Introduction

Top Abstract Introduction Subjects and methods Results Discussion References

Identification and treatment of patients with left ventricular systolic dysfunction improves survival and reduces morbidity.^1-5 About 3% of the adult population have systolic dysfunction,^6-8 but half of them are asymptomatic and can be identified only by objective methods, usually echocardiography. Identification consequently rests on selecting the right patients for echocardiography. Patients with a history of ischaemic heart disease are an obvious risk group,⁶ but greater sensitivity is achieved by examining anyone with signs suggestive of heart disease. The purpose of the study was to examine how primary care doctors might identify patients at increased risk of systolic dysfunction in order to decide who to refer for echocardiography. We evaluated simple and inexpensive clinical methods that can be used in general practice.

	Subjects and methods

Top Abstract Introduction Subjects and methods Results Discussion References

Patients
This study is a substudy of a cross-sectional survey that aimed to determine the prevalence of heart failure in general practice. The required population size was calculated as 2200 to give a standard error of about 0.5% on the estimate of heart failure prevalence. We asked the National Health Insurance Register for names and addresses of all patients over 40 years of age who were registered with three general practices on three separate dates in 1993 to 1995. The practices were chosen because they had used the same computer program for case notes (Docbase, Roskilde, Denmark) over two to four years and because they were situated in geographically and socially different areas of Copenhagen. The study was approved by the local ethics committee, and all examined patients gave informed consent.

Clinical examination
All patients had a physical examination, chest radiography, echocardiography, supine 12 lead electrocardiography, blood pressure measurements, and blood samples taken. OW and JFH independently coded the electrocardiograms using the Minnesota system⁹ without knowing the patient's clinical status.

Echocardiography
Left ventricular ejection fraction was indirectly estimated from the fractional shortening ^{10 11} or from a nine segment model for assessing wall motion index score. ^{12 13} Left ventricular systolic dysfunction is indicated by a wall motion index score below 1.5 or a fractional shortening below 0.26that is, roughly an ejection fraction below 0.45. Left ventricular ejection fraction values were determined from videotapes and photoechocardiograms at the end of the study by OW, who was blinded to other clinical data. The intraobserver and interobserver (CTL v OW) standard deviations for a single ejection fraction estimate were equal (0.05 units, coefficient of variation=8%).

N-terminal atrial natriuretic peptide
Concentrations of N-terminal atrial natriuretic peptide were determined from frozen stored plasma samples with a commercial radioimmunoassay (OY NT-pro-ANP ¹²⁵I radioimmunoassay, Biotop, Helsinki, Finland). Plasma samples were drawn after 10 minutes' supine rest and analysed within 12 months. The combined between assay and within assay coefficient of variation was 3.7%. The normal range was obtained from 22 patients who, after examination, were concluded to have normal cardiac function. The mean concentration plus 2 SD in these 22 patients was 0.77 nmol/l, so values above 0.8 nmol/l were defined as abnormal.

Statistical analysis
Predictive markers of left ventricular systolic dysfunction (ejection fraction 0.45) were identified by the ² test with Yates's continuity correction and a multiple linear regression model against the natural logarithm of ejection fraction. The logarithmic transformation was used to weight variables operating on the important lower ejection fraction spectrum. To reduce the risk of chance findings, we maintained only variables that were significant in two equally sized random subsets after sex and systolic dysfunction had been stratified for. Finally, a logistic regression model was used to make a clinically useful algorithm. The computer package Statistica (Statsoft, Tulsa, USA) was used for all calculations, and P values <0.05 in two sided tests were considered significant. Confidence limits around percentages were derived from the binomial distribution.

	Results

Top Abstract Introduction Subjects and methods Results Discussion References

Study population
We screened 2158 patients from three general practices; the demographic and socioeconomic characteristics were similar to those of the total Copenhagen population. About 2.5% (55) had a blank or missing case record. These were coded as cardiac healthy unless the questionnaire indicated otherwise. The questionnaire was returned by 87% (1504/1757) of patients aged 40 to 80 years. But among patients aged over 80 years, 48% (191/401) either did not respond or were resident in nursing homes. Nursing home residents had more heart disease than respondents, but it was often of unknown cause. The prevalence of a history of heart disease was similar in non-respondents and respondents.

Univariate markers of systolic dysfunction
Table 3 shows that a normal electrocardiogram indicates a normal systolic function and that systolic dysfunction can occur despite a normal plasma concentration of N-terminal atrial natriuretic peptide. The interaction term "heart rate greater than diastolic blood pressure" was superior to both heart rate and blood pressure individually. We initially evaluated "heart rate minus diastolic blood pressure" in the ² test using various cut off values because we had noted that heart rate and diastolic blood pressure were oppositely related to ejection fraction with numerically similar coefficients. A zero difference predicted systolic dysfunction best, hence we used a "greater than" relation. A history of heart failure and myocardial infarction were not as significantly associated with systolic dysfunction. Dyspnoea, chest radiography, and results of the physical examination were not significantly associated with systolic dysfunction (P>0.1).

Multivariate regression analysis
The logarithm of ejection fraction was inversely associated with N-terminal atrial natriuretic peptide concentration, heart rate, history of myocardial infarction (P<0.001), left bundle branch block in the electrocardiogram (P<0.002), history of heart failure, and Q waves in the electrocardiogram; it was positively associated with diastolic blood pressure and female sex (P<0.05). Sex and history of heart failure were eliminated because they were not significant in both patient subsets. The remaining significant variables were then used to form three clinically useful "yes or no" questions (table 4). Table 5 shows how to combine questions to assess the risk of systolic dysfunction. The probability of systolic dysfunction was very low if the patient had a normal electrocardiogram, regardless of the other questions. The probability increased if an abnormal electrocardiogram was accompanied by another positive answer, with a yes to all three questions being most specific.

	Discussion

Top Abstract Introduction Subjects and methods Results Discussion References

We evaluated a range of simple diagnostic tests to assess the risk of left ventricular systolic dysfunction in 126 patients from primary care with past or present signs or symptoms of heart disease. A normal electrocardiogram was the only clinically useful test to rule out systolic dysfunction with a sufficiently high accuracy. No single test could diagnose systolic dysfunction. Combined use of heart rate, diastolic blood pressure, and N-terminal atrial natriuretic peptide concentration in patients with an abnormal electrocardiogram, however, was accurate enough to be used to identify patients who should or should not be referred for echocardiography. Those with "yes" to all questions may be treated as if they have systolic dysfunction while waiting for the results of echocardiography, since they almost certainly will have it.

The electrocardiogram was the most useful test in this stable population from primary care. This has also been found true for patients referred for open access echocardiography,¹⁴ in a general population,¹⁵ and in referred patients with chest pain,¹⁶ dyspnoea,¹⁷ and myocardial infarction. ^{18 19} As in a recent study of stable myocardial infarction patients from primary care,²⁰ we found that a single measurement of plasma N-terminal atrial natriuretic peptide could not discriminate between minor degrees of systolic dysfunction and preserved systolic function. Measuring brain natriuretic peptide instead would probably not have changed our conclusions, since it has only a slightly higher association with systolic dysfunction than N-terminal atrial natriuretic peptide. ^{15 20 21} The close relation between heart rate and blood pressure gives reason for a joint assessment. Although the combined variable was highly significant (tables 3 and 4), it should be used with caution until it has been validated in other settings.

Application of results
Our results are robust because the two independent regression models minimised risk of finding significant variables by chance and because we used unbiased ejection fraction values and a clinically relevant study population for screening. We would probably have identified more variables with a higher significance if we had studied more patients and used a more reproducible measurement of ejection fraction than echocardiography. The algorithm will probably identify fewer false positives results if applied to a hospital outpatient population, where the prevalence of systolic dysfunction is higher. This screening method needs to be compared with other screening methods in a prospective randomised controlled trial with different samples of patients; the trial should incorporate a health economic evaluation. Other screening methods could be simplified echocardiography,²² computerised heart auscultation, or new biochemical markers.