Cohort eligibility

Using ICD-9-CM codes, we identified a patient cohort free of diagnosed cardiovascular disease in the fiscal years FY1999 and FY2000. From 1 380 433 patients using VA healthcare in 1999 and 2000 we excluded those with at least one primary or secondary diagnosis of hypertensive heart disease (ICD-9-CM 402–405), ischaemic 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-CM 420–429) as well as patients with cerebrovascular disease (ICD-9-CM 430–438). We then selected all patients with a primary diagnosis of major depressive disorder (ICD-9-CM 296.2, 296.3 or 311) and a set of 300 000 randomly selected, cardiovascular disease-free patients without depression. This resulted in a sample of 536 415 unique patients free of cardiovascular and cerebrovascular disease in both FY1999 and FY2000. The 7-year follow-up period began 1 October 2000 and ended 30 September 2007.

Exclusion criteria

We excluded patients with psychotic and bipolar disorders to reduce the risk of misclassifying major depressive disorder. Patients also had to be between the ages of 25 and 80 at the beginning of follow-up to allow for variability in risk for myocardial infarction and mortality. In order to limit the sample to regular users of VA healthcare, patients who did not have at least one out-patient visit in both FY1999 and FY2000 were excluded. Patients were also excluded if an acute myocardial infarction occurred within the first month of follow-up. After applying these criteria, a final sample of 96 612 patients with major depressive disorder and 259 387 patients without major depressive disorder remained. From this cohort of 355 999 veterans, we identified 12 304 patients with an incident myocardial infarction during the follow-up period. Of these patients, 4242 patients with depression were eligible for analysis. We excluded 126 patients who did not have 12 weeks of follow-up time after myocardial infarction because this was the minimum follow-up time necessary to define the primary independent variable (i.e. guideline-concordant antidepressant use). Also, 79 patients of unknown ethnicity or marital status were excluded. This resulted in a sample of 4037 patients available for analysis.

Predictor variables

Covariates

Covariates were selected because of their known association with mortality and high co-occurrence with depression. Nicotine dependence was defined by ICD-9-CM code 305.1 or V15.82 indicating personal history of tobacco use prior to mortality or censorship. Alcohol misuse/dependence was defined by ICD-9-CM code 305.0 for misuse and/or 303 indicating dependence. We adjusted for the following anxiety disorders: generalised anxiety disorder (ICD-9-CM 300.02), panic disorder (ICD-9-CM 300.1), anxiety disorder not otherwise specified (ICD-9-CM 300.0), obsessive–compulsive disorder (ICD-9-CM 300.3), social phobia (ICD-9-CM 300.23) and post-traumatic stress disorder (ICD-9-CM 309.81). Because beta-blocker use may be associated with symptoms of depression we adjusted for the medication possession ratio for atenolol, bisoprolol, carvedilol, labetalol, metoprolol, propranolol, sotalol, nadolol and pindolol. Obesity was defined as ICD-9-CM codes 278.0–278.02 or as a body mass index of >30 from height and weight data obtained from the Vital Signs data file and classified as obese v. not obese according to the Centers for Disease Control and Prevention guidelines. ¹² The Romano adaptation of the Charlson Comorbidity Index, ^{Reference Romano, Roost and Jollis13,Reference Schneeweiss, Wang, Avorn and Glynn14} a list of 17 health conditions associated with morbidity and mortality, was included to adjust for underlying risk for all-cause mortality. We have found that health service utilisation is inversely associated with mortality, therefore we adjusted for clinic stops per month.

Outcome variable: all-cause mortality

All deaths during the period 1 October 2000 to 30 September 2007 were obtained from the VA Vital Status file which tracks deaths by incorporating information from the Beneficiary Identification and Records Locator Subsystem (BIRLS) Death File created by the Veterans Benefits Administration (VBA), the Medical SAS Inpatient Datasets that track mortality and death dates that occur during a hospital stay, and the Social Security Administration (SSA) Death Master File.

Analytic design

Bivariate analyses included t-tests for continuous variables and chi-squared tests for categorical variables. We first characterised the association between covariates and mortality. Then we investigated the relationship between covariates and major depressive disorder treatment status. Survival models were then computed to analyse time to mortality. Hazard ratios (HRs) for mortality were estimated using Cox proportional hazards models with time-dependent covariates. Owing to a nonlinear relationship between age, mortality and major depressive disorder treatment status, both a linear and a quadratic age term were included in all multivariable survival models. Sociodemographics were modelled from their status at baseline, clinic stops were modelled continuously as number of clinic stops per month, and the remaining adjustments were based on time-dependent covariates that could occur any time before the end of follow-up. Analyses were performed using SAS version 9.1.3 for Windows, with α set at 0.05. Two-tailed tests were used to allow for both risk factors and protective effects. The PROC PHREG procedure was used to compute Cox proportional hazards models. Month was the unit of time for survival analyses and the index myocardial infarction date was the beginning of follow-up.

This project was approved by the institutional review boards of the St Louis Veterans Affairs Medical Center and Washington University.