Effects of airway obstruction and hyperinflation on electrocardiographic axes in COPD

The study was performed using a subset of the baseline data of the German COPD cohort COSYCONET, which is a prospective, observational, multi-center cohort study in patients with stable COPD that aims to evaluate the role of comorbidities [13‐15], including the relationship between lung and cardiovascular disease by ECG analysis and echocardiography [16, 17]. All study participants gave their written informed consent. The criteria of airflow limitation proposed by the Global Initiative for Obstructive Lung Disease (GOLD) [18] were applied to define spirometric GOLD grades 1–4.

For the present analysis, we used data from the recruitment phase and excluded patients with more than moderate heart valve disease, heart valve replacement, or other cardiac devices such as pacemakers/cardioverter-defibrillators. The analysis was restricted to patients with sinus rhythm and several criteria of completeness and plausibility of lung function, echocardiographic and ECG data were applied (see Additional file 1: Methods and Figure E1) [16, 17].

Spirometry and body plethysmography were performed following the recommendations of the American Thoracic Society (ATS)/European Respiratory Society (ERS) [19] and Deutsche Gesellschaft für Pneumologie und Beatmungsmedizin (DGP) [20‐23], after inhalation of 400 μg salbutamol and 80 μg ipratropium bromide [13]. As a measure of lung hyperinflation, we chose functional residual capacity (FRC_pleth; intra-thoracic gas volume, ITGV), the residual volume (RV), total lung capacity (TLC), and their ratio RV/TLC and forced expiratory volume in 1 s (FEV₁) for airway obstruction. The diffusing capacity for carbon monoxide (TLCO) was determined via duplicate assessments of the single-breath method, and the transfer coefficient (KCO) as ratio of TLCO and alveolar volume (VA). Echocardiography was performed as recommended by the American Society of Echocardiography and the European Association of Cardiovascular Imaging [24]. The assessments included the left ventricular end-diastolic and end-systolic diameter (LVEDD, LVESD), LV mass and the right ventricular (RV) wall thickness as indicator of RV hypertrophy as well as heart rate lowering medication. Besides the electrical axes, we selected the ECG derived RR interval as measure of heart rate, and QT duration as measure of repolarization. Standard ECG were obtained and analyzed using the recorder EL10 (VERITAS™, 9515–001-50-ENG REV A1, Mortara Instruments, Inc., Milwaukee, Wisconsin, USA).

FEV₁ and FRC were evaluated as percent predicted values [25‐27]. Cardiac size was expressed as LV mass normalized to body surface area [g/m²]. The RR interval was obtained as the mean of 10.88 ± 2.08 (mean ± SD) consecutive QRS complexes. The QT duration was used as measured, i.e. without heart rate correction, since heart rate was considered as distinct parameter.

For descriptive purposes mean values and standard deviations (SD) or standard errors of the mean (SE) were computed. Differences between groups were evaluated via analysis of variance (ANOVA) and by Tukey-HSD post-hoc comparisons. Univariate multiple linear regression analyses were employed to determine the influences of sex, age and medication on the different variables. Variables were adjusted for these three influencing factors via computation of non-standardized residuals and used for further analyses. Multivariate multiple linear regression analyses were used to determine the associations between FEV₁ % predicted, FRC % predicted, BMI and diastolic blood pressure as predictors, and LV mass, RR interval, QT duration, P wave axis, QRS axis and T wave axis as dependent variables. For all estimates of regression coefficients, 95% confidence intervals were computed.

To disentangle the multiple relationships between the measured variables, structural equation modelling (SEM) was employed [14, 16, 17, 28, 29]. The construct named “ECG axes” comprised the P wave, QRS and T wave axes. The goodness of fit was evaluated by the comparative fit index (CFI) and the root mean square error of approximation (RMSEA). Chi-square data are also given. For all computations the software IBM SPSS Statistics 24.0.0.1 and Amos 24.0.0 (Wexford, PA, USA) was used. Statistical significance was assumed for p < 0.05.

A total of 1195 stable COPD patients were analyzed. The selection process of the cohort is depicted in Additional file 1: Figure E1, and baseline characteristics are shown in Table 1. LV mass decreased significantly from GOLD grades 1–4 (mean ± SD: 111.5 ± 34.0, 109.5 ± 34.1, 103.0 ± 36.1, 97.6 ± 34.9 g/m²; p = 0.002), whereas no differences in RV wall thickness were observed (mean ± SD: 6.2 ± 6.1, 5.7 ± 3.3, 5.9 ± 2.3, 6.3 ± 4.4 mm).

The mean orientation of the P wave axis according to the spirometric GOLD grades 0–4 is illustrated in the left panel of Fig. 1a, while the right panel shows the values plotted against mean values of FRC % predicted observed for each GOLD grade. The rotation of the P wave axis significantly increased across the GOLD grades (p < 0.001). Pairwise post hoc comparisons of the axis orientations between GOLD grades revealed significant (p < 0.05 each) differences, except between grade 0 and 1 and between grade 1 and 2.

In a similar manner, mean QRS axes are illustrated in Fig. 1b. Again, values significantly differed across GOLD grades (p < 0.001). There was a clear trend towards an increased clockwise rotation in more severe airflow limitation. Post hoc comparisons revealed significant (p < 0.05 each) differences between a disease severity not exceeding moderate grades (GOLD 0 to 2) compared with severe to very severe COPD (GOLD 3 and 4). The relationship of the QRS orientation to FRC % predicted across GOLD grades is illustrated.

The results for the mean T wave axis are analogously shown in Fig. 1c, with a significant difference across all GOLD grades (p < 0.001). There were significant (p < 0.05 each) differences between all GOLD grades, except between grade 0 and 1 and between grade 3 and 4. Again, the relationship to the mean values of FRC % predicted for the different GOLD grades is shown.

We assessed the magnitude of the relationship between ECG axes and lung function using multivariate multiple linear regression analysis, with the three ECG axes as dependent variables against FEV₁ % predicted and FRC % predicted as covariates. In accordance with the GOLD definition of COPD [18], this subanalysis was purely restricted to GOLD grades 1–4 (n = 1020). Additional file 1: Table E1 shows regression coefficients of FEV₁ and FRC as predictors of the electrical axes. Since both predictors are cross-linked with each other and FRC is not always available in clinical practice, the analysis was rerun using FEV₁ as predictor only. The estimated incremental rotation of the QRS axis as a function of FEV₁ (univariate analysis) and as function of both FEV₁ and FRC (bivariate analysis) is illustrated in Fig. 2. This analysis demonstrates that airway obstruction and hyperinflation are significant predictors of the electrical axes (for regression analyses including the P and T wave axis see Additional file 1: Figure E2).

The measured distribution of the QRS axis across standard sectors is shown in Additional file 1: Figure E3. It is noteworthy that when influences of FEV₁ and FRC were subtracted, the distribution of the QRS axes shifts from a vertical type (sector 60° to 90°, upper panel) to a normal (sector 30° to 60°) as the most frequent type (lower panel).

To account for possible effects of confounders on measured variables, we also evaluated their relationship to sex, age and heart rate-lowering medication using univariate multiple linear regression analyses. All parameters showed a significant dependence on sex except FEV₁ % predicted and diastolic blood pressure, whereas age was significantly associated with FEV₁ and FRC % predicted, diastolic blood pressure, LV mass and QRS and T wave axis. Heart rate lowering medication (including betablockers, verapamil-type calcium channel blockers [phenylalkylamines] and ivabradine), was significantly related only to FEV₁ and FRC % predicted (p < 0.05 each). In all following analyses we used the values that were adjusted for sex, age and medication according to these results.

The relationship between the selected ECG and echocardiographic LV mass as dependent variables, and FEV₁ % predicted, FRC % predicted, BMI and diastolic blood pressure as covariates was determined by multivariate multiple linear regression analysis. FEV₁ % predicted was correlated with the RR interval, the QT duration and all three electrical axes. FRC % predicted correlated with the RR interval, QT duration and the three axes. BMI was associated with all dependent variables, with the exception of QT duration. Diastolic blood pressure correlated with all variables except LV mass and the T wave axis (Additional file 1: Table E2).

Given these multiple interdependences between parameters, we aimed to determine their relative importance in a network of associations via SEM, which is an extension of multiple regression and factor analysis [14, 16]. The SEM that showed the best fit and which represented a consistent and interpretable network of relationships is shown in Fig. 3; the estimates of the respective regression coefficients and covariances are given in Additional file 1: Table E3. The model comprised a latent variable named “ECG axes” which summarizes the information from the P wave, QRS and T wave axis. Although the mean values of the QRS axis were different from those of the P and T wave axes (Fig. 1), they could be summarized in one latent variable, since all of them were highly correlated with each other and depended in a similar manner on the covariates. LV size was represented by LV mass, which was related to the QT duration. The RR interval was connected to the QT duration, and this was connected to the ECG axes. This pattern of relationships fitted the data very well which was confirmed by the high values of critical ratios in Additional file 1: Table E3. The model showed a chi-squared value of 45.5, with 27 degrees of freedom (p = 0.014); the CFI was 0.992, with an RMSEA of 0.024 (90%CI 0.011; 0.036), which indicates an acceptable model that does not significantly deviate from the data. A detailed sensitivity analysis is given in Additional file 1: Results.