Impact of Diabetes on Long-Term Survival After Acute Myocardial Infarction: Comparability of risk with prior myocardial infarction

The Onset Study was conducted in 45 medical centers in the U.S. (21). Between August 1989 and September 1994, 1,935 patients (601 women and 1,334 men) were interviewed a median of 4 days after sustaining an AMI. Trained research interviewers identified eligible patients by reviewing coronary care unit admission logs and patient charts. For inclusion, patients were required to have CK-MB isoenzyme levels above the upper limit of normal for each center, an identifiable onset of symptoms of infarction, and the ability to complete a structured interview. The institutional review board of each center approved this protocol, and each patient gave informed consent.

Interviewers used a structured data abstraction and questionnaire form. Information collected from each interview and chart review included patient age, sex, medical history, and medication use (prescription and nonprescription). During the chart review, interviewers recorded complications of congestive heart failure or ventricular tachycardia.

We defined diabetes as a history of diabetes obtained during chart review or the current use of any hypoglycemic medication. When available, we also determined the type of diabetes (type 1 versus type 2) (n = 381) and the duration of time from diagnosis of diabetes to onset of AMI (n = 310).

We defined aspirin use as the use of aspirin or any aspirin-containing product in the 4 days before the index AMI. We used 1990 U.S. Census data to derive median household income from U.S. Postal Service zip codes (22). We considered patients to have had a previous myocardial infarction if it was noted on either the patient interview or the chart review; the agreement rate between these two sources was 98.6%.

We searched the National Death Index for deaths of Onset Study patients through 31 December 1995 and requested death certificates from state offices of vital records for all probable matches, using a validated algorithm (23). Three physicians independently reviewed each death certificate to verify the match. Two physicians categorized each death as due to cardiovascular or noncardiovascular causes. Disagreements among raters were resolved by discussion. For all analyses, total mortality was the primary end point.

We analyzed continuous and binary variables using Student’s t tests and exact tests, respectively. We compared unadjusted Kaplan-Meier survival plots using the log-rank test. We used Cox proportional hazards models to examine the effect of diabetes on survival after adjustment for potentially confounding factors. The factors we included were age, sex, previous myocardial infarction, angina, hypertension, medication use before hospitalization (aspirin, β-adrenergic antagonists, calcium-channel blockers, and angiotensin-converting enzyme inhibitors), current smoking, previous smoking, BMI, use of thrombolytic therapy, usual frequency of exertion (in three categories), alcohol consumption (in three categories), household income (in quartiles), education (in three categories), and complications of congestive heart failure or ventricular tachycardia during hospitalization. In the smaller treatment-specific models, we adjusted for age, sex, smoking status, BMI, previous myocardial infarction, angina, use of angiotensin-converting enzyme inhibitors, use of thrombolytic therapy, alcohol abstention, usual frequency of exertion, and congestive heart failure; results from the complete sample were essentially identical when the truncated model was used.

Patients with missing information on household income (n = 56) or educational attainment (n = 58) were assigned indicator variables. For all other covariates, patients missing a specific variable (≤22 patients for any variable) were assigned mean levels of continuous covariates and modal levels of binary covariates. Models that deleted patients with any missing information yielded similar results. We tested the proportionality of hazards using time-varying covariates and found no significant violations. We present hazard ratios (HRs) from Cox models with their 95% CIs. All probability values presented are two-sided.

The characteristics of the Onset Study patients have been reported (21) and are shown in Table 1. Diabetic participants were generally older and more likely to be women, obese, sedentary, and abstainers from alcohol. They had more comorbidity and used cardiac medication more frequently. They were less likely to be white or current smokers, and they reported lower educational attainment and median household income. During the index hospitalization, they were more likely to develop congestive heart failure but less likely to develop ventricular tachycardia.

Of the 399 patients with diabetes, 121 took insulin alone, 184 took oral hypoglycemic agents alone, 3 took insulin and oral agents, and 91 took no hypoglycemic medications. For the 381 patients with available information, 49 (13%) had type 1 diabetes and 332 (87%) had type 2 diabetes.

During a mean follow-up of 3.7 years, 116 (29%) of the diabetic patients died, as compared with 204 (13%) of the nondiabetic patients. The unadjusted HR for total mortality for diabetes was 2.4 (95% CI 1.9–3.0; P < 0.001). After controlling for potentially confounding factors, we found an HR of 1.7 (1.3–2.1; P < 0.001). Exclusion of deaths within 30 days of the index infarction did not change the results. The association of diabetes with cardiovascular mortality was similar (adjusted HR 1.7, 95% CI 1.3–2.3; P < 0.001).

The unadjusted HR for total mortality for a previous myocardial infarction was 2.4 (95% CI 1.9–2.9), identical to that for diabetes. Figure 1 shows the comparable effects of diabetes and previous myocardial infarction on estimated survival. Table 2 shows unadjusted and adjusted HRs for total and cardiovascular mortality among patients with diabetes, previous AMI, both diabetes and previous AMI, or neither. In every model, the effect of diabetes was at least as strong as the effect of a previous infarction, and the two conditions seemed to be independently associated with risk of mortality.

In adjusted analyses, we found a statistically significant interaction between diabetes and sex (P = 0.01). The HR for diabetes among women (2.7; 95% CI 1.8–4.2) was approximately twice as high as the HR for diabetes among men (1.3; 1.0–1.8). Among women, the HR for diabetes was actually greater than that for a previous AMI (2.7 vs. 1.4; P = 0.03), whereas the HRs for diabetes and previous AMI were comparable in men (1.3 vs. 1.5; P = 0.59). In other adjusted analyses, diabetes had a consistent effect on mortality in subgroups defined by age, obesity, thrombolytic therapy, hypertension, smoking, and income (data not shown).

We found that the effect of diabetes was similar regardless of treatment type, even after controlling for confounding factors. Compared with nondiabetic patients, the HR for mortality of diabetic patients was similar among those reporting use of insulin (1.8, 95% CI 1.2–2.6), oral agents (1.5, 1.1–2.1), or neither therapy (1.9, 1.3–2.9) (P = 0.49 for comparison across three groups).

Among diabetic patients, the median time period between the diagnosis of diabetes and the index myocardial infarction was 7 years (range 0–65). We found no relationship between the duration of diagnosed diabetes and the risk of long-term mortality (adjusted HR for every 10-year period of diabetes = 1.0, 95% CI 0.8–1.3; P = 0.82).

Among early survivors of AMI, we found that diabetes was associated with nearly twofold higher long-term mortality after infarction. The magnitude of the risk associated with diabetes was similar in magnitude to that associated with a previous myocardial infarction. This association was not substantially altered by treatment type, duration of diabetes, or patient subgroup, but the effect of diabetes was greater in women than men.

In a population-based study, Haffner et al. (19) found that diabetes confers a similar risk of mortality as a previous myocardial infarction, itself a powerful determinant of mortality. The generalizability of that study may have been limited by restriction to Finnish subjects, who have strikingly high rates of type 1 diabetes (24) and coronary mortality, particularly among men (25). Among patients with unstable angina and non–Q-wave infarction, Malmberg et al. (20) found that diabetes and previous vascular disease confer similar risks of mortality, although the authors did not distinguish between patients with and without confirmed infarctions. In confirming the equivalence of diabetes and previous AMI as risk factors for mortality among patients following AMI, we highlight the adverse prognostic impact of diabetes at all stages of CHD.

Our results generally agree with previous studies of diabetes and long-term mortality after AMI (Table 3). However, none of those studies controlled for potentially important lifestyle characteristics that differ between patients with and without diabetes. For example, Onset Study participants with diabetes were more sedentary, less likely to consume alcohol, lived in lower income areas, and reported lower educational attainment than participants without diabetes (Table 1); all of these factors can influence survival after AMI (8,9,10). Our results confirm that diabetes is associated with nearly twofold higher long-term mortality among AMI survivors, as suggested by Table 3, even after adjustment for confounding factors.

Our finding that diabetes is associated with a greater risk of long-term mortality among women than men after AMI agrees with other studies of this topic (26,27). Mortality during hospitalization for AMI may not differ between men and women with diabetes (28).

As much as diabetes and prior myocardial infarction confer similar risks of mortality after AMI, the mechanisms that mediate the adverse effects of diabetes resemble the detrimental effects of a previous myocardial infarction, particularly related to the left ventricle, as previously described (2,29,30). For many diabetic patients, a silent infarction may have preceded their first recognized AMI, partly related to more extensive coronary atherosclerosis. Diabetic patients may have cardiac autonomic neuropathy, with attendant systolic and diastolic dysfunction. Diabetes leads to exaggerated cardiac fibrosis among patients with hypertensive heart disease. Diabetic individuals have lower left ventricular fractional shortening than nondiabetic individuals. Metabolic derangements of diabetes may cause myocardial dysfunction by depressing adenosine triphosphate production. Finally, patients with diabetes have an increased risk of sudden death, like those with a previous myocardial infarction, in part related to sympathovagal imbalance.

Diabetes may also impair prognosis in ways that do not resemble a previous myocardial infarction. For example, glycosylated end products may generate oxygen free radicals, depleting nitric oxide and impairing vasodilatation. Endothelial dysfunction accompanies diabetes, as do higher left ventricular mass and wall thickness and higher arterial stiffness. Diabetic individuals have impaired fibrinolytic potential, higher platelet aggregability, and higher fibrinogen levels. These hematologic factors can contribute to recurrent infarction, a common and important complication among diabetic individuals.

A possible limitation of our study is inaccuracy in the identification of diabetes. We relied on clinical diagnoses made by treating clinicians in the medical record and may have misclassified patients with unrecognized diabetes. Such misclassification would tend to minimize the effect of diabetes, so the relative risks reported here may be overly conservative.

In this analysis, follow-up was limited to an average of ∼3.7 years, and longer follow-up would improve how precisely we can measure the impact of diabetes on long-term mortality. However, this length of follow-up is longer than many previous studies in the thrombolytic era, and inspection of Fig. 1 shows no evidence that the effect of diabetes would change with longer follow-up.

We do not have information on what secondary prevention strategies Onset Study patients received after their index hospitalization. Considering the established benefits of aspirin, statins, and β-blockers for diabetic individuals (31,32,33), this information could help further clarify the long-term prognosis among patients with diabetes. Nonetheless, our results reflect the usual care of diabetic patients in medical centers across the U.S. in the thrombolytic era.

Our patients were hospitalized before studies confirmed that intensive blood glucose control lowers the risk of AMI and decreases mortality during and after AMI among patients with diabetes (34,35). Therefore, the benefits of strict management of blood glucose were not as apparent during the time period of our study as they are now. Unfortunately, glucose control remains suboptimal throughout the U.S. even today (36).

Given our findings, we believe that attention to the short-term prognosis of diabetic patients after AMI is insufficient. Even after hospitalization, diabetic patients carry an adverse prognosis that fails to narrow over several years. Clinicians should ensure that diabetic patients with AMI receive all recommended secondary prevention measures (37). We also encourage research on specialized primary and secondary prevention strategies for diabetic patients, such as targeted screening for CHD (38) or the preferential use of bypass surgery for revascularization (39). Specialized strategies for the acute treatment of AMI among diabetic patients, such as insulin-glucose infusions, have already shown promise (35).

In summary, diabetes is associated with a nearly twofold higher long-term mortality after AMI among Onset Study participants. This finding is consistent across subgroups, although diabetes carries a particularly ominous prognosis for women. The magnitude of risk associated with diabetes is as large as that associated with a previous myocardial infarction, highlighting the powerful negative impact of diabetes among patients surviving the early phase of AMI.

This study was supported by grants from the National Heart, Lung, and Blood Institute (R01HL41016), Bethesda, MD; the National Institute on Alcohol Abuse and Alcoholism (K23AA00299), Bethesda, MD; and the American Heart Association (9630115N), Dallas, TX.

We thank Kate Dwyer, BS, for administrative assistance and Alexander Georgakis, MA, MBA, for determining household income.