This study was designed as a historical, registry-based cost-of-illness analysis. Healthcare provision in Denmark is publicly funded with equal access and includes no co-payment by the patient in the hospital sector and only a small amount of co-payment for some services in the secondary healthcare sector. Hence, except for some minor co-payment for out-of-hospital pharmacological therapy, all services related to AF (and the complications related to AF care, including bleeding and thromboembolic disease) are free of charge from the patient perspective. In Denmark, there is a long tradition of systematic recording of health service provisions, for example, and it is possible to access and combine a number of high-quality, exhaustive registries at the individual-patient level, with the use of the patient’s encrypted social security number. Danish legislation permits researchers and others to access the databases. Ethics committee approval and written informed consent are not required for registry-based research, according to Danish law.
Patients with atrial fibrillation and general population controls
New (incident) AF patients diagnosed in the period 2001–2012 (inclusive of both years) were identified in the National Patient Registry (
International Classification of Diseases, 10th edition [ICD-10] code I48) [
12]. The National Patient Registry contains information on all patients discharged from all Danish non-psychiatric hospitals since 1977 and all emergency room and outpatient specialty clinic visits since 1995. Patients with contacts (admissions or outpatient visits) with AF registered as primary or secondary diagnosis were defined as AF patients. In order to identify only new cases of AF, a washout period of 5 years was used, implying that individuals who have been in contact with a hospital and diagnosed with AF in the period 1996–2000 were excluded from the AF population.
A gender- and age-matched control group of individuals free of AF was identified using the Danish population registry (the Danish Civil Registry) [
13]. For each AF patient, three controls were identified. The same individual was allowed to be used as a control multiple times in different years, but not in the same year. Furthermore, controls for AF patients were only between 18 and 89 years old at incidence as there was lack of potential controls for older AF patients (aged ≥90 years). Therefore, the cost analyses, which depend on the matching, only include 18- to 89-year-old patients.
All included patients had at least 1 year of history and follow-up included in the registries (i.e. a patient diagnosed in November 2012 was ‘followed’ in the registries until November 2013).
Costs of AF
A societal perspective was applied including the following six cost types:
Primary sector costs were obtained from the National Health Service Registry, which holds data on all contacts in the primary care sector, including contacts with general practitioners and private practice specialists [
14]. The fee paid to the healthcare professional (e.g. general practitioner) was applied as unit cost estimate for services provided in the primary care sector.
Data on hospital costs were obtained from the National Patient Registry. We used the Danish outpatient charges (DAGS charges) as unit cost estimate for each outpatient contact (including emergency room visits) and the Diagnosis Related Groups (DRG) charges for admissions.
Data on medicine costs were obtained from the Registry of Medicinal Product Statistics, which holds information on all acquisitions of prescription medicine at Danish pharmacies, including prices [
15]. The market price was applied as unit cost estimate for each acquisition.
Home care costs included both costs at own home and from nursing homes. The Danish municipalities provide these home care services, and the data originate from the municipalities’ own registry, which was accessed via Statistics Denmark. The data were available from 2008 onwards. Data on care in nursing homes were, however, only available until 2012. The registry includes information about the services (measured in minutes) each individual has been granted every month with respect to both nursing and practical help. In order to approximate this into a cost estimate, gross wages for nurses and home-helpers were applied from the Salary Data Office of Municipalities and Regions. In order to estimate an hourly salary, the annual salary was divided by 1628 h, which is the number of effective working hours per year given a 37-h working week. To translate the wage into a cost, a 75% overhead charge was added covering administration costs and transportation.
Productivity losses were estimated using weekly employment data from the DREAM database (The longitudinal database of The Danish Agency for Labour Market and Recruitment). The DREAM database includes all individuals who have received labour market-related transfer payments since 1991 [
16]. With DREAM, it was estimated what fraction of the baseline year as well as the 1st, 2nd, and 3rd year after AF an individual was employed. Individuals who have never received public transfers do not appear in DREAM. Individuals, who do not appear in DREAM and who are aged ≤65 years, were treated as individuals who have worked for the entire year. Furthermore, individuals aged >65 years were treated as if they were not employed (i.e. retired) independent of their employment status according to DREAM. This was done in order to create consistency in the way individuals who appear in DREAM and those who do not were treated.
The productivity value for each individual was estimated by multiplying the estimated yearly employment fraction with a gender-specific gross average wage, adjusted for the number of effective working hours per week performed by each gender. With the productivity value of each individual, the productivity loss in year t was estimated as the productivity value in year t minus the productivity value in the baseline year. This was defined as the total productivity loss (productivity costs). The attributable cost in terms of productivity loss was the productivity loss for the AF patient minus the productivity loss for the controls. All costs were inflated to 2013 price levels. Primary sector costs, outpatient costs and hospital admission costs were inflated using the price and salary index from the Danish Regions. Similarly, medicine costs were inflated using the medicine price index from Danish Regions. For the estimates of home care costs and productivity losses, constant 2013 prices were used for salaries and earnings, respectively. The salaries for nurses and home-helpers were identified using yearly gross wages for municipally employed nurses and home-helpers, whereas the productivity values were estimated using per-hour earnings from the private sector. Future costs in year 2 and in year 3 were discounted by 4% per year.
Data analysis
We first determined the yearly incidence and prevalence of AF. The AF prevalence estimates were based on the AF patients diagnosed between 2001 and 2012. The number of prevalent AF patients in a given year (by the end of the year) was, thus, the number of new cases plus the number of AF patients from previous years minus the number of deaths. The denominator was the total Danish population aged 18 years or older in the given year. The mortality among AF patients relative to the controls was measured as the percentage of both AF patients and controls who died in the 1st, 2nd and 3rd year after AF diagnosis. Information about deaths was retrieved from the National Causes of Death Registry. Information on the general population, including age- and sex-distribution, was obtained from Statistics Denmark.
We then estimated the costs of AF, including the costs in the 1st, 2nd and 3rd year after the AF diagnosis, using both a total cost and attributable cost approach.
The total costs for every individual
i in year t was defined as:
$$ Total\ {cost}_{i,t}^{AF}={\sum}_{x=1}^X{Cost}_{i,x,t}^{AF}-{Cost}_{i,x,0}^{AF}, $$
where AF is an abbreviation for “AF patient”, t is the year of interest (i.e. 1st, 2nd or 3rd year after AF diagnosis), 0 is the year leading up to the AF diagnosis, x is the type of cost (e.g. outpatient cost, admission cost, etc.). If an AF patient was diagnosed on 1 April 2008, the total costs in the 1st year after AF would be his or her costs from 1 April 2008 to 31 March 2009 minus his or her costs from 1 April 2007 to 31 March 2008.
The attributable costs are the total costs of AF minus the total costs among the controls in accordance with previous publications [
17,
18]. Hence, attributable costs were estimated as a difference-in-difference in which the difference between costs for cases (AF patients) and controls was regarded as being attributable to the AF.
$$ Attributable\ {cost}_t^{AF}=\sum \limits_{x=1}^X\left({Cost}_{x,t}^{AF}-{Cost}_{x,0}^{AF}\right)-\left({Cost}_{x,t}^C-{Cost}_{x,0}^C\right) $$
Since AF patients and controls were not necessarily part of the analysis during the entire 3 years, due to either death or the conclusion of the study, their costs were weighted by the fraction of the time they were part of the analysis.
Finally, we performed a subgroup analysis among AF patients subsequently diagnosed with ischaemic stroke (ICD-10 codes I63, I64). This was done in order to assess the cost impact of ischaemic stroke in AF patients. That is, in the National Patient Registry for the 1st, 2nd or 3rd year following the AF diagnosis, we identified all AF patients, with a subsequent hospital contact for ischaemic or non-specified stroke. In order to ensure that only new cases of stroke were included, a 5-year wash out period was applied.
The total cost of AF and subsequent stroke was, for every patient, a measure of his or her costs in year t (after the AF diagnosis) minus his or her costs in the baseline year (i.e. the year up to the AF diagnosis). In this subanalysis, patients were also categorized according to the incidence year of stroke (i.e. AF patients, stroke year 1; AF patients, stroke year 2; and AF patients, stroke year 3 – where stroke year 1 indicates that the patient were diagnosed with stroke within the first year following AF).
The attributable cost was estimated as the total cost of an AF patient with subsequent stroke minus the total cost among AF patients without stroke (i.e. AF patients without stroke are the controls). Again, attributable costs were estimated as a difference-in-difference.