Left ventricular hypertrophy (LVH) is a pathological reaction to underlying cardiovascular disease and is a strong determinant for cardiovascular morbidity and mortality [1‐4]. In recent years, research has reported that LVH can be reversed, and associated adverse clinical outcomes can be prevented or delayed with prompt therapy [5]. Therefore, it is important to make early and correct diagnosis for LVH. Currently, LVH is diagnosed either using the standard 12-lead electrocardiogram (ECG) or the more expensive ultrasonic cardiogram (UCG) combined with magnetic resonance imaging, but each protocol has certain limitations [6].

For the diagnosis of LVH, ECG is used much more often than UCG and magnetic resonance imaging because the former is less costly, easier to perform and has high reproducibility [7]. For application of ECG, thirty-six ECG criteria have been recommended [7]. Among them, 22 are based on measurements of QRS voltages that are conventional ECG criteria (Table 1). However, the American Electrocardiogram Committee opined: “Published studies are currently insufficient to indicate whether any of the more recently proposed criteria are clearly superior to the others or are simply redundant.” [7] Furthermore, the various ECG criteria have high specificity, but low sensitivity [8]. Consequently, different modifications in the use of ECG have been proposed [9, 10]. With the various uncertainties, the use of ECG can be compromised. Hence, it is imperative to develop more useful ECG criteria for detection of LVH. The aim of our study was to establish new ECG criteria with higher diagnostic accuracy.

The study was approved by the First Affiliated Hospital of Shantou University Medical College Institutional Review Board and complied with the local laws (No.071). All subjects provided written informed consent before enrollment in the study.

From October 2013 to April 2014, 828 patients who have performed UCG examination in the First Affiliated Hospital, Shantou University Medical College, were sequentially recruited for participation in the study. Patient demographics (sex, age, height and weight) were collected, and the standard 12-lead ECG and UCG examinations were conducted at the same time. All entrants in our study were Chinese and older than 35 years, as recommended in the 2009 American ECG guidelines [7]. Other exclusion criteria were: having any sign of Q-wave myocardial infarction in the ECG, artificially paced rhythms and pre-excitation syndrome, complete left bundle branch block, complete right bundle branch block, or left anterior fascicular block, as these abnormalities may interfere with ECG determination of LVH [11‐13].

For further analyses, the population was divided into six original groups: male and female groups, <60 and ≥60 years old groups (the value of median age was chosen as the cut off for age) [11], and BMI <25 kg/m² and BMI ≥25 kg/m² groups [11, 14]. In addition, the population was organized into male and female groups with further subdivision into <60 years old and ≥60 years old groups. The gender groups were randomly divided into A and B subgroups, in which the characteristics of participants did not differ. Male subgroup A accounted for the 85% of male group and female subgroup A accounted for the 85% of female group. A and B contained 364 and 64 males, and 340 and 60 females, respectively.

Among our patients, 32.9% of males and 42.8% of females were positive for UCG-LVH, with the overall prevalence of UCG-LVH positive cases being 37.7%. Our results are consistent with that in a systematic review which showed ranges from 33% to 65% [8], and 36.6% in males and 53.4% in females in a general population [11]. Among soldiers and pre-soldiers, who should be healthier than the general population, the prevalence of UCG-LVH-positive cases ranged from 1.6% to 11.3% [17].

As reported by other studies, the diagnostic accuracy of the 22 traditional ECG criteria varied from each other in the present study [7]. The sensitivity of ECG12 (SV_1,2 + RV_5,6), which had the highest diagnostic accuracy in our groups, is higher than that from other reports showing 1.3% to 21% [14], but close to that of another report showing 43% to 46% sensitivity [18]. Several studies have reported that sensitivity of the conventional ECG criteria would increase as the prevalence and severity of LVH increased [10]. Therefore, the higher sensitivity in our study is likely due to the normal-to-high prevalence of LVH among our patients. Moreover, the population of this study was separated into 10 groups according to gender, age and BMI, which could make our samples more homogeneous, leading to greatly improved sensitivity. The result in the present study was corresponded to previous report [19], with a high sensitivity and a low specificity on ECG15 (RV5:RV6). In the case of LVH, the heart would be rotated which leading an early transition in the precordial leads, and presenting an increased value of RV5:RV6 with a high sensitivity for diagnosis. However, the rotated heart may also exist in many other diseases such as myocardial infarction, the low specificity was showed on ECG15.

In terms of ROC curve analysis: the values of AUC of ECG11 and ECG12 were identified as the highest among the most groups. Nevertheless, in the group of BMI ≥ 25 kg/m² with age ≥ 60 years male and female groups, the values of AUC were shown as the largest on ECG10 (0.641), ECG21 (0.696) and ECG22 (0.651). The values of ECG11 of the above 3 groups were 0.640, 0.685 and 0.649, while on ECG12 were 0.637, 0.681 and 0.638, respectively. The values of AUC of ECG11, ECG12 were very closed to the ones of ECG10, ECG21 and ECG22 in the above three groups. It is indicated that the values of AUC of ECG11 and ECG12 in all groups was relatively high, which is consistent with the results of 2 × 2 tables.

Daniel Pewsner et al. [8] collected 5608 patients from 21 studies to estimate the accuracy of conventional ECG criteria for LVH screening in 2007. The sensibility of S V₁ + R V_5,6, R aVL + SV3 and R I + S III was 4–52%, 2–41% and 0–39%, and the specificity of above three criteria was 53–100%, 89–100% and 80–100%, respectively. In addition, the recommendation of the American Electrocardiogram Committee [7] also showed a generally quite low sensitivity and high specificity of the conventional criteria. The results in our study were in accordance with above reports. Low sensitivity of conventional ECG criteria may cause difficulties on clinical application for differentiate diagnosis. Due to the complicated electrophysiological interpretations of LVH [20], controlling some of the important affected factors, such as age, sex and BMI, could improve the diagnosis accuracy of ECG for LVH.

The diagnostic accuracy of the 22 conventional ECG criteria was influenced by gender, age and BMI, as indicated in other reports [10, 21‐24]. Under the same criterion, the diagnostic accuracy of the female group was lower than that of the male group. This observation cannot be accounted for by just the heart size between men and women, but also by the thickness of subcutaneous fat layer, various hormones, conductivity of organization and other unknown factors [25].

One significant finding of our study is the demonstration that the diagnostic accuracy of the same conventional ECG criterion is higher in the <60 age group than that of the ≥60 age group. This observation is consistent with two prior reports [11, 25]. This phenomenon may be due to the fact that LVM increases with age, during which muscle tissues are gradually replaced by fibrous tissues, so transmission of electrical activity becomes perturbed. The influence of age may also differ between males and females.

The value of the QRS complex amplitude can be affected by the contact distance between heart and electrode, and the conduction of electrical activity which is also influenced by the thickness of subcutaneous fat layer. The diagnostic accuracy of conventional ECG standard is low in obese people. Similar results were derived from the use of the Sokolow-Lyon standard [26].

The other main finding of this study is that modeling by gender improves performance of ECG for diagnosis of LVH. The above analysis on seven different indicators which suggests diagnostic accuracy of the 22 traditional ECG criteria indicating that gender differences play an important role in ECG criteria. Therefore we modeled based on gender in our study. The test results of Xn and Y show that the diagnostic accuracy of the two models is higher than that of both ECG11 and ECG12, particularly model Y. The model Xn sensitivity for diagnosis LVH was 45.9%, higher than ECG11, and 18.9% higher than ECG12. The model Y sensitivity was 64.7% higher than ECG11, 48.3% higher than ECG12. Y’s I, PPV, NPV, accuracy and AUC of Xn and Y were also higher than that of ECG11 and ECG12.

SV₂, SV₃ and R V₅, R V₆ respectively represent the value of the cardiac electrical activity in the posterior left and anterior left of horizontal plane [19]. So we regard SV₂ and SV₃ as S V_2,3 (the higher between S V₂ and S V₃), and regard RV₅ and RV₆as R V_5,6 (the higher between RV₅ and RV₆) to simplify the models. We referred to the report [7] about ECG criteria for diagnosis of LVH from the earliest traditional ECG criteria proposals to more recent innovative ECG criteria. Which indicator of QRS complex amplitude of 12-lead ECG play main responsible for the diagnosis accuracy of ECG criteria, and how do the indicators reflect the differences between males and females, are questions which remain unanswered.

Previous reports do not provide sufficient information on factors that can influence the accuracy of the ECG criteria in the diagnosis of LVH, nor on practical standards after the correction factors have been put forward. In 2012, Sumche Man et al. repeated on 196 cases of outpatients [12]. Univariate analysis and linear discriminant analysis were conducted according to basic information (age, gender, and height) and clinical examination (ECG and vectorcardiogram indicators). The author obtained an ECG diagnosis standard,

(SA = vectorcardiogram-derived spatial QRS-T angle), with a diagnostic accuracy of 79%. However, sample size in that study was small, and the vectorcardiogram was not easy to operate. In 2014, Fabio Angeli et al. [10] included 2747 patients with untreated high blood pressure, and recommended correction for the effect of BMI using Cornell voltage standard and multiplication by BMI. Their results showed the prevalence of LVH, sensitivity, specificity and accuracy of the new Cornell standard was 18%, 36.1%, 90.5% and 73.1%, respectively. However, the sensitivity of the new standards is limited. In general, there is no simple solution to correct all variable factors, to improve the diagnostic accuracy, while providing simplicity of calculation. In terms of sensitivity and AUC, we demonstrated that our models performed better than conventional criteria and other new criteria that have been used in clinical practice.

ECG11 and ECG12 have the highest diagnostic accuracy, while ECG15 has the lowest among the 22 conventional ECG criteria. In addition, gender, age and BMI should be taken into consideration for the ECG criteria of LVH. Two models developed based on gender, have significantly higher diagnostic accuracy than the 22 conventional ECG criteria. Therefore, these models can be considered for use in the preliminary assessment of the probability of LVH.