patients. The resource use data are linked to unit costs from standard references such as the Medicare Benefits Schedule,16 the Pharmaceutical Benefits Schedule,17 and the Manual of Resource Items and Associated costs,18 to calculate estimates of direct annual costs. Linked spreadsheets are then used to calculate lifetime costs of stroke for the Australian population, based on direct and indirect costs incurred over the remaining lifetime of first-ever stroke survivors. The MORUCOS models have been primarily based on the smaller NEMESIS pilot study,5 in addition to long-term survival beyond 5 years in the Perth Community Stroke Study.19 Because there are now 10-year NEMESIS data available for all epidemiological and resource use model input variables, there was an opportunity to describe long-term resource use and re-estimate lifetime costs using current data from this larger cohort of patients.
We updated cost-of-illness models for IS and ICH from a 2004 to a 2010 reference year and included new resident population estimates and life tables of the Australian population. For the current analysis, all epidemiological data were derived from the larger 22-postcode cohort of NEMESIS. This population similarly represents the Australian population as did the pilot cohort but has more robust estimates of incidence, survival, and recurrence having a larger sample size. The unit costs in MORUCOS were updated to 2010 and where new unit costs were not available, the Total Health Price Index was used to adjust the costs to the new reference year.20 The Total Health Price Index method was also used to inflate the estimates from the 2004 model to enable comparison with the new results derived from this current analysis. To convert the 2010 costs presented in the results section to equivalent AUD values, multiply USD by 1.506 which is the recommended purchasing power parity conversion.6
To model the ongoing costs after the first year, the direct costs observed in years 3, 4, and 5 after stroke were averaged and assumed to have occurred annually from the second to the sixth year after stroke. This approach is similar to the previous update where the average from 3 to 5 years was used to model lifetime costs because this was deemed the best estimate of long-term annual costs.5 In this current update, direct costs observed at 10 years were annually applied from the seventh year over the remaining life expectancy. These costs are based on estimates of resource use gathered from interviews with 292 patients with first-ever stroke, 10 years after their stroke. These interviews took place between 2006 and 2009. The resource categories included in the questionnaire administered at these interviews covered most resources included in the previous MORUCOS update, such as the number of outpatient rehabilitation visits, hospitalizations because of stroke recurrence, and prescription medication as a consequence of stroke (Table I in the online-only Data Supplement). To reduce responder burden, we omitted some categories of resources from the 10-year interview. When a certain resource use category was not covered by the 10-year questionnaire (note these omitted categories of resource use had contributed only 7.3% to the average annual direct costs of IS and 2.7% to ICH in years 3–5; Table II in the online-only Data Supplement), the resource use observed at 5 years was used to approximate the 10-year costs.
The lifetime costs per case were calculated in MORUCOS by summing together the total costs incurred in the first year with the direct and indirect costs incurred over the rest of life. All costs incurred after the first year were discounted using a 3% discount rate.21
All analyses for this study were conducted using Intercooled STATA version 10 (Stata Corporation, 2005) and Microsoft Excel 2007 (Microsoft Corporation, 2007). One-way sensitivity analyses were used to vary the discount rate between 0% and 5%. Multivariable probabilistic uncertainty analyses using 3000 Monte Carlo sampling simulations were conducted using @Risk software version 5.0 (Palisade Corporation, 2007, Ithaca, NY) for the main input variables (Table III in the online-only Data Supplement).
For first-ever IS, 243 of 283 NEMESIS participants who were alive at 10 years were interviewed (response rate 85.9%). For first-ever ICH, 43 of 50 patients alive at 10 years were interviewed (response rate 86%). The response rates for the resource use interviews at each time point are provided in Table IV in the online-only Data Supplement.
In 2010, 25 351 first-ever IS cases and 5356 first-ever ICH cases were estimated to have occurred in Australia (Table 1). Compared with the 2004 MORUCOS model,5 there were 8% fewer incident ISs and 25% more ICHs. The increase in the number of ICH cases results from an increase in the incidence rates among people aged 45 to 84 years, combined with an increase in population estimates for these age groups. The survival estimates also differed, being greater for IS at 5 years (52% versus 47%) and 10 years (31% versus 21%). For ICH, the survival rate in 2010 was 38% (2004: 48%) at 5 years and 27% (2004: 21%) at 10 years.
For 2010, ≈53% of all cases were men and the average age at onset was 75 years for IS and 72 years for ICH. Among the 292 first-ever stroke patients who completed a resource use interview at 10 years, 243 (83%) had an IS and 43 (15%) had an ICH. Compared with survivors not interviewed at 10 years, those who were interviewed had similar age, sex, and stroke severity levels at baseline.
First-Year Direct Costs
Per incident case, the average direct costs incurred in the first year were USD19 992 for IS (AUD30 110) and USD11 796 for ICH (AUD17 767), with acute hospitalization making up 35% (IS) and 24% (ICH) of these costs. The cost of a recurrent stroke was estimated to be USD21 482 (AUD32 354) for both subtypes.
Ongoing Direct Costs
The average annual direct costs observed between years 3 and 5 were USD3682 for IS and USD4051 for ICH and USD5438 for IS and USD5807 for ICH when including informal care and out-of-pocket costs (Table 2). The average direct costs observed at 10 years were USD3598 for IS and USD5997 for ICH and USD5207 for IS and USD7607 for ICH when informal care and out-of-pocket costs were included.
For IS, the total annual direct costs, as well as the cost estimates in most cost categories, were similar at 10 years as they were for the average of 3, 4, and 5 years (Table 2). However, medication costs were slightly greater at 10 years (USD616) than at 3 to 5 years (USD458). In contrast, costs of inpatient rehabilitation were lower at 10 years (USD155) than 3 to 5 years (USD517), this being a relative decline of 70%. The results of the z-tests indicate that the direct cost categories contribute similarly to the overall direct cost across both costing methods. Only the proportional cost of inpatient rehabilitation for IS is significantly less at 10 years than the average of 3 to 5 years (3% versus 9.5% of total direct costs; P<0.01). The most important direct cost categories at 10 years were aged care facilities, informal care, and medications (Figure).
For ICH, the total annual direct costs were 24% greater at 10 years (USD7607) than that calculated using the 3 to 5 year average (USD5807). In absolute terms, aged-care-facilities costs were 64% greater at 10 years (USD3900) than at 3 to 5 years (USD2385), whereas medication costs were 1.7-fold (USD572) that in 3 to 5 years (USD328). Despite this, the percentage contribution of these cost categories to the total annual direct costs remained similar between the 2 costing methods, reflecting the fact that the total annual direct costs were also greater.
Indirect and Lifetime Costs
Indirect costs were USD23.3 million for IS (USD920 [AUD1385] per case) and USD9.8 million for ICH (USD1826 [AUD2750] per case; Table 3). The total lifetime costs were USD1743.4 million for IS and USD294.4 million for ICH. This corresponds to total costs per incident case of USD68 769 (IS; AUD103 566) and USD54 956 (ICH; AUD82 764). In comparison with the previous 2004 model, indirect costs decreased by USD14 million for IS (2010: 23.3 million versus 37.3 million) and slightly decreased for ICH by USD0.4 million (2010: 9.8 million versus 10.2 million). The total lifetime costs and average costs per case were significantly greater for both subtypes. The estimates that include long-term costs differ slightly when 0% or 5% discounting is applied.
When important input variables were varied within specified ranges of uncertainty, the median lifetime cost of first-ever IS was found to be USD68 738 (95% uncertainty interval [UI], USD54 718–USD83 703; Table V in the online-only Data Supplement). For first-ever IS cases, the first-year direct costs were USD494.2 million (95% UI, USD424.5–USD563.6 million), and total costs were USD1.7 billion (95% UI, USD1.34–USD2.08 billion). This was based on the median number of first-ever IS strokes in 2010 being 24 716 (95% UI, 24 113–25 319). For ICH, the uncertainty analysis provided a median lifetime cost per case of USD55 417 (95% UI, USD43 764–USD68 130). For first-ever ICH cases, the first-year direct costs were USD61.7 million (95% UI, USD53.2–USD70.27 million) and total costs were USD289.2 million (95% UI, USD227.4–USD357 million). This was based on the median number of first-ever ICH strokes in 2010 being 5222 (95% UI, 5096–5350). For interested readers, we also provide total quality-adjusted life years lost derived from each of cohorts modeled obtained from comparisons to age- and sex-matched population without stroke2 in Table V in the online-only Data Supplement.
This is the first cost-of-illness study in stroke, conducted in a population-based setting that includes 10 years of follow-up data of resources used by survivors of stroke. This novel work contributes to a better understanding of the long-term costs of stroke than when a shorter follow-up timeframe is used. We found that use of resource use and epidemiological data collected ≤10 years after stroke provided better estimates of the lifetime costs of stroke. The uncertainty analyses provided evidence that our point estimates were robust. Comparing our results to other studies is complicated because long-term costs are unavailable from prospective cohort studies and different methodological approaches have been used. However, a common finding of most prior work, including the 2004 MORUCOS model, was that annual direct costs declined after the first year.5,6,8,11,12 A cost-of-illness study using a comparable methodology to MORUCOS and that included 5-year resource use observations from the perspective of the social health insurance was conducted in Germany.11 The investigators identified lifetime costs per case of €50 507 (reference year: 2004-purchasing power parity: 1€=1.11 USD)6 for first-ever IS which is less than our estimates. However, they also identified a significant decline in total costs after the first year and, similar to our results, a sharp decline in rehabilitation costs. In the Oxford Vascular Study of 729 stroke patients, Luengo-Fernandez et al13 estimated 5-year hospital care costs (including outpatient visits) of USD26 347 for IS and USD19 641 for ICH (reference year: 2008). Although the investigators did not discount future costs, the distribution of costs over 5 years appeared to be similar to our estimates with over half of the direct costs occurring in the first year.
Our study significantly extends prior work and demonstrates that there are no further declines in direct costs after the fifth year for IS. Therefore, overall direct costs at 5 years may be considered a good predictor of long-term direct costs. The cost drivers at 10 years highlighted greater costs for medications and aged care facilities and lower costs for inpatient rehabilitation and informal care.
In comparison with the previous 2004 MORUCOS model,5 we found significantly greater lifetime costs for IS, despite similar estimates of long-term direct costs. The main driver for these differences is attributed to the use of different estimates of epidemiological data in the current update. For IS, the longer survival and more frequent risk of recurrent events imply that, on average, patients use more resources over a longer period of time. The greater survival in patients with IS aged <65 years also impacts on indirect costs. The productivity losses because of premature death were lower, and these outweighed the productivity losses of more hospitalizations for stroke recurrence, yielding slightly lower indirect costs in total.
For ICH, we found considerably greater direct costs at 10 years than 3 to 5 years, but this result should be interpreted with caution because only 43 stroke survivors provided interview data. To generate more reliable data for ICH, larger patient samples are needed. Similar to IS, we found greater lifetime costs. This increase is partly explained by higher estimates of long-term direct costs. By incorporating the observations at 10 years, we assumed that on average, an ICH patient incurs 1.3-fold increased direct annual costs after the sixth year. For ICH, the new survival estimates are less than estimated in earlier versions of MORUCOS, but greater recurrence rates still drive the total costs.
Although the estimates of indirect and total costs per case differ significantly from the 2004 model, they may also be more valid. Rather than using survival and recurrence beyond 5 years poststroke from the Perth Community Stroke Study,19 we now use 10-year data from NEMESIS. This represents a more internally consistent approach. The use of 10-year data derived from the expanded study region may also explain differences in epidemiological estimates (eg, the incidence rate of ICH that differed among the current and previous model update).
Although the epidemiological data in NEMESIS were longitudinally and continuously collected using a large population, the region has a slightly different socioeconomic status compared with the pilot-region and with Australia as a whole. This may explain differences in the incidence estimates between the current study and the 2004 model.5 Therefore, the epidemiological data in MORUCOS may not be perfectly representative of Australia. Uncertainty analyses were used to account for some of this likely variation.
Another limitation concerns the collection of resource use data. Although interviews were conducted ≤10 years after stroke onset, they only took place at certain time points and could involve different patients. The resource use estimates are from multiple cross-sectional datasets, and the patients surveyed are assumed to be representative of all NEMESIS patients. For patients who died, no interviews with their next-of-kin were conducted. Therefore, the time interval between death and next possible date of interview is subject to selection bias. As resource use in people who die in a particular year might be greater than for someone who does not die, our results may underestimate the total costs. To account partially for this issue, we assigned a half year value of the annual costs obtained from stroke survivors to those who died in a particular year to avoid underestimating their costs.
A further limitation was that, to reduce responder burden, we did not collect the same amount of detail on resource use at the 10-year interview. To address potential underestimation of resource use from this circumstance, we approximated the costs for 10 years using the resource use data observed between 3 and 5 years for these cost categories and applied current unit costs for these. This was deemed the best approach given the limitations of our data and the fact that the cost categories omitted were those that contributed to <8% of total direct annual costs in IS and <3% in ICH. Another limitation is that few patients reported use of some resources. This may have introduced reporting bias for infrequently used resources, and because all resource use data are based on self-reports by patients or their next-of-kin, recall bias is possible. However, our bottom-up approach has enabled all kinds of resources, eg, informal care, that would not have been covered by other data sources such as claims data. Our detailed multivariable uncertainty analysis modeling was undertaken to account for issues such as these in providing the likely range of costs given uncertainty for our input variables, including resources used.
This research makes a significant contribution to better understand the long-term costs of IS and ICH. We show that there is no further decline of direct costs beyond 5 years from stroke for IS but a significant increase for ICH. The high lifetime costs for both subtypes emphasize the significant economic implications of stroke. Our results provide important information for health policy and planning. Use of medications and rehabilitation is the resources that change the most in the long term, although further work is required to better determine these changes for ICH.
We acknowledge Cathy Mihalopoulos who contributed to the development of the original Model of Resource Utilization Costs and Outcomes for Stroke (MORUCOS) model and the National Stroke Foundation who commissioned the original model. We thank the North East Melbourne Stroke Incidence Study (NEMESIS) research nurses for their diligence regarding data collection and patient follow-up.
Sources of Funding
T. Gloede was supported by a scholarship from the Koeln Fortune Program, Faculty of Medicine, University of Cologne, Germany. Dr Cadilhac (1063761 cofunded by National Heart Foundation) and Dr Thrift (1042600) were supported by fellowships from the National Health and Medical Research Council (NHMRC; Australia). S.M. Halbach was the recipient of a PROMOS grant (1000 euro) from the German Academic Exchange Service used to support work on this project in Australia. Dr Thrift received funding for North East Melbourne Stroke Incidence Study (NEMESIS) with grants from the NHMRC (154600, 307900, and 526601), VicHealth, the Foundation for High Blood Pressure Research, and the National Stroke Foundation.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.114.006200/-/DC1.
- Received May 27, 2014.
- Revision received September 10, 2014.
- Accepted September 15, 2014.
- © 2014 American Heart Association, Inc.
- 1.↵Australian Institute of Health and Welfare. Australia’s Health 2010. Canberra: Australian Institute of Health and Welfare; 2010.
- Dewey HM,
- Thrift AG,
- Mihalopoulos C,
- Carter R,
- Macdonell RA,
- McNeil JJ,
- et al
- Dewey HM,
- Thrift AG,
- Mihalopoulos C,
- Carter R,
- Macdonell RA,
- McNeil JJ,
- et al
- Cadilhac DA,
- Carter R,
- Thrift AG,
- Dewey HM
- 6.↵Organization for Economic Co-Operation and Development (OECD). PPPs and exchange rates. Paris: OECD; 2014. http://stats.oecd.org/Index.aspx?datasetcode=SNA_TABLE4. Accessed June 22, 2014.
- Luengo-Fernandez R,
- Gray AM,
- Rothwell PM
- Evers SMAA,
- Struijs JN,
- Ament AJHA,
- van Genugten MLL,
- Jager JHC,
- van den Bos GAM
- Scholte op Reimer WJM,
- de Haan RJ,
- Rijnders PT,
- Limburg M,
- van den Bos GAM
- Kolominsky-Rabas PL,
- Heuschmann PU,
- Marschall D,
- Emmert M,
- Baltzer N,
- Neundörfer B,
- et al
- Luengo-Fernandez R,
- Gray AM,
- Rothwell PM
- 16.↵Australian Government Department of Health and Ageing. Medicare Benefits Schedule Book. Canberra: Commonwealth of Australia; 2010. http://www.health.gov.au/internet/mbsonline/publishing.nsf/Content/Downloads-201011. Accessed May 9, 2014.
- 17.↵Australian Government Department of Health and Ageing. Schedule of Pharmaceutical Benefits. Canberra: Commonwealth of Australia; 2010. http://www.pbs.gov.au/info/publication/schedule/archive#2011. Accessed May 9, 2014.
- 18.↵Australian Government Department of Health and Ageing. Manual of Resource Items and Their Associated Costs. Canberra: Commonwealth of Australia; 2009. http://www.pbs.gov.au/info/industry/useful-resources/manual. Accessed May 9, 2014.
- Hardie K,
- Hankey GJ,
- Jamrozik K,
- Broadhurst RJ,
- Anderson C
- 20.↵Australian Institute of Health and Welfare. Health Expenditure Australia 2009-10. Canberra: Australian Institute of Health and Welfare; 2011. http://www.aihw.gov.au/publication-detail/?id=10737420435. Accessed May 9, 2014.
- Drummond M,
- Sculpher MJ,
- Torrance GW,
- O´Brien BJ,
- Stoddart GL