The Heart and Soul Study is a prospective cohort study of psychosocial factors and health outcomes in patients with stable CAD. Design of the study has been published previously. In summary, patients were recruited from outpatient clinics of 12 different centers in the San Francisco Bay area if they met one or more of the following inclusion criteria: prior myocardial infarction, angiographic evidence of ≥50% stenosis in one of the coronary arteries, prior coronary revascularization, and exercise-induced ischemia (treadmill or nuclear scintigraphy). Exclusion criteria were acute coronary syndrome within the past 6 months, the inability to walk one block, and plans to move out of the area within 3 years.

Between September 2000 and December 2002, 1,024 patients were enrolled in the study. For the current investigation, 185 patients (18.1%) were excluded because they had a history of heart failure (n = 179) or heart failure status was unknown (n = 6).

Baseline study variables

All patients completed a daylong baseline study visit that included a medical history interview, physical examination, questionnaire, laboratory analysis, exercise test, and echocardiogram. Diabetes was defined as self-reported diabetes or the use of antidiabetes medication. Alcohol use was determined by questionnaire. Participants rated their physical activity during the previous month using a 6-point Likert scale. Those responding “not at all active” or “a little active” were classified as physically inactive. An estimate of chronic glycemia was provided by serum A1C measurement. Serum glucose level, A1C, LDL cholesterol, and C-reactive protein (CRP) were assessed by standard routine biochemistry analysis after an overnight fast (except for taking their regularly prescribed medication with water) using a venous blood sample, drawn via a 21-gauge butterfly needle. Subjects were considered to have metabolic syndrome if they met the criteria of the National Cholesterol Education Program. Echocardiography was performed with an Acuson Sequoia Ultrasound System (Siemens Medical Solutions USA, Malvern, PA), with a 3.5-MHz transducer. Left ventricular ejection fraction (LVEF) was calculated using the modified Simpson rule as recommended by the American Society of Echocardiography. In addition, a full description of diastolic function was performed according to predefined established criteria (normal, impaired, pseudonormal, or restrictive diastolic function). An exercise treadmill test (standard Bruce protocol) was performed. Immediately after exercise, echocardiographic analysis was performed to investigate exercise-induced wall motion abnormalities, which served as an indicator of myocardial ischemia. Details pertaining to acquisition and analyses of echocardiographic data were reported elsewhere. The institutional review board at each of the sites approved the study protocol, and all participants provided written informed consent.

End points

The main study outcome was time to hospitalization for heart failure, as was previously reported for the whole cohort in detail. Heart failure was diagnosed according to established criteria using clinical and radiological evaluation. Potential events were recorded annually by telephonic interviews. Additional information (e.g., medical records and death certificates) was collected and reviewed by two independent and blinded adjudicators. Discrepancies were discussed, and decisions were made by unanimity. In case of disagreement, a third blinded adjudicator was consulted. Follow-up was completed for all patients.

Statistical analysis

The study sample comprised 839 patients. Baseline differences between participants with diabetes and without were compared using t tests for continuous variables and χ2 tests for dichotomous variables.

In addition to the association of diabetes per se with hospitalization for heart failure, we investigated the role of glycemic control in the development of heart failure. A1C was used as a proxy measure for glycemic control (both dichotomized and continuous, per 1% change). For the former categorization, a cutoff of ≥6.5% and <6.5% was used because this cutoff was recently used to redefine the diagnosis of diabetes. The following analyses were conducted with diabetes and A1C as independent variables. First, Kaplan-Meier analysis was used to estimate the time from baseline to heart failure hospitalization in patients with or without diabetes and in patients with low or high A1C. The log-rank test was used for bivariate significance testing. In addition, given the influence of antidiabetes medication on A1C levels, we compared the effect of glycohemoglobin on heart failure in patients taking antidiabetes medication. Second, Cox proportional hazard regression analyses were performed to investigate the impact of diabetes and A1C level, respectively, on the time to first hospitalization for heart failure. To study the impact of diabetes and A1C on heart failure in the context of several potential confounders, we made a selection of the most important risk factors for heart failure based on recent guidelines. We then applied the following series of a priori determined Cox regression models in which we sequentially controlled for the following groups of confounders: model 1: age, sex, and race; model 2: smoking, physical inactivity, BMI, LDL cholesterol, and systolic blood pressure; model 3: myocardial infarction during follow-up; model 4: LVEF; model 5: exercise-induced wall motion abnormalities (i.e., ischemia); model 6: diastolic dysfunction; model 7: logCRP; and model 8: ACE inhibitor/angiotensin receptor blocker (ARB) and β-blocker-use. All models included age, sex, and race. In the final model (model 9) we include all the variables that were used in models 1–8. Finally, in sensitivity analyses, the relationship between several other definitions of diabetes and time to onset of heart failure were tested.

These definitions were 1) self-reported diagnosis of diabetes (irrespective of antidiabetes medication use, 2) fasting blood glucose >126 mg/dl, and 3) fasting blood glucose >126 mg/dl or use of antidiabetes medication. Moreover, the presence of metabolic syndrome was tested in sensitivity analyses. P < 0.05 was used for all tests to indicate statistical significance. Hazard ratios (HRs) with 95% CIs are reported. All statistical analyses were performed using SPSS (version 17.0 for Windows; SPSS, Chicago, IL).