Obesity reduces the accuracy of voltage-based electrocardiographic (ECG) criteria for diagnosis of left ventricular (LV) hypertrophy. We developed a new ECG score for diagnosis of LV hypertrophy, defined either by a typical strain pattern or a product of the Cornell voltage (R wave height in lead aVL plus S wave depth in lead V 3 ) by body mass index >604 mm∙kg/m 2 . We examined a population of 2,747 untreated hypertensive subjects (mean age 49 ± 11 years) with good quality ECG and echocardiographic tracings. Several traditional ECG criteria for LV hypertrophy were compared with the new score, with echocardiographic LV mass taken as reference. Among the tested criteria, the highest sensitivity combined with specificity was yielded by the new score (sensitivity 36.1%, 95% confidence interval [CI] 32.9 to 39.4; specificity 90.5%, 95% CI 89.1 to 91.8; and accuracy 73.1%, 95% CI 71.5 to 74.8). Prevalence of ECG LV hypertrophy with the new score was 18%. On the basis of comparisons between areas under the receiver operating characteristic curves, the best performance was achieved by the new score with respect to other ECG criteria for LV hypertrophy (all p <0.0001). In conclusion, correction of Cornell voltage by body mass index as a marker of obesity improves the performance of traditional electrocardiography for diagnosis of LV hypertrophy in patients with hypertension.
Left ventricular (LV) hypertrophy is an established predictor of major cardiovascular events including sudden cardiac death, acute myocardial infarction, stroke, and congestive heart failure. Although echocardiography is a sensitive tool to identify LV hypertrophy, the standard electrocardiography remains widely used. Despite a generally high specificity, most electrocardiographic (ECG) criteria for LV hypertrophy lack sensitivity. As a consequence, false-negative diagnoses of LV hypertrophy are common. Body habitus, as estimated by the magnitude of body mass index (BMI), is not only an important determinant of increased LV mass but also a confounder, because its increase may reduce accuracy of voltage-based ECG criteria for LV hypertrophy. Sensitivity decreases in obese patients, possibly because the ECG voltages at the skin level are attenuated by the subcutaneous adipose tissue. Conversely, high ECG voltages may be frequent in young and thin subjects, potentially leading to false-positive results. It is not known whether correction of ECG voltages by measures of body size may improve the accuracy of electrocardiography for diagnosis of LV hypertrophy. Therefore, the present study was designed to examine whether accounting for body mass in a new ECG criterion could improve sensitivity for the diagnosis of LV hypertrophy in hypertension, without compromising specificity.
Methods
We analyzed data of the Progetto Ipertensione Umbria Monitoraggio Ambulatoriale (PIUMA) study. The PIUMA study is a prospective observational registry of morbidity and mortality in initially untreated subjects with essential hypertension. The study design and selection criteria have been reported previously. Briefly, all patients were consecutively referred to our laboratories by their family doctors. The initial evaluation included a detailed clinical examination, 12-lead electrocardiography, 24-hour ambulatory blood pressure (BP) monitoring, screening laboratory tests, funduscopic examination, and an echocardiographic study. Entry criteria included an office BP of ≥140 mm Hg systolic and/or ≥90 mm Hg diastolic pressure on at least 3 visits at 1-week intervals, no previous antihypertensive treatment or treatment withdrawn for at least 4 weeks, no clinical or laboratory evidence of heart failure, valvular defects, secondary causes of hypertension, previous cardiovascular disease, and life-threatening conditions.
We also included in the present analysis a control group of normotensive healthy subjects (office BP <140 mm Hg systolic and 90 mm Hg diastolic pressure) without known cardiovascular risk factors or previous cardiovascular disease. They were members of the hospital staff, training fellows, or subjects examined in our echocardiographic laboratories for reasons other than hypertension (innocent systolic murmurs, general checkup) and found healthy. Obesity was defined as a BMI of ≥30 kg/m 2 and overweight as BMI from 25.0 to 29.9 kg/m 2 .
Standard 12-lead electrocardiogram was recorded during brief end-expiratory apnea. Subjects with conditions potentially precluding a correct ECG assessment of LV hypertrophy (i.e., complete right bundle branch block, left bundle branch block, atrial fibrillation, pathologic Q waves due to previous myocardial infarction, and Wolff-Parkinson-White syndrome) were excluded. The following criteria were used as comparators for testing the performance of the new criterion: original Cornell voltage ; Sokolow-Lyon index ; Romhilt-Estes score ≥5 ; typical strain ; and Perugia score. Briefly, the Sokolow-Lyon index was defined by the sum of the S wave in lead V 1 plus the tallest R wave in leads V 5 or V 6 ≥3.5 mV (35 mm); the Cornell voltage by the sum of the S wave in lead V 3 plus the R wave lead in aVL >2.8 mV (28 mm) in men and >2.0 mV (20 mm) in women; the Romhilt-Estes point score system was computed giving different weights to specific findings; the Perugia score was defined by the presence of a typical strain pattern and/or a modified Cornell voltage (sum of the S wave in V 3 plus the R wave in aVL >2.0 mV in women and >2.4 mV in men); and typical strain pattern was defined by a ≥0.5 mm depression of the J point, T-wave inversion with asymmetric branches, and rapid return to baseline. Tracings were coded and interpreted by 2 investigators without knowledge of other patient data. Interobserver differences occurred in <5% of readings and were resolved in conference.
The M-mode echocardiographic study of the LV was performed under 2-dimensional guide. Only frames with optimal visualization of interfaces and showing simultaneous visualization of septum, LV internal diameter, and posterior wall were used for reading. We calculated LV mass using a necropsy validated formula and defined LV hypertrophy by an unadjusted LV mass of >215 g. After correction for body surface area, we defined LV hypertrophy by an LV mass of >125 g/m 2 . We made a further adjustment by height and defined LV hypertrophy by an LV mass/height 2.7 of >51.0 g/m 2.7 . In our laboratory, the intra- and inter-observer coefficients of variation for LV mass were 6.33% and 7.65%, respectively. Other details about reading procedures and reproducibility in our laboratory have been published elsewhere.
We used Stata 13 (StataCorp LP, College Station, Texas) and R software, version 3 (R Foundation for Statistical Computing, Vienna, Austria; URL http://www.R-project.org ). Data are presented as mean ± SD for continuous variables and proportions for categorical variables. Differences in proportions between groups were analyzed using the chi-square test. Mean values of variables were compared by independent sample t test. The strength of the relations between variables was assessed by ordinary regression and partial correlation analysis. To identify a breakpoint where a significant change in the slope might have taken place, we modeled the relation between Cornell voltage and BMI by piecewise regression analysis with an unknown knot location. To compare the performance of different ECG criteria of LV hypertrophy (sensitivity, specificity, predictive value of a positive and negative test, and accuracy), echocardiographic LV hypertrophy was used as the reference standard. Definitions of test sensitivity, specificity, accuracy, and predictive values conform to standard use. To compare sensitivities and specificities of binary diagnostic tests we used the McNemar test. Accuracy of different ECG criteria was also estimated by receiver operating characteristic curve analysis to identify differences in test performance. The receiver operating characteristic curves were compared statistically by means of a 2-tailed univariate z test of the difference between the areas under 2 performance curves. In 2-tailed tests, p values <0.05 were considered statistically significant.
Results
The main characteristics of control subjects and hypertensive patients are listed in Table 1 . Of the 3,394 hypertensive patients recruited in the PIUMA study, 485 (14.3%) had suboptimal or poor quality echocardiographic tracings and 162 (4.8%) had ECG exclusion criteria (see methods). Hence, analysis was restricted to 2,747 hypertensive patients with complete clinical data, 12-lead electrocardiograms, and echocardiographic tracings of adequate technical quality in a nonpaced rhythm. As expected, the 485 patients with technically inadequate echocardiographic studies were older (57 ± 12 vs 49 ± 11 years) and more likely to be obese (25.5% vs 18.4%) than subjects forming the study population. Gender distribution and 24-hour ambulatory BP profile were similar (all p >0.05).
| Variable | Healthy Controls (n = 224) | Hypertensive Patients | |||
|---|---|---|---|---|---|
| Overall (n = 2747) | LV Hypertrophy | ||||
| No (n = 1872) | Yes (n = 875) | p | |||
| Age (years) | 42 (13) | 49 (11) | 49 (11) | 50 (11) | 0.070 |
| Women | 53% | 44% | 57% | 17% | <0.0001 |
| Diabetes mellitus | none | 6.3% | 5.4% | 8.1% | 0.009 |
| Body mass index (kg/m 2 ) | 23.8 (3.5) | 26.7 (3.8) | 25.9 (3.6) | 28.4 (3.8) | <0.0001 |
| Office systolic BP (mm Hg) | 126 (9) | 155 (19) | 153 (17) | 161 (21) | <0.0001 |
| Office diastolic BP (mm Hg) | 81 (6) | 97 (10) | 96 (9) | 100 (11) | <0.0001 |
| Office heart rate (bpm) | 74 (10) | 75 (11) | 76 (11) | 73 (11) | 0.713 |
| Creatinine (mg/dl) | 0.95 (0.18) | 0.97 (0.22) | 0.94 (0.18) | 1.04 (0.27) | 0.002 |
| Total cholesterol (mg/dl) | 205 (44) | 215 (42) | 218 (42) | 211 (41) | 0.267 |
| Uric acid (mg/dl) | 4.6 (1.3) | 4.7 (1.4) | 4.50 (1.3) | 5.29 (1.3) | 0.432 |
| Electrocardiography | |||||
| R wave in lead aVL (mm) | 4.4 (3.2) | 6.0 (3.6) | 5.4 (3.3) | 7.3 (3.9) | <0.0001 |
| S wave in lead V 3 (mm) | 7.2 (4.3) | 9.6 (4.9) | 8.8 (4.3) | 11.4 (5.5) | <0.0001 |
| S wave in lead V 1 (mm) | 8.9 (3.5) | 10.1 (4.0) | 9.8 (3.7) | 10.8 (4.4) | <0.0001 |
| R wave in lead V 5 (mm) | 13.4 (5.0) | 14.8 (5.9) | 14.2 (5.5) | 15.9 (6.6) | <0.0001 |
| R wave in lead V 6 (mm) | 11.7 (4.0) | 12.9 (4.8) | 12.6 (4.5) | 13.5 (5.4) | <0.0001 |
In the hypertensive group, prevalence of normal-weight, overweight, and obese subjects was 35.2%, 46.4%, and 18.5%, respectively. Prevalence of LV hypertrophy at echocardiography was 31.9% (LV mass >215 g), 20.5% (LV mass >125 g/m 2 ), and 35.5% (LV mass >51.0 g/m 2.7 ). The Cornell voltage showed a direct association with LV mass (r = 0.46, p <0.0001), but such association was attenuated after accounting for the effect of BMI (partial correlation r = 0.40, p <0.0001). Whereas the association between LV mass and BMI was nearly constant for nonobese (r = 0.42) and obese (r = 0.45, p value for the comparison 0.3550) subjects, the Cornell voltage showed a progressively waning relation with increasing BMI categories (nonobese r = 0.20 vs obese r = 0.05, p value for the comparison 0.0004). Using piecewise regression analysis (see Methods ), the value of BMI 26.3 kg/m 2 was identified by statistical minimization criteria as the breakpoint where a significant change in slopes (Davie’s test for change in slopes p <0.0001) took place. Accordingly, the relation ( Figure 1 ) between Cornell voltage and BMI was modeled with 2 “best fit” segments joined at a knot located at BMI = 26.3 kg/m 2 . The relation between Cornell Voltage and BMI observed in patients with BMI <26.3 kg/m 2 (slope 0.62, 95% confidence interval [CI] 0.44 to 0.80) was lost in patients with BMI ≥26.3 kg/m 2 (slope 0.06, 95% CI −0.44 to 0.17). To offset the weakening of the relation between voltage and BMI, we applied the following formula:
In this way, the Cornell voltage was amplified proportionally to BMI, thereby providing a simple correction for voltage attenuation at the skin surface.
LV hypertrophy at electrocardiography by the new score (BMI-corrected Perugia score) was defined by a Cornell-BMI product of >604 mm∙kg/m 2 or typical strain pattern. We used the ninety-fifth percentile (604 mm∙kg/m 2 ) of distribution of Cornell-BMI product in normotensive healthy subjects as a partition value. In the hypertensive group, prevalence of LV hypertrophy by the BMI-corrected Perugia score was 18%. Its prevalence was 7.5%, 18.4%, and 37.3% among normal-weight, overweight, and obese subjects, respectively (p <0.0001). The comparison of receiver operating characteristic curve areas among the different ECG criteria for LV hypertrophy is shown in Figure 2 . The BMI-corrected Perugia score was associated with significantly higher area under the curve values compared with other ECG criteria of LV hypertrophy, whatever the echocardiographic reference (all p <0.0001). The operational characteristics and the diagnostic performance of different ECG criteria are reported in Table 2 . Regardless of the echocardiographic reference, the BMI-corrected Perugia score showed the highest sensitivity.
| Criterion | Sensitivity | Specificity | Positive Predictive Value | Negative Predictive Value | Accuracy |
|---|---|---|---|---|---|
| LV mass >215 g as reference for LV hypertrophy | |||||
| Romhilt-Estes (≥5 points) | 10.4 (8.5–12.6) | 98.6 (97.9–99.0) | 77.1 (68.5–84.3) | 70.2 (68.4–71.9) | 70.5 (68.8–72.2) |
| Strain | 10.3 (8.4–12.5) | 97.1 (96.2–97.8) | 62.1 (53.6–70.0) | 69.8 (68.0–71.6) | 69.4 (67.7–71.1) |
| Sokolow-Lyon | 19.3 (16.7–22.1) | 91.1 (89.8–92.4) | 50.4 (45.0–55.9) | 70.7 (68.9–72.5) | 68.3 (66.5–70.0) |
| Cornell voltage | 19.7 (17.1–22.4) | 92.8 (91.6–94.0) | 56.2 (50.4–61.8) | 71.2 (69.4–73.0) | 69.5 (67.8–71.3) |
| Perugia score | 25.9 (23.1–29.0) | 90.4 (89.0–91.7) | 55.9 (50.9–60.8) | 72.3 (70.5–74.1) | 69.9 (67.2–70.6) |
| BMI-corrected Perugia criterion ∗ | 36.1 (32.9–39.4) | 90.5 (89.1–91.8) | 64.0 (59.6–68.2) | 75.2 (73.4–77.0) | 73.1 (71.5–74.8) |
| LV mass >125 g/m 2 as reference for LV hypertrophy | |||||
| Romhilt-Estes (≥5 points) | 16.3 (13.4–19.7) | 98.8 (98.3–99.2) | 78.0 (69.4–85.1) | 82.1 (80.6–83.5) | 81.9 (80.5–83.4) |
| Strain | 14.6 (11.8–17.8) | 97.1 (96.3–97.8) | 56.6 (48.1–64.8) | 81.5 (80.0–83.0) | 80.2 (78.7–81.7) |
| Sokolow-Lyon | 23.1 (19.7–26.8) | 90.6 (89.3–91.8) | 38.8 (33.6–44.3) | 82.0 (80.5–83.6) | 76.7 (75.2–78.4) |
| Cornell voltage | 25.9 (22.4–29.8) | 92.7 (91.5–93.7) | 47.7 (42.0–53.5) | 82.9 (81.4–84.4) | 79.0 (77.5–80.5) |
| Perugia score | 34.5 (30.5–38.5) | 90.3 (89.0–91.5) | 47.8 (42.8–52.8) | 84.2 (82.7–85.7) | 78.8 (77.3–80.4) |
| BMI-corrected Perugia criterion ∗ | 43.2 (39.0–47.4) | 88.5 (87.0–89.8) | 49.1 (44.6–53.6) | 85.8 (84.3–87.2) | 79.2 (77.7–80.7) |
| LV mass >51 g/m 2.7 as reference for LV hypertrophy | |||||
| Romhilt-Estes (≥5 points) | 10.6 (8.7–12.7) | 99.2 (98.6–99.5) | 87.3 (79.9–92.7) | 66.9 (65.0–68.7) | 67.7 (66.0–69.5) |
| Strain | 10.9 (9.0–13.0) | 97.8 (97.0–98.4) | 73.1 (65.1–80.1) | 66.6 (64.8–68.5) | 66.9 (65.2–68.7) |
| Sokolow-Lyon | 17.4 (15.0–19.9) | 90.6 (89.2–92.0) | 50.4 (45.0–55.9) | 66.6 (64.7–68.5) | 64.6 (62.9–66.4) |
| Cornell voltage | 20.1 (17.6–22.8) | 93.8 (92.6–94.9) | 64.1 (58.4–69.4) | 68.1 (66.2–70.0) | 67.7 (65.9–69.4) |
| Perugia score | 27.1 (24.3–30.0) | 92.0 (90.6–93.2) | 65.0 (60.2–69.7) | 69.7 (67.8–71.5) | 69.0 (67.3–70.7) |
| BMI-corrected Perugia criterion ∗ | 35.4 (32.4–38.5) | 91.5 (90.1–92.8) | 69.7 (65.4–73.7) | 72.1 (70.2–73.9) | 71.6 (69.9–73.3) |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
