Does ECG have additive benefit for predicting atrial fibrillation in patients with or without hypertension?

In the current issue, Lehtonen et al.[1] have investigated the additive predictive value of ECG changes for incident atrial fibrillation. Identification of a new predictor(s) associated with the risk of developing cardiovascular disease and estimation of this risk via multivariate analysis has been frequently used in preventive medicine and cardiovascular epidemiology [2,3]. Furthermore, risk-stratifying systems created through various combinations of identified predictors enables more precise predictions [3]. In addition to clinical risk factors, blood or urine tests, specialized imaging techniques, or special investigations, so-called biomarkers ('biological markers'), are commonly used as an individual predictor or as an incorporated component(s) of risk scoring systems. The more comprehensive the scoring system becomes, the more accurately (at least statistically) we can improve prediction of a certain disease or event, given the more detailed information extracted from several biomarkers. Indeed, the addition of biomarkers will always improve upon risk prediction based simply on clinical factors. However, a complex prediction system including detailed biomarkers, whether single or multiple, is not only difficult (and costly) to apply, but it might have limited additive clinical or practical value in busy everyday clinical practice As atrial fibrillation usually develops in the electrically remodelled heart, electrical remodelling precedes the disease onset and these electrical changes can be identified in the ECG, often before the clinical diagnosis of atrial fibrillation [4]. Up to now several ECG abnormalities have been reported as ECG predictors for atrial fibrillation, such as prolonged PR interval, prolonged P-wave duration, left ventricular hypertrophy (LVH) by ECG criteria, corrected QT interval prolongation, various ST-T wave abnormalities, and so forth [5–7]. Some are related with atrial electrical remodelling per se, whereas others suggest left ventricular diastolic dysfunction onset, which poses pressure overload to left atrium and consequently, contributes to left atrial electrical remodelling and atrial fibrillation development [8]. In the current issue of the Journal of Hypertension, Lehtonen et al.[1] report an analysis of incident atrial fibrillation in 5813 Finnish participants with or without hypertension (2665 hypertensive participants and 3148 nonhypertensive participants), compare the predictive power of ECG abnormalities and the conventional risk prediction model, based on the clinical risk factors. During an average follow-up of approximately 12 years, 301 and 111 participants were diagnosed as atrial fibrillation in hypertensive and nonhypertensive group, respectively. Negative T wave in lateral leads predicted atrial fibrillation in both hypertensive (hazard ratio 1.81; 95% CI 1.16–2.84) and nonhypertensive (hazard ratio 4.59; 95% CI 1.84–11.44) participants. Moreover, PR interval prolongation, increased P wave duration, LVH by Sokolow–Lyon criteria, and poor progression of R wave were statistically significant in hypertensive participants and prolongation of corrected QT interval and T-wave amplitude in aVR were statistically significant ECG predictors in nonhypertensive participants. As each ECG change is directly or indirectly related to atrial or ventricular electrical remodelling, analysis of each appeared to be related to the incident atrial fibrillation in previous studies [5,6,9]. In spite of using the established ECG abnormalities for incident atrial fibrillation, only PR prolongation addition to the conventional model demonstrated statistically significant but minimal increase in the area under curve or c indexes (0.806 vs. 0.809; P = 0.032). A few limitations can be considered in relation to the minimally additive value of ECG abnormalities in predicting incident atrial fibrillation. First, the conventional 12-lead ECG recording method is not sensitive enough for the detection of early electrical remodelling processes in left atrium. Minimal electrical remodelling, in the subclinical stages of atrial fibrillation, are so tiny that they are rarely detectable by conventional surface ECG recording methods. Strictly speaking, the PR interval is a summation of P wave duration and PR segment. Therefore, prolongation of PR interval can be caused either by the prolongation of P wave duration (delayed intra-atrial activation time by atrial remodelling) or by the prolongation of atrioventricular nodal conduction time (can be caused by diseased atrioventricular node or change of autonomic nervous system tone, etc.). In the conventional 12-lead ECG, there might be no clear discernible point between the end of terminal delayed potential of P wave and the exact beginning of pure PR segment in the initially remodelled left atrium. PR interval prolongation, which indicates the degree of inadequately assessed left atrial remodelling in the 12-lead ECG recording, would, therefore, not be able to precisely predict atrial fibrillation. As an alternative, P-wave signal averaged ECG (P-SAECG) methods could be chosen as a tool for such a small electrical change detection [10]. A few studies using P-SAECG have shown significant association between the electrical signal changes recorded in P-SAECG and development of atrial fibrillation [10–12]. However, as the recording procedure is too complex and requires specialized equipment and personnel, it would be challenging for routine clinical practice use or large scale screening studies. Second, pathogenesis of atrial fibrillation is far too complex to be summarized merely as being because of atrial electrical remodelling per se. Pressure overload associated with diastolic dysfunction or LVH is one of important contributing factor in the development of atrial fibrillation [8]. The preclinical ventricular electrical changes suggesting relaxation abnormalities or repolarization abnormalities such as corrected QT prolongation, various ST-T wave abnormalities, and LVH by voltage criteria are also meaningful changes preceding atrial remodelling and atrial fibrillation development [13]. But the exact causes of those ECG changes cannot be differentiated by a single ECG recording as such ECG changes can be influenced by numerous factors such as systemic illness, electrolyte imbalance, or ischemic heart disease, and so forth. In order to reinforce the statistical power of these ECG changes, exact causes of ECG changes should have been differentiated clearly by repeated ECG recordings and differential methods. Potentially, the predictive value of ECG changes will increase if selected participants, whose ECG changes are truly caused by ventricular electrical remodelling, are included to the prediction model. Finally, measuring a baseline variable and determining the risk of an outcome event many years later may be confounded by ageing, various drug therapies and the development of various comorbidities or risk factors, given that risk assessment is not 'static' but a dynamic process. In conclusion, because atrial fibrillation is an electrical disorder caused by numerous cardiovascular causes, choosing ECG as a biomarker for its prediction may be a reasonable approach, as ECG is a cheap and useful test that can easily document the electrical state of heart. However, considering that the ECG changes contributing to the development of atrial fibrillation are too small to be recorded by conventional surface 12-lead ECG, especially in the preclinical stage of atrial fibrillation, it would have been more meaningful if different recording and analysis methods were adopted. In the meantime, it should also be acknowledged that biomarkers that are too difficult to quantify are of limited value. For the prediction of atrial fibrillation with biomarkers using ECG, it is necessary to develop and research new technologies that can easily and precisely record minute electrical signals to overcome these limitations. When we are adopting biomarkers as a part of prediction model, a balance between their usefulness and simplicity to apply in everyday clinical practice should always be considered. ACKNOWLEDGEMENTS Conflicts of interest There are no conflicts of interest.