Risk assessment of left ventricular systolic dysfunction in primary care: cross sectional study evaluating a range of diagnostic tests

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.

We included patients with any past or present signs or symptoms of heart disease. Suspected cases were identified within the cross sectional study population by reviewing all general practice case notes and prescription lists and by a mailed questionnaire. We searched and asked for details of past or present palpitations, chest pain, or dyspnoea; treatment for heart disease; referral to coronary care unit;and recognised cardiac abnormalities on previous electrocardiography, chest radiography, or other cardiac imaging. If this screening raised any suspicion of heart disease, patients were given a personal interview, and their hospital discharge letters and hospital records were checked.

We excluded patients who lived in a nursing home, were receiving inpatient or outpatient treatment for advanced heart failure, did not respond to the questionnaire, or could not or refused to be examined.

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 shortening10 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 0.26—that 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 (CTLv 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 theχ² 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.

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.

We conducted telephone interviews with almost 500 patients. We identified 357 with past or present signs or symptoms of heart disease and deliberately excluded 38 nursing home residents, 36 non-respondents to the questionnaire, and 10 patients with advanced heart failure. In the early stages of the study 21 patients were not invited to participate because of a preliminary decision not to invite patients with a vague indication of heart disease; this decision was dropped in later stages, and 14 such patients were invited. A further 126 patients were excluded for various reasons: 32 declined the invitation, 1 died,37 were disabled due to various medical and psychosocial conditions, and 56 seemed to be interested but did not turn up for examination. These 126 patients were more likely to be over 80 years of age and less likely to have a registered history of ischaemic heart disease than the 126 patients who were examined. Hypertension, diabetes, chronic obstructive pulmonary disease, and sex were, however, equally distributed in the two groups.

The 126 examined patients were aged 49 to 93 years (5th and 95th percentiles, 53 and 83 years) and considered clinically stable from the case notes. Table 1 shows that 64 patients had good evidence of heart disease and 62 had suggestive evidence. A specialist in internal medicine or cardiology had seen 33 (26%) of the patients once or twice within the past year. Mean ejection fraction was normal, and most patients were in New York Heart Association classes I or II. The prevalence of left ventricular systolic dysfunction was 12%(nine men and six women). Table 2 describes these 15 patients.

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 χ² 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 withN-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.

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.

Congestive heart failure and left ventricular systolic dysfunction are not identical. In patients with ongoing symptoms suspected by a general practitioner to be due to heart failure, plasma brain natriuretic peptide concentration seems to be a useful indicator of which patients require further clinical assessment.23 Screening for stable systolic dysfunction, however, should be targeted at patients with symptoms or signs suggestive of heart disease,6 15 hypertension, or diabetes, who are easily identified in general practice.

Our algorithm could considerably improve the identification of patients with systolic dysfunction in primary care. However, not everyone with left ventricular systolic dysfunction in the community would be identified. Firstly, some patients with systolic dysfunction may not have complained of or have been registered with cardiac symptoms in general practice. Secondly, many patients over 80 years of age were unwilling to attend hospital for examinations, and a home echocardiographic service would probably have identified more patients.24