Table 1. Specifity, sensitivity, reported prevalence and correct fraction in the reported precvalence by two sets of ECG-LVH criteria with echo-LVH as the standard *.

3.2. Reliability of LVH Prevalence Estimates

The assessment of the functional relationships between the reported prevalence RP, SE and SP (Eqs.1 and 2 in the Methods section) using graphics displays (not shown) from data extracted from a previous publication (Rautaharju et al. 1996) revealed some rather striking observations. Considering that the sensitivity of ECG-LVH criteria varies from 15% to at most 55% at varying levels of specificity and that echo-LVH was relatively close to 15% in various gender and racial subgroups of the study cited above, the first conclusion was that it is impossible to obtain ECG-LVH prevalence 20% or higher as reported in some studies unless the specificity of the criteria is below 85%. The second conclusion was that it is in fact possible to get estimated LVH prevalence values agreeing within +/- 10% with the true LVH prevalence with a variety of combinations of SE and SP when the fraction of true cases in the reported prevalence unacceptably low. A seemingly good prevalence estimate may be based on false information! On the other hand, estimates with a reasonably high correct fraction, such as at least 50%, are in most instances associated with a substantial over- or underestimate of the true prevalence. Actual data are summarized in Table 1 demonstrating that even with better criteria such as the Cornell voltage, reported prevalence deviated by more than 10% of the expected prevalence of 15% in all subgroups by race and gender, and the correct fraction in the reported prevalence was only approximately 40% except 66% in black males. It can be concluded that the comparison of LVH prevalence data from contrasting populations with the present LVH criteria is an exercise in futility.

3.3. Reasons for Large Fraction of False Negatives by ECG-LVH Criteria

There is a multitude of reasons for the large fraction of echo-LVH missed by ECG-LVH models. One rational explanation attractive to frustrated electrocardiographers is that these false positives really represent a true diagnostic finding as asserted by Bacharova in this session of the Congress in her attractive evidence-based medicine presentation. Pathophysiologically valid and clinically important reasons for false negatives include situations where fibrotic tissue adds to the anatomical LVM and diminishes body surface potentials. CHD, clinical and subclinical, and LVH commonly coexist. Ischemic injury in CHD and myocardial damage can interfere with normal ventricular conduction and diminish body surface potentials. Left ventricular remodeling in CHD and MI is different from at least initially more physiological adaptation in hypertension. The discrepancy between echo and ECG findings calls for an explanation when causative pathological condition can be suspected in a clinical situation. There are numerous other explanations, however, to account for the large fraction of false negative ECG-LVH findings and suggest that either our “gold standard” or our ECG-LVH model or both have serious deficiencies.

Imperfect “Gold Standard”

The median interreader was variability 17% and intrareader variability 14% in older adults in the CHS study (Gardin et al. 1995). Partial error analysis with evaluation of the cumulative LVM distribution in the vicinity of LVH threshold for LVMI reveals the implications of the weakness of the “gold standard.” Assuming that all echo measurements are performed by one highly qualified reader and that one half of the 14% measurement variations deviate on the high and one half on the low side, the maximum sensitivity and specificity for instance in white males that can be expected for ECG-LVH classifier with no variability is 63% and 93%, respectively. Assuming total variability in combined echo and ECG measurements as just +/-10%, the maximum sensitivity that can be expected for ECG-LVH classifier is 47% and specificity 88%. Ultimately, the answer to the “gold standard” question lies in the use of refined LVM determination from better cardiac imaging methods than echocardiography.

Imperfect ECG-LVH Models; Extracardiac Factors

Obesity is one of the primary culprits for the problems with our ECG-LVH models. Although not shown here, the prevalence of definite echo-LVH at level LVMI >149 g.m^-2 was over two-fold in overweight compared to normal-weight white men and over five-fold in black men. Smaller but still quite notable overweight effect was noted in women. Echo-LVH is drastically overestimated by Sokolow-Lyon criteria in normal-weight black men and equally drastically underestimated in overweight black men. In white men, the underestimate of echo-LVH (false negatives) is nearly equally drastic in normal-weight and overweight subjects. Obesity-influence was small for the Cornell-voltage in comparison with the Sokolow-Lyon criteria, and the underestimate of echo-LVH is large in both normal-weight and overweight subgroups, particularly in white men. The results in women differ from those in men to variable degrees. They were more uniform for the Cornell voltage than for the Sokolow-Lyon criteria for overweight and normal-weight groups of women. Of concern was the observed overestimation of echo-LVH (false positives), with particularly large proportion in black women.

The association between ECG-LVH and possibly confounding extracardiac and demographic factors is best examined using multivariate models. Fig. 3 examines the odds ratios for ECG-LVH by the Cornell voltage and for Minnesota Code 3.1, 3.3 high QRS voltage criteria that include the Sokolow-Lyon voltage. Compared to Caucasians and Hispanics, black race was associated with approximately a two-and-half-fold likelihood of ECG-LVH by the Cornell voltage criteria and as high as four-fold excess by the Minnesota code criteria, although echocardiographic data do not support such difference. Male gender had a two-fold excess of ECG-LVH by Minnesota Code criteria and unexpectedly, the odds ratio was less than one-half for the Cornell voltage, reflecting in all likelihood suboptimal LVH threshold in women. Obesity had no influence on the odds ration with the Cornell voltage criteria and ECG-LVH had significantly reduced likelihood by the Minnesota Code criteria. This is in contrast with echocardiographic findings showing a definite excess of LVH with obesity. Increased distance of cardiac excitation fronts from body surface (inverse square effect) is the most likely factor diminishing the sensitivity of particularly the Sokolow-Lyon criteria. ECG-LVH likelihood is reduced with more vertical QRS axis and thus an increased likelihood with a more horizontal QRS axis and the consequent increase in RaVL amplitude improves the sensitivity of the Cornell voltage criteria in obesity. Finally, an increase in age by 25 years was associated with a 21/2-fold increase in the odds ratio for ECG-LVH by the Cornell voltage criteria, with no increase in age by the Minnesota Code criteria. Age-related increase in echo-LVM is mostly associated with increased obesity and the evolution of CHD since the increase in weight-adjusted echo-LV M is less than one gram per year, at least in older men and women (Gardin et al. 1995).

3.4. More Promising Applications; Risk Prediction

Most risk evaluation studies have found a significant association between ECG-LVH and mortality risk only for those criteria that combine high QRS amplitudes with repolarization abnormalities. As expected, mortality risk is highest in hypertension clinic populations selected for intervention trials. An example is the report from the Italian PIUMA study (Verdecchia et al. 1998). In that study, cardiovascular disease (CVD) mortality risk was not significantly increased for the Sokolow-Lyon, Cornell voltage. The risk was not significant also for the Framingham criteria that included left ventricular strain with high QRS amplitudes. CVD mortality risk was over four-fold for authors’ Perugia Score that in essence is a logic combination of three criteria (Cornell voltage >2,400 μV in men and >2,000 μV in women OR Romhilt-Estes score > 5 OR left ventricular strain). The promising feature of the reported risk for this new criterion was that its prevalence in the study population was high, 17.8%, with concomitant population attributable risk of 37.0%.

Is Reduction of ECG-LVH Beneficial ?

A high population attributable risk has connotations of etiological implications. Encouraging reports about the feasibility to use ECG-LVH in identification of subgroups at increased mortality risk have raised hopes that monitoring of the reduction of ECG-LVH may open doors for a convenient monitoring of the success of hypertension intervention efforts. An early example comes from MRFIT with the observation that there was a significant decrease in the Cornell voltage in the special intervention group compared with the usual care group of the study. A more recent analysis of MRFIT data (Prineas et al. 1989) revealed that regression exceeding the limit of short-range variability of the Sokolow-Lyon voltage was actually associated with independent excess CVD mortality risk, and there was no evidence for reduced risk by other criteria evaluated. The expectation of reduced risk with reduction of ECG-LVH also conflicts with recent echocardiographic evidence that has failed to demonstrated improved risk with LV mass reduction with ACE-inhibitors, at least in high-risk hypertensive patients with diabetes (Lindholm at al. 2002). The concept of attributable risk has to be used with caution, and it is unrealistic to assert that ECG-LVH in itself would be a causative factor in mortality. It is plausible to speculate, however, that certain repolarization wave features (not necessarily isolated negative T waves) combined with high QRS amplitudes may be a marker for a substrate for adverse cardiac events.

The last question to be considered is the apparent discrepancy between the rather gloomy results on the ECG-LVH classification accuracy in his communication and the considerably better classification accuracy data in some clinical reports, including data coming from body surface map (BSPM) studies. The obvious answer is the loss of information in 12-lead ECG compared to BSPM. The second obvious reason is that evaluation studies using standard leads with exceptionally good LVH discrimination come from highly selected polarized groups of clinically normals and patients with more advanced LVH rather than from more homogeneous populations with a more continuous distribution of echo-LVM.

References

Gardin JM, Siscovick D, Anton-Culver H, Lynch JC, Smith VE, Klopfenstein S, Bommer WJ, Fried L, O’Leary D, Manolio T. Sex, age, and disease affect echocardiographic left ventricular mass and systolic function in the free-living elderly. The cardiovascular Health Study. Circulation 91(6):1739-1748,1995.

Prineas RJ, Rautaharju PM, Grandits G, Crow R for the MRFIT Research Group. Independent risk for cardiovascular disease predicted by modified continuous score electrocardiographic criteria for 6-year incidence and regression of left ventricular hypertrophy among clinically disease free men:16 year follow-up for the Multiple Risk Factor Intervention Trial. Journal of Electrocardiology 34(2):91-101,2001.

Rautaharju PM, Manolio TA, Siscovick D, Zhou SH, Gardin JM, Kronmal R, Furberg CD, Borhani NO, Newman AB. The CHS Collaborative Research Group. Classification accuracy of electrocardiographic criteria for left ventricular hypertrophy in older adults. Annals of Noninvasive Electrocardiology 1(2):121-132,1996.

Verdecchia P, Schillaci G, Borgioni C, Ciucci A, Gattobigio R, Zampi I, Porcellati C. Prognostic value of a new electrocardiographic method for diagnosis of left ventricular hypertrophy in essential hypertension. Journal of the American College of Cardiology 31(2):383-390,1998.