Study Population
Individuals with stage 4–5 CKD not on dialysis and prevalent non-valvular AF or atrial flutter on warfarin prior to Jan 1,2012 were identified from the USRDS12 CKD files. Prior to January 1, 2012, individuals without continuous Medicare Parts A&B coverage for a year, a warfarin prescription within six months, or continuous Medicare Part D coverage for six months were excluded. Additional exclusion criteria included the following: valvular heart disease, dialysis start date prior to January 1, 2012, or history of kidney transplantation (Fig. 1). The presence of any prescription claim for warfarin within the 6 months prior to Jan 1, 2012 was used to identify warfarin use since this criteria would include all stroke events, major bleeds, and deaths that occur before the next prescription refill and reduce the risk of bias towards a healthier population that would occur if clinical events resulted in discontinuation and the absence of subsequent prescriptions.
Individuals were followed from Jan 1,2012 through Dec 31, 2015 for warfarin to apixaban switching and endpoints. Switching to apixaban was defined by the presence of at least 1 claim for apixaban during the study period January 1, 2012 – December 31, 2015. To be consistent with common prescribing patterns, anticoagulation discontinuation was defined for both the warfarin and apixaban switch groups using the date of the last available warfarin or apixaban prescription plus 90 days to account for a typical 3-month drug supply. Follow up was censored at the 2nd switch in anticoagulant class.
Comorbidity diagnoses were based on the presence of at least one inpatient or two outpatient claims within 1 year prior to January 1, 2012 using condition-specific codes from the 9th revision of the International Statistical Classification of Diseases and Related Health Problems. Baseline characteristics and relevant medication prescriptions were retrieved from the institutional and physician claims of the non-ESRD database in the USRDS12. The CHA 2DS2 -Vasc (congestive heart failure, hypertension, age, diabetes, stroke, vascular history)13 was calculated along with the modified HAS-BLED score (hypertension, renal disease, liver disease, stroke, prior bleeding or predisposition to bleeding, age, medication, alcohol use)14,15. The modified HAS-BLED score utilizes the original score but removes the component for labile INRs as INR values are not typically available in administrative data such as the USRDS.15
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 effectiveness outcome was hospital admission for a composite of new stroke or embolism which was defined as a composite of ischemic stroke, transient ischemic attack, systemic thromboembolism, or hemorrhagic stroke16. The primary safety outcome was major bleeding defined as any bleeding resulting in death and occurring at a critical site (intracranial, intraocular, retroperitoneal, intra-articular, pericardial, airway), or resulting in hospitalization. ICD 9 and 10 codes were utilized (Supplemental Table 1). Patients were followed up to the date of death, dialysis, kidney transplantation, or on August 1, 2017, the last date for which mortality was available. In addition, follow-up was censored at the time of anticoagulation discontinuation, discontinuous Medicare part D, dialysis, kidney transplant, or a 2nd switch in anticoagulant type.
Statistical Analyses
Poisson regression was used to estimate the incidence rates of stroke, the individual components of the stroke endpoint (hemorrhagic or ischemic), and major bleeding events. The primary analysis was conducted to determine the comparative effectiveness and safety of switching from warfarin to apixaban compared to continuing warfarin. Fine and Grey’s models for competing risk survival analyses were used to compare the time to first all cause stroke or major bleeding event with death as competing risk between the two groups adjusted for potential covariates with an inverse probability treatment weighting (IPTW) approach17. Eight covariates were selected due to their potential to confound a clinician’s decision to initiate treatment with apixaban and their association with mortality: age, sex, CHA 2DS2 -Vasc score, HAS-BLED score, history of stroke, history of bleeding, history of diabetes mellitus, history of heart disease (defined as either myocardial infarction, coronary artery disease, or congestive heart failure). Predicted probabilities for switching to apixaban (versus continuing warfarin) conditional on these 8 variables were used to calculate the standardized IPTW. For apixaban-switch patients, standardized weights were generated by taking the reciprocal of the estimated probability of switching to apixaban and multiplying it by unadjusted overall probability of apixaban treatment. For patients staying on warfarin, we took the reciprocal of the probability of not switching and multiplied it by the unadjusted probability of not switching. These weights were then used in the aforementioned primary analysis along with the robust sandwich variance estimator to account for potential dependency introduced by using IPTW. Results are reported in hazard ratios (HR) with 95% confidence intervals (CIs). The follow up period started on the date of switch for warfarin and January 2, 2012 for the warfarin continuation group.
Due to the high mortality risk in advanced CKD, a sensitivity analysis was conducted to account for the potential effects of death from factors other than stroke or bleeding. Individuals who died within 90 days of January 1,2012 for the warfarin continue group and died within 90 days of the switch date for the apixaban switch group were excluded from the survival analyses. Fine and Grey’s models were used to compare the time to first stroke or major bleeding event with death as competing risk (if death occurred after 90 days since the start date of the follow up period) between the two subgroups adjusted for potential covariates with IPTW described above. The SAS version 9.4 was used for all analyses. P < 0.05 was considered significant.