Because more than 600 000 patients in the United States are hospitalized with myocardial infarction each year [5], the aggregate expenditure on assessing LVEF after myocardial infarction is substantial. In attempts to minimize the unnecessary use of expensive technologies for this purpose, investigators [1, 6-10] have identified clinical variables that predict LVEF after myocardial infarction. Several studies have yielded clinical predictors of LVEF that are difficult to use at the bedside. Others have developed prediction rules with substantial misclassification rates [11]. Still other predictive schemes have never been validated [9]. In sum, these studies have been largely unsuccessful in providing validated clinical prediction rules that are easy to use.

We hypothesized that simple clinical variables could be used to develop a prediction rule to identify patients after myocardial infarction with a high likelihood of having preserved left ventricular systolic function. The purpose of developing such a prediction rule is to restrict assessment of LVEF to those patients most likely to benefit from the resulting information. Therefore, we first attempted to identify statistically significant, simple clinical correlates of LVEF after myocardial infarction. We then sought to combine these variables into a simple prediction rule that segregated patients after myocardial infarction into two groups: 1) a group likely to have preserved LVEF, in which further testing would be unnecessary; and 2) a group with less predictable LVEF (not necessarily decreased ejection fraction), in which technology-based assessment of LVEF might offer more prognostic information or influence therapy.

Methods

Top Methods Results Discussion Author & Article Info References

Patients

The study was a retrospective analysis of clinical data obtained as part of a prospective cohort study of 379 consecutive patients diagnosed with acute myocardial infarction at the Massachusetts General Hospital.

The 379 patients hospitalized on the medical service between January 1992 and October 1992 who had increases in the creatine kinase MB index to above 3% of total MB and either had a history compatible with myocardial infarction or had new electrocardiographic abnormalities (defined below) were eligible for the study. Patients were identified by daily review of all inpatient charts on patient floors with cardiac monitoring, including coronary and medical intensive care units and intermediate care units. Data were obtained using special data collection forms and were maintained in the Myocardial Infarction Registry. Informed consent was obtained for each patient as a requirement for inclusion in the study. The formation of the Myocardial Infarction Registry was approved by the Human Studies Committee in 1991 and has been reapproved annually.

Of 379 consecutive patients entered into the myocardial infarction database, 314 (83%; the study group) had one or more tests for assessing LVEF between days 2 and 21 after their index myocardial infarction. Of the initial 192 consecutive patients, 162 who had assessment of LVEF (methods described below) during the defined period after their myocardial infarction composed the derivation set for the clinical prediction rule. Of the subsequent 187 consecutive patients, 152 who had LVEF assessment during the defined period after myocardial infarction served as the validation set.

Data Collection and Definitions

Data on clinical history, physical examination, and laboratory tests were recorded by trained research assistants. Information was obtained primarily from patient interviews and chart reviews. Patient interviews largely eliminated the possibility of missing data for the historical information desired for this study. Those patients for whom historical data were missing were still included in the study.

The following electrocardiographic criteria were used: 1) Q waves were defined as a negative initial deflection in the QRS complex of at least 1 mV in amplitude and 40 ms in duration; 2) ST-segment elevation and depression were defined as a deflection of at least 1 mm from the baseline PR segment, 80 ms after the J point; 3) T-wave inversion was defined as a complex of at least 1 mm below the baseline PR segment; 4) left bundle-branch block was defined as a QRS duration of at least 110 ms, with a typical QRS morphologic pattern in leads V₁ and V₆; and 5) left ventricular hypertrophy with QRS widening was defined as a QRS duration of at least 110 ms with associated typical repolarization abnormalities consistent with strain in the presence of standard voltage criteria for left ventricular hypertrophy [12]. Electrocardiographic changes were classified as anterior in location if changes appeared in leads V₁ to V (₄); as inferior if changes occurred in leads II, III, aVF; as apical if changes occurred in leads V₅ to V₆; and as lateral if changes occurred in leads I and aVL.

Electrocardiograms obtained within the first 48 hours of admission were reviewed by one of the investigators (MS), who was blinded to clinical data. Electrocardiograms showing a left bundle-branch block pattern, ventricular pacing, or left ventricular hypertrophy with strain were classified as not interpretable (this occurred in 72 of 379 [19%] patients). Electrocardiograms with Q waves or ST-segment elevation in at least two limb leads or in contiguous precordial leads not known to be old were classified as Q-wave infarctions, and the location of these changes was recorded. If Q waves were present in leads in a location other than the ischemic ST-segment or T-wave changes, the patient was classified as having had a previous Q-wave myocardial infarction. Patients whose electrocardiograms showed ST-segment depression or T-wave inversion in at least two limb leads or two contiguous precordial leads were classified as having non-Q-wave infarctions, and the location of these changes was recorded. Patients whose electrocardiograms did not show clear ischemic changes were also classified as having non-Q-wave infarctions, but no location was recorded.

Congestive heart failure was defined as the report in the medical record of a previous episode of congestive heart failure or the presence of alveolar edema on a current chest radiograph.

Assessment of Left Ventricular Ejection Fraction

The ejection fraction was assessed by one or more of the following three modalities: transthoracic echocardiography, contrast ventriculography, and radionuclide ventriculography. The decision to order these tests was made by the attending physicians of the patients. In this analysis, patients studied by more than one modality were assigned a value for LVEF according to the following hierarchy: 1) echocardiographic LVEF, if not available; then 2) contrast LVEF, if not available; then 3) radionuclide ventriculographic ejection fraction.

Echocardiographic and radionuclide LVEFs were obtained from the appropriate reports available to the patients' clinicians. Each contrast LVEF was ascertained by one of the investigators (MS, GR, and CO), blinded to clinical data, using the single-plane modified area-length ellipsoid method of estimating left ventricular chamber volume [13]. The LVEFs were dichotomized as either 40% or more or less than 40%. This cutoff point for left ventricular function was preselected because of its well-recognized clinical significance [4]. For patients who had more than one test to assess LVEF, a statistic and 95% CI was calculated to assess the comparability of the various methods that assigned patients to the two categories (LVEF 40% or LVEF <40%.

Identification of Clinically Significant Echocardiographic Findings

The following clinically significant echocardiographic findings were sought in the clinical echocardiography report: 1) suspected ventricular thrombus, 2) severe valvular regurgitation, 3) ventricular or aortic aneurysm, 4) ventricular septal defect, and 5) circumferential pericardial effusion. Reports for each patient having echocardiography were reviewed for these findings by one of the investigators (GR), who was blinded to all other data.

Derivation of the Prediction Rule

The significance of the association between 20 preselected clinical variables and LVEFs of 40% or more was tested in the derivation set using Fisher exact tests, with a two-tailed of 0.05. These variables were then entered into a stepwise logistic regression model (BMDP Statistical Software, Los Angeles, California) to determine multivariate predictors of preserved LVEF. These independent predictors were subsequently used to develop the clinical prediction rule in order to maximize prediction of LVEFs of 40% or more. Predictors adding little discriminatory information, even if statistically significant, were eliminated in order to avoid adding complexity to the rule. In addition, variables found to add substantial discriminatory value to the prediction rule were included even if they were not independent predictors of preserved LVEF. After deriving the clinical prediction rule in the first 162 patients, it was tested in the validation set of 152 patients. Positive and negative predictive values were determined, along with their respective 95% CIs [14].

Results

Top Methods Results Discussion Author & Article Info References

Clinical characteristics of 314 patients included in the derivation and validation sets are listed in Table 1. In these 314 patients, a total of 200 echocardiograms, 187 contrast left ventriculograms, and 37 radionuclide ventriculograms were obtained. A total of 207 patients had only one of the above tests to assess LVEF, and 107 patients had two or more tests. Comparison of LVEFs determined by echocardiography and contrast left ventriculography ( = 0.55; 95% CI, 0.35 to 0.75) and by echocardiography and gated blood pool scan ( = 0.72; CI, 0.39 to 0.99) showed fair to good agreement beyond chance alone. Agreement between ejection fractions by contrast left ventriculography and gated blood pool scan did not exceed that expected by chance alone ( = 0.22; CI, 0 to 0.71), although only 13 patients had both of these tests.

Sixty-five patients who did not have any formal assessment of LVEF between days 2 and 21 after myocardial infarction differed somewhat from 314 patients who did have assessment of LVEF (Table 1). Most of these patients presented with small infarctions: Thirty-six patients (55%) did not have any electrocardiographic evidence of ischemia or infarction, and 33 patients (51%) had a peak total creatine kinase level of less than 8.34 µkat/L. Ten patients (15%) had evidence of a Q-wave infarction. A smaller percentage of these 65 patients died early in their hospital course and, thus, never had LVEF assessment.

Univariate and multivariate predictors of a normal LVEF in the 162 patients composing the derivation set are listed in Table 2, with identification of those determined to be multivariate predictors. Of multivariate predictors of LVEFs that were 40% or more, a peak total creatine kinase level was of borderline statistical significance and did not add discriminatory value in the prediction rule; therefore, it was not included in the prediction rule. Congestive heart failure with the index infarction, on the other hand, was a strong univariate predictor but not an independent predictor of a preserved ejection fraction; however, it did add discriminatory value to and was incorporated into the following clinical prediction rule. The LVEF is likely to be 40% or more Figure 1 in patients who have 1) an interpretable electrocardiogram, 2) no previous Q-wave myocardial infarction, 3) no history of congestive heart failure, and 4) a clinical event that is not a Q-wave anterior myocardial infarction.

View larger version (61K): [in this window] [in a new window]   Figure 1. The prediction rule. A "yes" response to any of the four questions places a patient into the "unpredictable LVEF" category. Four consecutive "no" responses suggest that a patient has a LVEF of 40% or more. CHF = congestive heart failure; LBBB = left bundle branch block; LVEF = left ventricular ejection fraction; LVH = left ventricular hypertrophy; and MI = myocardial infarction.

In the derivation set of 162 patients, this algorithm predicted that 65 patients would have preserved left ventricular systolic function (LVEF 40%). The remaining 97 patients were placed into the category for unpredictable LVEF (not necessarily decreased ejection fraction). Of those predicted to have a preserved LVEF, 64 patients (98%) actually had an ejection fraction of 40% or more (positive predictive value, 0.98; CI, 0.90 to 0.99). The one other patient had a LVEF of 35%. Of 97 patients with an unpredictable LVEF based on this algorithm, 48 patients (49%) had an ejection fraction of 40% or more (Figure 2).

View larger version (20K): [in this window] [in a new window]   Figure 2. Flow of patients in the derivation and validation sets through the prediction rule. Positive predictive values (95% CIs) are given on the extreme right. Negative predictive values are provided even though patients not meeting criteria for LVEF of 40% or more are actually assigned the label "unpredictable LVEF," not "LVEF less than 40%" (using this definition, a negative predictive value is less meaningful). LVEF = left ventricular ejection fraction; PVN = predictive value negative; and PVP = predictive value positive. *Assessed by transthoracic echocardiography, contrast left ventriculography, or radionuclide ventriculography. 95% CI for the positive or negative predictive value.

This prediction rule was then applied to the 152 patients composing the validation set. Of the 70 patients predicted to have an LVEF of 40% or more, 69 (99%) actually had an LVEF of 40% or more. The one remaining patient had an LVEF of 38%. Thus, the positive predictive value of the prediction rule was 0.99 (CI, 0.90 to 0.99). The negative predictive value was 0.43 (Figure 2).

Review of all echocardiograms (n = 200) showed that other clinically significant findings, as defined above, were more common in patients with impaired left ventricular systolic function (Table 3). In this group with measured LVEFs less than 40% as determined by echocardiography, 12 of 59 patients (20%) had other clinically significant findings, whereas in the group with ejection fractions of 40% or more, only 8 of 141 patients (6%) had these findings (P < 0.02). However, with respect to predicted LVEFs of 40% or more, other clinically significant echocardiographic findings were exceedingly rare. When 8 patients with one of these clinically significant findings and a measured LVEF of 40% or more were analyzed using the prediction rule, 6 were subsequently classified as having an unpredictable ejection fraction. Thus, only 2 patients with predicted and measured LVEFs greater than 40% had at least one of these unexpected echocardiographic findings. One patient had a mitral valve vegetation with severe mitral regurgitation, whereas the other patient had a small inferior wall aneurysm. Overall, in 80 patients undergoing echocardiography who were also predicted to have LVEFs of 40% or more, just 2 (2.5%) had other clinically significant echocardiographic findings.

Discussion

Top Methods Results Discussion Author & Article Info References

The results of our study show that simple clinical and electrocardiographic data can be combined to form a prediction rule that identifies two populations of patients after myocardial infarction: those with a high likelihood of having LVEFs of 40% or more and those with unpredictable ejection fractions. Patients likely to have a preserved LVEF may not require further testing solely for the purpose of assessing left ventricular systolic function. On the other hand, because the group with an unpredictable LVEF includes patients with normal and impaired left ventricular systolic function, technologic assessment of ejection fractions would be required to identify those with clinically significant left ventricular dysfunction.

Knowledge of the exact LVEF after myocardial infarction may not be essential for patient management. Instead, distinguishing those with relatively preserved left ventricular function from those with at least moderate dysfunction may appropriately influence the decision to recommend revascularization, angiotensin-converting-enzyme inhibitors, ß-blocking agents, or anticoagulants. For the most part, these decisions are dictated by a general sense of left ventricular function. If LVEF is relatively preserved after myocardial infarction, benefits of anticoagulation or a survival advantage from either revascularization (discounting left main disease) or institution of angiotensin-converting-enzyme inhibitors is by no means established. If LVEF is moderately or severely decreased, further diagnostic and therapeutic intervention is often warranted. Therefore, decision making in the patient who has had a myocardial infarction is reasonably guided by qualitative rather than exact quantitative information.

Clinically useful prediction rules such as ours should meet at least three criteria: 1) ease of use, 2) wide applicability, and 3) validity. Our clinical prediction rule is easy to use. It requires only one historical variable, the absence of congestive heart failure either in the past or with the index myocardial infarction. The necessary electrocardiographic variables are readily obtainable and apply to all patients with myocardial infarction. The electrocardiogram must first be interpretable (absence of ventricular pacing, left bundle-branch block, or left ventricular hypertrophy with strain) and free of Q waves, unless these Q waves are new and not in an anterior distribution. Patients meeting these criteria are likely to have preserved left ventricular systolic function (LVEF 40%), with a near perfect positive predictive value. If any one of the above criteria is not met, this rule predicts an indeterminate LVEF, with a negative predictive value of 0.43. This low negative predictive value is acceptable because the rule seeks to define a subset of patients who should have further testing, not to classify these patients in terms of their LVEF. Thus, our prediction rule is easy to use, widely applicable, and valid.

According to this prediction rule, invasive or noninvasive testing or both might be unnecessary in as many as 40% of patients after myocardial infarction if determination of LVEF is the major indication for the study. In our study group, determination of LVEF was the most common indication for ordering radionuclide ventriculograms and echocardiograms. Although these techniques, especially echocardiography, can provide additional information about cardiac structure and function, our review of all of the echocardiogram reports suggests that unexpected clinically significant findings are unusual, particularly in those patients predicted to have a preserved LVEF. Of 80 patients, only 2 had unanticipated findings on their echocardiograms. This is especially noteworthy because the choice of echocardiography instead of other modalities may have been made in part because of a previous suspicion of the existence of other significant clinical abnormalities.

With respect to rates of misclassification, our study incorrectly categorized only 1 of 70 patients in the validation set and 1 of 65 patients in the derivation set. This very low rate of failure appears acceptable when making clinical decisions because LVEFs of patients misclassified by this prediction rule are nearly 40% (in the two misclassified patients, the ejection fractions were 35% and 38%). Because we did not aim to predict patients with impaired ejection fractions, we did not have "misclassifications" in the unpredictable LVEF group (the rule does not predict ejection fractions of these patients). We calculated negative predictive values to provide clinicians with estimates of the proportion of patients in the unpredictable group who have low ejection fractions. Because the purpose of developing this prediction rule was only to predict one group of patients with a high likelihood of a preserved ejection fraction, we do not feel that our low negative predictive values are a limitation of our prediction rule or our study.

Several potential study limitations do deserve comment:

1. The study was done at a single institution, which may limit the generalizability of the results. However, our study group was large and diverse, consisting of referred patients and emergency department admissions. The possibility that our prediction rule may not do as well in smaller community hospitals must be considered; testing our rule in such a setting and with many patients would be necessary before the model could be applied widely in clinical practice.

2. Data collection was retrospective, and information bias may have occurred. This form of bias was minimized in our study because the hypothesis was conceived and tested after the database already had been created. Because all nonelectrocardiographic and non-LVEF data were collected before ascertainment of LVEF as part of a prospective registry, these data should not be biased by our hypothesis.

3. Not all patients in the database had some form of LVEF assessment. From review of clinical, electrocardiographic, and patient outcome data, it does not appear that excluding those patients who did not have assessment of LVEF from the derivation and validation sets should have favorably influenced the accuracy of the prediction rule. Of 379 consecutive patients, only 65 patients (17%) did not have an assessment of LVEF. Most of these patients appeared to have had small infarctions, as shown by low levels of creatine kinase and the low incidence of ischemia or infarction patterns on electrocardiogram review. The few patients who appeared to have sustained sizable infarctions had high early mortality; the lack of assessment of LVEF in these patients was likely because of their exceedingly poor prognosis. Therefore, we expect the prediction rule to work well in both groups because they represent opposite and readily apparent extremes in the spectrum of left ventricular function. Clinical variables in these groups correlated well with left ventricular function. In general, we expect the prediction rule to be most valuable when applied in clinical situations in which left ventricular function is more uncertain on clinical grounds.

4. Not all patients had the same test to assess LVEF. However, we prospectively designed our hierarchy for assigning a value to LVEF if more than one test was done. Our decision to use echocardiographic LVEF in preference to ventriculographic LVEF was based on our desire to study the more commonly used method for assessing LVEF at our institution and to investigate the assertion that echocardiographically derived ejection fractions are on average smaller than ventriculographically derived ejection fractions [15]. By primarily using the anticipated smaller LVEFs derived from echocardiography, we aimed to minimize the rate of misclassification of ejection fraction. In addition, reasonable agreement beyond that expected by chance was observed for LVEFs assessed by echocardiography compared with either of the two other modes of assessment. The lack of agreement beyond that expected by chance for ejection fractions assessed by contrast left ventriculography and gated blood pool scans is not of substantial concern.

Patients in whom one mode of assessment yielded an LVEF of more than 40% and the other mode yielded an LVEF of less than 40% were invariably placed in the "unpredictable ejection fraction" category by the prediction rule. As it turned out, in only one patient would use of a different strategy (for example, first using angiographic LVEFs followed by echocardiographic or radionuclide LVEFs if angiographic LVEFs were not done) have resulted in an additional misclassification. In the derivation set, one patient had an echocardiographic LVEF of 45% and an angiographic LVEF of 35% and was predicted to have an LVEF of 40% or more.

5. A potential methodologic weakness of our study design was the inconsistent timing of assessment of LVEF after myocardial infarction. Although left ventricular function may vary over time after a myocardial infarction, it is usually in the direction of improved function [16]. In the absence of reinfarction, left ventricular function typically improves in the period after myocardial infarction because of resolution of ischemia and stunning and the effects of revascularization. Because more than 85% of LVEF assessments in our study were done between days 2 and 7 after myocardial infarction, we were more likely to select for decreased ejection fractions given the relatively early assessment of ejection fractions. If this introduced systematic bias into our study, it should have only decreased the average LVEF, increasing our chances of a higher misclassification rate. Nonetheless, prospective testing of this prediction rule in a smaller group of patients with consistent timing of LVEF assessment would help evaluate the possible statistical significance of this study limitation.

6. A final potential concern for the prediction rule is the size of the validation set. It is possible that a larger validation set might contain more patients with large inferoposterior infarctions who did not have previous myocardial infarction or congestive heart failure. On the basis of the prediction rule, these rare patients would have a preserved LVEF, but they may in fact have ejection fractions less than 40%. The only way to gauge how well the prediction rule does in such patients is to test it in a sufficient number of these patients, a limitation of this study.

Overall, we do not believe there was systematic bias favoring the prediction rule with regard to data acquisition, outcome determination, or patient exclusion. However, this prediction rule should be revalidated in other settings, especially in community hospitals, before it is adopted into general clinical practice.

In our study, we sought to identify a subgroup of patients after myocardial infarction with a high likelihood of having a "preserved" LVEF. Our prediction rule identified this group with a high degree of accuracy (positive predictive value > 98%).

Although our study method did not allow precise classification of all patients, it appears to be a powerful tool in identifying a clinically relevant subgroup. This type of study design may be useful for generating cost-saving clinical prediction rules surrounding other clinical problems. Most clinical studies of diagnostic testing have attempted to identify a small group of high-risk patients from a large population. However, prediction rules that can decrease the size of the population requiring screening are also of value. Given the increasing societal imperative for physicians to define rational strategies for cutting costs while preserving quality of care, we believe that such prediction rules will have an increasingly prominent role in clinical practice.

Abbreviation

LVEF: left ventricular ejection fraction