Newly diagnosed atrial fibrillation and hospital utilization in heart failure: a nationwide cohort study.

AIMS: Atrial fibrillation (AF) constitutes a major burden to health services, but the importance of incident AF in patients with heart failure (HF) is unclear. We examined the associations between incident AF and hospital utilization in patients with HF. METHODS AND
RESULTS: In a nationwide matched-cohort study of HF patients, we identified patients diagnosed with incident AF between 2008 and 2018 in the Danish Heart Failure Registry (N = 4463), and we compared them to matched referents without AF (N = 17 802). Incident AF was associated with a multivariable-adjusted 4.8-fold increase (95% CI 4.1-5.6) and 4.3-fold increase (95% CI 3.9-4.8) in the cumulative incidence of inpatient and outpatient contacts within 30 days, respectively. At 1 year, the cumulative incidence ratios were 1.8 (95% CI 1.7-1.9) and 1.4 (95% CI 1.4-1.5). Incident AF was also associated with increases in the total numbers of inpatient and outpatient hospital contacts within 30 days (multivariable-adjusted rate ratio 1.4, 95% CI 1.4-1.5, and 1.6, 95% CI 1.6-1.7, respectively). At 1 year, the ratios were 2.2 (95% CI 2.1-2.3) and 2.0 (95% CI 1.9-2.1). The multivariable-adjusted proportion of bed-day use among HF patients with incident AF was 10.9-fold (95% CI 9.3-12.9) higher at 30 days and 5.3-fold (95% CI 4.3-6.4) higher at 1 year compared with AF-free referents.
CONCLUSIONS: Incident AF in HF is associated with earlier hospital contact, more hospital contacts, and more hospital bed-days. More evidence on interventions that may prevent the risk and subsequent burden of AF in HF is urgently needed.

Introduction

Heart failure (HF) is a growing worldwide epidemic with a global prevalence of 64 million cases in 2017, and projection studies suggest that the prevalence will increase in the future. , Among Europeans, the lifetime risk of HF is 28% in women and 33% in men at the age of 55 years, and among Americans 20% and 21%, respectively, at the age of 40 years. HF is associated with considerable morbidity and high mortality, is a leading cause of hospitalizations and outpatient visits, and is associated with substantial healthcare expenditures that will likely rise. Therefore, HF constitutes a major burden to public health and health services.

Patients with incident HF have a two‐fold increased risk of incident atrial fibrillation (AF), and ~10% of patients with HF developed AF after ~2.5 years of follow‐up according to UK primary care data. Incident AF in HF patients is considered an adverse prognostic indicator of HF, with an approximately two‐fold increased rate of death compared with HF patients without AF. , AF constitutes a major burden to health services, including hospitalizations and health expenditures, , but the consequence of incident AF on hospital utilization in the HF setting is unclear.

Therefore, the objective of the present study was to quantify hospital utilization associated with incident AF among patients with HF.

Methods

Setting and data sources

The Danish Heart Failure Registry (DHFR) is a nationwide clinical quality database established in 2003 that comprises inpatients and outpatients with incident HF. The objectives of the DHFR are to monitor and improve the quality of care for Danish patients with HF. HF has to be diagnosed or validated by a cardiologist, and reporting of HF patients from all Danish hospitals is mandatory. The source population for our investigation included patients diagnosed with incident HF from 2008 to 2018. The inclusion criteria of the registry include a first‐time diagnosis of HF and diagnostic criteria from the National Society for Cardiology and European Society of Cardiology: HF symptoms and objective signs of HF, and/or a possible clinical improvement on HF treatment. Exclusion criteria include HF caused by uncorrectable structural heart disease, HF caused by valvular heart disease, HF caused by rapid heart rhythm (including AF), isolated right‐sided HF, HF diagnosed concurrently with a primary diagnosis of acute myocardial infarction, or HF patients diagnosed and treated by a private practitioner of cardiology. The cardiologist identified the conditions in the patient's medical records.

We linked data of the DHFR to data from three national registries. The Danish National Patient Registry was established in 1977 and contains prospectively registered information on all inpatients and after 1995 also all outpatients. Data included individual‐level information on dates of admission and discharge, surgical procedures performed, and one primary diagnosis and one or more secondary diagnoses per discharge. The coding of diagnoses followed the Danish version of the International Classification of Diseases Eighth Revision (ICD‐8) before 1994 and the 10th Revision (ICD‐10) from 1994 and onwards. The physician who discharged a patient coded all diagnoses for that patient.

The Danish National Prescription Registry contains individual‐level data on all dispensed prescriptions since 1994 and was used to retrieve information on pharmacological treatments. The coding of medications follows the Anatomical Therapeutic Chemical Classification System.

Statistics Denmark provided information on family income and highest completed education.

The Danish Civil Registration System contains daily updated individual‐level information on sex, date of birth, vital statistics, and migration. All Danish citizens are assigned a unique 10‐digit civil registration number that enables unambiguous linkages of data between registries.

Supporting Information, Table presents the definition of variables and data sources.

Design and study population

We designed a nationwide registry‐based matched‐cohort study. From the DHFR population, we first excluded patients with any diagnosis of AF (including AF and atrial flutter, and referred to as AF hereafter) identified in the Danish National Patient Registry prior to or on the same day as the diagnosis of HF to ensure that only individuals at risk of incident AF were included in the study population (Supporting Information, Table ). Furthermore, to ensure sufficient time for the registry‐based identification of history of diseases, we excluded HF patients who lived in Denmark for <5 years. We identified all HF patients with newly diagnosed AF. We matched each HF patient with newly diagnosed AF to HF referents without AF and followed all patients for hospital contacts, admissions, and bed‐day use.

Heart failure patients with newly diagnosed atrial fibrillation

We identified all HF patients with a new primary or secondary hospital diagnosis of AF (or atrial flutter) from the National Patient Registry after their HF diagnosis (Supporting Information, Table ). Both inpatient and outpatient diagnoses were included. The index date was the date of newly diagnosed AF, which we defined as the discharge day for inpatients and the first date of the diagnosis for outpatients.

Matched referents

We matched each AF case with up to four HF patients without AF (Supporting Information, Figure ). Matching variables included age at HF diagnosis, sex, and age at index date. Including both age at HF and age at AF index date prevented selective survival bias due to the impossibility of selecting HF patients with high short‐term mortality as referents. The index date for each matched referent was the case index date, that is, the date of newly diagnosed AF. Sampling of referents was with replacement, so a referent may have been sampled more than once. A referent that became an AF case at an older age led to censoring at the time of AF diagnosis.

Hospital utilization

We examined three measures of hospital‐related activity, each stratified into inpatient and outpatient care. A hospital contact was any inpatient or outpatient hospital contact with a registered primary diagnosis. We did not include emergency room contacts because the diagnostic validity in the setting is low for many diagnoses. First, we examined the risk of having at least one hospital contact within 30 days and 1 year, respectively, after the index date. Second, we counted the number of hospital contacts within 30 days and 1 year after the index date, respectively. Third, we examined the number of bed‐days, which included any hospitalization with at least one overnight stay.

Covariates

Covariates included sex, age, lifestyle factors, clinical data, co‐morbidities, and socio‐economic factors (Supporting Information, Tables and ).

Lifestyle factors included weekly alcohol consumption and smoking. We applied the definitions of elevated alcohol consumption pertained to the DHFR. The definition was more than 14 drinks per week for women and 21 drinks per week for men until 1 July 2015. After that date, the registry applied a lower threshold of elevated drinking, namely, more than 7 drinks per week for women and 14 drinks per week for men. Smoking status was classified as current smoker, former smoker, or never smoker. Information on alcohol consumption and smoking status was reported at the diagnosis of HF.

Clinical data included left ventricular ejection fraction (LVEF) and New York Heart Association (NYHA) Classification. Categorization of LVEF followed the universal LVEF classification groups, but to account for the patients with very low LVEF, we added a category <25% (≥50%, >40–49%, 25–40%, or <25%). The patients underwent echocardiography between 7 days after the diagnosis of HF and up to 6 months before the diagnosis of HF if the cardiologist considered the examination relevant. NYHA classification was categorized as I, II, or III/IV and reported at the diagnosis of HF or up to 12 weeks after the diagnosis.

Co‐morbidities included history of myocardial infarction, ischaemic stroke, diabetes mellitus, chronic obstructive pulmonary disease, hypertension, chronic kidney disease, and overweight/obesity. Furthermore, we calculated a modified Charlson Comorbidity Index that did not include AIDS because of a very low prevalence. History of co‐morbidities was ascertained up to and including the day before the index date.

Socio‐economic factors comprised family income and highest completed education. Family income was the yearly disposable equivalent income, which is a comparable measure between families that accounts for the number of family members that live together and their ages. Statistics Denmark generated the measure for a family by adding the income of all the family members and dividing by a weighted average number of people in the family. We categorized level of education into the following International Standard Classification of Education (ISCED) groups. Group 1 included early childhood, primary education, and lower secondary educations (ISCED 0–2). Group 2 included general upper secondary education and vocational upper secondary education (ISCED 3). Group 3 included short‐cycle tertiary, medium‐length tertiary, bachelor's level educations or equivalent, second cycle, master's level or equivalent, and PhD level (ISCED 5–8). ISCED 4 does not exist in Denmark.

Medications included angiotensin‐converting enzyme (ACE) inhibitors/angiotensin II receptor blockers (ARBs), beta‐blockers, and mineralocorticoid receptor antagonists (MRAs) (Supporting Information, Table ). The definition of use was at least one redeemed prescription within the last 6 months before the index date. Furthermore, we examined use of oral anticoagulants within 3 months after AF and the use of loop diuretics 3 months prior to index date.

Statistical methods

We used time‐to‐event analyses to examine time to the first hospital contact. Time at risk began on the first day after the index date and ended on the earliest of date of death, last follow‐up, heart transplantation, at 30 days or 1 year, whichever came first. We used the Aalen–Johansen estimator, which accounted for the competing risk of death, to estimate the cumulative incidence of first hospital contact with 95% confidence interval (95% CI). We used the pseudo‐value approach to estimate multivariable‐adjusted cumulative incidence ratios between HF patients with and without AF.

We used negative binomial regression to estimate the number of hospital contacts per 1000 risk‐days and the percentage of risk‐days spent in hospital bed, respectively. Time at risk began on the first day after the index date and ended on date of death, date of end of follow‐up, heart transplantation, or date of emigration, at 30 days or 1 year, whichever came first. To account for different lengths of risk periods, days at risk was logarithmized and included in the models to provide a rate per risk‐day. We estimated the ratio between AF cases and referents at 30 days and 1 year.

We performed five pre‐specified subgroup analyses. First, we stratified by sex. Second, we stratified by age group ≥75 vs. <75 years at index date. Third, we stratified by LVEF level categorized into ≤40%, >40–49%, and ≥50%. Fourth, we stratified by time with HF ≤ 1 vs. >1 year. Fifth, we stratified by educational level. We tested for interactions by adding interaction terms to the multivariable models.

We used Stata statistical software (StataCorp. 2019: Release 16.1, College Station, TX: StataCorp LLC) for all data management and statistical analyses.

Missing data

Variables with missing values included LVEF, NYHA class, alcohol use, and smoking status. We used multiple imputation to account for missing values in all analyses except the analyses of LVEF groups and level of education in which we excluded patients with missing LVEF and educational level, respectively. The imputation model included case/referent status, age at HF, sex, age at matching, time since diagnosis of HF, highest level of education, family income, history of acute myocardial infarction, stroke, diabetes, chronic obstructive pulmonary disease, hypertension, chronic kidney disease, overweight/obesity, use of ACE inhibitor/ARB treatment, beta‐blocker, MRA, the Charlson Comorbidity Index, and an outcome indicator. We created 10 imputed datasets and combined the estimates with Rubin's rules. We used the mi command in Stata to perform multiple imputation.

Ethics

The Danish Health Data Authority, Statistics Denmark, and the Danish Data Protection Agency approved this study. Registry‐based studies do not require approval from an ethics committee according to Danish law.

Results

Characteristics of patients

We identified 27 947 patients diagnosed with incident HF from 2008 through 2018 with no history of AF. Among all HF patients, 4471 patients were diagnosed with incident AF during follow‐up. At 10 years after the diagnosis of HF, the cumulative incidence of AF was 23.5% when accounting for the competing risk of death. The incidence rate of AF was 45.4 per 1000 person‐years. We excluded eight of the 4471 AF cases (mean age at HF diagnosis 96.1 years; mean age at AF diagnosis 98.6 years) because no AF‐free referents could be matched. The study population consisted of 4463 AF cases and 17 802 AF‐free matched referents (balance of covariates presented in Supporting Information, Table ). Of the 17 802 matched referents, 4327 patients were selected as referents more than once and 1455 referents later became AF cases. At the diagnosis of AF, the mean age was 74 years, the proportion of female patients was 29%, and the median time with HF was 1.2 years (Table ). Among AF cases, 47% had a history of hypertension, 32% had a history of acute myocardial infarction, and 21% had a history of diabetes. The median Charlson Comorbidity Index was 3 (inter‐quartile range 2–5) among AF cases and 2 (inter‐quartile range 1–4) among referents (Supporting Information, Table ). The respective use of ACE inhibitors/ARBs, beta‐blockers, and MRA (classified at the time of incident AF diagnosis) was higher among AF cases at the index day (Table ). Within 3 months after the diagnosis of AF, 44.2% of patients diagnosed between 2008 up to and including 2013 redeemed at least one prescription of an oral anticoagulant, while 53.8% of patients diagnosed from 2014 up to and including 2018 redeemed at least one prescription. The mortality rate was 177.1 per 1000 person‐years among AF cases and 88.7 per 1000 person‐years among matched referents. In a multivariable‐adjusted Cox model, the hazard ratio for death between AF cases and matched referents was 1.91 (95% CI 1.81–2.02).

Table 1

Characteristics of HF patients with incident AF and matched referents at index date

	AF cases N = 4463	Referents N = 17 802
Socio‐demographics
Age (years), mean (SD)	73.7 (11.4)	73.6 (11.3)
Female sex, N (%)	1295 (29.0)	5153 (29.0)
Family income a (€), median (Q1–Q3)	23 097 (19 283–31 120)	23 439 (19 434–31 453)
Highest completed education, N (%)
Group 1	2199 (51.5)	8947 (52.4)
Group 2	1544 (36.2)	6170 (36.1)
Group 3	528 (12.4)	1964 (11.5)
Lifestyle factors, N (%)
Elevated alcohol consumption	362 (9.5)	1237 (8.1)
Smoking status
Never	1065 (26.7)	4101 (25.2)
Former	1857 (46.6)	7532 (46.3)
Current	1064 (26.7)	4636 (28.5)
Clinical data, N (%)
LVEF
<25%	1076 (24.7)	3628 (20.8)
25–40%	2671 (61.4)	11 161 (64.1)
>40–49%	296 (6.8)	1314 (7.6)
≥50%	311 (7.1)	1302 (7.5)
NYHA class
I	499 (12.7)	2729 (16.7)
II	2361 (59.9)	9989 (61.2)
III/IV	1082 (27.5)	3593 (22.0)
Time since HF (years), median (Q1‐Q3)	1.2 (0.2–3.5)	1.2 (0.4–3.4)
Age at HF, mean (SD)	71.5 (11.4)	71.4 (11.4)
History of co‐morbidity, N (%)
Myocardial infarction	1404 (31.5)	5946 (33.4)
Any stroke/TIA	697 (15.6)	1950 (11.0)
Diabetes mellitus	956 (21.4)	3149 (17.7)
COPD	696 (15.6)	2029 (11.4)
Hypertension	2100 (47.1)	7562 (42.5)
Chronic kidney disease	460 (10.3)	1228 (6.9)
Overweight/obesity a	394 (8.8)	954 (5.4)
Modified CCI, N (%)
0	0	0
1–2	2070 (46.4)	10 430 (58.6)
3–4	1185 (26.6)	4292 (24.1)
5–6	655 (14.7)	1979 (10.1)
7–8	316 (7.1)	732 (4.1)
≥9	237 (5.3)	551 (3.1)
Current HF medication b , N (%)
ACE inhibitors/ARBs	2806 (62.9)	9475 (53.2)
Beta‐blockers	2669 (56.8)	8888 (49.9)
MRAs	1036 (23.2)	3269 (18.4)
Loop diuretics	1925 (43.1)	4707 (26.4)

ACE, angiotensin‐converting enzyme; AF, atrial fibrillation; ARBs, angiotensin receptor blockers; CCI, Charlson Comorbidity Index; COPD, chronic obstructive pulmonary disease; HF, heart failure; LVEF, left ventricular ejection fraction; MRAs, mineralocorticoid receptor antagonists; NYHA, New York Heart Association; SD, standard deviation; TIA, transient ischaemic attack.

We matched each HF patient with incident AF with up to four referents on age at diagnosis of HF, sex, and time since diagnosis of HF. Characteristics were collected at the time of AF diagnosis or corresponding age for referents, except alcohol use, smoking, LVEF, and NYHA, which were collected around the time of HF diagnosis. Missing, N (%): income 36 (0.2%); education: 913 (4.1); alcohol consumption: 3164 (14.2); smoking: 2010 (9.3); LVEF: 506 (2.3); and NYHA: 2012 (9.0).

Defined by the International Classification of Diseases codes.

Defined as fulfilling at least one prescription within 6 months prior to the index date, except loop diuretics for which the period was 3 months. Modified CCI did not include AIDS.

Time to first hospital contact

Half of the AF cases had encountered at least one inpatient contact 7.8 months after AF, while half of the referents encountered at least one contact 4.5 years after index date. The cumulative incidences of first inpatient contact were 25.3% among AF cases and 5.2% among referents at 30 days (Figure and Table ). At 1 year, the cumulative incidences were 56.3% and 23.5%, respectively. The multivariable‐adjusted cumulative incidence ratio associated with incident AF was 4.8 (95% CI 4.1–5.6) at 30 days and 1.8 (95% CI 1.7–1.9) at 1 year. When examining outpatient contacts, half of the AF cases had encountered at least one contact 2.3 months after AF, while half of the referents encountered at least one contact 1.5 years after index date. At 30 days, the cumulative incidence of first outpatient contact was 38.3% among AF cases and 10.6% among referents. At 1 year, the cumulative incidences were 69.8% and 43.1%, respectively. The multivariable‐adjusted cumulative incidence ratio associated with incident AF was 4.3 (95% CI 3.9–4.8) at 30 days and 1.4 (95% CI 1.4–1.5) at 1 year.

Figure 1

Cumulative incidence of first hospital contact after index date. The cumulative incidence accounted for the competing risk of death. AF, atrial fibrillation; HF, heart failure.

Table 2

Incident AF and time to first hospital contact

	Inpatient contact			Outpatient contact
	Cumulative incidence, % (95% CI)	Cumulative incidence ratio		Cumulative incidence, % (95% CI)	Cumulative incidence ratio
		Model 1	Model 2		Model 1	Model 2
30 days after index date
AF cases	25.3 (24.0–26.7)	5.36 (4.71–6.11)	4.76 (4.07–5.58)	38.3 (36.8–39.7)	4.73 (4.31–5.18)	4.33 (3.91–4.81)
Referents	5.2 (4.9–5.5)	1.00 (ref)	1.00 (ref)	10.6 (10.1–11.0)	1.00 (ref)	1.00 (ref)
1 year after index date
AF cases	56.3 (54.8–57.9)	2.03 (1.95–2.11)	1.78 (1.71–1.85)	69.8 (69.4–69.8)	1.58 (1.54–1.62)	1.43 (1.39–1.47)
Referents	23.5 (25.8–27.1)	1.00 (ref)	1.00 (ref)	43.1 (42.4–43.9)	1.00 (ref)	1.00 (ref)

AF, atrial fibrillation; CI, confidence interval.

Model 1: adjusted for age at heart failure, time with heart failure, and sex. Model 2: adjusted as Model 1 + family income, educational level, alcohol use, smoking status, left ventricular ejection fraction, New York Heart Association class, history of myocardial infarction, history of stroke, history of diabetes, history of chronic obstructive pulmonary disease, history of hypertension, history of chronic kidney disease, history of obesity, Charlson Comorbidity Index, use of angiotensin‐converting enzyme inhibitors/angiotensin receptor blockers, use of beta‐blockers, and use of mineralocorticoid receptor antagonists.

Number of inpatient and outpatient contacts

During the initial 30 days, AF cases had 28.1 inpatient contacts per 1000 risk‐days, while the matched AF‐free referents had 19.9 inpatient contacts (Table ). The multivariable‐adjusted rate ratio associated with incident AF was 1.4 (95% CI 1.4–1.5). At 1 year after incident AF, the number of inpatient contacts per 1000 risk‐days was 8.8, while the matched referents had 3.7 contacts. The multivariable‐adjusted rate ratio was 2.2 (95% CI 2.1–2.3). Counting outpatient contacts, AF cases had 30.3 per 1000 risk‐days compared with 18.7 among referents. The multivariable‐adjusted rate ratio was 1.6 (95% CI 1.6–1.7). At 1 year, the number of outpatient contacts per 1000 risk‐days was 10.6 among AF cases and 3.9 among referents, which corresponded to a multivariable‐adjusted rate ratio of 2.0 (95% CI 1.9–2.1).

Table 3

Incident AF and number of hospital contacts

	Inpatient contacts			Outpatient contacts
	Number of contacts per 1000 risk‐days (95% CI)	Rate ratio (95% CI)		Number of contacts per 1000 risk‐days (95% CI)	Rate ratio (95% CI)
		Model 1	Model 2		Model 1	Model 2
30 days after index date
AF cases	28.1 (27.1–29.1)	1.42 (1.36–1.47)	1.41 (1.36–1.47)	30.3 (29.4–31.3)	1.62 (1.57–1.68)	1.63 (1.57–1.68)
Referents	19.9 (19.6–20.2)	1.00 (ref)	1.00 (ref)	18.7 (18.5–19.0)	1.00 (ref)	1.00 (ref)
1 year after index date
AF cases	8.8 (8.4–9.3)	2.39 (2.28–2.51)	2.16 (2.06–2.27)	10.6 (10.2–11.0)	2.13 (2.05–2.22)	1.97 (1.89–2.05)
Referents	3.7 (3.6–3.8)	1.00 (ref)	1.00 (ref)	4.9 (4.9–5.0)	1.00 (ref)	1.00 (ref)

AF, atrial fibrillation; CI, confidence interval.

Model 1: adjusted for age at heart failure, time with heart failure, and sex. Model 2: adjusted as Model 1 + family income, educational level, alcohol use, smoking status, left ventricular ejection fraction, New York Heart Association class, history of myocardial infarction, history of stroke, history of diabetes, history of chronic obstructive pulmonary disease, history of hypertension, history of chronic kidney disease, history of obesity, Charlson Comorbidity Index, use of angiotensin‐converting enzyme inhibitors/angiotensin receptor blockers, use of beta‐blockers, and use of mineralocorticoid receptor antagonists. Outpatient contacts within 30 days: one patient with four referents was excluded because of an unlikely high number of hospital contacts within 30 days.

Total hospital bed‐day use

During the initial 30 days after AF, AF cases spent 8.5% of risk‐days as hospital inpatients, while the proportion was 0.9% for referents (Table ). The multivariable‐adjusted ratio of proportions was 10.9 (95% CI 9.3–12.9). At 1 year, the proportion of days was 5.8% among AF cases and 1.1% among referents. The multivariable‐adjusted ratio of proportions was 5.3 (95% CI 4.3–6.4).

Table 4

Incident AF and proportion of days spent in hospital bed

	Proportion of days spent in a hospital bed, % (95% CI)	Ratio of proportions (95% CI)
		Model 1	Model 2
30 days after index date
AF cases	8.5 (7.5–9.5)	10.36 (8.81–12.18)	10.94 (9.25–12.94)
Referents	0.9 (0.8–1.0)	1.00 (ref)	1.00 (ref)
1 year after index date
AF cases	5.7 (4.8–7.0)	5.65 (4.55–7.01)	5.26 (4.32–6.40)
Referents	1.1 (1.0–1.2)	1.00 (ref)	1.00 (ref)

AF, atrial fibrillation; CI, confidence interval.

Model 1: adjusted for age at heart failure, time with heart failure, and sex. Model 2: adjusted as Model 1 + family income, educational level, alcohol use, smoking status, left ventricular ejection fraction, New York Heart Association class, history of myocardial infarction, history of stroke, history of diabetes, history of chronic obstructive pulmonary disease, history of hypertension, history of chronic kidney disease, history of obesity, Charlson Comorbidity Index, use of angiotensin‐converting enzyme inhibitors/angiotensin receptor blockers, use of beta‐blockers, and use of mineralocorticoid receptor antagonists.

Subgroup and supplementary analyses

In the sex‐specific analysis (Supporting Information, Table ), we found evidence of a higher cumulative incidence ratio of first hospital contact for outpatients among men vs. women at 1 year (1.5, 95% CI 1.4–1.5 vs. 1.3, 95% CI 1.3–1.4, P < 0.001, Supporting Information, Table ). Statistical evidence supported a higher rate ratio of number of inpatient contacts among women at 30 days (1.5, 95% CI 1.4–1.6 vs. 1.4, 95% CI 1.3–1.4, P < 0.01) and 1 year (2.4, 95% CI 2.2–2.7 vs. 2.1, 95% CI 1.9–2.2, P = 0.002) but not when considering outpatients care (Supporting Information, Table ). There was no substantial difference between men and women in the proportion of days in hospital bed (Supporting Information, Table ).

In the age‐stratified analysis (sample characteristics, Supporting Information, Table ), we found statistical evidence of a higher cumulative incidence ratio of first hospital contact among the younger age group at 1 year for both inpatients (1.9, 95% CI 1.8–2.1 vs. 1.6, 95% CI 1.5–1.7, P < 0.001) and outpatients (1.5, 95% CI 1.4–1.5 vs. 1.4, 95% CI 1.3–1.4, P = 0.02), and we noted no substantial differences at 30 days (Supporting Information, Table ). We found statistical evidence of a higher rate ratio of number of inpatient contacts at 30 days (1.5, 95% CI 1.5–1.6 vs. 1.3, 95% CI 1.2–1.4, P < 0.001) and 1 year (2.3, 95% CI 2.2–2.5 vs. 2.0, 95% CI 1.9–2.2, P = 0.03) among patients aged 75 or older, but not for outpatient contacts (Supporting Information, Table ). There were no substantial differences in the analysis of days in hospital bed (Supporting Information, Table ).

When stratifying by LVEF group (Supporting Information, Table ), we noted no substantial differences in cumulative incidence ratios of first contact across the groups, but the statistical precision was low (Supporting Information, Table ). We found no substantial difference in number of hospital contacts at 30 days or 1 year (Supporting Information, Table ) nor in the analyses of bed‐day use (Supporting Information, Table ). In a supplemental analysis, we noted no substantial differences between patients with LVEF ≤ 40% and patients with LVEF <25% (Supporting Information, Tables – ).

In the analysis of time to first contact stratified by time with HF (Supporting Information, Table ), we found statistical evidence of a higher cumulative incidence ratio at 30 days among patients with HF > 1 year for both inpatient (3.5, 95% CI 2.7–4.5 vs. 8.2, 95% CI 6.3–10.9, P < 0.001) and outpatient care (3.3, 95% CI 2.9–3.9 vs. 6.1, 95% CI 5.1–7.2, P < 0.001, Supporting Information, Table ). Furthermore, the cumulative incidence ratio at 1 year was also higher for patients with HF > 1 year and inpatient care (1.6, 95% CI 1.5–1.7 vs. 2.0, 95% CI 1.9–2.1, P < 0.001). In the analysis of number of contacts, we found evidence of a higher rate ratio at 30 days among patients with HF ≤ 1 year for inpatient (1.5, 95% CI 1.4–1.6 vs. 1.4, 95% CI 1.3–1.4, P = 0.03) and outpatient care (1.7, 95% CI 1.6–1.8 vs. 1.5, 95% CI 1.5–1.6, P < 0.001, Supporting Information, Table ). At 1 year, the rate ratio was higher among patients with HF ≤ 1 year for inpatient care (1.9, 95% CI 1.8–2.1 vs. 2.0, 95% CI 1.9–1.2, P = 0.002). There was evidence of a higher ratio of proportions of bed‐days for patients with HF > 1 year at 30 days (7.8, 95% CI 6.5–9.2 vs. 16.2, 95% CI 12.2–21.4, P < 0.001) and 1 year (3.5, 95% CI 3.1–4.1 vs. 7.5, 95% CI 5.3–10.5, P < 0.001, Supporting Information, Table ).

In the analysis stratified by highest educational level (Supporting Information, Table ), we noted no substantial differences across the groups (Supporting Information, Table ). In the analysis of number of hospital contacts, we noted a lower rate ratio among patients with the highest level of education at 30 days for inpatient care (Group 1: 1.5, 95% CI 1.4–1.6 vs. Group 2: 1.4, 95% CI 1.3–1.5 vs. Group 3: 1.3, 95% CI 1.2–1.4, Supporting Information, Table ). At 30 days, the proportion of days spent in hospital bed was higher among patients in Group 3 (Group 1: 9.9, 95% CI 8.1–12.1 vs. Group 2: 11.2, 95% CI 8.8–14.3 vs. Group 3: 23.0, 95% CI 15.3–37.7), but the statistical precision was low (Supporting Information, Table ).

In an analysis stratified by inpatient/outpatient registration of AF, the results showed that patients diagnosed with AF in inpatient settings had shorter time to first hospital contact, had more hospital contacts, and spent more time in hospital bed (Supporting Information, Tables – ).

We performed an additional analysis, in which we examined HF‐related contacts, defined as all ICD‐10 I50 contacts. We noted that incident AF was associated with increased rate of HF‐related hospital contact, larger number of HF‐related hospital contacts, and increased proportion of HF‐related days spent in hospital bed (Supporting Information, Tables – ).

Discussion

In this nationwide study of patients with incident HF, we found that incident AF was associated with considerable increases in hospital utilization, particularly within the initial 30 days after the diagnosis. After 30 days of follow‐up, incident AF was associated with an almost 5‐fold increased risk of at least one inpatient contact, an ~4‐fold increased risk of at least one outpatient contact, a 41% increase in number of inpatient contacts, a 63% increase in number of outpatient contacts, and an ~11‐fold increase in hospital bed‐day use. At 1 year, incident AF was associated with an ~4‐fold increased risk of at least one inpatient contact, a 43% increased risk of at least one outpatient contact, an ~2‐fold increase in respective numbers of inpatient and outpatient contacts, and >5‐fold increase in bed‐day use. In our subgroup analyses, we found statistical evidence of higher risk of first outpatient contact among men at 1 year, higher risk of first hospital contact among the younger patients at 1 year, and a higher number of inpatient contacts among women and patients aged 75 years or older at both 30 days and 1 year. Additionally, patients with HF for more than 1 year had higher risk of first hospital contact and higher utilization of bed‐days, but the number of hospital contacts was higher for patients with HF for up to 1 year. We noted no differences of clinical significance by baseline LVEF and educational level. Our analysis of oral anticoagulant use showed that only about half of the patients redeemed a prescription of oral anticoagulant within 3 months after AF, but the proportion increased by almost 10 percentage points in the study period.

A diagnosis of incident AF in HF is likely a risk marker that represents several interrelated clinical and non‐clinical aspects rather than being a strong causal factor for hospital needs per se. First, incident AF may represent progression of HF disease associated with clinical and haemodynamic deterioration, such as pulmonary congestion and/or hypotension, and AF may be diagnosed before or after acute decompensation. Second, HF and AF share several risk factors, such as hypertension, obesity, and ischaemic heart disease, and incident AF may mark the intensity and/or the duration of exposure. Third, frailty and prevalent AF commonly coexist, , , and incident AF may be a manifestation of decreased physiological reserve or a consequence of frailty‐related diagnostic evaluation. Frailty in AF is associated with high mortality , ; however, no study has to our knowledge examined the association between frailty and hospital utilization in AF. Fourth, HF patients often suffer from multimorbidity, which may lead to frequent hospital contact and earlier detection of undiagnosed AF.

Data on hospital utilization associated with incident AF in HF are sparse. Results from the Beta‐blocker Evaluation of Survival Trial have shown that incident AF was associated an ~2‐fold increase in all‐cause hospitalizations measured as days per patient. However, hospitalization days did not account for the risk time pertained to each patient, and the analysis was not based on real‐world data. Analyses of the HF Long‐Term Registry of European Society of Cardiology showed that prevalent AF was associated with HF hospitalizations with the following hazard ratios by LVEF: 1.04 (95% CI 0.89–1.21) if LVEF < 40%, 1.43 (95% CI 1.09–1.88) if LVEF 40–49%, and 1.49 (95% CI 1.20–1.85) if LVEF ≥ 50%. However, information on AF was limited to a prevalent disease, and it is likely that such patients represent a preselected survivor group. Moreover, information of HF hospitalizations was only available for 84% of the study participants. Finally, the definition of HF hospitalization only included the first event during follow‐up and did not reflect the total burden.

Analyses based on data from the PARADIGM‐HF study reported that incident AF in HF was associated with a hazard ratio of 2.11 (95% CI 1.58–2.81) for HF hospitalization. Based on data from the CHARM trial, incident AF was associated HF hospitalization both among patients with preserved ejection fraction (LVEF > 40%) and patients with low ejection fraction (LVEF < 40). A study based on data from the COMET trial has reported an association between baseline AF and the composite of hospitalizations and death. However, the analyses from PARADIGM‐HF and CHARM did not include other reasons for hospitalizations, such as AF‐related contacts. None of the studies included separate data on outpatient contacts. Furthermore, the analyses only included first HF hospitalization.

Our results demonstrated important time‐related insights into the clinical course. Strength of the association between incident AF and the risk of having at least one contact attenuated from 30 days to 1 year. Furthermore, the association with days spent in hospital bed declined substantially at 1 year of follow‐up. However, the respective number of inpatient and outpatient contacts increased from 30 days to 1 year. Our data suggest that most incident AF patients rapidly needed hospital contact that often requires admission for several days, and the number of contacts was higher after 30 days up to 1 year but with fewer bed‐days. The distinction between inpatient and outpatient care is important because the location of care represents the severity of disease, as the prognosis associated with inpatients with HF is poor, and the economic costs are higher for inpatient care among HF patients worldwide. We observed a higher risk of at least one inpatient contact at 1 year and a higher number of outpatient contacts at 30 days.

Our study demonstrates that incident AF constitutes a substantial disease burden among patients with HF from a short‐term and long‐term perspective. The high proportion of patients that develops AF after HF place demands on health service resources. The high numbers of excess hospital contacts and days spent in hospital bed are central to inform policy decisions and planning of strategies to improve care pathways and expectedly reduce strain on healthcare. National spending data from the USA in 2016 have shown that the healthcare spending of HF was $51.2bn for inpatient care and $6.9bn for outpatient care, whereas the spending of AF and atrial flutter was $29.8bn for inpatient care and $29.4bn for outpatient care. However, the economic burden of incident AF in HF is yet unclarified and still needs examination to complement decisions.

Our results also underline the importance of identifying targets for preventive efforts of AF in HF. AF and HF have complex interrelations, with both conditions sharing risk factors and adverse prognosis, and both conditions predisposing to each other. We have recently reported analysing observational data that HF patients that received guideline‐based care of high quality have lower risk of incident AF. As recently underscored in a National Heart, Lung, and Blood Institute research statement, there is an imperative to investigate strategies for diagnosing and preventing AF in individuals with HF. For instance, after a diagnosis of HF, there is uncertainty regarding the most appropriate strategies to screen for AF, and it is unclear whether early detection of asymptomatic AF will prevent hospitalizations and complications and reduce mortality rate. Further, the evidence base to prevent AF, particularly in individuals with HF with preserved LVEF, is lacking. The excess hospital utilization in HF patients with new‐onset AF provides further motivation to pursue randomized controlled trials of standard and novel treatments to prevent AF in patients with HF in the setting of reduced and preserved LVEF. Incident AF in HF may also suggest the need to optimize and/or intensify medical treatment for HF. In a recent scientific statement on HF patients with reduced LVEF and AF, the American Heart Association underscores that the treatment strategy should include maximally tolerated guideline‐directed HF therapy and lifestyle and aggressive risk factor management, in addition to therapy targeted for AF.

Limitations

When excluding patients with history of AF at the diagnosis of HF, we missed patients with AF whose diagnosis was not recorded in the registries. The time of the development of AF may be wrong, as we only have information on the time of the diagnosis, a point in time that could mark symptom worsening or onset of another morbidity. Furthermore, we were unable to classify the type of AF. Information on clinical characteristics including LVEF was ascertained at the diagnosis of HF and may have changed over years.

As we analysed observational, non‐randomized, data; we cannot prove causal relations. Even though we included and adjusted for several important covariates, we cannot rule out residual confounding. For instance, we were unable to adjust for body mass index, severity of concomitant disease, systolic blood pressure, natriuretic peptides, and measures of renal function.

Lastly, generalizability may be limited by the rather restrictive inclusion and exclusion criteria of the DHFR and the fact that the population consisted mainly of European ancestry. The prevalence of HF with preserved LVEF was lower in our study than reported in most population‐ or community‐based studies. For example, the Framingham Heart Study has demonstrated a higher burden of prevalent AF among HF patients with preserved LVEF and that may account for the lower proportion of patients with preserved LVEF. Furthermore, different care practices and organizations between countries may influence whether patients are referred to inpatient or outpatient care. The low proportion of users of loop diuretics may reflect mild disease and thereby reduce the generalizability of our findings.

Conclusions

This study underscores that incident AF in HF patients is associated with a high level of inpatient and outpatient hospital service utilization. Patients with incident AF have a shorter time to first hospital contact, more hospital contacts, and greater utilization of hospital bed‐days. The high proportion of patients that develop AF after HF places demands on health service resources. Future studies need to address the screening and prevention of AF in HF and optimization of care pathways.

Conflict of interest

N.V., P.C., E.J.B., L.T., and S.P.J. declared no conflicts of interest. G.Y.H.L. is a consultant and speaker for BMS/Pfizer, Boehringer Ingelheim, and Daiichi Sankyo. No fees are received personally. L.F. is an advisory board for Pfizer, BMS, and MSD.

Funding

This project was supported by grants from the Health Research Fund of Central Denmark Region (R38‐A1385‐B1044) and the Danish Heart Foundation (Hjerteforeningen) (16‐R107‐A3987). E.J.B. is supported in part by the National Institutes of Health, National Heart, Lung, and Blood Institute (2R01 HL092577, 1R01 HL141434, and 2U54HL120163), the National Institutes of Health National Institute on Aging (1R01AG066010, and 1R01AG066914), and the American Heart Association, (AHA_18SFRN34110082). L.F. is supported by the Health Research Fund of Central Denmark Region. L.T. is supported by the American Heart Association, 18SFRN34150007.

Table S1. Definitions of covariates and data sources.

Table S2. Definition of diagnoses according to ICD.

Table S3. ATC codes used to define medications

Figure S1. Illustration of matching.

Covariates

Table S4. Balance of covariates in main analysis.

Table S5. Comorbidities of the Charlson Comorbidity Score.

Subgroup analyses by sex

Table S6. Characteristics of HF patients with incident AF and matched referents at the index date, by sex.

Table S7. Incident AF and time to first hospital contact, by sex.

Table S8. Incident AF and number of hospital contacts, by sex.

Table S9. Incident AF and proportion of days spent in hospital bed, by sex.

Subgroup analyses by age

Table S10. Characteristics of HF patients with incident AF and matched referents at the index date, by age group.

Table S11. Incident AF and time to first hospital contact, by age group.

Table S12. Incident AF and number of hospital contacts, by age group.

Table S13. Incident AF and proportion of days spent in hospital bed, by age group.

Subgroup analyses by LVEF

Table S14. Characteristics of HF patients with incident AF and matched referents at the index date, by LVEF group.

Table S15. Incident AF and time to first hospital contact, by LVEF group.

Table S16. Incident AF and number of hospital contacts, by LVEF group.

Table S17. Incident AF and proportion of days spent in hospital bed, by LVEF group.

Table S18. Incident AF and time to first hospital contact in patients with LVEF <25%.

Table S19. Incident AF and number of hospital contacts in patients with LVEF <25%.

Table S20. Incident AF and proportion of days spent in hospital bed in patients with LVEF<25%.

Subgroup analyses by time with HF

Table S21. Characteristics of HF patients with incident AF and matched referents at the index date, by time with HF.

Table S22. Incident AF and time to first hospital contact by time with HF.

Table S23. Incident AF and number of hospital contacts by time with HF.

Table S24. Incident AF and proportion of days spent in hospital bed by time with HF.

Subgroup analyses by level of education

Table S25. Characteristics of HF patients with incident AF and matched referents at the index date, by level of education.

Table S26. Incident AF and time to first hospital contact by level of education.

Table S27. Incident AF and number of hospital contacts by level of education.

Table S28. Incident AF and proportion of days spent in hospital bed by level of education.

Incident AF registered in an outpatient or inpatient setting

Table S29. Incident AF (inpatient diagnosis) and time to first hospital contact.

Table S30. Incident AF (outpatient diagnosis) and time to first hospital contact.

Table S31. Incident AF (inpatient diagnosis) and number of hospital contacts.

Table S32. Incident AF (outpatient diagnosis) and number of hospital contacts.

Table S33. Incident AF (inpatient diagnosis) and proportion of days spent in hospital bed.

Table S34. Incident AF (outpatient diagnosis) and proportion of days spent in hospital bed.

Incident AF and HF‐related hospital contacts

Table S35. Incident AF and time to first HF‐related hospital contact.

Table S36. Incident AF and number of HF‐related hospital contacts.

Table S37. Incident AF and proportion of HF‐related days spent in a hospital bed.

Click here for additional data file.