Data from our study were obtained from the Cardiac Care Follow-Up Clinical Study (CCFCS) which is derived from the External Peer Review Program for quality monitoring of acute myocardial infarction and unstable angina in the Veterans Health Administration. This program includes data on 100% of patients diagnosed with acute myocardial infarction and a random sample of 10% of patients diagnosed with unstable angina treated at VA medical centers during the study period. Patients were identified by International Classification of Diseases, Ninth Revision (ICD-9)
, diagnosis codes 410.xx (myocardial infarction) and 411.xx (unstable angina) from administrative data. Resulting patient lists were then transmitted to Veterans Affairs (VA) facilities, where both paper and electronic medical records were manually abstracted by trained abstractors using standard reporting forms into the CCFCS data repository. The study was approved by the Institutional Review Board of VA Puget Sound. Additional details regarding CCFCS have been described previously [6
From patients in CCFCS admitted between January 1, 2005 and November 22, 2006, we identified patients with a prior history of myocardial infarction, unstable angina, or coronary atherosclerosis as determined by review of ICD-9 diagnosis codes (410.xx, 411.xx, 412.xx, or 414.xx) for all inpatient and outpatient visits in the 3 years prior to the admission date. Thus, all patients within the study cohort had a history of CAD within the 3 years preceding the ACS hospitalization captured by CCFCS. For patients with multiple hospitalizations captured within the CCFCS study period, we considered the most recent hospitalization as the index event. We excluded patients with missing data regarding outpatient medications for coronary prevention at the time of index admission. Patients with an absolute (i.e. medication allergy) or relative contraindication to a medication were excluded from analyses of that medication. Relative contraindications for aspirin included use of warfarin prior to admission, history of anemia, or history of ulcer. History of renal failure was considered a relative contraindication to angiotensin-converting enzyme inhibitor (ACE-I) or angiotensin II receptor blocker (ARB) and history of liver disease was considered a relative contraindication to use of lipid lowering medications.
The main outcomes of interest were receipt of outpatient prescriptions for preventive medications (aspirin, β-blocker, lipid-lowering medication, and ACE-I or ARB) as determined from chart review of the admission records at the time of index hospitalization for ACS in the CCFCS registry. The exposures of interest were the time interval from prior non-ACS and ACS hospitalizations in the VA to the index event. For each patient, a time since prior non-ACS hospitalization and time since prior ACS hospitalization was determined separately. These intervals were determined by review of all discharges from VA health care in the two years preceding the index event and measured from the date of discharge to the date of admission for the index event. Previous hospitalizations were determined to have been for ACS if the primary diagnosis code was myocardial infarction, unstable angina, or coronary atherosclerosis (ICD-9 of 410.XX, 411.XX, 412.XX, or 414.XX) [7
]. When the discharge date for the previous hospitalization and index admission date were the same, as in the case of patient transfer from another facility, the preceding hospitalization was considered for this interval. We grouped these intervals into ≤ 6 months, 6-12 months, 12-24 months, and > 24 months since last hospitalized. These intervals were chosen to maximize sample size in the comparison intervals prior to 2 years.
We used data from the CCFCS data registry to describe demographic and personal characteristics (i.e. age, sex, current smoker, hypertension, diabetes, heart failure, cerebral vascular disease, peripheral vascular disease, chronic kidney disease, and chronic obstructive pulmonary disease). Ethnicity was based on a combination of self-report and the electronic record [8
]. Socioeconomic status was based on the VA means test [9
We present unadjusted descriptive summaries of baseline patient characteristics for the study cohort. We report unadjusted descriptive statistics of the proportion of eligible patients on individual medications and multiple medications by patient characteristics. We conducted analyses separately for each individual medication and cumulative medications of interest as the eligible population varied due to contraindications to therapies.
We anticipated missing values for our ethnicity covariate [8
]. Instead of ignoring missing data using complete-case analysis, we conservatively chose to impute missing covariate values with chained equations. This method provides less biased estimates than complete-case analysis [10
]. From the imputed data, multivariate logistic regression estimated the effect size of time interval since hospitalization for non-ACS and ACS episodes on the prescription of preventive medications prior to recurrent event after adjustment for other covariates [11
]. Our model included both exposures of interest and the following covariates: age (< 55, 55-65, 65-75, ≥ 75 as indicator variables), sex, ethnicity, socioeconomic status (low, middle, high as indicator variables), current smoker, hypertension, diabetes, heart failure, cerebral vascular disease, peripheral vascular disease, chronic kidney disease, and chronic obstructive pulmonary disease. We did not perform hypothesis testing to determine patient characteristics included in the regression analyses, but instead included standard covariates that have been associated with preventive medication use in a prior analysis [12
]. We also included age and socioeconomic status as indicator variables to allow for non-linear associations between these variables and preventive medication use. From our logistic regression results, we report the proportion of patients prescribed medications for the exposure of interest adjusted to reflect the mean for each covariate in the cohort. This was determined for each level of the predictor of interest by calculation of the log odds and standard error within each imputed dataset, combination of log odds and standard errors across datasets using methods described by Rubin,[13
] and transformation of log odds and standard errors to proportions and confidence intervals. We report adjusted proportions or percentages in place of odds ratios as our outcome was common and odds ratios do not approximate relative risks in this setting [14
We conducted a number of sensitivity analyses to evaluate the robustness of our findings. First, we repeated our analyses after excluding patients with a history of dementia or cancer as providers may have anticipated less benefit to preventive therapies in this setting. We then repeated our analyses after stratifying on past history of myocardial infarction. This stratified analysis was to ensure any observed effect size in the primary analysis was not related to inclusion of low-risk patients with a remote history of ACS. We next explored the potential for bias related to exclusion of patients with incomplete ascertainment of preventive medication history. Finally, we completed a complete-case analysis to estimate the influence of imputation on our results. All statistical analyses were conducted with Stata, version 10.0 (StatCorp LP, College Station, TX). All hypotheses were evaluated at a two-sided significance level of 0.05, with calculation of 95% confidence intervals.