Methods

We conducted a population-based, retrospective, cohort study of patients aged ≥65 years admitted to a hospital with a primary or secondary diagnosis of AF from 1998 to 2012, in Quebec province, Canada. Quebec residents have universal access to hospital care and physician services and those aged ≥65 years have universal prescription drug coverage. We obtained Institutional Review Board approval from McGill University Faculty of Medicine, Canada.

In our study cohort, we identified patients with a primary or secondary diagnosis of AF according to the International Classification of Diseases —ninth and/or tenth revision codes (427.3, 427.31, or 427.32/I48). For patients with >1 eligible admission with a diagnosis of AF, we considered the date of the first admission with a diagnosis of AF as the index date of entry into the study cohort. To avoid selecting patients with transient AF, we excluded patients for whom AF was listed as a postadmission complication who were likely to have perioperative AF (defined as having pericardial surgery, coronary artery bypass surgery, or structural cardiac repair within 30 days before AF hospitalization) or who had thyrotoxicosis or hyperthyroidism within 12 months before and including AF hospitalization. We also excluded patients who were residents of long-term care facilities and who did not have a valid health card number.

We determined patients’ baseline characteristics, outcome data, and drug prescriptions from linkage between the provincial hospital discharge database (Maintenance et Exploitation des Données pour l’Étude de la Clientèle Hospitalière) and the provincial physician and prescription claims database (la Régie de l’assurance maladie du Québec [RAMQ]). The Quebec prescription claims database (RAMQ) provides highly accurate information on dispensed outpatient medications. We used drug identification numbers to identify drug prescriptions from the RAMQ.

We grouped selected AF cohort into patients with and without HF using validated database codes. International Classification of Diseases -9 and -10 revision codes for HF have been previously validated with an overall positive predictive value ranging from 84% to 100% and 81% to 93% respectively, where most of the studies determined HF with a primary hospital discharge diagnosis. We further grouped patients with and without HF into digoxin and no-digoxin treatment groups according to the first prescription filled for digoxin within 30 days after AF hospital discharge. We selected a 30-day window period to capture most patients with the first prescription for digoxin after AF hospital discharge, while minimizing the potential for survival bias. For the main analyses, we started the follow-up period 30 days after AF hospital discharge (from the first day after the 30-day window period), in which we only included patients who remained alive at 30 days after AF hospital discharge. We also performed sensitivity analyses in which we started follow-up from the first day after AF hospital discharge. The outcome of interest was all-cause mortality during the follow-up period. For all patients, we determined demographic characteristics and co-morbidities at and within 1 year before AF hospitalization using validated codes (whenever possible). We obtained information on prescriptions filled for digoxin, β blocker, calcium channel blocker (diltiazem and verapamil), rhythm control drugs (amiodarone and sotalol), diuretic, angiotensin-converting enzyme inhibitor (ACEi), angiotensin receptor blocker (ARB), statin, warfarin, aspirin, clopidogrel, and nonsteroidal anti-inflammatory drugs within 30 days after AF hospital discharge.

Descriptive analyses were used to compare demographic characteristics, co-morbidities, and prescription for medications between digoxin and no-digoxin users in the group of patients with and without HF. We presented continuous variables as mean ± SD and dichotomous variables as number (%). We also performed descriptive analyses for propensity score–matched digoxin and no-digoxin users in the group of patients with and without HF.

The propensity score matching was used to select digoxin and no-digoxin treatment groups that were well balanced on various patient-related baseline characteristics. The propensity score indicated the likelihood of receiving digoxin given that a particular patient-related characteristic is present. We used multivariable logistic regression models to derive individual propensity scores for the group of patients with and without HF. For each patient in the digoxin or no-digoxin treatment group within the group of patients with and without HF, we derived a propensity score for receiving digoxin from the following variables: age (years), gender, type of AF admission (primary diagnosis vs secondary diagnosis), co-morbidities 1 year before and at AF admission (acute myocardial infarction [AMI], coronary artery disease [CAD], hypertension, diabetes, valvular heart disease, chronic kidney disease [CKD], cancer, hypothyroidism, chronic obstructive pulmonary disease, and liver disease), history of stroke and/or transient ischemic attack, history of bleeding, and use of β blocker, diltiazem, verapamil, amiodarone, sotalol, diuretic, ACEi, ARB, statin, warfarin, aspirin, clopidogrel, and nonsteroidal anti-inflammatory drugs. We used the nearest available pair-matching method with a greedy algorithm. In greedy matching, if a patient prescribed digoxin is selected, matching is attempted with the “nearest” patient without a digoxin prescription according to their propensity score.

In both the propensity score–matched group of patients with and without HF, we conducted unadjusted and adjusted multivariable Cox proportional hazards regression analyses to determine association between digoxin use and all-cause mortality (primary analyses). In the multivariable analyses, we adjusted for the following variables: age (years), gender, AMI, CAD, hypertension, diabetes, valvular heart disease, CKD, and use of β blocker, diltiazem, verapamil, amiodarone, diuretic, ACEi, ARB, and warfarin. In both the propensity score–matched group of patients with and without HF, we also performed tests of interaction between digoxin use and gender, CKD, and use of various cardiovascular medications such as β blocker, diltiazem, verapamil, amiodarone, diuretic, ACEi, and ARB, in which we adjusted for all factors mentioned previously except the subgroup factor. In both the propensity score–matched group of patients with and without HF, a significant test of interaction with digoxin use (p value <0.05) was followed by subgroup analyses for that variable (i.e., subgrouping of the relevant cohort [HF or no-HF] according to the presence or absence of that variable and multivariable Cox proportional hazards regression analyses). In Cox proportional hazards models, we considered digoxin use versus no digoxin use as a time-fixed binary variable, where we assumed that patients who were prescribed digoxin within 30 days of AF hospital discharge remained on the digoxin prescription throughout the follow-up period. This approach, using a time-fixed binary variable, is trying to mimic an intention-to-treat analysis in randomized controlled trials.

Results are expressed as hazard ratios (HRs) for Cox proportional hazards regression with their 95% confidence intervals (CIs). A p value <0.05 for the test of interaction was considered statistically significant. We conducted all statistical analyses using SAS 9.2 (SAS Institute, Cary, North Carolina).

Results

In the unmatched cohort of patients with concomitant AF and HF, digoxin users (n = 15,181), compared with no-digoxin users (n = 24,331), were predominantly women; had less AMI, CAD, hypertension, diabetes, and CKD; were more frequently prescribed diltiazem, diuretics, ACEi, and warfarin; and less frequently prescribed β blocker, amiodarone, statin, aspirin, and clopidogrel ( Table 1 ). The propensity score–matched cohort of patients with concomitant AF and HF comprised 13,986 digoxin users and 13,986 no-digoxin users, which were well balanced on baseline characteristics ( Table 1 ).

In the unmatched cohort of patients with AF and without HF, digoxin users (n = 23,200), compared with no-digoxin users (n = 77,399), were predominantly women; had more valvular heart disease and chronic obstructive pulmonary disease; had less AMI, CAD, hypertension, and CKD; were more frequently prescribed diltiazem, verapamil, diuretics, ACEi, and warfarin; less frequently prescribed β blocker, amiodarone, sotalol, ARB, statin, aspirin, and clopidogrel ( Table 2 ). The propensity score–matched cohort of patients with AF and without HF comprised 23,131 digoxin users and 23,131 no-digoxin users, which were well balanced on baseline characteristics ( Table 2 ).

Table 2

Baseline characteristics of patients with atrial fibrillation and without heart failure

	Unmatched Cohort		Propensity Matched Cohort
	Digoxin, N = 23,200	No Digoxin, N = 77,399	Digoxin, N = 23,131	No Digoxin, N = 23,131
Patients diagnosed with atrial fibrillation:
Atrial fibrillation as a main diagnosis	6,244 (26.9%)	21,569 (27.9%)	6,232 (26.9%)	6,302 (27.2%)
Patients characteristics:
Age at the index atrial fibrillation admission in years, mean ± standard deviation	79.4 ± 7.2	78.6 ± 7.2	79.4 ± 7.2	79.4 ± 7.3
Male	9,920 (42.8%)	37,564 (48.5%)	9,908 (42.8%)	10,026 (43.3%)
Length of hospitalization in days, mean ± standard deviation	13.9 ± 19.7	12.9 ± 23.8	13.9 ± 19.7	13.3 ± 20.6
Co-morbidities:
Acute myocardial infarction	2,406 (10.4%)	9,973 (12.9%)	2,403 (10.4%)	2,429 (10.5%)
Coronary artery disease	8,406 (36.2%)	30,482 (39.4%)	8,390 (36.3%)	8,393 (36.3%)
Hypertension	12,852 (55.4%)	48,484 (62.6%)	12,844 (55.5%)	12,876 (55.7%)
Diabetes	5,396 (23.3%)	16,717 (21.6%)	5,365 (23.2%)	5,394 (23.3%)
Valvular heart disease	3,769 (16.3%)	10,282 (13.3%)	3,745 (16.2%)	3,756 (16.2%)
Chronic kidney disease	2,535 (10.9%)	10,945 (14.1%)	2,535 (11.0%)	2,535 (11.0%)
Cancer	2,867 (12.4%)	9,273 (12.0%)	2,856 (12.4%)	2,818 (12.2%)
Hypothyroidism	3,612 (15.6%)	12,557 (16.2%)	3,602 (15.6%)	3,610 (15.6%)
Chronic obstructive pulmonary disease	6,152 (26.5%)	16,042 (20.7%)	6,093 (26.3%)	6,228 (26.9%)
Liver disease	739 (3.2%)	2,360 (3.1%)	738 (3.2%)	743 (3.2%)
History of bleeding event	1,409 (6.1%)	4,669 (6.0%)	1,406 (6.1%)	1,413 (6.1%)
History of stroke	1,680 (7.2%)	6,513 (8.4%)	1,680 (7.3%)	1,653 (7.2%)
Medication prescription, 30 days post discharge:
Beta-blocker	9,287 (40.0%)	33,752 (43.6%)	9,279 (40.1%)	9,284 (40.1%)
Diltiazem	4,556 (19.6%)	12,023 (15.5%)	4,541 (19.6%)	4,566 (19.7%)
Verapamil	1,168 (5.0%)	2,081 (2.7%)	1,133 (4.9%)	1,183 (5.1%)
Amiodarone	1,321 (5.7%)	8,435 (10.9%)	1,321 (5.7%)	1,334 (5.8%)
Sotalol	824 (3.6%)	5,609 (7.3%)	824 (3.6%)	859 (3.7%)
Diuretics	11,723 (50.5%)	29,600 (38.2%)	11,655 (50.4%)	11,667 (50.4%)
Angiotensin converting enzyme inhibitor	7,262 (31.3%)	20,712 (26.8%)	7,214 (31.2%)	7,162 (31.0%)
Angiotensin receptor blocker	3,025 (13.0%)	12,646 (16.3%)	3,025 (13.1%)	2,989 (12.9%)
Statin	5,540 (23.9%)	25,342 (32.7%)	5,540 (24.0%)	5,533 (23.9%)
Warfarin	14,884 (64.2%)	43,704 (56.5%)	14,817 (64.1%)	14,823 (64.1%)
Acetylsalicylic acid (aspirin)	6,302 (27.2%)	25,324 (32.7%)	6,297 (27.2%)	6,287 (27.2%)
Non-steroidal anti-inflammatory drugs	381 (1.6%)	1,250 (1.6%)	381 (1.7%)	375 (1.6%)
Clopidogrel	647 (2.8%)	4,422 (5.7%)	647 (2.8%)	671 (2.9%)

Only gold members can continue reading. Log In or Register to continue

Share this:

• Click to share on Twitter (Opens in new window)
• Click to share on Facebook (Opens in new window)

Related

Related posts:

1. Effect of Renin-Angiotensin System Inhibitors on Long-Term Survival in Patients Treated With Beta Blockers and Antiplatelet Agents After Acute Myocardial Infarction (from the MONICA/KORA Myocardial Infarction Registry)
2. Effect of Left Ventricular Diastolic Dysfunction on Outcomes of Atrial Fibrillation Ablation
3. Should Repair of Partial Atrioventricular Septal Defect Be Delayed Until Later in Childhood?
4. Ischemic Stroke Occurring During Intercourse in Young Women on Oral Contraceptives

Stay updated, free articles. Join our Telegram channel

Join