Nonparametric comparison of entire ROC curves for computerized ECG left ventricular hypertrophy algorithms using data from the Framingham Heart Study

Performance of classic electrocardiographic criteria for left ventricular hypertrophy in an African population

ECG criteria for left ventricular hypertrophy (LVH) have been almost exclusively elaborated and calibrated in white populations. Because several interethnic differences in ECG characteristics have been found, the applicability of these criteria to African individuals remains to be demonstrated. We therefore investigated the performance of classic ECG criteria for LVH detection in an African population. Digitized 12-lead ECG tracings were obtained from 334 African individuals randomly selected from the general population of the Republic of Seychelles (Indian Ocean). Left ventricular mass was calculated with M-mode echocardiography and indexed to body height. LVH was defined by taking the 95th percentile of body height-indexed LVM values in a reference subgroup. In the entire study sample, 16 men and 15 women (prevalence 9.3%) were finally declared to have LVH, of whom 9 were of the reference subgroup. Sensitivity, specificity, accuracy, and positive and negative predictive values for LVH were calculated for 9 classic ECG criteria, and receiver operating characteristic curves were computed. We also generated a new composite time-voltage criterion with stepwise multiple linear regression: weighted time-voltage criterion=(0.2366R(aVL)+0.0551R(V5)+0.0785S(V3)+ 0.2993T(V1))xQRS duration. The Sokolow-Lyon criterion reached the highest sensitivity (61%) and the R(aVL) voltage criterion reached the highest specificity (97%) when evaluated at their traditional partition value. However, at a fixed specificity of 95%, the sensitivity of these 10 criteria ranged from 16% to 32%. Best accuracy was obtained with the R(aVL) voltage criterion and the new composite time-voltage criterion (89% for both). Positive and negative predictive values varied considerably depending on the concomitant presence of 3 clinical risk factors for LVH (hypertension, age >/=50 years, overweight). Median positive and negative predictive values of the 10 ECG criteria were 15% and 95%, respectively, for subjects with none or 1 of these risk factors compared with 63% and 76% for subjects with all of them. In conclusion, the performance of classic ECG criteria for LVH detection was largely disparate and appeared to be lower in this population of East African origin than in white subjects. A newly generated composite time-voltage criterion might provide improved performance. The predictive value of ECG criteria for LVH was considerably enhanced with the integration of information on concomitant clinical risk factors for LVH.

Automatic learning of rules. A practical example of using artificial intelligence to improve computer-based detection of myocardial infarction and left ventricular hypertrophy in the 12-lead ECG

The authors developed a computer program that detects myocardial infarction (MI) and left ventricular hypertrophy (LVH) in two steps: (1) by extracting parameter values from a 10-second, 12-lead electrocardiogram, and (2) by classifying the extracted parameter values with rule sets. Every disease has its dedicated set of rules. Hence, there are separate rule sets for anterior MI, inferior MI, and LVH. If at least one rule is satisfied, the disease is said to be detected. The computer program automatically develops these rule sets. A database (learning set) of healthy subjects and patients with MI, LVH, and mixed MI+LVH was used. After defining the rule type, initial limits, and expected quality of the rules (positive predictive value, minimum number of patients), the program creates a set of rules by varying the limits. The general rule type is defined as: disease = lim1l < p1 < or = lim1u and lim2l < p2 < or = lim2u and ... limnl < pn < or = limnu. When defining the rule types, only the parameters (p1 ... pn) that are known as clinical electrocardiographic criteria (amplitudes [mV] of Q, R, and T waves and ST-segment; duration [ms] of Q wave; frontal angle [degrees]) were used. This allowed for submitting the learned rule sets to an independent investigator for medical verification. It also allowed the creation of explanatory texts with the rules. These advantages are not offered by the neurons of a neural network. The learned rules were checked against a test set and the following results were obtained: MI: sensitivity 76.2%, positive predictive value 98.6%; LVH: sensitivity 72.3%, positive predictive value 90.9%. The specificity ratings for MI are better than 98%; for LVH, better than 90%.

Relation between electrocardiography and echocardiography for left ventricular mass in mild systemic hypertension (results from Treatment of Mild Hypertension Study)

Clinical recognition of hypertensive cardiac involvement depends primarily on use of noninvasive methods. The performance of 8 electrocardiographic (ECG) criteria sets were compared with an echocardiographic standard in the treatment of Mild Hypertension Study. Electrocardiograms were computer processed to define the following ECG criteria sets: (1) Casale/Devereux, (2) Cornell product, (3) Cornell voltage, (4) 12-lead voltage product, (5) sum of 12-lead voltage, (6) Rautaharju, (7) Sokolow-Lyon, and (8) Romhilt-Estes. Echocardiographic left ventricular (LV) mass index was calculated by using the Penn convention on a biracial population of 834 men and women. Correlations between ECG and echocardiographic LV mass index were modest (<0.40). ECG-LV hypertrophy sensitivity at 95% specificity was < 34%. The Casale/Devereux ECG criteria showed the highest average sensitivity (17%) at 95% specificity for all race-sex groups. Whites had significantly higher correlation values than blacks. ECG correlations with LV mass index were consistently improved by including systolic blood pressure and body mass index. ECG criteria sets appear to be optimized for white men. The study findings confirm the poor ECG sensitivity and correlation with echocardiographic LV mass and suggest: (1) further refinement of ECG criteria alone in white men is unlikely to improve its relationship with LV mass; and (2) combining the electrocardiogram with other non-ECG variables or noninvasive measurements offers the best strategy for improving ECG sensitivity and its prognostic value.

Improved electrocardiographic diagnosis of left ventricular hypertrophy

This study was aimed at improving the performance of standard electrocardiographic criteria of left ventricular hypertrophy (LVH) in essential hypertension using echocardiographic left ventricular mass as reference. In 923 white, untreated hypertensive subjects (mean age 51, prevalence of echocardiographic LVH 34%), sensitivity of electrocardiographic criteria of LVH varied between 9% and 33% and specificity was generally > or = 90%. The sum of Sv3 + RaVL (Cornell voltage) showed the closest association with echocardiographic left ventricular mass (r = 0.48, p < 0.001), and its performance was superior to that of Sokolow-Lyon voltage in a receiver-operating characteristic curve analysis. A modified partition value of the Cornell voltage was tested (> 2.4 mV in men and > 2.0 mV in women), that yielded a good combination between sensitivity (26% in men and 19% in women, overall 22%) and specificity (96% in men and 95% in women, overall 95%). When LVH at electrocardiography was defined as the positivity of at least 1 of the following 3 criteria--Sv3 + RaVL > 2.4 mV in men or > 2.0 mV in women, a typical strain pattern, or a Romhilt-Estes point score > or = 5--sensitivity increased to 39% in men and 29% in women (overall 34%) and specificity decreased to 94% in men and 93% in women (overall 93%). Sensitivity of electrocardiography progressively increased from the first to the fourth quartile of left ventricular mass in subjects with echocardiographic LVH.(ABSTRACT TRUNCATED AT 250 WORDS)

Improved detection of echocardiographic left ventricular hypertrophy using a new electrocardiographic algorithm

OBJECTIVES

The purpose of this study was to use the Framingham data base to devise and test an improvement in an electrocardiographic (ECG) voltage criterion for detecting left ventricular hypertrophy that is gender specific and adjusts for age and obesity.

BACKGROUND

Electrocardiographic detection of left ventricular hypertrophy has been receiving increasing attention. The "Cornell" ECG voltage, defined as the sum of voltages for the R wave of lead aVL and S wave of lead V3, has been shown to correlate strongly with echocardiographically estimated left ventricular mass. Because the magnitude of this voltage varies with both age and obesity, we have proposed a simple formula for its adjustment for these two variables.

METHODS

Using linear regression, the adjustment formula was estimated from data on 1,468 men and 1,883 women from the Framingham Heart Study cohort who were free of myocardial infarction and who had both an ECG and an echocardiogram recorded during the same clinic examination. A modified receiver operating characteristic curve method was used to compare sensitivities at the same specificity levels. The adjustment formula was estimated from one randomly chosen half of the study cohort and applied to the other half for evaluation.

RESULTS

Significant improvement in sensitivity for the detection of left ventricular hypertrophy was realized at all levels of specificity. At a specificity level of 98%, the adjustment increased the sensitivity of the Cornell voltage from 10% to 17% in men and from 12% to 22% in women. For severe hypertrophy, defined as a left ventricular mass > 3 SD above the gender-specific mean, the sensitivity increased from 23% to 38% for men and from 22% to 55% for women at a specificity level of 95%.

CONCLUSIONS

This approach can substantially enhance the utility of the ECG for the detection of left ventricular hypertrophy. If these results are validated in other population groups, this approach may prove valuable in the screening of hypertensive populations and for the monitoring of patients undergoing treatment for hypertension.