Echocardiography for Assessing Cardiac Risk in Patients Having Noncardiac Surgery

Methods

Patients

The study sample was drawn from a prospective cohort of 474 male veterans who were scheduled to have elective noncardiac surgery that required general anesthesia at the San Francisco Veterans Affairs Medical Center as part of the study of Perioperative Ischemia [4]. These men had definite coronary artery disease or a high risk for developing this disease. Consecutive patients who met entry criteria were enrolled from January 1987 to September 1989. The Committee on Human Research approved the study protocol, and all patients gave informed consent. Patients were considered to have definite coronary artery disease if they had previously had myocardial infarction, had typical angina, or had atypical angina with evidence of inducible ischemia shown by exercise testing [18] or evidence of myocardial perfusion defects shown by scintigraphy [19]. Patients were classified as having a high risk for coronary artery disease if they had previously had or were scheduled to have vascular surgery or if they had two or more of the following cardiac risk factors (in addition to male sex): diabetes mellitus, hypertension, age of 65 years or greater, current smoking, or a serum cholesterol level of at least 6.2 mmol/L (240 mg/dL). We excluded patients with paced rhythms or complete left bundle-branch block.

Echocardiography was successfully done before surgery in 368 patients (78%). Echocardiograms were unavailable for the other patients because of scheduling problems, technical failures, or inadequate examinations. Our cohort comprised the 339 patients (92%) whose echocardiograms could be interpreted.

Data Collection and Measurement

A study physician did a routine clinical evaluation and reviewed the medical record of each patient before surgery. Clinical variables of interest were information from the history, physical examination, electrocardiography, and laboratory studies. All cardiovascular medications were recorded. The Canadian Cardiovascular Society classification of angina [20], the New York Heart Association classification of heart failure [21], the Goldman cardiac risk index [3], the Detsky risk index [5], and the American Society of Anesthesiologists classification of anesthetic risk [22] were also determined for each patient. Surgery was classified as major vascular (108 patients), intra-abdominal (76 patients), intrathoracic (21 patients), and other (134 patients; this category consisted of orthopedic, neurosurgical, general, plastic, and head and neck procedures).

Preoperative testing included routine laboratory studies and 12-lead electrocardiography. The 12-lead electrocardiogram was analyzed for evidence of previous myocardial infarction using the Minnesota Code criteria [23] and was analyzed for left ventricular hypertrophy using the Sokolow-Lyon [24] and Romhilt-Estes [25] criteria.

Standard two-dimensional and M-mode echocardiography were done in each patient before surgery. Standard parasternal long- and short-axis views, apical two- and four-chamber views, and subxiphoid views were obtained. A General Electric Pass II machine (Milwaukee, Wisconsin) equipped with 2.5- and 3.5-MHz transducer probes was used. Videotapes of the echocardiograms were read by one of two research cardiologists who were blinded to the clinical history, research data, and outcome events. The variability with which echocardiograms are interpreted in our laboratory has been reported previously; the interobserver, intraobserver, and interexamination agreement rates all exceed 90% [26].

Echocardiographic variables of interest were left ventricular systolic ejection fraction, regional wall motion abnormalities, and the presence of left ventricular hypertrophy. Ejection fraction was assessed visually, and numeric estimates were assigned when the endocardial resolution of both short-axis and four-chamber views or four- and two-chamber views was adequate. Left ventricular wall motion was assessed using standard techniques [26]. We used a standard system to assign wall motion scores for the anterobasal, anterolateral, apical, posterobasal, and diaphragmatic walls. Each segment was graded as 1 (normal wall motion), 2 (mild hypokinesis), 3 (severe hypokinesis), 4 (akinesis), or 5 (dyskinesis). The total wall motion score was calculated as the sum of the regional wall motion scores (range, 5 to 25). Left ventricular hypertrophy, determined by measuring wall thickness in the parasternal views, was deemed to be present if either the septal or the posterior wall was at least 11 mm thick.

Clinical Care and Outcomes

Research data were collected by independent study physicians in conjunction with routine clinical care. Patients were interviewed and examined by a study physician before surgery and then daily from the day of surgery until hospital discharge. Physicians directly involved in patient care were blinded to the research data, including the echocardiographic information. A 12-lead electrocardiogram was obtained before surgery, each day for the first 7 days after surgery, on days 10 and 14 after surgery, at hospital discharge, and when clinically indicated. Serum cardiac enzyme levels were measured before surgery, on days 1 and 5 after surgery, and when clinically indicated by symptoms or electrocardiographic changes [4]. Clinicians had full independence in medical decision making.

Perioperative cardiac outcomes were documented by study physicians and were validated by two independent investigators who were blinded to the clinical and echocardiographic data obtained before surgery. The following adverse cardiac outcomes were considered hierarchically, in descending order of severity: cardiac-related death, nonfatal myocardial infarction, unstable angina, congestive heart failure, and ventricular tachycardia. If a patient had several outcomes, we considered only the most severe event. Cardiac-related death, nonfatal myocardial infarction, and unstable angina were classified as ischemic events.

Cardiac-related death was defined as death caused by myocardial infarction, dysrhythmia, or congestive heart failure. Diagnosis of myocardial infarction required an elevation of the creatine-kinase MB isoenzyme level (≥ 0.83 µmol/L per second, equivalent to 50 U/L) and at least one of the following: development of new Q waves (as defined by Minnesota Code I.1 or I.2 [23]); development of persistent ST-T wave changes (as defined by Minnesota Code IV or V [23]); or evidence of acute infarction on necropsy. Unstable angina was defined as severe precordial chest pain that was not related to the surgical incision, lasted 30 minutes or longer, did not respond to standard therapies (rest and nitroglycerin), and was associated with transient ST-segment and T-wave changes without the development of Q waves or elevated enzyme levels.

Diagnosis of congestive heart failure required 1) symptoms or signs of pulmonary edema [shortness of breath and rales]; 2) signs of new left or right ventricular failure [cardiomegaly, a third heart sound, jugular venous distention, and peripheral edema]; 3) abnormal results on chest radiography [vascular redistribution and interstitial or alveolar edema]; and 4) a change in medication that involved at least treatment with diuretics. Ventricular tachycardia was defined as five or more consecutive premature ventricular beats at a rate of at least 100 beats/min [6] detected by continuous two-lead Holter monitoring. Monitoring was done in all study patients during a period of at least 48 hours after surgery.

Statistical Analysis

Potential univariate predictors of adverse cardiac outcomes were identified using the chi-square or the Fisher exact test (for dichotomous variables) and logistic regression (for categorical and continuous variables). Continuous variables are reported with means and SDs. Odds ratios and 95% CIs were based on the SE of the logistic coefficients. Two-sided P values and 95% CIs are reported. Likelihood ratios were calculated as the probability that the echocardiographic result would occur in patients with adverse events divided by the probability that the test result would occur in patients without adverse events [27]; CIs for likelihood ratios were calculated according to the method of Simel and colleagues [28]. To evaluate the incremental value of echocardiography, we used the likelihood ratio test to determine whether adding the echocardiographic variables to multivariable models that contained known clinical risk factors produced a better model fit [29]. The effect of adding the echocardiographic data to the prediction models on sensitivity, specificity, and predictive values was examined by comparing classification tables for the nested logistic regression models. We used SAS statistical software (SAS Institute, Cary, North Carolina) for all statistical analyses.

Results

Patient Characteristics and Outcomes

Most of the 339 study patients were middle-aged or elderly men with definite coronary artery disease or multiple risk factors (particularly smoking and hypertension) for coronary artery disease (Table 1). Half had vascular disease. Clinical, demographic, and surgical characteristics at baseline did not significantly differ in patients who had echocardiography and those who did not (P > 0.07 for all comparisons).

Fourteen percent of patients had an ejection fraction less than 40%, and 41% had some degree of depressed left ventricular function (Table 2). Ejection fraction ranged from 15% to 75%. Wall motion scores ranged from 5 (normal) to 20 (severely dyskinetic). Sixty patients (20%) had echocardiographic evidence of left ventricular hypertrophy. Echocardiography showed left ventricular hypertrophy twice as often as did electrocardiography.

Adverse cardiac outcomes occurred in 65 patients (19%). Ten patients (3%) had ischemic events (cardiac-related death [4 patients], nonfatal myocardial infarction [5 patients], and unstable angina [1 patient]), 26 (8%) had congestive heart failure, and 29 (8%) had ventricular tachycardia. All cases of ventricular tachycardia were asymptomatic, and none resulted in sustained ventricular tachycardia or ventricular fibrillation. The rate of adverse cardiac events was not statistically significantly different between patients who had echocardiography and those who did not (data not shown).

Univariate Associations between Echocardiographic Data and Outcomes

Univariate analyses showed that when all events were considered together, ejection fraction and total wall motion score were associated with negative perioperative cardiac outcomes (Table 3). Patients with an ejection fraction less than 40% were more than 3 times as likely to have an adverse cardiac event as were those with greater ejection fractions (95% CI, 1.8 to 6.7 times as likely). An increasing wall motion score, which indicates more extensive or severe abnormalities, also predicted all cardiac complications (odds ratio for each 3-unit increase in wall motion score, 1.6 [CI, 1.3 to 2.1]). The presence of left ventricular hypertrophy was not associated with poor cardiac outcomes (odds ratio, 1.2 [CI, 0.7 to 2.0]; P > 0.2).

No echocardiographic measurement significantly predicted ischemic outcomes. Ejection fraction, however, was associated with heart failure after surgery. For every 10% decrease in ejection fraction, the odds of postoperative heart failure increased by 1.5 (CI, 1.1 to 2.0). When episodes of ventricular tachycardia alone were analyzed, both ejection fraction (odds ratio per 10-unit decrease, 1.6 [CI, 1.2 to 2.2]) and wall motion score (odds ratio per 3-unit increase, 1.6 [CI, 1.2 to 2.2]) were significant univariate predictors. Ejection fraction was also associated with congestive heart failure and ventricular tachycardia when dichotomized at cut-points of less than 30% and less than 40%, but not at a cut-point of less than 50%.

Multivariable Predictors of Adverse Events

In multivariable models that adjusted for preoperative clinical variables, ejection fraction and wall motion score remained significant predictors of all cardiac outcomes. However, the adjustment attenuated the magnitude of the associations (compare Table 4 with Table 3). Every 10-unit decrease in ejection fraction indicated a 1.4-fold increased risk for a postoperative cardiac event (CI, 1.1 to 1.9). Similarly, a 3-unit increase in the wall motion score increased the chance of a negative event by 1.3 (CI, 1.0 to 1.7).

However, when adverse outcomes were examined according to clinical category, the echocardiographic measurements no longer added any significant information to known clinical risk factors (Table 4). No echocardiographic measurements were associated with ischemic events. After adjustment for history of congestive heart failure, neither ejection fraction (odds ratio, 1.3 [CI, 0.9 to 1.9]) nor wall motion score (odds ratio, 1.1 [CI, 0.8 to 1.6]) was a statistically significant predictor of postoperative congestive heart failure. Similarly, in models of ventricular tachycardia that considered history of definite coronary artery disease and digoxin use, the echocardiographic variables were no longer significant (odds ratio of ejection fraction, 1.3 [CI, 0.9 to 1.8]; odds ratio of wall motion score, 1.3 [CI, 0.97 to 1.8]). Dichotomization of ejection fraction into more clinically relevant cut-points (<30%, <40%, or <50%) did not change the results of the multivariable analyses.

Sensitivity, Specificity, Predictive Values, and Likelihood Ratios

We also assessed the usefulness of echocardiography by examining the characteristics of echocardiography as a diagnostic test. An ejection fraction less than 40% had a sensitivity of 0.28 to 0.31 and a specificity of 0.87 to 0.89 for all categories of cardiac complications studied (Table 5). The positive predictive values of an ejection fraction less than 40% were also low for patients in this sample: 0.40 for all cardiac events, 0.17 for congestive heart failure, 0.17 for ventricular tachycardia, and 0.06 for ischemic outcomes.

The limited usefulness of ejection fraction alone as a prognostic test also became apparent when we calculated likelihood ratios for each distinct clinical outcome (Table 6). For all outcomes examined, discrimination was poor across the range of systolic function from ejection fraction of 30% to 60%; the CIs were wide and in most cases overlapped with 1 (indicating a nonsignificant value). The finding of a normal ejection fraction (50% to 59%), mildly depressed ejection fraction (40% to 49%), or moderately depressed ejection fraction (30% to 39%), with likelihood ratios statistically no different than 1.0, means that the post-test probability of an adverse cardiac event would not significantly differ from the pretest probability. Only extreme ejection fractions (<30% or ≥ 60%) tended to be associated with likelihood ratios that differed from 1.0; this is consistent with the findings of our univariate analyses.

Because in clinical practice echocardiography is done in the context of a patient with certain risk factors, we constructed Table 7 to show how various ejection fraction and wall motion score findings would change the predicted probabilities of negative cardiac outcomes after adjustment for known clinical predictors. Adding the echocardiographic variables to a clinical model that contained history of vascular surgery, dysrhythmias, coronary artery disease, and use of digoxin as risk factors only slightly increased the sensitivity for identifying persons at risk. Specificity and negative predictive value did not change. The incremental increases in positive predictive value were small to modest. Overall, adding the echocardiographic information minimally enhanced the ability to distinguish patients who had postoperative events from those who did not, as indicated by a minimal change in the c-statistic listed in Table 7.

Discussion

Our principal finding was that echocardiography routinely obtained before surgery provided limited clinical information on cardiac risk in patients having noncardiac surgery. The echocardiographic measures of ejection fraction and wall motion score did not identify patients at high risk for cardiac-related death, myocardial infarction, or unstable angina. As expected, low ejection fraction was a univariate predictor of congestive heart failure and ventricular tachycardia. However, ejection fraction was not a significant multivariable predictor of congestive heart failure in models that adjusted for known clinical risk factors. Similarly, the association between wall motion score and ventricular tachycardia seen with univariate analysis was no longer significant after multivariable analyses. Although ejection fraction and wall motion score were statistically significant multivariable predictors of all cardiac outcomes combined, the clinical value of these variables was limited by poor operating characteristics [30, 31]. Ejection fraction had low sensitivities and positive predictive values for all clinical conditions examined. Even if clinicians used the best-performing cut-point (ejection fraction <40%), they would still fail to identify 71% of all adverse outcomes. Conversely, only 40% of patients with an ejection fraction less than 40% would have an negative event, and most of these events would be isolated episodes of nonsustained ventricular tachycardia.

When analyzed by using likelihood ratios, ejection fraction also provided poor discrimination. In most instances, a powerful diagnostic test should have a likelihood ratio close to 0 for a negative result and a ratio of at least 10 to 20 for a positive result [27, 31]. The likelihood ratios for ejection fraction were too close to 1.0 (a value indicating a useless test) to help clinicians identify patients at very high or very low risk for adverse outcomes.

Testing done before surgery using such a procedure as echocardiography is not done in a vacuum; rather, it is done in addition to traditional clinical risk stratification. The echocardiographic data, however, did not substantially change the sensitivity, specificity, negative predictive value, or accuracy of risk prediction models after adjustment for known clinical risk factors. Because noninvasive testing before surgery is usually done to “clear” a patient for surgery (that is, to rule out a high-risk situation), the inability of echocardiography to modify the negative predictive values of the prognostic models underscores the test's disappointing performance as a tool for assessing risk before surgery. The echocardiographic results modestly influenced the positive predictive value; for example, an ejection fraction less than 40% would increase the positive predictive value from 0.50 (on the basis of a model containing only clinical data) to 0.73. Whether a change of this magnitude would be sufficient to alter a clinician's threshold foregoing a procedure or changing a patient's management (by altering surgery, type of anesthesia, medications, or monitoring of hemodynamics) is unknown.

Our results confirm and extend the findings of three smaller studies of echocardiography that reported conflicting associations between ejection fraction and postoperative complications [17, 32, 33]. Takase and colleagues [32] reported that decreased ejection fraction was not a significant predictor of death or myocardial infarction in 53 patients having major nonvascular surgery. In this retrospective study, however, decreased ejection fraction was a multivariable predictor of all cardiac events and perioperative pulmonary edema. McEnroe and colleagues [33] made similar observations after analyzing the findings of 59 echocardiograms that were obtained before surgery in patients having repair of an abdominal aortic aneurysm. In this historical case series, ejection fraction did not predict cardiac-related death, myocardial infarction, or myocardial ischemia. All 13 ischemic events occurred in the 42 patients who had an ejection fraction of 50% or greater. A recent unblinded study of a consecutive series of 250 patients having peripheral vascular surgery [17] also reported a univariate association between low ejection fraction and total cardiac events. This study compared the operating characteristics of ejection fraction alone with those of clinical risk indices alone but did not determine whether echocardiography added any incremental information to the clinical risk indices.

Our data are also consistent with the findings of most studies that measured ejection fraction by radionuclide angiography. For example, Baron and coworkers [11] reported that in a prospective cohort of 457 patients, ejection fraction measured by radionuclide angiography was associated with postoperative congestive heart failure but not ischemic events or death. The findings of several older studies have been mixed. Two small studies [12, 13] found an association between ejection fraction and cardiac risk in patients having vascular surgery, but subsequent studies [14, 15, 34, 35] found no relation between ejection fraction measured by radionuclide angiography and perioperative cardiac complications.

Although ejection fraction and wall motion abnormalities are known predictors of illness and death in many other settings [36-39], these variables may not be as useful as indicators of perioperative cardiac-related death or risk for myocardial infarction in patients having noncardiac surgery. Ejection fraction, as measured by two-dimensional echocardiography, shows left ventricular systolic function at rest but provides little information on ventricular function or myocardial blood flow during conditions of added cardiac stress, such as surgery and the period immediately after surgery. Several investigators have recommended the use of stress techniques with echocardiography. Dobutamine, exercise, or dipyridamole can be used to obtain information on ventricular performance during increased cardiac workload, increased myocardial oxygen demand, or changes in coronary blood flow [40-42]. The application of these stress echocardiography techniques to the assessment of risk before surgery may merit further investigation.

Our study had several strengths and limitations. Because the clinicians directly involved in patient care (including the anesthesiologists and surgeons) were blinded to the echocardiographic findings, knowledge of the test results could not have influenced perioperative management or determination of outcomes, as was possible in previous studies [43]. In addition, by presenting operating characteristics, likelihood ratios, and the manner in which the echocardiographic information might modify risk predictions on the basis of standard clinical assessment, we allow the readers to be the ultimate judge of the test's potential usefulness in their own practices.

Ours is the largest study of echocardiography to date, and event rates are similar to those cited in previous reports [2, 11, 44]. However, our finding of a significant multivariable association between the echocardiogram and all cardiac outcomes, but not the individual categories of events, may indicate that our study had insufficient power to detect weak associations for events that occurred infrequently. This caveat is common to all studies of postoperative cardiac complications [45, 46]. The cardiac troponin I assay, which was not available during our study, may help improve the detection of myocardial infarction after surgery [47]. It is also worth noting that the association between the echocardiographic data and all outcomes appeared to be primarily the result of weak associations between low ejection fraction and wall motion abnormalities and ventricular tachycardia and congestive heart failure. However, whether these complications are serious enough to merit such additional screening is debatable. In our cohort, ventricular tachycardia and congestive heart failure were not associated with such serious or lasting adverse consequences as ischemic outcomes [4], silent ischemia [48, 49], or illness and death at 2 years of follow-up [50].

All patients in our study were male veterans known to have or suspected of having ischemic heart disease. Although these characteristics may limit the generalizability of our findings, it is likely that if echocardiography would be clinically useful in any group, that group would be an intermediate- and high-risk group, such as ours. Finally, because only a few patients in our study had clinically significant valvular heart disease (<5%), we cannot comment on the potential benefit of echocardiography in these patients. Because previous studies indicate that patients with severe aortic stenosis have increased cardiac risk [3, 5], the use of echocardiography to evaluate patients who are suspected of having high-grade aortic stenosis or other clinically significant valvular disease remains appropriate. However, routine echocardiographic screening in our cohort detected no clinically important unsuspected valvular disease.

Overall, measurements from transthoracic echocardiography were associated with cardiac complications when events were broadly defined to include all episodes of ischemia, congestive heart failure, and ventricular tachycardia. However, echocardiography was not useful in predicting cardiac-related death, myocardial infarction, or unstable angina. In addition, the echocardiographic data added no incremental prognostic information to simple predictive models of congestive heart failure or ventricular tachycardia that included known clinical risk factors. When evaluated rigorously as a diagnostic test, ejection fraction also did poorly. Thus, we found no evidence to support the routine use of preoperative echocardiography for assessing cardiac risk in patients having noncardiac surgery. These findings may help foster a more evidence-based, cost-effective approach to assessing cardiac risk and may influence the development of guidelines for the appropriate use of echocardiography.

Appendix

The Study of Perioperative Ischemia Research Group is part of the Multicenter Study of Perioperative Ischemia Research Group, a collaboration of multicenter investigators in anesthesiology, cardiology, surgery, epidemiology, and other disciplines. The following persons were directly involved with the current project: Dennis T. Mangano, PhD, MD (principal investigator); Warren S. Browner, MD, MPH, Milton Hollenberg, MD, Julio F. Tubau, MD, and Joseph A. Rapp, MD (associate investigators); Martin G. Wong, RDMS, Elizabeth Layug, MD, and Juliet Li (technical staff and research associates); Ida M. Tateo, MS (data management); Barry Massie, MD (consultant); and Kanu Chatterjee, MD, H. Barrie Fairley, MD, Lawrence W. Way, MD, and Warren Winkelstein, MD, MPH (policy and data monitoring board; Dr. Winkelstein was the chair of this board).

Presented in part at the Society of General Internal Medicine 17th Annual Meeting held in Washington, D.C., 24 April 1994.

Dr. Browner: General Internal Medicine, Veterans Affairs Medical Center 111A1, 4150 Clement Street, San Francisco, CA 94121.

Dr. Tubau: Division of Cardiology, University of Southern California-Los Angeles County Hospital, 1200 North State Street, Los Angeles, CA 90033.

Ms. Tateo: Ischemia Research and Education Foundation, 250 Executive Park Boulevard, Suite 3400, San Francisco, CA 94134.

Dr. Mangano: Department of Anesthesia, Veterans Affairs Medical Center, 4150 Clement Street, San Francisco, CA 94121.