OUP user menu

Predictors of costs related to cardiovascular disease among patients with atrial fibrillation in five European countries

Emy Holstenson, Anna Ringborg, Peter Lindgren, Florence Coste, Francoise Diamand, Robby Nieuwlaat, Harry Crijns
DOI: http://dx.doi.org/10.1093/europace/euq325 23-30 First published online: 7 September 2010

Abstract

Aims To estimate predictors of direct costs and costs of hospitalization related to cardiovascular disease (CVD) in patients with atrial fibrillation (AF) recruited to the Euro Heart Survey on AF (EHS-AF) in Greece, Italy, Poland, Spain, and the Netherlands.

Methods and results Annual direct costs were modelled by country using ordinary least squares (OLS) regression. For costs of hospitalization related to CVD, logistic regressions followed by conditional OLS regression were employed. In each case, effects of the following potential explanatory variables were tested: age, sex, body mass index, type of AF, diabetes, hypertension, myocardial infarction (MI), angina pectoris (AP), valvular heart disease (VHD), congestive heart failure (CHF), stroke, and/or other underlying heart disease at the time of enrolment in the EHS-AF. Estimated direct annual costs for the reference EHS-AF patient (female aged <65 years with first-detected AF and no co-morbidities at baseline) were €933 in Greece, €1383 in Italy, €698 in Poland, €1316 in Spain, and €1544 in the Netherlands. The co-morbidities identified as predictors of direct costs were VHD in Greece, Italy, and Spain, AP in Italy and Spain, diabetes and stroke in Poland, CHF in Italy, MI in Spain and other underlying heart disease in Poland and the Netherlands. For costs of CVD-related hospitalization, the most important co-morbidity identified as a predictor was VHD.

Conclusion The results reported in this study increase the understanding of the cost structure of CVD in AF patients and may therefore inform the targeting of intervention therapy in selected groups of at-risk patients.

  • Arrhythmia
  • Economic
  • Burden
  • Europe

Introduction

Atrial fibrillation (AF) is a common cardiac arrhythmia associated with substantial morbidity and high societal costs.14 The prevalence of AF has been estimated at 7.8 and 11.7% in men aged 65–74 and 75–84 years, respectively.5 Since AF is more common among the elderly its prevalence is likely to increase over the coming years due to population ageing, thus increasing the public health burden of the disease.69

Moreover it is common for AF to be accompanied by other cardiovascular diseases (CVDs) such as congestive heart failure (CHF) and stroke.1012 The prevalence of AF among stroke patients is 18–31%, with the variation mainly depending on the patients' age.13,14 Data from the Framingham study also indicate that the elderly are particularly vulnerable to stroke when AF is present.15 AF-related strokes are associated with a poorer outcome than non-AF-related strokes and stroke patients with AF have higher levels of morbidity and incur higher inpatient costs than other stroke patients.10,16 Results from a recent cohort study furthermore showed that ischaemic stroke is about as common in paroxysmal AF as in permanent AF.17,18

Medical treatment of AF includes antithrombotic therapy, use of anti-arrhythmic agents to convert AF to sinus rhythm and rate control treatments to control the ventricular rate.1924 However, the presence of multiple diagnoses can make the optimal treatment hard to determine.25

Several studies that have estimated the economic burden of AF have identified direct costs and hospitalizations as major cost drivers. In the UK and in Germany, hospital admissions have been found to represent 44–50% of direct costs and in France hospitalizations have been found to represent 60% of direct costs among patients with AF.2629

The objective of the present study was to identify predictors of annual direct costs and annual costs of hospitalizations related to CVD for AF patients enrolled in the Euro Heart Survey on AF (EHS-AF) in the following countries: Greece, Italy, Poland, Spain, and the Netherlands.

Methods

Patients and data

The EHS-AF enrolled 5333 patients in 35 European countries. Recruitment and data collection took place in 2003 at outpatient cardiology clinics, cardiology wards, first (heart) aids, cardiac surgery wards, cardioversion departments, and/or device implantation departments at university, non-university, and specialized hospitals. The inclusion criteria were age 18 or older and AF on electrocardiogram or Holter monitoring during the enrolment admission, or in the preceding 12 months. A detailed description of the EHS-AF has been published previously.30 Briefly, data covering patient characteristics, cardiovascular risk factors, co-morbidities, an assessment of quality of life using the EuroQoL-5D (EQ-5D) instrument, medical treatment and management during the enrolment admission were collected from medical records and patient interviews. A follow-up visit was conducted at 12 months during which data on medical treatment, quality of life, the occurrence of adverse events and resource use in terms of inpatient days, consultations and work loss during the year were collected.31

We have previously estimated annual costs of patients enrolled in the EHS-AF in Greece (n = 251), Italy (n = 645), Poland (n = 203), Spain (n = 720), and the Netherlands (n = 686) by multiplying costs from published national sources with the quantities of resource use recorded at 1-year follow-up.32 The selected countries were those in which more than 200 patients were enrolled in the EHS-AF.

Presently, we aimed to identify the determinants of direct annual costs of these EHS-AF patients defined as costs of diagnostics (transthoracic and transoesophageal echocardiography, chest X-ray, Holter monitoring, exercise tests, electrophysiology, and event recorders), interventions [coronary artery bypass graft surgery (CABG), cardiac valve replacement, catheter ablation, pacemaker implantation, implantable cardioverter-defibrillator implantation, surgical therapy, and percutaneous coronary intervention (PCI)], drug therapy (vitamin K antagonist treatment, other antithrombotic treatments, and anti-arrhythmic/rate control treatments), consultations, and inpatient days incurred during the year of follow-up.

Owing to their economic importance as drivers of costs, the probability of hospitalization due to CVD and the costs of CVD-related inpatient care and interventions were additionally subjected to separate analysis for determination of predictors. Inpatient care and interventions are both intimately linked to the costs incurred at hospitalization; inpatient care is a consequence of hospitalization by definition and interventions of the type studied in the EHS-AF (CABG, valve replacement, catheter ablation etc.) are performed in a hospital setting. Henceforth, the sum of costs of CVD-related inpatient care and interventions will be referred to as the cost of CVD-related hospitalization.

Healthcare systems in the studied countries

Owing to differences in healthcare systems, social security coverage, price levels, and treatment patterns in the selected countries, annual costs and probabilities of hospitalization were not considered directly comparable and separate analyses of predictors were therefore conducted for each country. The aim of the study was not to conduct a comparison between the five countries but rather to present nation-specific results. To give a contextual overview, the following sections provide a brief description of the healthcare systems in the studied countries.

Greece

The Greek healthcare system features a mixture of public and private services and funding. It has been found to be relatively efficient in international comparisons of health outcomes, but dissatisfaction among the population is more pronounced than in other countries. This is related to the high proportion of private health spending on health. In spite of the healthcare system having been designed to guarantee free access to public medical services for the entire population, over 38% of health spending was private in 2006.33

Italy

The healthcare system in Italy is predominantly publicly financed. The financing and delivery of care are divided, with the central government allocating regional budgets and the regions deciding how to organize and deliver care. This division between financing and delivery creates some tension, since the regions claim the government under-budgets and the government claims the regions need greater cost control. Hospitals dominate the Italian health system, accounting in 2006 for 45.2% of overall health spending—the third-largest proportion reported among OECD countries.34

Poland

Poland has a mixed system for public and private healthcare financing. Mandatory social health insurance contributions represent the major public source of healthcare financing. Since 2003, the centralized National Health Fund (NHF) administers the social health insurance scheme. The NHF with its regional branches maintains the responsibility for planning and purchasing public financed health services.35

Spain

The Spanish healthcare system is similar to the Italian one in that governance of the system is decentralized, with the 17 regions comprising the Spanish state responsible for the delivery of health services. The Spanish health system has a strong public provision sector that dominates the healthcare market and a smaller fairly dependent private sector. Like many other countries in Europe, the Spanish healthcare system confronts continued pressures to provide high-quality universal care in the face of ever-increasing costs and competing uses for financial resources.36

The Netherlands

The Dutch healthcare system differs markedly from the systems in the other countries since all residents in the Netherlands as of 2006 are required to purchase their own health insurance coverage. Insurers are private and are governed by private law but are legally required to provide a standard benefit package to insures. In addition, all Dutch citizens are covered by the statutory Exceptional Medical Expenses Act (AWBZ) scheme for a wide range of chronic and mental healthcare services such as home care and care in nursing homes. Physicians practise directly or indirectly under contracts negotiated with private health insurers.34

Statistical analysis

Annual direct costs were modelled using five separate ordinary least squares (OLS) regressions, one for each country. In each case, the relationship between direct annual costs and the following potential explanatory variables was tested in univariate analyses: age (<65 years/65–74 years/≥75 years), sex, body mass index (BMI) (≥30 kg/m2/<30 kg/m2), type of AF (first-detected/paroxysmal/persistent/permanent), diabetes mellitus, hypertension, myocardial infarction (MI), angina pectoris (AP), valvular heart disease (VHD), CHF, history of stroke, and other underlying heart disease.

All co-morbidities were indicator variables referring to the patients' status at the time of the EHS-AF enrolment admission, with MI including both acute MI at baseline and old MI, history of stroke including both haemorrhagic and ischaemic stroke and underlying heart disease including history of CABG, PCI, sick sinus syndrome, congenital heart disease, cardiomyopathy, ventricular fibrillation, and/or sustained tachycardia.

Variables significant at the 20% level in univariate analysis were selected to enter the regression model. A backward method (removal of the variables step-by-step starting from the least significant) was applied, with the final model restricted to variables significant at the 10% level.

The probability of hospitalization related to CVD was modelled by country using logistic regression with the dependent variable set to 1 for all patients who reported a CVD hospital admission during the year of follow-up and 0 for all other patients. To determine the impact of the different baseline characteristics on the probability of being hospitalized for CVD, the explanatory variables listed above were employed and univariate analysis (P < 0.2) followed by backward selection was performed (P < 0.1).

For the annual costs of CVD-related hospitalization, conditional OLS regression models were used (one for each country) to identify predictors. This implies modelling the annual cost of CVD-related hospitalization conditional on the patient being hospitalized for CVD and thus incurring CVD-related hospitalization costs. Thus, only the patients who were hospitalized for CVD during the year were included in the model, the aim being to infer what factors are associated with costs of CVD-related hospitalization among patients actually incurring costs. Conditional models are useful when a considerable proportion of patients incur zero costs (i.e. are not hospitalized) since they are excluded from the analysis.37,38

Similar to above, predictors of CVD-related hospitalization conditional on costs being incurred were inferred through univariate analysis of the listed explanatory variables followed by backward selection (P < 0.1).

Results of the OLS models were reported as parameter coefficients with 95% confidence interval (CI) whereas results of the logistic regression models were reported in terms of odds ratios (OR) with 95% CI.

Results

Patient characteristics

A total of 2505 patients were enrolled in the EHS-AF and had 1-year follow-up data in Greece (n = 251), Italy (n = 645), Poland (n = 203), Spain (n = 720), and the Netherlands (n = 686). Table 1 shows an overview of the patient characteristics at baseline, the clinical events that occurred during the year of follow-up and the estimated annual direct costs by resource category.

View this table:
Table 1

Patient characteristics, clinical events and costs during 1-year follow-up

Greece (n = 251)Italy (n = 645)Poland (n = 203)Spain (n = 720)The Netherlands (n = 686)
Baseline characteristics
 Age (years)
  <6592 (37)192 (30)97 (48)288 (40)209 (30)
  65–74109 (43)227 (35)76 (37)216 (30)197 (29)
  ≥7550 (20)226 (35)30 (15)214 (30)279 (41)
 Male patients151 (60)391 (61)113 (56)406 (56)403 (59)
 BMI ≥ 30 kg/m269 (27)132 (20)50 (25)191 (27)349 (51)
 AF-type
  First-detected59 (24)101 (16)27 (13)144 (20)89 (13)
  Paroxysmal82 (33)142 (22)88 (43)119 (17)250 (36)
  Persistent46 (18)245 (38)40 (20)140 (19)111 (16)
  Permanent53 (21)145 (22)48 (24)314 (44)202 (29)
 Co-morbidities/risk factors at enrolment admission
  Diabetes53 (21)102 (16)25 (12)155 (22)107 (16)
  Hypertension162 (65)452 (70)139 (68)404 (56)360 (52)
  Myocardial infarction (acute or history)39 (16)105 (16)35 (17)104 (14)140 (20)
  Angina pectoris40 (16)89 (14)66 (33)97 (13)144 (21)
  Valvular heart disease65 (26)173 (27)60 (30)194 (27)177 (26)
  Congestive heart failure39 (16)132 (20)69 (34)227 (32)147 (21)
  History of stroke9 (4)32 (5)11 (5)43 (6)41 (6)
  Other underlying heart diseasea62 (25)190 (29)70 (34)178 (25)218 (32)
CVD-related hospitalizationsb
 Patients with ≥1 hospitalization due to CVD81 (32)273 (42)74 (36)194 (27)169 (25)
 Mean (SD) costs of CVD-hospitalizations (inpatient care and interventions related to CVD) among pts with ≥1 hospitalization952 (3770)3947 (5658)1518 (1599)3322 (3241)3099 (5396)
Annual direct costs incurred during the year of follow-up, mean (SD)
 Diagnostics46 (124)150 (227)29 (156)107 (142)164 (392)
 Interventions797 (3157)869 (2493)176 (466)723 (2601)816 (3563)
 Drug therapy162 (302)204 (318)98 (332)242 (366)89 (153)
 Consultations37 (47)45 (90)24 (27)38 (68)58 (57)
 Inpatient care359 (2296)1816 (4162)665 (1298)1008 (2282)852 (3101)
 Total direct costs1401 (4937)3083 (5800)992 (1669)2118 (4263)1977 (5786)
  • Data are presented as observed number (%) within country unless otherwise indicated, all costs are expressed in 2007€.

  • BMI, body mass index; AF, atrial fibrillation; CVD, cardiovascular disease; SD, Standard deviation.

  • aIncludes CABG, PCI, cardiomyopathy, congenital heart disease, sick sinus syndrome, sustained ventricular tachycardia, and ventricular fibrillation.

  • bInpatient admissions during the year of follow-up caused by AF, stroke/transient ischaemic attack, or other cardiac reason.

A majority of the patients included in the EHS-AF in all five countries were aged 65 years or older and were male. Aside from this, baseline characteristics were not very consistent between the countries. In the Netherlands a majority of patients had a BMI ≥ 30 kg/m2 while only 20% were overweight in the Italian cohort. The distribution of AF-types (first-detected, paroxysmal, persistent, or permanent) and co-morbidities varied considerably between the countries. For example, 44% of the Spanish cohort had permanent AF while only 17% had paroxysmal AF. In contrast, paroxysmal AF was the most prevalent AF-type in the Greek, Polish, and Dutch cohorts, accounting for 33, 43, and 36% of patients, respectively.

Ten to 20% of the patients experienced at least one clinical event during the year of follow-up. In most cases the event was classified as related to CVD. Cardiovascular disease-related events included MI, stroke (both haemorrhagic and ischaemic), transient ischaemic attack (TIA), unstable angina, peripheral embolism, syncope, pulmonary embolism, heart failure, and asystole. Events not considered related to CVD included malignancy, other major bleeding and other major adverse event.

The share of patients that underwent hospitalization related to CVD during the year of follow-up was 32% in Greece, 42% in Italy, 36% in Poland, 27% in Spain, and 25% in the Netherlands. CVD-related reasons for hospitalization were defined as AF, stroke/TIA, or other cardiac reason.

Regarding the annual direct costs, the costs of interventions and inpatient care accounted for the bulk of costs among EHS-AF patients in all five countries.

Annual direct costs

Predictors of annual direct costs (P < 0.1) identified in the regression analyses are reported in Table 2 by country.

View this table:
Table 2

Results from ordinary least squares regression models for annual direct costs, by country

VariablesCoefficients (95% CI)
GreeceItalyPolandSpainThe Netherlands
Constant933 (219; 1648)1383 (696; 2070)698 (374; 1023)1316 (925; 1707)1544 (1029; 2060)
Age (reference group <65 years)
 65–74 years−385 (−836; 67)
 ≥75 years
Male gender
BMI ≥ 30 kg/m2
AF-type (reference group first-detected AF)
 Paroxysmal1245 (128; 2362)
 Persistent
 Permanent1554 (445; 2661)
Co-morbidities/risk factors at enrolment admission
Diabetes744 (61; 1427)
Hypertension
Myocardial infarction (acute or history)963 (38; 1888)
Angina pectoris2181 (869; 3492)1569 (619; 2518)
Valvular heart disease1644 (251; 3037)1430 (398; 2461)1704 (1015; 2393)
Congestive heart failure1941 (799; 3084)
History of stroke1668 (702; 2635)
Other underlying heart diseasea767 (303; 1232)1170 (261; 2080)
  • Note: All costs expressed in 2007€; P < 0.1 for all coefficients.

  • OLS, ordinary least squares; CI, confidence interval; BMI, body mass index; AF, atrial fibrillation.

  • aRecord of CABG, PCI, cardiomyopathy, congenital heart disease, sick sinus syndrome, sustained ventricular tachycardia, and/or ventricular fibrillation.

The constant term of the model reflects the annual direct cost of the reference patient, i.e. a patient for which values of all model parameters are equal to zero. Estimated direct annual costs for such a patient (female aged <65 years with first-detected AF and no co-morbidities at baseline) were €933 in Greece, €1383 in Italy, €698 in Poland, €1316 in Spain, and €1544 in the Netherlands.

In the Greek cohort, the only parameter with an impact on annual direct costs was VHD. Annual direct costs for Greek EHS-AF patients with this co-morbidity were on average €1644 higher compared with patients without VHD.

In the Italian cohort, the co-morbidities AP, VHD, and CHF were associated with higher direct costs: €2181, €1430, and €1941, respectively. Additionally, having paroxysmal or permanent AF predicted direct annual costs that were on average €1245 and €1554 higher compared with first-detected AF.

In both Poland and the Netherlands, other underlying heart disease was found to be associated with higher annual direct costs—€767 and €1170, respectively—and in Poland, diabetes and history of stroke were also shown to be predictors of annual direct costs, the estimated excess cost associated with these complications being €744 and €1668, respectively. In Spain MI, AP, and VHD predicted annual direct costs that were on average €963, €1569, and €1704 higher, respectively.

Probability and cost of cardiovascular disease-related hospitalization

Table 3 reports the odds ratios from the logistic regression model for the annual probability of hospitalization related to CVD (i.e. of incurring costs of CVD-related hospitalization) and the regression coefficients from the OLS regression model for annual costs of CVD-related hospitalization, conditional on such costs being incurred.

View this table:
Table 3

Results from models for the annual probability of cardiovascular disease-related hospitalization and the annual costs of cardiovascular disease-related hospitalization (conditional model)

Part 1: Model for the annual probability of hospitalization related to CVD and incurring costs of CVD-related hospitalizationPart 2: Model for annual costs of CVD-related hospitalization, conditional on costs being incurred
Odds ratios (95% CI)Coefficients (95% CI)
GreeceItalyPolandSpainThe NetherlandsGreeceItalyPolandSpainThe Netherlands
Constant2042 (103; 3981)3257 (2085; 4430)1834 (1026; 2642)3232 (2183; 4281)5294 (3515; 7073)
Age (reference group <65 years)
 65–74 years2.22 (1.17; 4.19)
 ≥75 years
Male gender
BMI ≥ 30 kg/m2
AF-type (reference group first-detected AF)
 Paroxysmal4.07 (2.06; 8.08)1.94 (1.32; 2.85)5.45 (1.51; 19.66)−859 (−1779; 62)
 Persistent7.52 (1.93; 29.32)
 Permanent0.41 (0.16; 1.07)4.19 (1.09; 16.13)0.49 (0.32; 0.76)3153 (1074; 5233)−1029 (−2194; 134)
Co-morbidities/risk factors at enrolment admission
Diabetes2.25 (0.93; 5.42)1049 (−52; 2150)
Hypertension
Myocardial infarction (acute or history)1.81 (1.16; 2.83)
Angina pectoris5.05 (2.26; 11.25)2125 (−124; 4374)
Valvular heart disease4338 (409; 8267)2321 (334; 4308)4652 (2975; 6328)2710 (−487; 5907)
Congestive heart failure2.48 (1.05; 5.87)1.51 (0.99; 2.31)1.52 (1.07; 2.15)2578 (479; 4678)
History of stroke2.44 (1.13; 5.27)1581 (−32; 3194)
Other underlying heart diseasea1.59 (1.09; 2.31)1.39 (0.95; 2.04)902 (97; 1707)2865 (1059; 4672)
  • Note: All costs expressed in 2007€; P < 0.1 for all odds ratios and coefficients; Part 1 is derived through a logistic regression model, Part 2 is derived through an OLS regression model.

  • CI, confidence interval; BMI, body mass index; AF, atrial fibrillation.

  • aRecord of CABG, PCI, cardiomyopathy, congenital heart disease, sick sinus syndrome, sustained ventricular tachycardia, and/or ventricular fibrillation.

The estimated probability of hospitalization related to CVD was 12% in Greece, 32% in Italy, 10% in Poland, 23% in Spain, and 26% in the Netherlands for the reference patient (i.e. a female patient, age <65 years with first-detected AF, BMI <30 and no co-morbidities at baseline). The estimated probability of the reference patient was thus, somewhat surprisingly, slightly higher than the general probability of CVD-related hospitalization in the Netherlands, but lower in all other countries.

In the Greek, Italian, and Polish cohorts, patients with paroxysmal AF had higher probabilities of undergoing CVD-related hospitalization compared with patients with first-detected AF: 4.07, 1.94, and 5.45 times higher, respectively. For patients in Poland with persistent or permanent AF, the probability of CVD-related hospitalization was on average 7.52 and 4.19 times higher compared with patients with first-detected AF. Conversely, for patients with permanent AF in Greece and the Netherlands, the estimated probabilities of CVD-related hospitalization were 0.41 and 0.49 compared with patients with first-detected AF.

Patients with CHF had significantly higher predicted probabilities of being hospitalized due to CVD in Greece (OR 2.48), in Italy (OR 1.51), and in Spain (OR 1.52) compared with patients free of this co-morbidity. Patients with other underlying heart disease also had higher probabilities of undergoing CVD-related hospitalization in Italy (OR 1.59), and the Netherlands (OR 1.39) while patients with a history of stroke were identified as having a higher probability of CVD-related hospitalization in Italy alone.

Annual costs of CVD-related hospitalization for the reference patient who incurred such costs, i.e. who underwent hospitalization due to CVD during the year of EHS-AF follow-up were €2042 in Greece, €3257 in Italy, €1834 in Poland, €3232 in Spain, and €5294 in the Netherlands.

There was a clear trend of average annual conditional costs of CVD-related hospitalization being higher (P < 0.1) for patients with VHD in the Greek, Italian, Spanish, and Dutch cohorts (estimated coefficients €4338, €2321, €4652, and €2710). Patients with permanent AF in Italy who were hospitalized for CVD incurred on average €3153 higher annual costs of CVD-related hospitalization compared with patients with first-detected AF. In addition, Italian EHS-AF patients with AP and CHF had annual costs of CVD-related hospitalization that were higher by an estimated €2125 and €2578, respectively. In Poland and Spain, patients hospitalized due to CVD who suffered from other underlying heart disease had on average €902 and €2865 higher costs of CVD-related hospitalization, respectively, compared with patients free of other underlying heart disease. Furthermore, hospitalized EHS-AF patients in Poland with a history of stroke had higher annual costs of CVD-related hospitalization (coefficient €1581). As for AF-type, permanent AF was associated with higher costs on average compared with first-detected AF in Italy (coefficient €3153) but lower costs in Poland (coefficient −€1029).

Discussion

This study identified predictors of annual direct costs, probabilities of hospitalization related to CVD and conditional annual costs of CVD-related hospitalization for patients recruited to the EHS-AF in Greece, Italy, Poland, Spain, and the Netherlands based on data gathered within the survey. Results are nation-specific and there was substantial variation across the studied countries. In general, we found that co-morbidities rather than demographic characteristics constituted predictors of costs. For example, BMI and gender did not emerge as predictors of costs in any of the studied countries. The co-morbidities identified as predictors of direct annual costs among the studied AF patients were VHD in Greece, Italy, and Spain, AP in Italy and Spain, diabetes and stroke in Poland, CHF in Italy, MI in Spain and other underlying heart disease in Poland and the Netherlands. On the contrary, the presence of hypertension at the time of EHS-AF enrolment never emerged as a predictor of cost in any of the five countries.

For the probability of being hospitalized for CVD (including AF recurrences), the expected risk factors such as MI, AP, stroke, CHF, and other CV conditions intermittently emerged as predictors. When interpreting these results the sample sizes reported in Table 1 are important to take into account since they are sometimes quite small, in particular for the patients with stroke.

For costs of CVD-related hospitalization (conditional models) observed in the sub-set of patients who experienced at least one hospitalization, it is important to note that the annual costs of CVD-related hospitalization for the reference patients were far above the total annual direct costs observed in the entire cohort. This result illustrates the substantial impact of CVD-related hospitalization on costs. The most important co-morbidity identified as a predictor was VHD, which was associated with average additional costs of €4338, €2321, €4652, and €2710 among EHS-AF patients hospitalized due to CVD in Greece, Italy, Spain, and the Netherlands, respectively. Atrial fibrillation-type only emerged as a predictor of direct costs in Italy, although it impacted on the probability of hospitalization in all the included countries except Spain. Compared with first-detected AF, having paroxysmal AF increased the probability of being hospitalized in three out of the five countries and having permanent AF was actually associated with a lower probability of hospitalization in two out of the five countries. The latter may be due to the fact that patients with permanent AF on average are less symptomatic than patients with first-detected AF.30

Even though AF is associated with high societal costs that are expected to increase in the coming years,7 there is little knowledge about the drivers of costs of AF and the factors leading to CVD-related hospitalizations among AF patients. Previous studies on the costs of AF have shown that direct costs constitute a significant part of the cost burden in patients with AF and that hospitalization is the main cost driver.26,27 The present study aimed to identify the underlying predictors of these costs as they account for the largest part of the societal burden of AF today. By studying the details of costs in the management of AF we aimed to create a more comprehensive picture of the cost structure, and thereby potentially facilitate the effective targeting of intervention. Although generalizability may be limited to AF patients in centres resembling those participating in the EHS-AF, the predictors identified in each country should be interpreted as providing an indication of subgroups of AF patients in which risk factor management and CVD prevention is particularly warranted from a health economic perspective. For example, since VHD was in many cases identified as a predictor of annual direct costs as well as costs of CVD-related hospitalization, this indicates that from a health economic perspective, particular attention should be paid to AF patients with VHD and that treatments preventing further CVD for patients with this comorbidity is highly likely to be cost-effective.

The models used to derive the predictors of costs were subjected to methodological evaluation. Since cost data tends to be positively skewed with a large proportion of zero observations and a long right-hand tail, the underlying assumption of the OLS regression equation may be violated suggesting the use of another model.37,38 We considered using a generalized linear model with a gamma distribution in order to better match the typical distribution of cost data. However, histograms displaying the residuals of direct annual costs and the costs of CVD-related hospitalizations in the studied sample showed a symmetric pattern and were approximately normal, indicating that the underlying assumption of a normal distribution of the residuals in the OLS model was not violated. Hence, after careful consideration the OLS approach was used. One of the major advantages in using the OLS regression is that it facilitates interpretation of the results and gives the reader an intuitive understanding of the findings. For the costs of CVD-related hospitalization, a conditional model approach was employed to handle the large number of zero values.

This study has a few limitations worth mentioning. First of all, due to country-specific variations in the organization and delivery of healthcare, results were limited in generalizability. Second, as described in previous studies of the EHS-AF, there was an overrepresentation of patients enrolled at highly specialized centres.30,32 It may be the case that these centres enrol more severely ill patients that incur costs above the general average in the respective countries. This would also cause higher frequencies of hospitalizations than would be found in a population-based sample of AF patients, making the results hard to generalize to the country level. There is also a risk that the highly specialized centres that are overrepresented in the EHS-AF employ more sophisticated clinical procedures compared with the regular hospitals. This could exaggerate the predictors of both direct costs and the costs of CVD-related hospitalizations that we estimated in this study. On the other hand, patients followed in such types of centres may be better monitored and controlled for the traditional CVD risk factors such as diabetes or hypertension, which as a result do not consistently show up as significant predictors of costs of hospitalization.

Since to the best of our knowledge, this is the first study conducted on predictors of costs of CVD in patients with AF, the opportunity of judging our results in the light of previous research is limited. There is no data with which our results can be compared, leaving us with uncertainty regarding the impact of issues related to representativity. Another matter worth highlighting regarding the costs in this study is the large standard deviations of mean costs, indicating large variability of the underlying data due to the limited sample size especially for patients who had a CVD-related hospitalization.

A final question worth discussing is our definition of a CVD-related hospitalization. We included admissions for all cardiovascular reasons, including AF, stroke/TIA, and other cardiac reasons. An option could have been to choose a more conservative approach and limit the definition to only include AF-related admissions. The broader definition was chosen since cardiovascular events can be seen as linked to the presence of AF. Co-morbidities related to all forms of CVD are very common in patients with AF making it hard to separate the costs for AF-specific treatments from costs of treatment of the CVD-related co-morbidities. This was also the reason for studying the total costs of CVD-related hospitalizations (including both CVD-related inpatient care and interventions) in the conditional regression models.

In conclusion, since AF is the most common cardiac arrhythmia there is a need to understand its underlying cost structure in order to assess the cost-effectiveness of alternative strategies of disease management. The EHS-AF data have provided us with a valuable opportunity to investigate the determinants of costs of AF patients in five selected countries. Although patients in highly specialized centres were overrepresented, the results of the EHS-AF reported in this study further increase the understanding of the cost structure of CVD in AF patients in Europe and may therefore inform the targeting of intervention therapy in selected groups of at-risk patients with AF.

Conflicts of interest: E.H., A.R., P.L., and F.D. have received funding from and acted as consultants to Sanofi-Aventis Recherche & Développement. F.C. is an employee of Sanofi-Aventis Recherche & Développement.

Funding

This work was supported by Sanofi-Aventis Recherche & Développement.

References

View Abstract