1. Introduction

The resting ECG is the most widely used cardiovascular diagnostic test. Approximately 75 million are performed a year in the US alone and probably twice that number around the world. Currently, approximately one half are performed by physicians without special training in cardiology. Medicare payment rate for the technical and professional components is a total of $29 with most health insurances reimbursing at a similar rate. Under capitation, the exact cost cannot be calculated. There are $200 million worth of ECG recorders sold yearly in the US and most of the current models include a computerized interpretive program at an inclusive price as low as $3000. There are over 3,000 ECG over reading and storage systems in use at a price ranging from $30,000 to $200,000. Guidelines and clinical competency statements are available [Schlant et al., 1992; Fisch, 1995]

1.1. Can the ECG be Used As a Screening Tool?

The validity of using the resting 12 lead electrocardiogram as a screening test for cardiovascular disease in asymptomatic individuals has never been convincingly demonstrated. A review of the epidemiological studies that assessed the prevalence and prognostic value of electrocardiographic abnormalities could direct both the primary and secondary prevention of cardiovascular disease. Since only a few reviews have attempted to reach consensus on this topic [Sox et al., 1989; Rautaharju, 1989] we recently wrote a monograph on this subject [Ashley et al., 2000]. In this brief we will try to present our findings more succinctly and via two new figures (Fig. 1 and Fig. 2).

1.2. Criteria for a Screening Test

The value of any screening test depends critically on four key principles: its cost; the prevalence of the abnormalities detected in the population assessed; the relationship of the abnormalities to morbidity and mortality; and the possibility of reducing or avoiding future morbidity or mortality given the information provided by the test. In particular, to be worth the additional expense, the ECG must add significantly to the ability of standard risk factors to identify asymptomatic individuals with sub-clinical disease.

2. Methods

Using MEDLINE we reviewed the literature over a period of 33 years from 1966 to 1999. We attempted to identify studies where a randomly selected population of asymptomatic patients with no history of ischemic heart disease, underwent resting 12 lead ECG before a follow up of at least 5 years with respect to mortality. We identified very few studies that exactly met our criteria, so have included several studies where random sampling was not employed, or where symptomatic patients were not excluded, or where soft endpoints were used. All studies were critically assessed according to standard criteria [Sackett, 1997]. The studies included are listed in Table 1.

TABLE 1. The ECG Screening Studies

The Framingham Heart Study [Kannel et al., 1969; Kannel et al., 1970; Schneider et al., 1979; Schneider et al., 1980; Kannel and Abbott, 1984; Kannel et al., 1984; Kannel et al., 1987; Kreger et al., 1987; Kreger et al., 1987; Kannel, 1991; Wolf et al., 1991; Kannel and Cobb, 1992; Kannel, 1996; Benjamin et al., 1998; Framingham Research Group, 1999]

The Seven Countries Study [Keys, 1970]

The US Pooling project [Pooling Project Research Group, 1978]

The Finnish Social Insurance study [Reunanen et al., 1983]

The Manitoba Study [Mathewson et al., 1987; Krahn et al., 1995]

The Busselton Health Studies, Busselton City, Australia [Cullen et al., 1974; Cullen et al., 1982; Cullen et al., 1983]

Chicago Heart Association Detection Project in Industry [Liao et al., 1988]

Chicago Western Electric Study [Daviglus et al., 1999]

Copenhagen City Heart Study [Ostor et al., 1981; Truelsen et al., 1997]

White Hall study [Rose et al., 1968]

British Regional Heart Study [Whincup et al., 1995]

Italian Risk Factors and Life Expectancy Pooling Project [Menotti and Seccareccia, 1997]

The Tecumseh community health study [Epstein et al., 1965; Ostrander et al., 1965; Chiang et al., 1970]

Belgian Inter-university Research on Nutrition and Health [Kornitzer and Dramaix, 1989; De Bacquer et al., 1994; De Bacquer et al., 1998; De Bacquer et al., 1998]

The WHO European study [Rose et al., 1978]

Multiple Risk Factor Intervention Trial (MRFIT) [The Multiple Risk Factor Intervention Trial Group, 1977; Rautaharju and Neaton, 1987; Crow et al., 1997]

The Honolulu Heart program [Knutsen et al., 1988; Knutsen et al., 1988; Yano and MacLean, 1989]

Evans County Study [Tyroler et al., 1984; Strogatz et al., 1987; Hames et al., 1993] Charleston Heart Study [Sutherland et al., 1993; Arnett et al., 1997]

The Cardiovascular Health Study [Furberg et al., 1992]

The ECG and survival in the very old [Rajala et al., 1984; Rajala et al., 1985]

2.1. Limitations of Study Analysis

The majority of subjects for whom ECG and prognostic data are available are male. This is partly due to two very large prevalence studies that screened young aviators in the United States Air Force (189,418 aircrewmen) [Averill and Lamb, 1960; Hiss and Lamb, 1962]. These two studies have not been included in the plotted data for fear of biasing the averages. Although most studies included a wide age range of participants, most information is available regarding subjects 50 to 59 years of age (Fig. 3). A small number of studies have focused on groups with different racial backgrounds. [Oopik et al., 1996; Strogatz et al., 1987; Sutherland et al., 1993; Miall et al., 1972; Knutsen et al., 1988; Keys, 1970].

In assessing the value of the ECG as a screening test, studies that excluded or analyzed separately, individuals with known cardiovascular disease were of key interest. Some studies made no exclusions [Miall et al., 1972; Rajala et al., 1985; Casiglia et al., 1991; Furberg et al., 1992; Sutherland et al., 1993]. The Manitoba study [Mathewson and Varnam 1960] followed an initially young and fit population over many years as they developed cardiovascular disease. The pooling project excluded all those with major Q waves [Pooling Project Research Group, 1978]. However, many more excluded participants on the basis of more than one criterion including physician history of MI or angina pectoris, medical examination, and ECG.

The technique of random sampling is preferred to draw conclusions based on probability estimates from a sampled population to the whole population. Few studies employed random sampling [Ostor et al., 1981; Whincup et al., 1995; De Bacquer et al., 1998] raising questions over impact of the demographic make up of participants and the participation rates for the studies.

2.2. ECG Classification Systems

The "Minnesota code" early on became the de facto standard for the measurement of ECG abnormalities in epidemiological trials. Computerization has solved many of the problems that the Minnesota code was designed to address [Savage et al., 1987]. The most commonly used computer coding system in epidemiological trials has been the NOVACODE system [Rautaharju, 1989]. The Minnesota Code has also been computerized as the MEANS system from the Netherlands and runs on a personal computer under the Windows operating system [Kors et al., 1996; de Bruyne et al., 1997]. The US pooling project categorized ECG findings into major and minor groupings. Because some authors have found this useful in achieving statistical significance where the individual ECG abnormalities fail to do so, it has intuitive appeal for epidemiology. However, the clinical utility of this simple dichotomization is uncertain and, remarkably, the final report of the Pooling Project [Pooling Project Research Group, 1978] does not make it clear why these particular abnormalities were chosen, or indeed why they opted to categorize at all. Despite this, the categorization proved popular and was adopted in numerous trials [Ostor et al., 1981; Strogatz et al., 1987; Liao et al., 1988; Kornitzer and Dramaix, 1989; Smith et al., 1990; Sutherland et al., 1993]. The presence of major Q waves on the ECG was an exclusion criterion; thus, studies that used the Pooling Project categorization did not consider major Q waves.

3. Results and Discussion

A critical factor in the adoption of any screening test is the prevalence of a positive test in the asymptomatic, apparently healthy population. There were few studies that fully presented ECG data on asymptomatic participants [Pedoe, 1978; Pooling Project Research Group, 1978; Rose et al., 1978; Strogatz et al., 1987; Whincup et al., 1995; Menotti and Seccareccia, 1997; De Bacquer et al., 1998]. In addition, there was some variation in the exclusion criteria for cardiac disease (symptoms, etc.) in those studies that did. However, despite the wide inter-population variation in prevalence, and despite some studies finding no intra-population difference in prevalence of ECG abnormalities between those with a diagnosis of heart disease and those without [Liao et al., 1988], findings from 18,403 British men in the Whitehall study suggest caution in the combination of these two groups for analytical purposes.

3.1. Left Ventricular Hypertrophy

Electrocardiographic left ventricular hypertrophy (LVH) has been recognized as a risk factor for cardiac events for some time. Most of the seminal data comes from the Framingham study [Kannel et al., 1969; Kannel, 1983] but assessment of the actual impact of LVH has been confounded by the use of different definitions. The most commonly used electrocardiographic definitions of LVH have been the voltage criteria of Sokolow and Lyon [Sokolow and Lyon, 1949], Gubner and Ungerleider [Gubner and Ungerleider, 1943], and that of Casale et al. There have been many attempts at improving the predictive value of the electrocardiogram by relating the 12 lead ECG to echocardiographic estimates of LV mass [Casale et al., 1986; Rautaharju et al., 1988; Okin et al., 1995]. Casale and colleagues found that augmenting the Cornell criterion with information from the T wave in V1 improved the performance of the ECG in estimation of LV mass. Okin [Okin et al., 1995; Okin et al., 1996] suggested the use of a time-voltage criterion for identification of LVH. However, considering sex, age, body surface area, the duration of the terminal P in V1, and the S voltage in V1 and V4 explains more of the variance than this model [de Vries et al. 1996], than a linear regression model of Wolf [Wolf et al., 1991], and significantly more than standard criteria. Okin et al. examined the test accuracy of the criteria for LVH in relation to body mass index in 250 patients and confirmed the need to consider BMI in LVH estimates [Okin et al., 1996]. This was reinforced by findings from the Framingham study where incorporation of obesity and age into ECG algorithms consistently improved their performance in the detection of hypertrophy [Norman and Levy, 1995]. More recently, Rautaharju, using data from the third National Health and Nutrition survey, and the Atherosclerosis Research in Communities study demonstrated that Sokolow-Lyon voltages decreased and Cornell voltages increased significantly with increasing breast tissue [Rautaharju et al., 1998]. However, the overall conclusion was that these effects were small, and that when entered into a multivariate equation, chest size was the dominant variable.

One study found a poor correlation overall with ECG and echo criteria. Crow et al. [Crow et al., 1996] studied the association between eight ECG criteria and echocardiographic LV mass estimates in men and women with mild hypertension. Electrocardiograms and echocardiograms were recorded at baseline, 3 months, and annually for 4 years. The ECGs were computer processed to define 8 different criteria, and the researchers found a poor correlation between ECG and the echocardiogram. However, this result may have been confounded by poorly reproducible echocardiographic measurements.

Although using echocardiographic LV mass as a gold standard to refine electrocardiographic estimates is valid, more important is the prognostic value of ECG detected LVH. In a prognostic study [Verdecchia et al., 1998], the value of electrocardiographic criteria for LVH in patients with essential hypertension was evaluated. Six methods were compared. A total of 1,717 white hypertensive subjects were prospectively followed for a mean of 3.3 years. At entry, the prevalence of LVH was highest with the Perugia score [Schillaci et al., 1994] (18%) and lowest with the Framingham score (4%). During follow-up there were 159 major cardiovascular events (33 fatal). The event rate was higher in the subjects with than in those without LVH. The Perugia score best predicted cardiovascular events, accounting for 16% of all cases, while the others only accounted for 7%. LVH diagnosed by the Perugia score was also associated with an increased risk of cardiovascular mortality (4x) and outperformed the classic LVH criteria. Notably, the Perugia score exhibits lower sensitivity ratings when related to echocardiographic LV mass compared to at least one multivariate continuous model.

3.2. Prevalence of ECG Left Ventricular Hypertrophy

The prevalence of ECG LVH has a wide variation. All studies showed increases in prevalence of ECG-LVH with increasing age. The high values seen in the young men can be readily explained by physical fitness and muscular hypertrophy associated with testosterone (see High R wave, Fig. 1). We could hypothesize that with aging men are less physically active and have correspondingly lower voltage R waves. Then, with further increasing age in both men and women (Fig. 2), pathological processes set in, and the size of the R wave increases again. In fact, recent studies in both humans and animals have emphasized gender differences in the response to pressure overload. Although degree of hypertrophy seems to be similar [Douglas et al., 1998; Weinberg et al., 1999], male animals exhibit earlier transition to heart failure, with cavity dilatation, loss of concentric remodeling and diastolic dysfunction. This falls into line with human echo studies that show that for obesity and hypertension, relative increase in left ventricular mass is similar [Kuch et al., 1998] among men and women, but that overall, other factors including risk [Liao et al., 1995] are not [Dimitrow et al., 1998].

3.3. Finnish Populations and LVH

The most startling finding from these studies is the high prevalence of LVH in the Finnish populations assessed both as part of the Finnish cohort of the Seven countries study and the Finnish Social Insurance Institution study (outliers). For the 50-59 year old males, the Finnish cohort of the Seven countries study had a mean prevalence of LVH (MC 3-1) of 19%; the Finnish Social Insurance study had a mean prevalence of 27.3% and this relative high prevalence even extended to females (mean 13.5%). The figures demonstrate how far these points are outliers. Of all the other countries with predominant Caucasian populations, only Copenhagen (12%) and the Moscow cohort of the European study (18.7%) came close to these estimates. Estimates were also high in the black population, both from the Jamaica study (29.9% in the 40 to 49 age group) and 19.8% in Evans County. The wide variation is demonstrated by studies such as the Whitehall study which found a prevalence of less than 1% in British civil servants aged 50-59, and the age-pooled, white male cohort of the Charleston study. These wide variations demand some explanation.

A clearer pattern is the lower prevalence of ECG-LVH when the criterion requires ST depression. The Belgian study found the age-pooled prevalence of LVH by this definition to be 0.8% male and 0.5% female. The Honolulu Heart program found the prevalence of high R wave-LVH to be 5.4% compared to the prevalence of 0.6% when both high R waves and ST-T depression is used. Although the Charleston study found a low prevalence of ST depression inclusive LVH (0.9%) in their age-pooled, male-only sample, they in fact found LVH by high R wave criterion to be only 0.3%. This contrasted with the findings in the black population where they found the prevalence of LVH by these criteria to be 7.8%. This was mirrored in the high R wave criteria of the Evans county study that found the prevalence in blacks to be over double that in whites (19.8% compared to 7.4%).

It is not clear why there have been such a wide variety of LVH prevalence estimates from electrocardiograms carried out in different populations using the same criteria. Many studies were rigorous in their training of coders and use of independent assessments. In particular, the Finnish social insurance study used 2 independent coders, and two or three independent medical readers at the University of Minnesota read all ECGs from the Seven Countries participating centers. Some important points can be made. Firstly, it seems likely that at least some of the differences noted are real. It would not seem unreasonable to conclude on the basis of the above, that black populations and the Finnish population have truly higher mean R wave amplitude than many others. As discussed above, this may not necessarily imply a greater prevalence of echocardiographic LVH, although comparison of the relative weight and skin fold thickness measurements from the Seven Countries study suggests no difference between the Finnish population and the others (Finnish relative weight: 92.5%, others: 92%. Finnish skin fold: 15, others: 17.7). However, Finland did have the highest rates of hypertension and the CHD death rate was higher than all other countries.

3.4. Regression of LVH

It is now clear from several trials, meta-analyses [Schlaich and Schmieder, 1998], and one meta-analysis review [Jennings and Wong, 1998] that there is a strong relationship between changes in blood pressure and LVH regression. The overall ranking of anti-hypertensives according to Jennings & Wong [Jennings and Wong, 1998] was: calcium antagonists, angiotensin converting enzyme inhibitors, diuretics, alpha-blockers, beta-blockers, and lifestyle change. Percent reductions in left ventricular mass were typically 12% for ACE inhibitors, 11% for calcium channel blockers, 5% for beta-blockers, and 8% for diuretics.

Evidence of LVH regression with anti hypertensive treatment also comes from population data. Mosterd et al. [Mosterd et al., 1999] presented data from 10,333 participants who were 45 to 74 years of age at entry. From 1950 to 1989, the rate of use of antihypertensive medications increased from 2.3% to 24.6% among men and from 5.7% to 27.7% among women, while the age-adjusted prevalence of a systolic blood pressure above 160mmHg or diastolic blood pressure above 100mmHg declined from 18.5% to 9.2% among men and from 28.0% to 7.7% among women. They report that this decline was accompanied by reductions in the prevalence of LVH from 4.5% to 2.5% in men and from 3.6 % to 1.1 % in women. In fact, the Framingham investigators removed LVH from the most recent version of their prognostic score (previously the most important factor in the score) since its prevalence has declined probably due to the improved treatment of HBP [Wilson et al., 1998].

Some important data from Framingham has shown that reduction of electrocardiographic LVH is associated with a decrease in risk. Levy [Levy et al., 1994] studied 274 men (mean age, 60 years) and 250 women (mean age, 64 years) who were free of overt cardiovascular disease but manifested ECG evidence of left ventricular hypertrophy. Logistic regression analyses of pooled biennial examinations were used to determine risk for cardiovascular disease as a function of baseline voltage (sum of R wave in aVL plus S wave in V3) and repolarization abnormality. Subjects with a serial decline in voltage were at lower risk for cardiovascular disease; those with a serial rise were at greater risk. An improvement in ST depression was associated with a marginally significant reduction in cardiovascular risk in men only. Worsening of ST depression was associated with increased risk for cardiovascular disease in both sexes.

While high R wave LVH may simply be a marker of physiological response to hypertension, ST depression inclusive-LVH is associated with an up to 15 fold increase in the risk of cardiac death, making it a more potent risk factor than any other, and suggesting that we take seriously its detection and reversal. Of the cross sectional studies which presented data on individuals with no history of cardiovascular disease, only the Belgian study provided an estimation of ST depression inclusive-LVH prevalence. This study estimated the age-pooled prevalence (25yrs-74yrs) at 0.8%. That is, screening asymptomatic individuals might pick up the one person with unrecognized ST depression inclusive-LVH out of one hundred screened.

3.5. Q Waves

The prevalence of both major and minor Q waves is low in the asymptomatic population (about 1%) but, as with LVH, it increases with age. In fact, in middle age, where the increase in prevalence is most marked, our data offer some support for the notion that women lag approximately 10 years behind men in their prevalence of cardiovascular disease (compare Q waves in Fig. 1 to Fig. 2). At all ages, women have a lower prevalence than men.

Q waves in screening electrocardiograms are important as markers for latent cardiovascular disease. In fact, the syndrome of painless myocardial infarction has been recognized for some time [Roseman, 1954]. Estimates vary as to the proportion of actual infarctions that go unrecognized (i.e. silent), but the average seems to be around 30% [Nadelmann et al., 1990].

Unrecognized myocardial infarction is a common and high-risk condition. Secondary prevention measures for recognized infarction are widely recommended and often represent significant life changes for individuals who can drastically cut their risk factor profiles. The long-term risk of infarction is likely to be similar whether recognized or not. Our data suggests that for the age group 40-59, we could expect to pick up one silent MI per 100 patients from routine screening.

3.6. ST Segment Abnormalities

That the most prevalent abnormality, ST depression, is a prognostic marker for cardiovascular disease is clear from all the studies. The age adjusted CHD morbidity and mortality occurred at about twice the rate in those with this abnormality. Our data also demonstrate that the prevalence of ST segment depression increases with increasing age. Inconsistently, although similar median prevalence values are seen in the youngest age groups, in both the 40-49 and 50-59 age groups, prevalence of ST depression was lower for males than for females. This is not easy to explain, and the pattern is inconsistent with virtually all other abnormalities. It may be that it relates to the loss of the protective effect of estrogen. It is possible that interplay between the loss of these effects on a population scale, and the younger-age mortality from cardiovascular disease in men, contributes to ST depression. Notably, Q wave prevalence does not display this pattern.

The low prevalence of ST elevation in the elderly has relevance for the difficult diagnosis of pericarditis. In fact, the data suggest that a lower index of suspicion for this diagnosis in older people would be appropriate. That is, ST elevation is much more likely to be due to pericarditis than early repolarization.

3.7. Bundle Branch Block

Similar to the other findings, the bundle branch blocks increase in prevalence with increasing age. Males have a higher prevalence of RBBB at all ages, whereas the reverse is true for LBBB (Fig. 1). The reason for this is not entirely clear. RBBB may relate to smoking and lung disease, and certainly, at the time these studies were carried out, men had higher smoking rates than women. Less easy to explain is why females should display higher rates of LBBB (Fig. 2). Certainly, the pattern is most marked for the oldest age group, and we might speculate that the lesser longevity and higher CHD mortality of men might reduce the pool of those men with LBBB, leaving the prevalence of cardiomyopathy and its associated LBBB [De Maria et al., 1993; Huang et al., 1995] to increase with age (LBBB is in fact an independent prognostic indicator for idiopathic dilated cardiomyopathy [Cianfrocca et al., 1992]). Reinforcing this possibility is Framingham data that suggest a trend for higher mortality in left rather than right bundle that is more apparent in men [Schneider et al., 1981].

The increasing prevalence of LBBB with age makes the prognostic character of the abnormality in the elderly population of interest. Rajala [Rajala et al., 1985] found no increased risk of death associated with either left or right bundle branch block in a population of 559 people over the age of 85 years - a finding consistent with the earlier finding of Kitchin and Milne [Kitchin and Milne, 1977] but contradicts the findings of Caird and colleagues [Caird et al., 1974].

Figure 3. Prevalence of ECG abnormalities among 50-59 year old males and females.

These results serve to illustrate a concept first delineated by Bayes - namely, that the pre test probability of disease is crucially important to the sensitivity of a test. As detailed above, population data suggest that LBBB is associated with a poor prognosis. However, in the follow up study of US aircrewmen, there was a very low mortality over 12 years [Rotman and Triebwasser, 1975]. In fact, only 9 out of 125 subjects with LBBB and14 out of 394 subjects with RBBB died during this period.

3.8. Atrial Fibrillation and the Elderly Population

In comparison with LVH, Q waves and ST depression, the prevalence of atrial fibrillation is low. Further, it can be seen in the figures that the prevalence remains fairly low in both men and women until 70 years of age when it increases markedly. Some studies suggest that this steep rise continues. Rajala [Rajala et al., 1984] reported prevalence as high as 19.2% and 17% in men and women over the age of 85yrs while other studies [Bonard and Sears, 1959; Bensaid et al., 1974; Golden and Golden, 1974] also found values above 10%. The pathophysiological mechanism for the increase in prevalence of AF with age is not entirely certain. The 'classical' causes such as rheumatic heart disease and thyrotoxicosis are declining. Most cases seem to be related to coronary or hypertensive heart disease while no cause is found [Luderitz, 1994] in about 15%.

As for all the abnormalities, the critical question for screening is what proportion of AF goes unrecognized? Anticoagulation can cut the stroke rate in half. The prevalence in asymptomatic individuals would appear to be low, but it is unknown what percentage of the higher numbers of people suffering with increasing age goes unrecognized. Our pooled data suggests a prevalence of approximately 1% in the 50-59 year old asymptomatic population.

3.9. Sensitivity and Specificity Estimates

Any test considered as a screening test should be considered in terms of its sensitivity, specificity and predictive value. If we are to assess the prognostic value of a screening ECG, we need to compare the test characteristics to the ultimate endpoint: mortality. In this way, we can gain some idea for the amount of variance we are able to account for using the ECG. Only one paper in the literature has previously attempted this [Whincup et al., 1995]. These calculations are displayed in Table 2. As is clearly seen, the sensitivity estimates of individual ECG abnormalities are very low. We know that attributable risk relates to population prevalence and that low prevalence will result in low sensitivity, and this seems to be what is happening here. The data is calculated only from those studies with stringent exclusion criteria, so we could be certain of assessing the true screening qualities of the test. The sensitivity values seem to be highest for LVH and this almost certainly relates to the higher prevalence of this abnormality and (at least when defined by the Framingham investigators using ST depression inclusive criteria) greater risk.

It is misleading to consider only individual abnormalities in isolation. The clinician carrying out the screening will do one ECG and look for several abnormalities and it is in just such a situation that the pooling project classification might prove helpful. Accordingly, we have included some estimates based on this data. As illustrated, the sensitivity values are higher when abnormalities are pooled but still do not reach levels where we might consider the ECG useful as a screening tool (for the ultimate 'gold standard' of mortality).

The only other authors to carry out similar analyses for ECG screening were Whincup and colleagues [Whincup et al., 1995]. Two important ECG abnormalities (definite myocardial ischemia and definite myocardial infarction) were analyzed separately in the presence or absence of symptomatic coronary disease. They note that the prevalence of these abnormalities was low in their asymptomatic population, especially below 50 years of age, and that these abnormalities in combination identified only about 10% of major coronary heart disease events in a 10-year follow up. Finally, they note that the rate of major coronary disease events occurring in men identified by the test was low and of the order 14/1000 per year. The fact that these two ECG abnormalities were able to identify only 10% of major events over 10 years agrees with our sensitivity estimates. However, the point made above in relation to individual ECG abnormality estimates is important. Although these authors have considered the two highest risk abnormalities (which would presumably include the very high risk ST depression inclusive-LVH), clinicians would consider more than two abnormalities on a screening ECG. In particular, we have previously noted the high risk for heart rate in the elder population, and significant risk associated with LBBB.

4. Conclusions

In light of recent changes in the approach to primary prevention in cardiovascular medicine [Shepherd et al., 1995], we comprehensively reviewed the electrocardiogram and its utility for screening asymptomatic individuals for the development of heart disease. In so doing, we have considered the seminal epidemiological studies carried out since the beginning of the specialties of electrocardiography and epidemiology. We have confirmed that all ECG abnormalities increase with age and that some are more prevalent in men (Q waves, RBBB) while others are more prevalent in women (ST depression, LBBB). We have identified several ECG abnormalities that are associated with significant risk. A striking finding is that ST depression inclusive-LVH (that is, LVH with strain) has a 33% five-year mortality in men and a 21% five-year mortality in women [Kannel et al., 1969]. Also, unrecognized Q wave infarction is associated with the same risk as symptomatic infarction [Kannel and Abbott, 1984]. Studies have shown that ST depression is poorly reproducible, more prevalent in women yet its prevalence is associated with increasing risk [Daviglus et al., 1999]. Our review emphasizes that early repolarization is benign [Schouten et al., 1992; Mehta and Jain, 1995] and that the risk of bundle branch block depends on the population in which it appears [Rotman and Triebwasser, 1975; Schneider et al., 1979]. The prevalence of atrial fibrillation rises exponentially with age and is associated with higher risk than any other ECG abnormality in this age group [Rajala et al., 1985]. We have confirmed the often-quoted observation that T wave inversion and high voltage QRS are more common in Blacks than whites, but in general, do not predict coronary heart disease to the same extent [Sutherland et al., 1993]. We have also noted that elevated heart rates, but not ventricular premature beats, are independent risk factors specific for sudden cardiac death [Wannamethee et al., 1995].

Finally, we have estimated the utility of the ECG as a screening tool by calculating sensitivity and specificity values from some of the studies that used stringent exclusion criteria. We found that, for individual ECG abnormalities as well as for pooled categories of abnormalities, the sensitivity of the ECG for future death was too low for it to be practical as a screening tool. This almost certainly relates to the low prevalence of these abnormalities in the populations considered. However, all abnormalities increase with age, and screening with electrocardiograms is a consideration in the elderly. There is clearly much to be gained from the use of statins, alteration of other risk factors in the secondary prevention of those who have suffered silent MI, anticoagulation for atrial fibrillation, and aggressive treatment of hypertension in the presence of ST segment inclusive-LVH. The cost of the ECG is minimal and likely to decrease further as stand-alone machines are replaced by integration into personal computers. Although diagnostic criteria have been improved by computerization, many of these techniques have not been widely applied.

Many health care researchers are recommending the use of multivariate equations to identify those who would benefit from additional therapy or health promotion [Whincup et al., 1995]. We conclude that in those who are already symptomatic with ischemic or myocardial disease, the ECG can identify a subset at particularly high risk. We hypothesize that this would extend to those at high risk from diabetes, HBP and those with high-risk scores. The Framingham data suggest that although conventional risk factors relate to long-term risk, ECG abnormalities are better predictors of short-term risk [Cupples et al. 1992]. This places the ECG at a pivotal point in the identification of those with the most to gain. Increased awareness of the prognostic implications of ECG abnormalities and the newer criteria for abnormal in selected populations should allow us to optimize one of the most useful tools in this new millennium.

References