the hospital costs of care for stroke in nine european countries

HEALTH ECONOMICSHealth Econ. 17: S21–S31 (2008)Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/hec.1329

THE HOSPITAL COSTS OF CARE FOR STROKE IN NINEEUROPEAN COUNTRIES

DAVID EPSTEIN*, ANNE MASON and ANDREA MANCA

Centre for Health Economics, University of York, UK

SUMMARY

Stroke is a major cause of mortality and morbidity, but the reasons for differences in costs of care within andbetween countries are not well understood. The HealthBASKET project used a vignette methodology to comparethe mean costs and prices of hospital care across providers in nine European Union countries. Data on resource use,unit costs and prices of care for female stroke patients without co-morbidity were collected from a sample of 50hospitals. Mean costs for each provider were analysed using multiple regression. Sensitivity analysis explored theeffects on cost of using official exchange rates, purchasing power parity (PPP) and proportion of national incomeper capita. The mean cost of a hospital episode per patient for stroke at PPP was h3813 (standard error 227) with anadditional day in hospital typically associated with 6.9% (95% CI: 4–9%) higher costs and thrombolysis associatedwith 41% higher costs (10–73%). After adjusting for explanatory factors, about 76% of the variation in cost couldbe attributed to between-country differences, and the extent of this variation was sensitive to the method ofcurrency conversion. There was considerable variation in the care pathways within and between countries, includingdifferences in the availability of stroke units and access to rehabilitative services, but only the length of stay and useof thrombolytic therapy were significantly associated with higher cost. The vignette methodology appears feasible,but further research needs to consider access to healthcare over a longer follow up and to include both costs andoutcomes. Copyright # 2008 John Wiley & Sons, Ltd.

KEY WORDS: cerebrovascular accident; costs and cost analysis; inpatients; regression analysis

INTRODUCTION

Stroke is a major cause of morbidity and mortality for both men and women in European Union (EU)countries (European Commission, 2002), and medical and social care consume considerable healthcareresources (Leal et al., 2006; OECD, 2007). Careful empirical work has shown wide variation in costsand outcomes across EU countries (Grieve et al., 2001a,b; Moon et al., 2003). The objectives of thisstudy were to (i) compare the costs of hospital inpatient services for stroke using a detailed vignettecosting methodology, (ii) assess whether prices are a good estimate of costs of individual services, and(iii) explore the reasons underlying variations between and within nine EU countries in the use ofresources and costs of hospital inpatient services for stroke. Comparisons of resource use betweencentres and between countries can be powerful tools for improving performance (Hakkinen andJoumard, 2007). The evidence in this study may help national decision makers to understand whyinternational cost differences exist and to distinguish between reasons for cost differences that arerelatively exogenous, such as input prices, and those that are potentially under the national decision-makers control. The nine countries participating in the HealthBASKET project were Denmark,England, France, Germany, Hungary, Italy, The Netherlands, Poland and Spain.

*Correspondence to: Centre for Health Economics, University of York, YO10 5DD, UK. E-mail: [email protected]

Copyright # 2008 John Wiley & Sons, Ltd.

BACKGROUND

A stroke is an acute neurological injury in which the blood supply to a part of the brain is abruptlyinterrupted (Klijn and Hankey, 2003). The region of the brain with disturbed perfusion nolonger receives adequate oxygen, which rapidly causes brain cells death or serious damage,impairing local brain function. Stroke is a medical emergency and prompt diagnosis and treatmentcan reduce the chances of permanent neurological damage or death. The causes of strokescan be classified into two major categories: ischaemic and haemorrhagic. In an ischaemic stroke,which is the cause of approximately 80–85% of strokes, a blood vessel becomes occluded and theblood supply to part of the brain is totally or partially blocked. Haemorrhagic stroke occurs when ablood vessel in the brain ruptures or bleeds. Haemorrhagic stroke must be ruled out with medicalimaging and requires a neurosurgical evaluation and sometimes intervention. Ischaemic stroke may betreated with anti-thrombotic medication (such as aspirin), anti-coagulants (such as warfarin) orthrombolysis.

The majority of patients will initially be managed in hospital, and inpatient care accounts for a highproportion of the total costs of stoke (Leal et al., 2006). Rehabilitation starts from the onset of strokeand may need to continue for a very long time following hospital discharge. In some organisations,patients can be discharged to a separate hospital or community rehabilitation unit at an early stage ofrehabilitation in order to avoid hospital-acquired infections and use healthcare resources moreefficiently (Hoffman et al., 2005; Cochrane Review, 2004).

A number of principles have been identified that could be considered ‘best practice’, given the currentevidence base, and expert opinion and consensus amongst international clinical guidelines is high (Klijnand Hankey, 2003). Stroke services should be organised so that patients are admitted under the care of astroke unit, consisting of a specialist, multidisciplinary team for their acute care and rehabilitation(Hoffman et al., 2005). Accurate early diagnosis is essential, specifically brain imaging should beundertaken within 24 h of onset. Thrombolytic treatment with alteplase should be administered byspecialists within 3 h of onset of symptoms and only when haemorrhage has been categorically excluded,preferably by computed tomography (CT).

Despite the international consensus on optimum care, the ‘wide variations in the pattern of care forstroke’ within Europe (Beech et al., 1996), identified over a decade ago, persist. There are significantdifferences in most aspects of clinical management including the availability of stroke units, specialistservices such as speech therapy, and access to medication (Grieve et al., 2001a; Bhalla et al., 2004; Grayet al., 2006). Grieve et al. (2001b) undertook a comparative study of the costs of strokeacross several European countries using a bottom-up costing methodology, selecting one ortwo hospitals from each country. Nearly 1900 patients (between 40 and 300 per hospital) wereprospectively recruited to the study and followed up for hospital and community costs and outcomes forthree months after stroke. Regression analysis was used to adjust for casemix, and costs converted to1998 price-year US $ using purchasing power parity (PPP). For a reference group of males over 74years, continent and conscious on admission, the study found considerable variation in mean costs – US$466 (Latvia) to US $8512 (Denmark). This range reduced to US $2132–$6579 if the unit costs of onecentre were applied to all centres, but the relative ranking of centres by cost was almost unchanged,indicating that both the unit cost and volume of resources explain differences in overall cost. Afterpatient mix, the study identified the availability of stroke units, length of stay, the use of separaterehabilitation facilities and the use of CT as the main factors contributing to differences in cost betweenproviders (Grieve et al., 2001b).

Although there are important differences in quality and outcomes of hospital stroke care (Moonet al., 2003) that may be correlated with the costs or organisation of health services (National AuditOffice, 2005), the assessment of health outcomes was not an objective of the HealthBASKET projectand was not included in this study.

D. EPSTEIN ET AL.S22

Copyright # 2008 John Wiley & Sons, Ltd. Health Econ. 17: S21–S31 (2008)

DOI: 10.1002/hec

METHODS

To ensure that patients and the treatment setting were comparable, patients were selected for inclusionin the study if their characteristics corresponded with a standard vignette. The vignette considers anotherwise healthy female, without comorbidity, 60–70 years old, with sudden severe right-sidehemiparesis, dependency and severe aphasia. Outcomes such as length of stay can differ between malesand females (Moon et al., 2003), and females were chosen to simplify the care pathway. The definitionof absence of comorbidity was at the discretion of the centre providing the data. Admission is tohospital (or accident and emergency, medical or neurological department depending on country orhospital) by ambulance vehicle. The vignette begins at the hospital door. All the interventions includingdiagnostic and treatment are delivered in the same hospital. The patient is diagnosed and treatedaccording to normal hospital standards (which may or may not include a stroke unit, earlyrehabilitation, etc.), and progress is average for the patient’s age. Transient (TIA), short and reversible(RIND) and prolonged and reversible (PRIND) ischaemic neurological deficits are excluded. Thevignette ends with the patient being discharged to a rehabilitative institution or her home.

The project protocol required researchers to identify a representative sample of providers and toexclude very small, very large and tertiary hospitals with cost structures that would be expected to differfrom those normally providing the service in that country. Resource use data on a sample of at least 10patients from each provider meeting the vignette description were collected by interviews or postalquestionnaires from medical staff and hospital administrators. Where actual patient data wereunavailable, expert opinion was substituted. Personnel costs included time spent directly with therespective case per day and time spent on patient-related administrative tasks. National wage rates wereused where available. The costing methodology used in this study required direct costs (diagnostic tests,medical and nursing and therapeutic staff, and drug costs) to be calculated. The costs of overheadfacilities (laundry, catering, power, etc.) were obtained from hospitals’ accounting departments(Mogyorosy and Smith, 2005).

This study focuses on the index hospital admission and does not include costs arising beforeadmission or after discharge from hospital. Because individual patient data were not available for allhospitals, data were aggregated on hospital level. The price year was 2005.

Multiple regression was carried out to investigate the differences in costs between countries and,where possible, identify hospital-level factors associated with costs. The literature review identified thekey factors associated with increased costs that were to be included in the regression as independentvariables: the availability of a stroke unit, the use of CT and magnetic resonance imaging (MRI) and themean length of stay for each hospital of the patients sampled in the vignette (Moon et al., 2003). Giventhe high cost of thrombolytic therapy (British Medical Association, 2006), the proportion of patients inthe hospital for whom thrombolysis was administered was also included. The number of beds in thehospital was included to test if economies of scale and scope influenced total costs. Country-levelfactors, other than gross domestic product (GDP) per head, were not included in the analysis. This isbecause at country-level there are many factors both within and outside the healthcare system that couldaffect costs, and it would be difficult to isolate an unconfounded effect of any single variable. Dischargeto separate rehabilitation facilities was not included in the regression model because data were notavailable from hospitals in England and Germany.

To test different assumptions about the relationship of the explanatory factors to total cost, threeregression models were constructed. The first was a model without covariates where the dependentvariable was log transformed to address its right-skewed distribution (model 1). The second modelextended the previous by including the independent variables described above (model 2). The thirdmodel was as model 2 but with log-transformation of the explanatory variables length of stay and thenumber of beds (model 3). In each case, clustering was taken into account using a fixed effectspecification (Wooldridge, 2006; Snijders and Bosker, 1999) whereby the error term in the regression is

THE HOSPITAL COSTS OF CARE FOR STROKE IN NINE EUROPEAN COUNTRIES S23


DOI: 10.1002/hec

decomposed to disentangle the extent to which the total variance is attributable to within- versusbetween-country variability. The fixed effects are preferred in this study to the random effects estimatorsince the former makes no assumptions about the correlation between the explanatory variables and thecountry-specific effects (Wooldridge, 2006). The correlation was tested using a Hausman test(Hausman, 1978). If the data are homoskedastic (Manning, 1998), then the slope coefficients of thecontinuous variables of model 2 represent the proportionate change in cost for a one unit change in theexplanatory variable and the slope coefficients of model 3 represent elasticities, for example, theproportionate change in cost for a 1% change in the length of stay. The assumption that the data werelog-normal was tested using the Skewness–Kurtosis test (D’Agostino, 1990), the Shapiro–Wilk test(Shapiro and Wilk, 1965) and by examination of P–P and Q–Q plots for the predicted within-countryresiduals. Heteroskedasticity was tested by a Breusch–Pagan–Godfrey test (Breusch and Pagan, 1979),regressing the squared within-country predicted residuals against the explanatory variables.

To investigate how different methods of converting local currencies into a common unit influencedthe estimate of between-country variability in cost, for each of the three regression models, threemethods were used to convert currencies: costs were converted into Euros at the official exchange rate,in Euros at PPP, and calculated as a proportion of GDP per head (Table I). PPP is used to adjust fordifferences in price levels between countries (Wordworth and Ludbrook, 2005), while expressing costs asa proportion of national income per capita provides a measure of resource use relative to nationalincome, rather than absolute expenditure (Moon et al., 2003). Therefore, in total, nine analyses werecalculated.

RESULTS

Nine countries provided data from a total of 50 hospitals. The number of patients sampled per providerranged from 3 to 109. Providers in France, Poland, Germany and Denmark had access to individualpatient data for most resource use items (in Denmark, access was to national databases); providers inother countries relied on hospital accounting departments and expert opinion. Table I shows thenumber of providers sampled, the size of the providers and the mean and the range of the number ofpatients sampled per provider. Denmark, England, Hungary, The Netherlands and Poland had acutestroke unit in all the hospitals sampled (Table I). In other countries, some providers had no acute strokeunit and admitted patients to neurological or other wards.

All providers reported that patients would undergo brain imaging with a CT or MRI scan aspart of their diagnosis. There was considerable variation in the proportion of patients meeting thevignette criteria treated with thrombolysis (Table I). No providers reported use of thrombolysis inDenmark, England, Hungary and Spain. In France the proportion ranged between 0 and 13%.In The Netherlands, Germany and Italy, some providers reported that 80–100% of patients sampledusing the vignette criteria received this therapy. All providers indicated that aspirin would begiven where appropriate.

Table I shows the mean lengths of stay reported by each provider in each country for the vignette andthe length of stay in the stroke unit (if applicable). Although Denmark contributed only one hospital tothe sample, the length of stay of 4.6 days is very similar to an OECD estimate (Moon et al., 2003),and in Denmark patients tended to be transferred early to other rehabilitation hospitals. In othercountries, the mean length of stay was between 9.3 and 15.9 days. There was also considerablevariation in the proportion of the hospital stay spent in the stroke unit. In Denmark and Hungary, thewhole hospital stay was within the stroke unit for the hospitals sampled. In The Netherlands, patientsspent 2–5 days in a stroke unit before being transferred to a general or neurological ward,whereas in Poland the majority of the hospital stay was in the stroke unit. In Italy, France and Spain,



DOI: 10.1002/hec

TableI.

Characteristics

ofhospitalcare

forstrokeprovided

ineach

country

Denmark

England

France

Germany

Hungary

Italy

TheNetherlands

Poland

Spain

Providersin

sample

N1

54

15

25

76

5Patients

sampledper

provider

Mean(range)

109

10

10.8

(6–16)

26.1

(3–109)

10

10

10

9.3

(7–10)

10

Bedsper

provider

Mean(range)

1086

737(400–1058)500(356–648)421(145–896)608(480–737)794(503–1218)571(300–1000)415(201–720)402(150–872)

Bedsper

department

Mean(range)

45

N/A

24.0

(17–30)

74.4

(36–144)

35.0

(30–40)

22.0

(18–25)

19.4

(4–45)

40.0

(40–40)

N/A

Physiciansper

department

Mean(range)

10

N/A

4.2

(2.8–7.0)12.3

(6.0–19.4)3.0

(1.0–5.0)11.0

(8.0–16.0)

5.9

(4.0–9.0)

12

N/A

Strokeunits

N(%

)1(100%)

5(100%)

1(25%)

6(40%)

2(100%)

3(60%

)7(100%)

6(100%)

1(20%)

Thrombolysisgiven

Rangeof%

0%

0%

0–13%

0–90%

0%

0–100%

a0–80%

0–100%

a0%

Length

ofstayin

strokeunit

Mean(SD)

4.6

(–)

N/A

0.5

(1)

N/A

9.3

(0.4)

2.8

(3.9)

2.8

(1.1)

8(4)

0.4

(0.9)

Totallength

ofstay

inhospitalb

Mean(SD)

4.6

(–)

10.8

(2.8)

10.4

(3)

13.1

(3.3)

9.3

(0.5)

10.5

(4.1)

15.9

(7.4)

14.7

(2.6)

10.2

(2.9)

Earlyrehabilitation

Rangeof%

100%

N/A

23–60%

N/A

5–20%

30–100%

0–100%

0%

0%

Currency

cKrone

Sterling

hh

Forint

hh

Zloty

hOfficialexchangerate

7.452

0.684

11

247.097

11

41

Purchasingpower

parity

d10.015

0.738

1.074

1.053

148.89

1.027

1.052

2.202

0.908

GDPper

head

e271674

19463

27382

26859

2019737

24132

30012

23144

19613

aOneofthePolish

hospitalsusuallyadministers

thrombolysistherapyto

allpatients

meetingthevignette

description.

bThetotallength

ofstayrefers

totheindex

hospitaladmission.

cMeanofficialexchangerate

per

hduring2005.

dPPPweights

adjusted

totheEU-25average(O

ECD,2005).

eGDPper

capitain

nationalcurrency

(OECD,2005).



DOI: 10.1002/hec

the care pathway varied between providers. Providers in England, Germany and France could notalways accurately determine the proportion of time spent in stroke units.

The mean costs of stroke care in each country are shown in Table II and the distribution in Figure 1,in PPPs. The mean cost across all hospitals in all countries was h3813 (standard error 227). Theproportion of overhead in total costs varied between countries, which may in part be due to thedifferences in the accounting methodology used. The ordering of countries by cost is similar using eitherPPPs, official exchange rates or as a proportion of GDP per head, apart from Poland which appears lowcost in absolute terms but high cost in relative terms when costs are expressed as a proportion ofnational income per head.

The relationship between total cost of hospital care for this vignette and the third partyreimbursement received by the provider is shown in Figure 2 (Pearson’s correlation coefficientr ¼ 0:619). All of the cases in this study were treated in public hospitals with no private co-payments.For most countries, reimbursement was made according to a diagnosis-related groups (DRG) tariff orsimilar, although prices did vary for this vignette within Italy, Poland and Germany. Not all thehospitals in Spain were reimbursed according to activity; therefore, these providers are not included inFigure 2. In all other countries, reimbursement was made close to the median cost across providers,indicating that in most countries payment systems there exists a DRG or equivalent that correspondsrelatively closely with the patient characteristics described by this vignette. The magnitude of thepotential ‘profits’ or ‘losses’ made by providers in each country where costs differ from reimbursementappears similar in percentage terms across countries.

Table III shows the results of the regression of the log-transformed costs against hospital-levelexplanatory factors. The null hypothesis that the residuals were normally distributed was not rejected bythe Skewness–Kurtosis test or the Shapiro–Wilks test, and the Breusch–Pagan test did not rejecthomoskedascity at the 5% level in any model. Model 2 calculated that on average an additional day isassociated with a 6.9% increase in costs (95% CI 4.1–9.1%) and thrombolysis with 41% higher costs(10–73%). Model 3, the log–log model calculating elasticities, estimated that a 1% increase in length ofstay was associated with a 1% increase in costs (95% CI 0.68 to 1.33%). Other variables were notstatistically significant at the 5% level. Variation in costs between countries is greatest in the modelsusing official exchange rates, which is unsurprising since this includes differences in cost due to bothinput prices and use of resources. In a log-linear model of costs at official exchange rates with a constantonly, 79% of the total variance (the sum of the within and the between-country variances) was due todifferences in costs between countries. After adjusting for explanatory variables, 88% of the variation incost could be attributed to between-country differences, because the explanatory variables at providerlevel have reduced the within-country variance. There is relatively less variation in cost betweencountries in the models using PPP since this has adjusted for at least some of the differences in inputprices between countries. There is least variation in cost between countries in the models using costs as aproportion of GDP per head. Variation in costs between countries are nevertheless significant at the 5%(but not 1%) level in all models.

DISCUSSION AND CONCLUSIONS

This article examined differences in the resources used and costs of stroke in nine EU countries. Allhospitals gave patients prompt brain imaging. Not all hospitals had acute stroke units in place. Wheresuch facilities were available, in some countries patients were often transferred to a non-stroke ward atsome point during their hospital stay. Length of stay at the acute hospital is influenced by policies forrehabilitation. In the Danish hospital, patients were on average transferred to a rehabilitation hospitalwithin 5 days, while other countries to a varying extent initiated rehabilitation therapies in the admittinghospital (Schreyogg et al., 2008; Grieve et al., 2005a). The availability of appropriate long-term



DOI: 10.1002/hec

TableII.Meancostsofstrokein

each

country,accordingto

category

ofresourceuse

(inPPP)

Denmark

England

France

Germany

Hungary

Italy

TheNetherlands

Poland

Spain

Diagnostic

procedures

443(18%)

366(6%

)467(12%)

471(14%)

69(7%

)550(12%

)244(4%

)197(9%

)449(21%

)Strokeunita

61(2%

)1414(25%)

168(4%

)170(5%

)364(35%)

0(0%

)479(7%

)556(25%

)117(6%

)Ward

a953(38%)

775(14%)

1471(36%)

1530(47%)

0(0%

)657(15%

)2982(46%)

562(25%

)1098(52%

)Drugs

0(0%

)37(1%

)143(4%

)139(4%

)161(15%)

375(8%

)N/A

219(10%

)33(2%

)Overheads

1045(42%)

3082(54%)

1789(44%)

974(30%)

448(43%)

2883(65%

)2829(43%)

715(32%

)430(20%

)Totalmeancost

hatPPP

2501(100%)

5674(100%)

4038(100%)

3283(100%

)1043(100%

)4465(100%

)6533(100%)

2249(100%

)2128(100%

)Order

(highcost

tolow)

62

45

93

17

8hatofficialexchangerates

3362

6122

4337

3457

628

4586

6873

1238

1932

Order

(highcost

tolow)

62

45

93

18

7As%

ofGDPper

head

9.2

21.5

15.8

12.9

7.7

19.0

22.9

21.4

9.8

Order

(highcost

tolow)

82

56

94

13

7

Note:N/A

,Notavailable.Drugcostsin

theNetherlandswerenotseparable

from

strokeunitandward

costs.

aRefersto

directcostsofmedical,nursingandtherapeuticstafftimewithstrokepatients

onstrokeunitandwards.



DOI: 10.1002/hec

residential or nursing care is likely to be a constraint though this was not examined in this study.Initiation of thrombolysis also varied between hospitals and countries: this is to some extent determinedby the organisation and effectiveness of pre-hospital ambulance services (Hoffmann et al., 2005).

Cos

t in

Eur

osa

t pur

cha

sin

gpo

wer

parit

y

0

5000

10000

15000

DenmarkEngland

FranceGermany

HungaryItaly

NetherlandsPoland

Spain

Figure 1. Differences in total cost within and between countries for stroke (Euros at PPP). No box plot is shown forDenmark since they only provided data for one hospital

0

4000

8000

12000

16000

0 160001200080004000

Cost (Euro)

Netherlands

Italy

Denmark

Germany

England

Poland

France

Hungary

Reimbursement=Cost

Rei

mbu

rsem

ent (

Eur

o)

Figure 2. Price (reimbursement) versus cost of care for stroke for each hospital in Euros at market exchange rates.The 458 line in Figure 2 represents points where mean cost incurred by hospital for patients in the vignette equals

the mean reimbursement received by the hospital from the third-party payer



DOI: 10.1002/hec

Length of stay and the administration of thromolytic drugs were found to be significant factors inexplaining variability in costs between providers. The extent to which costs differ between countries issensitive to the method of currency conversion chosen. There are significant differences in mean costsbetween countries when costs are calculated at official exchange rates. PPP adjustment reduces some,but not all, of this between-country variation. These analyses, using different measures of relative prices,showed that differences in cost appear to be due to both the use of resources and the prices of thoseresources. Adjusting costs for the GDP per head in the country further reduces the variation betweencountries. Poland, exceptionally, appears low cost by the standards of official exchange rates, butstroke care costs are a relatively high proportion of national income per head. This finding is incontrast with the low proportion of its national income that Poland spends on health relative to otherOECD countries (6.2% in Poland in 2005 compared with an OECD average of 9%) (Rodgers et al.,2003).

This study has made comparisons between countries by adjusting hospital costs based on officialexchange rates, PPPs and GDP per head. Further work might take into account that relative prices forspecific inputs used in stroke care might differ from those used to calculate PPPs. One option to addressthis might be to convert the prices of relatively tradable items such as drugs at the official exchange rateand convert the costs of non-tradeables using PPPs. Alternatively, the vignette methodology could itselfbe used to calculate episode-specific PPPs for hospital stroke care (Wordworth and Ludbrook, 2005;Schreyogg et al., 2008; Grieve et al., 2005b).

Table III. Results of the regressions of costs (expressed in Euros at official exchange rates, in Euros at PPP and as aproportion of GDP per head) against hospital-level variables

Model 1a: Mean (95% CI) Model 2b: Mean (95% CI) Model 3c: Mean (95% CI)

VariablesStroke unitd } 0.024 (�0.237 to 0.280) 0.059 (�0.199 to 0.316)Thrombolysis givene } 0.414 (0.099 to 0.727)nn 0.415 (0.11 to 0.739)nn

Length of stay (days)f } 0.069 (0.045 to 0.092)nn 1.009 (0.687 to 1.33)nn

Number of bedsg } 0.008 (�0.035 to 0.051) �0.009 (�0.222 to 0.203)

Between and within countryerror official exchange rateBetween-country variance 0.602 0.605 0.733Within-country variance 0.158 0.083 0.078r (p-value) 79.2% ðp50:0001Þ 88.0% ðp50:0001Þ 90.4% ðp50:0001Þ

Purchasing Power ParityBetween-country variance 0.294 0.255 0.331Within-country variance 0.158 0.083 0.078r (p-value) 65.0% ðp50:0001Þ 75.5% ðp50:0001Þ 81.0% ðp50:0001Þ

GDP per headBetween-country variance 0.125 0.065 0.099Within-country variance 0.158 0.083 0.078r (p-value) 44.2% ðp ¼ 0:02Þ 44.0% ðp ¼ 0:005Þ 56.2% ðp ¼ 0:002Þ

Note: r; Fraction of total variance due to between-country variance. The p-value for the significance of r was calculated usingANOVA (F-test).nnStatistically significant at the 5% level.aThis is a typical variance components model where log(cost) are regressed against a constant only.bThis model regressed log(cost) against the explanatory variables.cThis model (also known as ‘elasticity’ model) regressed log(cost) against log(length of stay), log(number of bed), thrombolysisand stroke unit.

dAvailability of acute stroke unit at the hospital (dummy variable).eProportion of patients in the vignette for whom thrombolysis was given at the hospital.fMean length of stay for patients in the vignette at the hospital.gThe number of acute beds at the hospital.



DOI: 10.1002/hec

The vignette methodology used in this study appears feasible to collect resource use and costs formost hospital providers of stroke services. It avoids the need for patients to be individually recruited toa prospective study and adjustment for casemix does not have to be undertaken using regression. Thereare some structural and administrative limitations with the methodology. Firstly, the vignettemethodology was successful at identifying the use of direct resources but the methods used to allocateoverheads varied between and within countries. Secondly, the vignette controlled for casemix byobtaining data on a relatively small number of typical or median patients from each provider, andtherefore does not capture the distribution in costs at the individual level. Thirdly, the vignette onlycaptures a part of the care pathway: outcomes after stroke depend on the coordination of emergency,hospital and rehabilitation services. Lastly, costs may be correlated with health outcomes, which werenot collected in this study (Grieve et al., 2001b). The detailed information obtained from this vignettemethodology complements, rather than substitutes for, data from prospective cohort studies oraggregate national sources. Further work might attempt to capture the resources used and costsincluding community services and rehabilitation providers and to integrate measures of quality andoutcomes with costs.

ACKNOWLEDGEMENTS

The results presented in this article are based on the project ‘Health Benefits and Service Costs inEurope - HealthBASKET’, which was funded by the European Commission within the SixthFramework Research Programme (grant no. SP21-CT-2004-501588). We would like to thank DrFederico Augustovski (Hospital Italiano, Argentina) and Dr Elliot Epstein (Walsall Manor Hospital,United Kingdom) for their valuable comments at the drafting stage.

CONFLICT OF INTEREST

No conflicts of interest declared.

REFERENCES

Bhalla A, Grieve R, Tilling K et al. 2004. Older stroke patients in Europe: stroke care and determinants of outcome.Age and Ageing 33: 618–624.

Beech R, Ratcliffe M, Tilling K et al. 1996. Hospital services for stroke care. A European perspective. Europeanstudy of stroke care. Stroke 27: 1958–1964.

Breusch T, Pagan A. 1979. A simple test for heteroscedasticity and random coefficient variation. Econometrica 47:1287–1294.

British Medical Association and the Royal Pharmaceutical Society of Great Britain. 2006. British NationalFormulary. Royal Pharmaceutical Society of Great Britain: London.

Cochrane Review. 2004. Early Supported Discharge Trialists. Services for Reducing Duration of Hospital Care forAcute Stroke Patients. Wiley: Chichester.

D’Agostino R, Belanger A, D’Agostino Jr RB. 1990. A suggestion for using powerful and informative tests ofnormality. American Statistician 44: 316–321.

European Commission Health and Consumer Protection Directorate-General. 2002. Age Standardised Death Ratefor 65 Causes of Death; Females. ECHI, 2002. http://www.ec.europa.eu [16 February 2007].

Gray LJ, Sprigg N, Bath PMW et al. 2006. Significant variation in mortality and functional outcome after acuteischaemic stroke between Western countries: data from the tinzaparin in acute ischaemic stroke trial (TAIST).Journal of Neurology; Neurosurgery and Psychiatry 77: 327–333.

Grieve R, Dundas R, Beech R et al. 2001a. The development and use of a method to compare the costs of acutestroke across Europe. Age and Ageing 30: 67–72.



DOI: 10.1002/hec

Grieve R, Hutton J, Bhalla A et al. 2001b. A comparison of the costs and survival of hospital: admitted strokepatients across Europe. Stroke 32: 1684–1691.

Grieve R, Nixon R, Thompson SG et al. 2005. Using multilevel models for assessing the variability of multinationalresource use and cost data. Health Economics 14: 185–196.

Hausman JA. 1978. Specification tests in econometrics. Econometrica 46(6): 1251–1271.Hakkinen U, Joumard I. 2007. Cross-country analysis of efficiency in OECD health care sectors: options for

research. OECD Economics Department Working Papers No. 554, OECD, Paris.Hoffman A, Lowe D, Rudd A et al. 2005. National Sentinel Stroke Audit 2004. Royal College of Physicians:

London.Klijn CJM, Hankey GJ. 2003. American stroke association and European stroke I. Management of acute ischaemic

stroke: new guidelines from the American Stroke Association and European Stroke Initiative. Lancet Neurology2: 698–701.

Leal J, Luengo-Fernandez R, Gray A et al. 2006. Economic burden of cardiovascular diseases in the enlargedEuropean Union. European Heart Journal 27: 1610–1619.

Manning W. 1998. The logged dependent variable, heteroscedasticity, and the retransformation problem. Journal ofHealth Economics 17: 283–295.

Mogyorosy Z, Smith PC. 2005. The Main Methodological Issues in Costing Health Care Services: CHE ResearchPaper 7. Centre for Health Economics, University of York: York, 2005.

Moon L, Moise P, Jacobzone S. 2003. ARD-Stroke Experts Group. Stroke Care in OECD Countries: A Comparisonof Treatment, Costs and Outcomes in 17 Countries. OECD: Paris.

National Audit Office. 2005. Reducing Brain Damage: Faster Access to Better Stroke Care. Department of Health:London.

OECD. 2005. Eurostat. Purchasing Power Parities and Real Expenditures. OECD: Paris.OECD. 2007. OECD Health Data 2007: Statistics and Indicators for 30 Countries. OECD: Paris.Rodgers H, Dennis M, Cohen D et al. 2003. British association of stroke physicians: benchmarking survey of stroke

services. Age and Ageing 32: 211–217.Shapiro S, Wilk MB. 1965. An analysis of variance test for normality (complete samples). Biometrika 52: 591–611.Snijders TAB, Bosker RJ. 1999. Multilevel Analysis. An Introduction to Basic and Advanced Multilevel Modeling.

Sage: London.Schreyogg J, Tiemann O, Stargardt T, Busse R. 2008. Cross-country comparisons of costs: the use of episode-

specific transitive purchasing power parities with standardised cost categories. Health Economics 17(S1):S95–S103.

Wooldridge J. 2006. Introductory Econometrics: A Modern Approach. South Western: Belmont, CA.Wordworth S, Ludbrook A. 2005. Comparing costing results in across country economic evaluation: the use of

technology specific purchasing power parities. Health Economics 14: 93–99.



DOI: 10.1002/hec

the hospital costs of care for stroke in nine european countries

Documents