Anxiety, Depression, and Adverse Clinical Outcomes in Patients With Atrial Fibrillation Starting Warfarin: Cardiovascular Research Network WAVE Study

Background Anxiety and depression are associated with worse outcomes in several cardiovascular conditions, but it is unclear whether they affect outcomes in atrial fibrillation (AF). In a large diverse population of adults with AF, we evaluated the association of diagnosed anxiety and/or depression with stroke and bleeding outcomes. Methods and Results The Cardiovascular Research Network WAVE (Community‐Based Control and Persistence of Warfarin Therapy and Associated Rates and Predictors of Adverse Clinical Events in Atrial Fibrillation and Venous Thromboembolism) Study included adults with AF newly starting warfarin between 2004 and 2007 within 5 health delivery systems in the United States. Diagnosed anxiety and depression and other patient characteristics were identified from electronic health records. We identified stroke and bleeding outcomes from hospitalization databases using validated International Classification of Diseases, Ninth Revision (ICD‐9), codes. We used multivariable Cox regression to assess the relation between anxiety and/or depression with outcomes after adjustment for stroke and bleeding risk factors. In 25 570 adults with AF initiating warfarin, 490 had an ischemic stroke or intracranial hemorrhage (1.52 events per 100 person‐years). In multivariable analyses, diagnosed anxiety was associated with a higher adjusted rate of combined ischemic stroke and intracranial hemorrhage (hazard ratio, 1.52; 95% confidence interval, 1.01–2.28). Results were not materially changed after additional adjustment for patient‐level percentage of time in therapeutic anticoagulation range on warfarin (hazard ratio, 1.56; 95% confidence interval, 1.03–2.36). In contrast, neither isolated depression nor combined depression and anxiety were significantly associated with outcomes. Conclusions Diagnosed anxiety was independently associated with increased risk of combined ischemic stroke and intracranial hemorrhage in adults with AF initiating warfarin that was not explained by differences in risk factors or achieved anticoagulation quality.

is possibly because of providers' concerns about noncompliance with a complicated medication regimen, 9 suggesting an uncertainty among healthcare providers about the risk-benefit profile of warfarin among these patients. Evidence on treatment outcomes related to warfarin in patients with mental health disorders is sparse, and understanding the impact of anxiety and depression on clinical outcomes in patients on anticoagulation treatment is crucial to improve their management by identifying those who might benefit from more intensive monitoring or alternative anticoagulation strategies. Therefore, we examined the association between diagnosed anxiety or depression with adverse clinical outcomes in adults initiating warfarin therapy for AF.

Study Design and Setting
The WAVE (Community-Based Control and Persistence of Warfarin Therapy and Associated Rates and Predictors of Adverse Clinical Events in Atrial Fibrillation and Venous Thromboembolism) project in the Cardiovascular Research Network 10 was a retrospective cohort study of adults with AF who were enrolled in 1 of 5 large integrated healthcare systems in the United States: Kaiser Permanente Northern California, which served >3.2 million members in Northern California; Kaiser Permanente Colorado, which served >460 000 members in the metropolitan area of Denver, CO; Geisinger, which served >1 million patients (of whom 450 000 members were covered by the Geisinger Health Plan) in central and northeast Pennsylvania; Marshfield Clinic, which served >550 000 members in central and northwest Wisconsin; and Harvard Pilgrim, which served >1 million members in New England during the study period. These healthcare systems represent demographically and socioeconomically diverse community-based populations in the specific geographic areas. 11 Research divisions from these healthcare delivery systems each created site-specific virtual data warehouses containing individual patient information from electronic medical records and administrative databases to promote interinstitutional research. 10 The virtual data warehouse was the primary data source for identification of patients and characterization of covariates in our study, and mortality data were obtained from State Death files and the National Death Index.
The study was approved by the institutional review boards of the participating organizations. Because of the nature of the study, a waiver of informed consent was obtained.
If requested, a complete, cleaned, deidentified copy of the data set can be made available through the Cardiovascular Research Network. For information on the data elements, format, and security, see CVRN.org. The Cardiovascular Research Network has a standard approach for data sharing that includes review of requests by the project Steering Committee based on the following: (1) high scientific merit; (2) consistency with the overall goals and objectives of the parent study; (3) provision of adequate resources to effectively complete the project, including sufficient budget to cover costs of personnel and data acquisition; (4) the requisite expertise to meet the objectives of the project; and (5) inclusion of the parent study team member(s) as part of the investigative team.

Cohort Assembly
Eligible adults had a diagnosis of AF or atrial flutter and were newly initiating warfarin therapy between January 1, 2004, andDecember 31, 2007. AF diagnoses were based on a primary hospital discharge diagnosis of AF (International Classification of Diseases, Ninth Revision [ICD-9] code 427.31 or 427.32) or ≥2 outpatient diagnoses of AF found in health plan databases. Information on warfarin prescription was obtained from dispensing data found in health plan pharmacy databases. The index date of warfarin therapy was defined as new warfarin dispensing without prior evidence of receiving warfarin within 3 years before the index dispensing date. Time on warfarin was based on the number of days supplied per prescription and intervening international normalized ratio (INR) tests using a previously validated algorithm. 12 We included adults who were ≥21 years at their index warfarin prescription date, had at least 1 outpatient INR measurement after the index date, and had ≥12 months of continuous health plan membership and pharmacy drug benefit before

Clinical Perspective
What Is New?
• The impact of anxiety and depression diagnoses on adverse outcomes in patients with atrial fibrillation anticoagulated with warfarin is unclear. • In a large cohort of adults with atrial fibrillation newly starting warfarin, diagnosed anxiety was associated with an increased risk of combined ischemic stroke and intracranial hemorrhage during follow-up, whereas depression or combined depression and anxiety was not. • This risk increase was not mediated by anticoagulation quality.
What Are the Clinical Implications?
• Diagnosed anxiety might be useful for further stratification of stroke and bleeding risk in patients with atrial fibrillation who newly start warfarin. • Future studies should investigate underlying mechanisms and assess whether different treatment approaches might relate to improved outcomes in these patients. study entry to ensure minimum necessary information on patient characteristics. Individuals with any warfarin exposure in the 3 years before study entry were excluded.

Primary Predictors: Diagnosed Anxiety and/or Depression
We considered anxiety and/or depression diagnoses that were present at entry. Study participants were considered to have prevalent anxiety or depression at baseline if they had ≥2 outpatient diagnoses within 12 months of each other or 1 inpatient primary diagnosis based on ICD-9 codes found in health plan databases. A diagnosis of anxiety was defined as the presence of any of the following ICD-9 codes: 300. Outcomes Follow-up started at the first warfarin dispensing date and ended on the date of an outcome event, withdrawal from the health plan, termination of warfarin use, death not attributable to an outcome event, or end of the study on December 31, 2007, whichever came first. Clinical events potentially related to AF or warfarin treatment were chosen as outcomes. The primary outcome was the rate of combined ischemic stroke and/or intracranial hemorrhage (ICH), the 2 most influential factors driving net clinical benefit of anticoagulation for AF. 14 We also examined as a secondary outcome hospitalizations for extracranial hemorrhage. Clinical outcomes were identified by searching computerized hospitalization and billing databases for relevant ICD-9 codes (Data S1), and only the first event of each type was recorded.  10,15 Hospitalizations for extracranial hemorrhage were identified by appropriate ICD-9 codes in the primary position. Because all patients in this cohort were enrolled in healthcare delivery systems, we were able to identify outcomes even when patients presented to institutions outside of the particular healthcare delivery network, because care provided outside of owned facilities is systematically tracked.

Covariates
We used the virtual data warehouse to obtain information on age, sex, race, relevant medical history, comorbid conditions, and other cardiovascular risk factors at baseline using previously validated approaches based on ICD-9 diagnosis and procedure codes, Current Procedure Terminology procedure codes, laboratory records, and pharmacy records. 15,16 Anticoagulation quality was defined as the percentage of time in a therapeutic INR range of 2.0 to 3.0. It was calculated as the total time that the patient had an INR between 2.0 and 3.0, estimated using a modified linear interpolation between measured INR points, divided by the total time the patient was on warfarin. 17 Baseline renal function using serum creatinine concentration values was obtained from health system laboratory databases and estimated by the glomerular filtration rate (mL/min per 1.73 m 2 ) using the Chronic Kidney Disease Epidemiology Collaboration equation. 18 We searched health plan pharmacy databases for dispensed prescriptions for anxiolytic medications (buspirone, alprazolam, chlordiazepoxide, clonazepam, clorazepate, diazepam, lorazepam, estazolam, halazepam, oxazepam, prazepam, and quazepam) and antidepressant medications available during the study period (amitriptyline, amoxapine, bupropion, citalopram, clomipramine, desipramine, doxepin, duloxetine, escitalopram, fluoxetine, fluvoxamine, imipramine, maprotiline, mirtazapine, nefazodone, nortriptyline, paroxetine, phenelzine, protriptyline, selegiline, sertraline, tranylcypromine, trazodone, trimipramine, and venlafaxine). We also searched for cardiovascular medications, including angiotensin-converting enzyme inhibitors, aldosterone receptor antagonists, b blockers, digoxin, diuretics, nitrates, and statins.

Statistical Analysis
All analyses were performed using SAS statistical software, version 9.3 (Cary, NC). Group differences for baseline characteristics in patients stratified by anxiety and/or depression diagnoses were compared using ANOVA test for continuous variables and the v 2 test for categorical variables. Multivariable extended Cox regression models using time-updated covariates were used to assess the relations between diagnosed anxiety and/or depression and incident adverse outcomes, adjusting for patient demographic factors, baseline medical conditions, renal function, and hemoglobin. We then further adjusted the analyses for the percentage of time in therapeutic INR range over each patient's follow-up period, and then performed an analysis adjusting for baseline anxiolytic and antidepressant as well as cardiovascular medication use. To assess whether results applied to the subgroup of patients without prior ischemic stroke, we added a sensitivity analysis restricted to patients without a history of stroke at baseline.

Results
We identified 25 570 eligible adults with diagnosed AF initiating a new course of warfarin therapy (Figure). A total of 897 participants (3.5%) had a diagnosis of anxiety only, 2322 (9.1%) had a diagnosis of depression only, and 904 (3.5%) had combined anxiety and depression (Table 1). Mean age was 72.3 years, and 43.5% were women in the overall cohort. Patients with a combined diagnosis of depression and anxiety were more likely to have a history of myocardial infarction, unstable angina, or percutaneous coronary intervention, whereas those without anxiety or depression were more likely to have diagnosed hypertension, but less likely to be affected by dementia or chronic lung or liver disease compared with the other subgroups. The percentage of time in therapeutic INR range differed among the 4 groups. It was 57% in patients without depression or anxiety and those with anxiety only, whereas it was 52% and 53% in patients with depression and those with combined depression and anxiety, respectively. P value < 0.001; results not shown.

Diagnosed Anxiety and/or Depression and Combined Ischemic Stroke and ICH
The median follow-up was 339 days (interquartile range, 131-708 days), with a maximum follow-up of 1460 days. In the overall cohort, 490 (1.9%) experienced an ischemic stroke or ICH event, for a rate of 1.52 events per 100 person-years (Table 2). In multivariable adjusted analyses, diagnosed anxiety was associated with a significantly higher adjusted rate of combined ischemic stroke and ICH (hazard ratio, 1.52; 95% confidence interval, 1.01-2.28), whereas depression was not (Table 3). Additional adjustment for patient-level time in therapeutic INR range did not materially affect the multivariable association between diagnosed anxiety and the risk of combined ischemic stroke and ICH (hazard ratio, 1.56; 95% confidence interval, 1.03-2.36). Further adjustment for the receipt of anxiolytic and antidepressant drugs or cardiovascular drugs did not lead to a meaningful change in the results (Table 3). In a sensitivity analysis with exclusion of patients with a history of ischemic stroke at baseline, a statistically significant association between diagnosed anxiety and combined ischemic stroke and ICH remained (adjusted hazard ratio, 1.56; 95% confidence interval, 1.03-2.36; Table S1).

Association Between Diagnosed Anxiety and/or Depression and Secondary Outcomes
Crude event rates of individual clinical outcomes are presented in Table 2. Diagnosed anxiety was associated with higher adjusted hazard ratios for ischemic stroke, ICH, and extracranial hemorrhage, but the associations were not statistically significant (Table 4). Neither diagnosed depression alone nor the combination of anxiety and depression was significantly associated with ischemic stroke, ICH, or hospitalization for extracranial hemorrhage (Table 4).

Discussion
In a large, multicenter, community-based cohort of adults newly treated with warfarin for AF, we found that having diagnosed anxiety was associated with an increased risk for worse outcomes independent of known risk factors and the quality of anticoagulation therapy. In contrast, diagnosed depression was not associated with an increased risk of adverse events in this population.
Other studies have noted that certain mental health conditions are associated with adverse cardiovascular outcomes, 4,19,20 but this has only been evaluated in patients with AF in a limited number of studies. A recent retrospective Journal of the American Heart Association cohort study among >100 000 patients on warfarin (≥65 000 with AF) from the Veterans Health Administration found that the risk of hemorrhage was slightly higher in those with mental health disorders (such as depression) compared with those without mental health disorders. 21 However, this previous study included only patients who were taking warfarin for at least 6 months and likely represented a selected subset of patients who tolerated anticoagulation better. 22 Another study using data from a Medicaid administrative claims database found an association between any psychiatric illness and an increased risk of thromboembolic and hemorrhagic events among patients with AF on warfarin, but did not specifically examine risk according to type of underlying condition. 23 The mechanism behind the association of anxiety and adverse outcomes is not clear. In our study, we examined whether the increased risk of adverse events in patients with diagnosed anxiety may have been related to anticoagulation control, which is potentially modifiable. Anticoagulation quality depends on multiple factors, including age, sex, the targeted INR, 24 the frequency of INR monitoring, 25 medication adherence, 26 diet, comorbidities, 24 and drug-drug interactions 27 ; some of these factors may be negatively affected by mental health disorders. [28][29][30] However, adjustment for patient-level time in therapeutic INR range did not materially affect the strength of association for anxiety and outcomes. Information on certain behavioral factors (eg, smoking and physical inactivity) 31 that may potentially confound the relationship between anxiety and outcomes was not consistently available and, thus, could not be analyzed. An intriguing avenue of further investigation may be how anxiety-related stress can lead to alterations in hemostasis. 32,33 Autonomic arousal in the context of anxiety may result in physiologic changes, such as an increase in blood pressure, 34 that are associated with the risk of both ischemic stroke and intracranial bleeding events.
We did not find a significant association between depression and adverse events in AF. Depression has been extensively studied as an independent risk factor for health outcomes, including cardiovascular disease and death, 2,3,35  not only among the general population, but also in patients receiving anticoagulation. 36,37 Our study relied on ICD-9 codes to define anxiety and depression, and it is possible that a diagnosis of anxiety may be less sensitive for underlying anxiety disorders compared with depression diagnoses. Generalized anxiety disorder may be more underrecognized than major depression. 38 A comparatively decreased sensitivity in anxiety diagnoses might lead to a selection of anxiety patients with higher disease severity, which could explain the increased risk of adverse outcomes in this population compared with patients with depression diagnoses or the combination of depression and anxiety. A strength of our study is that we were able to include a large number of patients with diagnosed anxiety and depression from a geographically and ethnically diverse population of anticoagulated patients with AF. By including patients newly treated with warfarin, our study reflects outcomes from the start of treatment in real-life clinical practice, where the choice of whether to anticoagulate a patient must be made without proven tolerance or patient adherence. We ascertained longitudinal warfarin exposure and assessed patient-level time in therapeutic INR range, which allowed us to examine the potential impact of anticoagulation control on our results.
Our study also had limitations. Although there was a consistent pattern of an increased rate of stroke and bleeding outcomes in patients with diagnosed anxiety compared with those without anxiety or depression, our results did not reach statistical significance for the individual secondary outcomes (ischemic stroke, ICH, and hospitalization for extracranial hemorrhage), possibly because of insufficient power. As noted previously, anxiety disorders and depression are often underdiagnosed, 39 which would likely lead to nondifferential misclassification that would bias our results towards the null. We also cannot exclude residual confounding by variables that are difficult to obtain from electronic health records, such as genetic factors, socioeconomic status and social network, 40,41 or stress, 32 and we were not able to assess the role of physical inactivity or smoking in the association between anxiety or depression and adverse outcomes. 31 Our study relied on anxiety and depression diagnosed in the medical record, which likely identified more severe cases. Finally, our results may not be completely generalizable to all populations and practice settings.
Given a consistent trend towards a higher risk of adverse events in patients with diagnosed anxiety, future research should investigate whether different treatment approaches might be associated with more favorable risk-benefit profile in these patients. The increased risk of adverse events did not seem to be mediated by poor INR control, so more intensive INR monitoring might, therefore, not be sufficient to prevent adverse events in these patients. Whether anticoagulation with direct oral anticoagulants reduces the risk of adverse events in patients with anxiety remains to be examined. Additional studies are needed to confirm our findings and further explore underlying mechanisms and optimal management strategies.
In conclusion, diagnosed anxiety, but not depression, was linked to a higher rate of combined ischemic stroke and ICH in adults with AF initiating anticoagulation. This information may be helpful in further risk stratifying patients with AF.