Physician-level variation in the diagnosis of myocardial infarction and the use of angiography among Veterans with elevated troponin

We conducted a single center retrospective cohort study at our Veterans Affairs (VA) Medical Center. Data were obtained from a quality improvement program conducted on patients with elevated Tn who were seen in our facility between February 2006 and February 2007. The study protocol was reviewed by the Institutional Review Board-01 Gainesville Health Science Center, and the requirement for informed consent was waived. During that time frame, a “troponin team” evaluated every patient with an elevated Tn, defined as being greater than 0.03 ng/ml on our Tn-T assay. The team was led by a clinical coordinator who received a daily list from the facility’s core laboratory of all patients who had an elevated Tn. This coordinator evaluated each patient based on their chart documentation and presented all patients to a cardiologist who determined if the patient’s presentation was consistent with a MI and whether ICA should be performed. No formal definition of MI was applied, and physicians were free to make the diagnosis based on their clinical assessment of the patient. Because our facility does not have an inpatient service led by a cardiologist, all other care decisions were left to the admitting team with input from the cardiology consultation service. Several cardiologists shared responsibility for oversight of the “troponin team” including general/noninvasive, interventional, and electrophysiology cardiologists. Data were blinded as to the physician, and therefore could not be analyzed at the level of subspecialty within cardiology. Data, including each patient’s baseline clinical characteristics, clinical course, thrombolysis in myocardial infarction (TIMI) score, electrocardiogram (ECG) results, and follow-up, were recorded by the clinical coordinator.

The co-primary outcomes were the rates of MI diagnosis and ICA use between individual physicians, which were compared using chi square analysis. We conducted a multivariate logistic regression to determine how the diagnosis of MI and the use of ICA were affected by the following variables: age by year, sex, coronary artery disease, hypertension, diabetes mellitus, hyperlipidemia, prior coronary intervention, prior coronary bypass surgery, ECG changes, new ECG changes, primary symptom of chest pain / dyspnea / other, serum creatinine, level of 1st Tn measurement. Each physician was included as an independent variable with the remainder of the physicians for comparison. We considered attempting an additional hierarchical multi-level model to compare odds ratios for MI diagnosis and ICA use between physicians after accounting for clustering of patients within their physician-based cohort. Due to the small sample sizes, however, this model was not considered statistically valid [12–14]. Odds ratios (OR) and 95 % confidence intervals (CI) were reported. Because we used an existing dataset, no formal a priori power calculation was performed. Baseline variables were compared by parametric and nonparametric tests as appropriate. Analysis was performed using SPSS version 21 (IBM, Armonk, NY).