Background

Atrial fibrillation is the most common cardiac arrhythmia and is associated with significant morbidity and mortality related to stroke and thromboembolism. The prevalence of atrial fibrillation is estimated to be 0.1% in persons younger than 55 years and as high as 10% in those older than 85 years. Incidence of atrial fibrillation is expected to increase as the population ages and the occurrence of risk factors increases.

The most devastating and potentially fatal consequence of atrial fibrillation is ischemic stroke. Strokes in atrial fibrillation patients are often more severe than other types of stroke, with more instances of disability or death. Assessing the atrial fibrillation stroke risk depends on the evaluation of patient characteristics, comorbidities, and clinical features. Older age and history of stroke, transient ischemic attack, or thromboembolism are the strongest independent predictors of stroke in atrial fibrillation patients. Additionally, diabetes, female sex, vascular disease, and congestive heart failure are associated with elevated stroke risk among persons with atrial fibrillation.

Risk assessment scores have been developed to quantify stroke risk in patients with atrial fibrillation. The European Society of Cardiology, the American College of Cardiology and American Heart Association, and the National Institute for Health and Care Excellence recommend the use of CHA ₂ DS ₂ -VASc (congestive heart failure/left ventricular dysfunction, hypertension, ≥75 years, diabetes mellitus, previous stroke or transient ischemic attack or thromboembolism, vascular disease, aged 65 to 74 years, sex female) as the preferred risk-scoring method to gauge stroke risk in atrial fibrillation patients. Using stroke risk scores allows targeted interventions to aid in the prevention of thrombotic events, namely, through the use of oral anticoagulants. Clinical studies provide evidence of therapeutic options at a population level. However, in clinical practice, treatment options must be considered from the perspective of the individual patient and require strategic personalization that considers the unique constellation of risks and benefits, of which few are typically represented in clinical trials.

With the availability of anticoagulation that does not require heparin bridging, consequent to medications that are effective within 2 hours, the need for hospitalization for anticoagulation is rapidly declining. Thus, in selected patients the emergency physician will need to choose between a hospital admission of little benefit (besides administration of a daily pill) or discharge with the patient receiving oral anticoagulation. Knowledge of the risks and benefits of oral anticoagulants is becoming critical. In treating atrial fibrillation patients at risk for stroke, clinicians must determine whether the net clinical benefit from anticoagulation therapy outweighs its risk, namely, bleeding.

Importance

Clinicians must weigh many factors when deciding whether a patient should receive anticoagulation therapy. Assessing the risk for any given patient requires a personalized assessment of his or her risk of stroke without therapy versus the risk of bleeding with therapy. CHA ₂ DS ₂ -VASc scores have been proven to quantify stroke risk ; these scores may also have a relationship with bleeding risk. Understanding the relationship between CHA ₂ DS ₂ -VASc scores and the incidence of major bleeding may provide another valuable tool in the clinical decisionmaking process.

Goals of This Investigation

The current evaluation was derived from an ongoing, 5-year pharmacovigilance study, designed to provide longitudinal safety data by actively obtaining information associated with rivaroxaban use in the postapproval setting, complementary to the clinical trial data and that being collected by the spontaneous adverse event reporting process. Our objective for this analysis was to examine the relationship between CHA ₂ DS ₂ -VASc scores and the incidence of nontrauma-related major bleeding in rivaroxaban users with nonvalvular atrial fibrillation.

Study Design

This observational study used a retrospectively assembled cohort of patients diagnosed with nonvalvular atrial fibrillation. Within this population, patients who were exposed to rivaroxaban were included in the study analyses.

Selection of Participants

The patient population for this analysis was derived from the US Department of Defense electronic medical records system. The Department of Defense Military Health System covers military service members and their families and represents one of the largest US health systems, with nearly 10 million active patients. The system is not linked with data streams from Veterans Affairs because it and the Military Health System are separate entities; therefore, this study population does not contain data from the Veterans Affairs patient population.

The Military Health System population has a relatively high representation of elderly, with patients older than 65 years making up more than 20% of the total compared with 13% of the overall US population. The Military Health System databases consist of longitudinal electronic medical records that are continually updated and contain administrative, pharmacy, laboratory, and clinical data. Inpatient and outpatient data are integrated, which ensures robust capture of medical encounters, including major bleeding events, along with associated pharmacy data. As described previously, the Department of Defense health care system is paperless, and all patient-level medical information and clinical data are kept in one electronic format. Incoming records are checked against a set of minimum standards for quality to ensure the integrity of the data, and duplicate records are removed before analysis.

Patients insured through the Military Health System are not required to use military medical facilities. In fact, many insured members use their Military Health System coverage to obtain care in nonmilitary (civilian) facilities. If a patient insured by the Military Health System is hospitalized anywhere for any reason, the claim and related clinical information are routed back to the Department of Defense Military Health System databases, thus limiting the probability of undetected events.

Study Interval and Patient Eligibility

This study analyzed data from the 2.5-year observation period of January 1, 2013, to June 30, 2015, among patients with nonvalvular atrial fibrillation diagnoses who were receiving rivaroxaban. The information included in the analyses for this study cohort was identified through relevant International Classification of Diseases, Ninth Revision, Clinical Modification ( ICD-9-CM ) diagnosis and procedure codes and Common Procedure Terminology and Healthcare Common Procedure Coding System procedure codes that were listed in any available procedure field within any medical encounter record. The ICD-9 code 427.31 was used to identify patients with atrial fibrillation, and a separate set of diagnosis and procedure codes was used to further distinguish nonvalvular atrial fibrillation patients.

All patients meeting the definition for nonvalvular atrial fibrillation were included, regardless of incident or prevalent rivaroxaban usage, as long as the patient was identified as having nonvalvular atrial fibrillation before or concurrent with rivaroxaban usage. Patients were evaluated for major bleeding during the study period or censored at the earliest occurrence of any of 4 events: a major bleeding event, death, loss of Military Health System eligibility, or end of study.

Outcome Measures

The primary outcome of interest was major bleeding, as defined by the Cunningham algorithm. Cunningham et al developed a validated database algorithm using administrative data for identification of major bleeding events that result in a hospitalization. Per the algorithm, bleeding events determined to be trauma related were excluded. The Cunningham algorithm identifies bleeding-related hospitalizations from the primary discharge diagnosis, with the types of major bleeding events considered including gastrointestinal bleeding, hemorrhagic strokes, other intracranial bleeding events, and bleeding at other sites. The use of bleeding diagnoses showed a positive predictive value of 89% to 99% in the validation study by Cunningham et al, and this algorithm has been used in other clinical studies to identify serious bleeding events.

Although the Cunningham algorithm is a comprehensive and validated case-finding tool, the definition for major bleeding in this study is not an exact match with the clinical trial (Rivaroxaban Once Daily Oral Direct Factor Xa Inhibition Compared with Vitamin K Antagonism for Prevention of Stroke and Embolism Trial in Atrial Fibrillation [ROCKET-AF]) definition because the algorithm is applied retrospectively and relies on the information available in the electronic medical record. As a comparison, the clinical trial definition of major bleeding was defined as clinically overt bleeding associated with a decrease in hemoglobin level of 2 g/dL or more, or a transfusion of 2 or more units of packed RBCs or whole blood, or major bleeding in a critical site, or a fatal outcome.

Major bleeding events were included if they occurred anytime during rivaroxaban exposure plus 7 days postdiscontinuation. The drug exposure period for patients was captured from the initiation date of rivaroxaban therapy until its discontinuation. For patients who experienced a major bleeding event, data were collected on major bleeding management, fatal outcomes, and discharge status.

Primary Data Analysis

Descriptive statistics are presented on patient demographics, comorbidities, rivaroxaban usage, hospitalization and bleeding-event management data, bleeding-event characteristics, and outcomes. Because the database contained limited data on smoking, alcohol use, race, and body mass index, these variables are not presented.

Categorical variables were summarized with counts and proportions. Mean values, along with SD, were calculated for continuous variables. No a priori hypothesis testing was planned or performed.

The incidence rates of major bleeding plus 95% confidence intervals (CIs) were calculated using a person-time approach: the number of patients with a first episode of major bleeding divided by the exposure time at risk, reported in person-centuries. Subsequent major bleeding events in patients experiencing multiple bleeding events were not included in the incidence rate calculations or related analyses. Baseline characteristics of the study population were stratified and evaluated by CHA ₂ DS ₂ -VASc scores and major bleeding status (major bleeding versus no major bleeding).

CHA ₂ DS ₂ -VASc scores were calculated for each study patient, using the schema presented in Table 1 , with age defined at bleeding for major bleeding cases and at the end of study participation for patients without major bleeding. Major bleeding incidence rates were also calculated by bleeding site and for each CHA ₂ DS ₂ -VASc score category.

Comorbidities were defined within a 6-month period before the date of bleeding event for major bleeding cases and at the end of study participation for patients without major bleeding. The major bleeding incidence rate was studied by score category (0 to ≥5), as well as by each component of CHA ₂ DS ₂ -VASc scoring.

Data captured for this study were subject to multiple quality control measures. The Department of Defense houses all of the queryable data sources within the Military Health System Data Repository. Incoming records transmitted to the repository are verified to be constructed in a standardized format respective to the data source, and all data elements are inspected for acceptable formats and to ensure that they are devoid of improbable values. Records that meet the minimum standards are then processed according to a set of published business rules and transformed into standardized data sets.

The Military Health System Data Repository is a SAS-computable environment, and all data sets within the repository are linked by a common, unique patient identifier. Two SAS programmers worked in parallel to independently draft code necessary to query and analyze all data pertinent to this study.

This postmarketing safety surveillance study was part of a postmarketing commitment, which was funded by Janssen Scientific Affairs, LLC and Bayer HealthCare. The study protocol, including the ascertainment method of major bleeding events, was reviewed and approved by the Food and Drug Administration before its finalization. The research data were derived from an approved Naval Medical Center, Portsmouth, Virginia institutional review board protocol, and the research was conducted in compliance with federal and state laws, including the Health Insurance Portability and Accountability Act of 1996.

Health ResearchTx, LLC carried out all analyses with SAS (version 9.2; SAS Institute, Inc., Cary, NC).

Study Design

This observational study used a retrospectively assembled cohort of patients diagnosed with nonvalvular atrial fibrillation. Within this population, patients who were exposed to rivaroxaban were included in the study analyses.

Selection of Participants

The patient population for this analysis was derived from the US Department of Defense electronic medical records system. The Department of Defense Military Health System covers military service members and their families and represents one of the largest US health systems, with nearly 10 million active patients. The system is not linked with data streams from Veterans Affairs because it and the Military Health System are separate entities; therefore, this study population does not contain data from the Veterans Affairs patient population.

The Military Health System population has a relatively high representation of elderly, with patients older than 65 years making up more than 20% of the total compared with 13% of the overall US population. The Military Health System databases consist of longitudinal electronic medical records that are continually updated and contain administrative, pharmacy, laboratory, and clinical data. Inpatient and outpatient data are integrated, which ensures robust capture of medical encounters, including major bleeding events, along with associated pharmacy data. As described previously, the Department of Defense health care system is paperless, and all patient-level medical information and clinical data are kept in one electronic format. Incoming records are checked against a set of minimum standards for quality to ensure the integrity of the data, and duplicate records are removed before analysis.

Patients insured through the Military Health System are not required to use military medical facilities. In fact, many insured members use their Military Health System coverage to obtain care in nonmilitary (civilian) facilities. If a patient insured by the Military Health System is hospitalized anywhere for any reason, the claim and related clinical information are routed back to the Department of Defense Military Health System databases, thus limiting the probability of undetected events.

Study Interval and Patient Eligibility

This study analyzed data from the 2.5-year observation period of January 1, 2013, to June 30, 2015, among patients with nonvalvular atrial fibrillation diagnoses who were receiving rivaroxaban. The information included in the analyses for this study cohort was identified through relevant International Classification of Diseases, Ninth Revision, Clinical Modification ( ICD-9-CM ) diagnosis and procedure codes and Common Procedure Terminology and Healthcare Common Procedure Coding System procedure codes that were listed in any available procedure field within any medical encounter record. The ICD-9 code 427.31 was used to identify patients with atrial fibrillation, and a separate set of diagnosis and procedure codes was used to further distinguish nonvalvular atrial fibrillation patients.

All patients meeting the definition for nonvalvular atrial fibrillation were included, regardless of incident or prevalent rivaroxaban usage, as long as the patient was identified as having nonvalvular atrial fibrillation before or concurrent with rivaroxaban usage. Patients were evaluated for major bleeding during the study period or censored at the earliest occurrence of any of 4 events: a major bleeding event, death, loss of Military Health System eligibility, or end of study.

Outcome Measures

The primary outcome of interest was major bleeding, as defined by the Cunningham algorithm. Cunningham et al developed a validated database algorithm using administrative data for identification of major bleeding events that result in a hospitalization. Per the algorithm, bleeding events determined to be trauma related were excluded. The Cunningham algorithm identifies bleeding-related hospitalizations from the primary discharge diagnosis, with the types of major bleeding events considered including gastrointestinal bleeding, hemorrhagic strokes, other intracranial bleeding events, and bleeding at other sites. The use of bleeding diagnoses showed a positive predictive value of 89% to 99% in the validation study by Cunningham et al, and this algorithm has been used in other clinical studies to identify serious bleeding events.

Although the Cunningham algorithm is a comprehensive and validated case-finding tool, the definition for major bleeding in this study is not an exact match with the clinical trial (Rivaroxaban Once Daily Oral Direct Factor Xa Inhibition Compared with Vitamin K Antagonism for Prevention of Stroke and Embolism Trial in Atrial Fibrillation [ROCKET-AF]) definition because the algorithm is applied retrospectively and relies on the information available in the electronic medical record. As a comparison, the clinical trial definition of major bleeding was defined as clinically overt bleeding associated with a decrease in hemoglobin level of 2 g/dL or more, or a transfusion of 2 or more units of packed RBCs or whole blood, or major bleeding in a critical site, or a fatal outcome.

Major bleeding events were included if they occurred anytime during rivaroxaban exposure plus 7 days postdiscontinuation. The drug exposure period for patients was captured from the initiation date of rivaroxaban therapy until its discontinuation. For patients who experienced a major bleeding event, data were collected on major bleeding management, fatal outcomes, and discharge status.

Primary Data Analysis

Descriptive statistics are presented on patient demographics, comorbidities, rivaroxaban usage, hospitalization and bleeding-event management data, bleeding-event characteristics, and outcomes. Because the database contained limited data on smoking, alcohol use, race, and body mass index, these variables are not presented.

Categorical variables were summarized with counts and proportions. Mean values, along with SD, were calculated for continuous variables. No a priori hypothesis testing was planned or performed.

The incidence rates of major bleeding plus 95% confidence intervals (CIs) were calculated using a person-time approach: the number of patients with a first episode of major bleeding divided by the exposure time at risk, reported in person-centuries. Subsequent major bleeding events in patients experiencing multiple bleeding events were not included in the incidence rate calculations or related analyses. Baseline characteristics of the study population were stratified and evaluated by CHA ₂ DS ₂ -VASc scores and major bleeding status (major bleeding versus no major bleeding).

CHA ₂ DS ₂ -VASc scores were calculated for each study patient, using the schema presented in Table 1 , with age defined at bleeding for major bleeding cases and at the end of study participation for patients without major bleeding. Major bleeding incidence rates were also calculated by bleeding site and for each CHA ₂ DS ₂ -VASc score category.

Comorbidities were defined within a 6-month period before the date of bleeding event for major bleeding cases and at the end of study participation for patients without major bleeding. The major bleeding incidence rate was studied by score category (0 to ≥5), as well as by each component of CHA ₂ DS ₂ -VASc scoring.

Data captured for this study were subject to multiple quality control measures. The Department of Defense houses all of the queryable data sources within the Military Health System Data Repository. Incoming records transmitted to the repository are verified to be constructed in a standardized format respective to the data source, and all data elements are inspected for acceptable formats and to ensure that they are devoid of improbable values. Records that meet the minimum standards are then processed according to a set of published business rules and transformed into standardized data sets.

The Military Health System Data Repository is a SAS-computable environment, and all data sets within the repository are linked by a common, unique patient identifier. Two SAS programmers worked in parallel to independently draft code necessary to query and analyze all data pertinent to this study.

This postmarketing safety surveillance study was part of a postmarketing commitment, which was funded by Janssen Scientific Affairs, LLC and Bayer HealthCare. The study protocol, including the ascertainment method of major bleeding events, was reviewed and approved by the Food and Drug Administration before its finalization. The research data were derived from an approved Naval Medical Center, Portsmouth, Virginia institutional review board protocol, and the research was conducted in compliance with federal and state laws, including the Health Insurance Portability and Accountability Act of 1996.

Health ResearchTx, LLC carried out all analyses with SAS (version 9.2; SAS Institute, Inc., Cary, NC).