Abstract
Background Depression is a risk factor for incident myocardial infarction (MI), but little is known about the independent or additive risk from anxiety disorders.
Methods In a 7-year retrospective cohort design, we identified a cohort free of cardiovascular disease in fiscal years 1999 and 2000 that contained 96,612 patients between 25 and 80 years of age who had an International Classification of Diseases, Ninth Revision, Clinical Modification code indicating a diagnosis of depression in 2000 (baseline) and 259,387 patients without depression. Cox proportional hazards models stratified by depression were computed to test for a main effect of anxiety disorder unspecified, generalized anxiety disorder, panic disorder, social phobia, obsessive-compulsive disorder, and posttraumatic stress disorder (PTSD) on risk of incident MI. The models were adjusted for multiple MI risk factors and sociodemographics.
Results Depressed as compared to nondepressed Veterans Administration patients were at increased risk for incident MI (HR 1.39; 95% CI 1.34–1.45). In nondepressed patients those with anxiety disorder unspecified (HR 1.34; 95% CI 1.21–1.47), panic disorder (HR 1.43; 95% CI 1.11–1.83), and PTSD (HR 1.25; 95% CI 1.16–1.36) were at increased risk for incident MI. The independent risk associated with anxiety disorders was reduced in patients comorbid for depression.
Conclusions In Veterans Administration patients free of heart disease in 1999 and 2000, those with depression, anxiety disorder unspecified, panic disorder, and PTSD were at increased risk of incident MI. Anxiety disorders are independent risk factors for MI. Depression partially accounts for the effect of anxiety disorders on risk of MI in patients with both conditions.
Background
Numerous epidemiological studies have established that depression is a risk factor for cardiovascular disease.[1] Most studies have demonstrated that a history of depression increases the likelihood of incident heart disease, including myocardial infarction (MI), by approximately 2-fold.[2–6] This effect remains after adjusting for common vulnerabilities including smoking, diabetes, and hypertension. Though less well studied, a growing literature has demonstrated that anxiety is a risk factor for incident heart disease.[7–15] Both nondiagnostic measures of phobic anxiety such as the Crown Crisp Index[10, 13] and Minnesota Multiphasic Personality Inventory anxiety trait measures[16] and diagnoses of panic attack[15] and posttraumatic stress disorder (PTSD) have been associated with incident heart disease.[17, 18]
Depression commonly co-occurs with anxiety disorders. The National Comorbidity Survey Replication reported significant correlations between major depressive episode in the preceding 12 months and the following anxiety disorders: generalized anxiety disorder (GAD), posttraumatic stress disorder (PTSD), specific phobia, social phobia, and obsessive-compulsive disorder (OCD).[19] Anxiety disorders, especially PTSD, often co-occur in veteran patient populations with depression.[20] Determining whether anxiety disorders have independent effects or worsen the cardiovascular prognosis of depressed patients could have implications for screening and management of an at risk population. There are few studies of the risk of incident heart disease in depressed patients with and without comorbid anxiety disorders. Additionally, to our knowledge, no reports have been published on the effect of the major anxiety disorders and depression status on risk of MI in patients initially free of heart disease at baseline. Thus, we sought to fill this gap in the literature by determining if the major anxiety disorders contribute to incident MI and whether the effect is increased in the presence of depression.
Using a national cohort of Veterans Administration (VA) patients free of heart disease at baseline, we tested the following hypotheses: (1) depression is an independent risk factor for incident MI; (2) individual anxiety disorders predict risk of MI; and (3) individual anxiety disorders that co-occur with depression increase the risk for MI above that due to depression or anxiety alone.
Methods
We used a retrospective cohort design. Data were obtained from inpatient and outpatient International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnoses, Current Procedural Terminology codes, Pharmacy Benefits Management records and sociodemographic information maintained by the Veterans Health Administration (VHA) national medical care data sets beginning in Fiscal Year (FY) 1999, the first year that national data is considered complete. These data are inclusive of all health utilization encounters in the VHA and are maintained by the VHA Office of Information at the Austin Information Technology Center (formerly Austin Automation Center; see www.virec.research.va.gov/datasourcesname/medical-sas-datasets).
Cohort Eligibility
Using ICD-9-CM codes, we constructed a cohort that appeared to be free of cardiovascular disease in FY 1999 and 2000 by first selecting all patients with any one ICD-9-CM code indicating a primary diagnosis for depression ICD-9-CM: 296.2, 296.3, 300.4 or 311 and a set of 300,000 controls free of cardiovascular disease randomly selected from all VA patients (n = 1,380,433) enrolled in 1999 and 2000 (see Figure 1). Patients with at least one primary or secondary diagnosis of hypertensive heart disease (ICD-9-CM 402–405), ischemic heart disease (ICD-9-CM 410–414), disease of the pulmonary circulation (ICD-9-CM 414–417), and other forms of heart disease (ICD-9 420–429) were excluded from the cohort as well as patients with cerebrovascular disease (ICD-9-CM 430–438). This resulted in a sample of 536,415 unique patients free of cardiovascular/cerebrovascular disease in both FY 1999 and 2000. The follow-up period began October 1, 2000, and ended September 30, 2007 (7-year follow-up period).
(Enlarge Image) Figure 1.
Diagram shows ascertainment of depressed and nondepressed VA patients.
[ CLOSE WINDOW ]
Figure 1.
Diagram shows ascertainment of depressed and nondepressed VA patients.
Depressed Cohort
To be included in the depressed group patients must have had either 1 inpatient ICD-9-CM code for depression in FY 2000 or at least 2 outpatient ICD-9 codes for depression in a 12 month span, with at least 1 of these ICD-9 codes occurring in FY 2000. This algorithm of 1 inpatient visit or 2 outpatient visits in a 12-month span was used to define all other psychiatric conditions. This algorithm has 99% positive predictive value for depression diagnoses in administrative claims data.[21]
Exclusion Criteria
We excluded conditions which are thought to be associated with poor treatment adherence and self care, i.e., ICD-9-CM codes 296.1 (affective psychoses and manic disorders), 296.9 to 296.99 (unspecified affective psychosis), and 296.4 to 296.8 (bipolar) and excluded hypothyroidism to reduce risk of misclassification of depression (n = 9,653 excluded). Patients must also have been between the ages of 25 and 80 years at the beginning of follow-up (n = 16,333 excluded). To establish that our subjects were regular users of VA health care, a subject must have had at least 1 outpatient visit in FY 1999 and 2000 (n = 8). Finally, subjects were excluded if a myocardial infarction occurred before the beginning of follow-up (n = 43). After applying these group definitions and exclusion criteria, a final sample of 96,612 depressed and 259,387 nondepressed subjects were available for analysis.
Outcome Variables
Incident MI. Incident MI during the period of October 1, 2000, and September 30, 2007, was identified if patient records contained ICD-9-CM codes 410 to 411. The duration of follow-up was calculated as the time from October 1, 2000, to the diagnosis of MI. Conversely, if coding for MI was not present, the follow-up time was calculated as the time from October 1, 2000, to the last recorded visit (date of last inpatient or outpatient ICD-9-CM code). Follow-up time was calculated in months.
Predictor Variables
Anxiety Disorders. We modeled the risk of MI as a function of the following anxiety disorders: anxiety disorder unspecified (ICD-9-CM: 300.0) GAD (ICD-9-CM 300.02), PTSD (ICD-9-CM: 309.81), social phobia (ICD-9-CM: 300.23), panic disorder (ICD-9-CM: 300.1), and OCD (ICD-9-CM: 300.3). As with depression, we required cases of anxiety to have a record of 2 outpatient or 1 inpatient diagnoses within a single 12-month period at any time preceding the onset of MI. It is logical that "anxiety disorder unspecified" will be used early in a diagnostic work-up or by a clinician in primary care less familiar with anxiety disorders. The diagnosis is used for an anxiety state that does not clearly and immediately fit the diagnostic criteria for a specific anxiety disorder.
Covariates
Cardiovascular Risk Variables. Conditions recorded by ICD-9-CM code and known to be established risk factors for heart disease included hypertension (ICD-9-CM 401–401.9), hyperlipidemia (ICD-9-CM 272–272.2), and type 2 diabetes (ICD-9-CM 250). Hypertension (ICD-9-CM 401), hyperlipidemia (ICD-9-CM 272, 272.0, 272.1, 272.2, 272.4), and obesity (ICD-9-CM 278.0–278.02) were considered present if an ICD-9 code for this disorder occurred prior to MI or censorship. Type 2 diabetes was considered present if a patient received an ICD-9 code (250.x0, 250.x2, 357.2, 362.0, 366.41) or a diabetic medication prior to MI or censorship. We considered alcohol abuse/dependence diagnoses if patients had either an ICD-9-CM code of 305.0 for abuse and/or 303 indicating dependence prior to MI or censorship. Nicotine dependence was defined by ICD-9-CM code 305.1 or V15.82 indicating personal history of tobacco use prior to MI or censorship.
Sociodemographic. Data available from the 1999 VA administrative electronic records included age, gender, race, insurance status and marital status. We adjusted for insurance to control for potential bias in our ability to detect heart disease outcomes. Patients with private insurance may be more likely to use private sector care in addition to their VA utilization, thus reducing the likelihood of detecting all of their health care use including specialized cardiovascular care. We adjusted for marital status to account for the correlation among depression, social support, and heart disease outcomes. Marital status was modeled as a three-level variable (married, divorced/widowed/separated and single/never married).
Reliability of Administrative ICD-9 Codes
Veterans Administration administrative records have been shown to have good reliability when compared with written patient charts. This is especially true of mental health and cardiology care which show >96% agreement.[22] A large body of literature from a broad range of medical specialties have utilized VA national databases and ICD-9-CM codes for clinical epidemiology and outcomes research including studies using definitions of depression similar to ours (ICD-9-CM codes 296.2, 296.3, 296.5, 300.4, 309.4 or 311),[23] anxiety disorders,[24] cardiovascular disease, and other chronic conditions.[25] In addition, we recently demonstrated an increased risk for MI in a cohort of depressed patients with rheumatoid arthritis by analyzing VA national databases.[26]
Analytic Design. χ2 Values were computed to test if the covariates were significantly associated with depression and/or MI. Bivariate analyses were conducted using χ2 tests for categorical variables and t tests for continuous variables. Bivariate relationships between depression, anxiety disorders and MI were estimated by first computing models only adjusted for age.
Based on plots of the rate of MI, depression, and anxiety disorders by 5-year age increments (results not shown), age was associated with predictor and outcome variables in a nonlinear pattern. To account for these nonlinear relationships, a quadratic age term was included in all multivariate analyses.
Cox-proportional hazard models with time-dependent covariates were computed to derive the hazard ratio for incident myocardial infarction. Modeling time-dependent variables fixed all covariates as occurring before incident MI. Separate adjusted models were computed for each individual anxiety disorder in depressed and nondepressed cohorts. Adjusted models accounted for the effect of age, gender, race, marital status, insurance, diabetes, hypertension, hyperlipidemia, alcohol abuse/dependence, and nicotine dependence.
All analyses were performed using SAS software version 9.1.3 (SAS Institute, Cary, NC) and α's were set at .05. Two-tailed tests were used to allow for the possibility of protective effects. Cox proportional hazards models were computed using the PROC PHREG procedure.[27]
Role of Funding Source. This study was supported by VA HSR&D Career Development Award–2 to JFS. The funding source had no role in the study design, conduct or reporting. The authors are solely responsible for the design and conduct of the study, all study analyses, and the drafting and editing of the manuscript and its final content. This project was approved by the institutional review board of the St. Louis VAMC and Washington University.
Patients with and without Depression
The distributions of the covariates in the depressed and nondepressed groups are shown in Table I. Patients with depression were significantly younger at baseline as compared to nondepressed patients (51.6 vs 57.2 years, P < .0001). About 90% of the cohort were male but females were disproportionally over-represented among depressed patients (P < .0001). Depressed patients were more likely (P < .0001) to be white, not married, and to have only VA insurance coverage.
Compared to nondepressed patients, depressed patients were more likely (P < .0001) to have a diagnosis of anxiety disorder unspecified, GAD, panic disorder, social phobia, OCD, and PTSD. Depressed patients were significantly more likely to have multiple anxiety disorders as compared to nondepressed patients (P < .0001).
Cardiovascular risk factors, with the exception of diabetes, were more prevalent in the depressed cohort. Compared to nondepressed patients, a higher proportion of depressed patients had hypertension (P < .01). Depressed patients were much more likely (P < .0001) to have a diagnoses of hyperlipidemia, obesity, alcohol abuse/dependence, and nicotine dependence.
Patients with and without MI
The distribution of covariates in patients with and without incident MI is shown in Table II. Patients who had an MI were significantly older (59.3 years vs 55.5 years; P < .0001). About 90% of the cohort were male and were over-represented among MI patients (P < .0001). Patients with MI were more likely (P < .0001) to have non-missing data on race and marital status. Missing marital status and race data is associated with amount of health care utilization such that patients with conditions requiring hospitalization (eg, MI) are less likely to have missing data on these two sociodemographic variables. There was no difference in the distribution of insurance coverage in patients with and without MI.
Depression was significantly (P < .0001) more common among MI patients than patients without MI. Compared to patients who did not develop MI, patients with MI were more likely (P < .001) to have a diagnosis of anxiety disorder unspecified, GAD, panic disorder, and PTSD. Social phobia and OCD were marginally more common among non-MI patients. Compared to non-MI patients, MI patients were more likely (P < .0001) to have hypertension, hyperlipidemia, diabetes, obesity, and nicotine dependence. Significantly (P < .0001), more MI patients had ≥2 anxiety disorders as compared to non-MI patients.
Relationship between Depression, Anxiety, and Incident MI
Age-adjusted HRs were computed to test associations among depression, major anxiety disorders, and incident MI before accounting for other covariates. As shown in Table III, depression was associated with a 1.39 fold (95% CI 1.34–1.45) increase in hazard of incident MI. Anxiety disorder unspecified, GAD, panic disorder and PTSD were each associated with significantly increased hazard of incident MI (range of HRs 1.28–1.53). Neither social phobia nor OCD diagnoses were associated with increased risk of MI.
Anxiety and Risk of Incident MI in Depressed and Nondepressed Patients
Among nondepressed patients, those with anxiety disorder unspecified (HR 1.34; 95% CI 1.21–1.47), panic disorder (HR 1.43; 95% CI 1.11–1.83) and PTSD (HR 1.25; 95% CI 1.16–1.36) were at increased risk for incident MI (see Table IV). Among nondepressed patients, those with OCD were less likely to have an MI (HR 0.32; 95% CI 0.12–0.85). Among depressed patients, the magnitude of risk for MI was lower for each anxiety disorder. However, anxiety disorder unspecified (HR 1.11; 95% CI 1.03–1.20) and panic disorder (HR 1.22; 95% CI 1.07–1.39) remained significant predictors of incident MI in depressed patients. The effect of PTSD was not significant in depressed patients (HR 1.07; 95% CI 0.99–1.14). GAD, social phobia and OCD diagnoses were not associated with incident MI in depressed patients.
Discussion
In a cohort of patients receiving care in the VA medical system who were free of cardiovascular disease at baseline, we observed an increased risk for incident MI among patients with depression, anxiety disorder unspecified, GAD, panic disorder, and PTSD. In time-dependent adjusted analyses, anxiety disorder unspecified, panic disorder, and PTSD were significant predictors of MI in non-depressed; however, the effect of these anxiety disorders on risk of MI was attenuated among depressed patients. This finding suggests depression partially explains the likelihood of incident MI in patients with anxiety disorder. Importantly, the risk for incident MI remained significant among patients with depression and anxiety disorders after adjusting for numerous cardiovascular and sociodemographic factors including age, gender, race, marital status, insurance, diabetes, obesity hypertension, hyperlipidemia, alcohol abuse/dependence, and nicotine dependence.
Few studies have examined the relationship between anxiety disorders and incident MI among initially healthy persons. However, our results are consistent with findings that PTSD is a risk for incident heart disease among Vietnam Veterans,[20] and symptoms of PTSD are a risk factor for incident coronary heart disease in women.[18]
In a prospective study of patients with stable coronary artery disease, depression and GAD both contributed to the rate of incident adverse cardiac events, but patients with both disorders were not at greater risk than those with depression or GAD alone.[9] Our results are consistent with these findings. However, the risk for incident MI as a function of a range of specific anxiety disorders in depressed and nondepressed patients has not been previously demonstrated.
We did not expect to find that OCD was negatively associated with MI. We can extrapolate some comparative information from a review of mental illness and all-cause mortality that showed the median relative risk of OCD was 1.1 (95% CI 1.1–1.7),[28] which was much lower than the relative risk for panic disorder (1.9) and depression (1.7). Among older adults with OCD, the number of chronic health conditions was similar to those with other anxiety disorders.[29] Studies of OCD and indicators of cardiovascular disease have found no reduced heart rate variability in OCD patients.[30] Whether persons with OCD have a greater occurrence of poor physical outcomes including cardiovascular disease is uncertain given limited and inconclusive evidence. Lastly, preparation for the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition may re-classify OCD outside the spectrum of anxiety disorders.
Strengths
The national VA cohort allows findings to be generalized to most VA patients. Data from large cohort studies produce stable estimates. In fact, our previous study (Scherrer et al., 2009) of a smaller cohort of VA patients with rheumatoid arthritis resulted in nearly identical point estimates for the risk of incident MI as a function of major depression (HR 1.38). The current large sample size allowed us to separately assess rare anxiety disorders such as OCD. Our retrospective cohort design and use of time-dependent covariates allows us to conclude that depression and anxiety are independent predictors of incident MI not explained by pre-existing standard cardiac risk factors. Lastly, sensitivity analysis indicated results did not differ when the cohort was limited to patients with only VA coverage. Thus, our results should generalize to veteran populations with additional health insurance options (eg, Medicare).
Limitations
Administrative data have inherent limitations as compared to in person assessments and clinical trials. Direct clinical assessments would be required to evaluate more sensitive markers of coronary disease (eg sub-clinical coronary disease) and, hence, a more comprehensive assessment of the relationship between depression, anxiety, and MI. Because our data are based on clinical diagnoses, we are unable to account for the cases of depression that are never diagnosed. However, Desai et al[31] examined the rate of screening for depression in VA patients without a diagnosis of depression. These researchers found 85% of patients had documentation of receiving a depression screen and 8.8% screened positive. Of those who screened positive, 54% received a follow-up evaluation and only 1.1% resulted in a diagnosis of depression. While the authors acknowledge follow-up screening and the screening paradigm may also explain evidence for the rarity of undetected depression in the VA, some evidence indicates that the percent of undetected depression is low in the VA system. It is possible that less severe cases of depression and anxiety are undetected; therefore, the present results may not generalize to sub-clinical anxiety and depression. In addition, receipt of antidepressant is very low among the non-depressed cases (<4%) as compared to 39% of patients in the depressed group providing further evidence that misclassification of depressed patients as nondepressed is unlikely. Even if misclassification resulted in coding depressed and anxious patients as unaffected, this would result in more conservative estimates of the association demonstrated between depression and anxiety disorders and incident MI.
Misclassification of covariates may confound our results if covariates were systematically under or overdiagnosed by depression or MI status. We anticipated possible misclassification for tobacco dependence at baseline because it is entirely dependent upon patient self-report as compared to laboratory or physical exam based diagnoses for our other covariates (eg, diabetes). However, clinical reminders are part of the electronic medical record system and include prompts for clinicians to assess patient smoking status. The prevalence of tobacco dependence, (47% in depressed and 29% in nondepressed patients, respectively) is consistent with the 23% to 33% prevalence of smoking reported from surveys of the general veteran population.[32, 33] However, because smoking histories are not available, it is not possible to model the effects of pack years in this cohort. Other major risk factors not available in our national VA electronic data base include weight, height, diet, and exercise. Residual confounding may be present due to the unknown effect of these and other variables not available for adjusted analyses.
It is possible that the accuracy of ICD-9-CM diagnoses may differ outside the VA system (eg, managed care health plans); however, we expect the automated and systematic method of maintaining electronic records in the VA improves diagnostic accuracy. Nonetheless, misclassification is a potential limitation of our study. ICD-9-CM codes for MI have very high agreement (>99%) with written medical records in the VA.[22] Although unlikely, we acknowledge that the classification of incident MI may be a recurrence or exacerbation of a pre-existing cardiovascular condition that was first recorded in non-VA records. Such data might be found via abstraction of physician notes from individual patient records. It is beyond the scope of the present study to examine individual patient records and we believe the absence of a diagnosis during the 2-year baseline period is a reasonable length of time to consider that patients were free of clinically significant cardiovascular disease prior to MI.
We lack data on non-VA healthcare utilization. After adjusting for insurance resources, we found depression and anxiety remained significant risk factors for incident MI. Receipt of care for acute MI among VA patients often occurs in emergency situations in a non-VA hospital;[34] however, based on information from the VA Health Services Research and Development Ischemic Heart Disease Quality Enhancement Research Initiative, VA patient records would include coding for MI upon their return to VA care.
Lastly, the relatively small number of female veterans could have biased results if higher prevalence of anxiety and depression in females combined with lower prevalence of MI reduced the association. To test this potential bias, we recomputed full models shown in Table IV in men only. Results did not differ from those shown for the full cohort.
We emphasize that the cohort under study did not have diagnostic evidence of cardiovascular disease at baseline. In this data set, depression and anxiety contributed to new-onset MI; however, we do not exclude the possibility that other cardiovascular disease may be in the pathway from affective disorders to MI. In addition, further research is needed to determine if anxiety disorders have similar or greater impact on risk of MI among patients with pre-existing heart disease.
Conclusion
A growing literature on the effect of psychological stress on adverse health outcomes clearly indicates a role for depression and anxiety disorders. Further research to determine if treating depression and anxiety may reduce risk for cardiovascular events is warranted. Currently, the American Heart Association[35] does not include anxiety and depression in its list of established risk factors for coronary heart disease despite the fact that the risk attributable to these disorders is similar in magnitude to other major established risk factors such as diabetes and smoking. Cardiologists are not, in general, trained to screen patients for psychiatric conditions and recognition of the cardiovascular consequences of anxiety and depression is not a routine discussion in many physician-patient encounters. Lastly, these data raise the question of whether psychiatrists, psychologists and other mental health care providers have a role in monitoring the cardiovascular health of patients with anxiety and depression. Future research is needed to determine if greater integration of primary and specialty care would reduce the burden of mental health and cardiovascular disease.
