Apixaban versus No Anticoagulation in Patients Undergoing Long-Term Dialysis with Incident Atrial Fibrillation

Study Population

This retrospective cohort study used US Renal Data System data from 2012 to 2015 (7). We identified all patients on hemodialysis or peritoneal dialysis on January 1, 2012 (prevalent population) or who were started on hemodialysis or peritoneal dialysis between January 1, 2012 and December 31, 2015 (incident population).

We next identified all individuals with a primary or secondary diagnosis of atrial fibrillation or flutter on the basis of at least one inpatient or two outpatient claims using condition-specific diagnostic codes (International Classification of Diseases, ninth and tenth revisions). To verify whether this was a new diagnosis or a preexisting comorbidity, the claims dataset (ESKD and pre-ESKD) was reviewed for the year preceding the date of entry in the cohort. We restricted our analysis to patients who (1) had a new diagnosis of atrial fibrillation after January 1, 2012; (2) were alive on day 30 postdiagnosis, in order to exclude patients who developed atrial fibrillation in the setting of a serious intercurrent illness; (3) had continuous Medicare Part A and B coverage for the year prior to diagnosis; (4) had part D enrollment for ≥6 months prior to diagnosis; (5) did not have valvular atrial fibrillation; and (6) did not have a prescription of any anticoagulant agent in the 6 months prior to diagnosis (8). Atrial fibrillation was considered to be valvular if the patient had a prosthetic valve or mitral stenosis.

Relevant baseline characteristics and medication prescription (new and prevalent) were retrieved for each patient from institutional and physician claims (the 2728 form was not used). The congestive heart failure, hypertension, age, diabetes, stroke, vascular history (CHA₂DS₂-Vasc) score and a modified hypertension, kidney disease, liver disease, stroke, prior bleeding or predisposition to bleeding, labile international normalized ratios, age, medication, alcohol use (HAS-BLED) score were calculated. The modified HAS-BLED score used all variables of the original score except labile international normalized ratios (9,10).

We compared patients treated with apixaban with patients who did not receive any anticoagulation.

Restrictions apply to the availability of these data, which were used under license for this study and, therefore, are not publicly available. The Partners Healthcare institutional review board approved the study (2016P001613/BWH) and ruled that informed consent was not necessary. This work was carried out in accordance with the Declaration of Helsinki.

Outcomes

The primary outcome was hospital admission for a new stroke (ischemic or hemorrhagic), transient ischemic attack, or systemic thromboembolism. The secondary outcomes were any stroke (ischemic or hemorrhagic), fatal or intracranial bleeding, clinically important bleeding, ischemic stroke, hemorrhagic stroke, a composite outcome of myocardial infarction or ischemic stroke, a composite outcome of all-cause mortality and the primary outcome, and all-cause mortality. We additionally examined two outcomes not expected to be related to use of apixaban to assess the specificity of our findings: pneumonia and hip fracture. The institutional inpatient claims dataset was used, and one condition-specific diagnostic code was requested for each outcome (Supplemental Table 1).

Clinically important bleeding was defined as any bleeding resulting in death; any bleeding occurring at a critical site (intracranial, intraocular, retroperitoneal, intra-articular, pericardial, airway); or any gastrointestinal, urinary tract, or gynecologic bleeding requiring hospitalization.

The follow-up period (time 0) started 30 days after atrial fibrillation diagnosis to allow enough time for patients to fill a prescription and to exclude patients who developed atrial fibrillation in the setting of a serious intercurrent illness. Patients were followed up to the date of death, kidney transplantation, or December 31, 2015. For the outcome of all-cause mortality, patients were followed up to the date of death, kidney transplantation, or August 1, 2017 because data for mortality were available until this date. Time to event data for each outcome in the apixaban and the control group are shown in Supplemental Table 2.

Statistical Analyses

Multivariable logistic regression was used to calculate a propensity score for apixaban prescription versus no anticoagulation including 23 parameters at baseline selected for their potential association with cardiovascular or bleeding outcomes (shown in Table 1). Apixaban users were matched 1:3 without replacement according to the propensity score (caliper width of 0.5%) to patients not treated with any anticoagulant agent. Standardized differences were calculated to assess balance of the matched cohorts, with a cutoff of ≤0.1 considered to indicate a negligible difference (11).

Cox proportional hazard models were used to examine the association between apixaban prescription and the clinical outcomes, considering death from any cause as a competing risk, as described by Fine and Gray (12). For the main “as-treated” analysis, we censored patients when Medicare Part A, B, or D coverage was lost. We also censored apixaban users at the date of the last available prescription plus drug supply days. To account for immortal time bias for patients prescribed apixaban at >30 days from atrial fibrillation diagnosis, we examined apixaban prescription as a time-varying covariate (13). For the primary analysis, hazard ratios (HRs) were not adjusted. For the secondary analysis, HRs were adjusted for (1) CHA₂DS₂-Vasc score for the primary outcome, ischemic stroke, any stroke, and all-cause mortality; (2) modified HAS-BLED score for fatal or intracranial bleeding, hemorrhagic stroke, and clinically important bleeding; and (3) history of hypertension, diabetes mellitus, dyslipidemia, and myocardial infarction for the composite outcome of myocardial infarction or ischemic stroke.

We performed a subgroup analysis including only patients with apixaban either 5 mg twice daily or 2.5 mg twice daily to study the effect of the standard dose and the reduced dose on relevant clinical outcomes. We also separately present outcomes for the subgroup of patients on maintenance hemodialysis.

We performed three sensitivity analyses: (1) an “intention-to-treat” analysis, where we assumed that each patient remained on apixaban throughout the whole follow-up period (14); (2) a prescription time-distribution matching to account for immortal time bias: the time to apixaban prescription was assessed for users (for each nonuser, an assigned “time to apixaban” was randomly selected from this distribution of observed times to apixaban start; the two groups were followed from the date of apixaban prescription [real or assigned] until the occurrence of an event or the end of follow-up, and nonusers with an event prior to assigned time 0 were excluded from this analysis [13]); and (3) a marginal structural model where stabilized inverse probability weights were estimated and then used in Cox proportional hazards regression. The primary analysis was also performed in a larger dataset to further include patients with prevalent atrial fibrillation at the time of dialysis initiation.

Statistical analyses were performed in SPSS Statistics (version 24.0, Released 2016; IBM Corp., Armonk, NY) or in Stata (version 14 IC; College Station, TX). P values =0.05 were considered statistically significant.