The Electrocardiogram and Pulmonary Hypertension: When Occam'S Razor Fails

Pulmonary hypertension (PH) is a potentially life‐threatening disease characterized by elevated pulmonary artery pressure (PAP) that can result in progressive worsening of right ventricular (RV) function leading to RV failure. Although the finding of elevated mean PAP (mPAP) on right heart catheterization (RHC) is the key common hemodynamic variable that defines PH this condition includes distinct clinical/hemodynamic profiles: pulmonary arterial hypertension (PAH), PH associated with left heart disease, PH secondary to lung diseases and/or hypoxia, PH associated with pulmonary artery obstructions and PH with undefined and/or multifactorial mechanisms, including the exposure to exogenous factors, including appetite suppressants or infectious agents [1, 2].

Currently, the mean mPAP threshold of 20 mmHg, supported from data collected in different clinical settings involving well phenotyped patients, is recommended to identify patients at risk for PH instead of the previous one setted at >25 mmHg. Among the various subgroups of PH, PAH, characterized by remodeling in the pulmonary circulation leading to increased pulmonary vascular resistance (PVR) and PAP, is the least frequent but not entirely rare. Indeed, in Europe, the prevalence of PAH ranges between 15 and 60 cases per million inhabitants and 5–10 cases per million per year. On the contrary, left heart disease is the most common cause of PH, which may be classified further as isolated post‐capillary or combined post‐ and pre‐capillary PH [3]. Despite the medical and social relevance of PH and the improvement of non‐invasive diagnostic techniques, there is persistent evidence of a serious delay in the diagnosis of PH reported by numerous real‐world studies revealing an average of over 2 years between symptom onset and diagnosis.

According to the European Society of Cardiology and European Respiratory Society (ESC/ERS) guidelines, initial evaluation of patients with suspected PH should include medical history, physical exam, resting electrocardiogram (ECG), pro‐brain natriuretic peptide (NT‐proBNP) levels and transthoracic echocardiogram (ECHO) [4].

Although it has been widely demonstrated that the contribution of ECG in the diagnosis of PH is limited by its low sensitivity, research in this area is continuously enriched with new contributions. This is the case of the study by Fuchigami [5], published in this Journal, aimed to assess whether machine‐reported right axis deviation (RAD) or RV hypertrophy (RVH) on standard ECG was associated with echocardiographic phenotypes suggestive of PH. For this purpose, the author analyzed data obtained from 68 905 ECG–ECHO pairs of hospitalized patients treated at Beth Israel Deaconess Medical Center in which the ECG was performed within the 7 days preceding the echocardiographic examination. Before reporting in detail the main findings of this study, available evidence of the diagnostic performance of ECG and ECHO criteria in the PH setting deserves to be summarized in order to shed light on this complex scenario.

ECG abnormalities providing supportive evidence of PH may include P pulmonale, right axis deviation, RV hypertrophy, RV strain, right bundle branch block (RBBB), and QTc prolongation. Prolongation of the QRS complex and QTc may suggest severe disease. A normal ECG does not exclude the diagnosis especially in the mild and moderate stages of PH, while this is less likely in more severe forms. The diagnostic performance of ECG in detecting patients with PH depends on the clinical and hemodynamic presentation characteristics underlying this condition. A seminal paper by Ahearn et al. [6] revealed a relatively high sensitivity of RAD and RVH in a cohort of severe primary PH or secondary to connective tissue diseases (i.e. mPAP ≥ 50 mmHg). Several ECG alterations such as sinus tachycardia, RBBB, the S1Q3T3 pattern, T‐wave inversions in precordial leads and supraventricular arrhytmias have been linked to RV strain in patients with acute pulmonary embolism (PE). A meta‐analysis of 24 studies, comprising 7467 patients with imaging‐confirmed acute PE provided a comprehensive information on the pooled prevalence of these ECG abnormalities [7]. The most common abnormal ECG findings were sinus tachycardia (31%, 95% CI: 22%–40%), clockwise rotation (28%, 95% CI: 12%–45%), T‐wave inversion in leads V1–V3 (18%, 95% CI: 13%–23%), S1Q3T3 pattern, (15%, 95% CI: 11%–19%), and RBBB (14%, 95% CI: 10%–17%). The pathophysiological background of these ECG changes is attributed to acute pressure overload and RV dilatation, resulting in altered depolarization vectors and conduction abnormalities. Artificial intelligence may play a valuable role in identifying patients with PH. For instance, Du Brock et al. [8] recently described an artificial intelligence (AI) algorithm based on analysis of a standard 12‐lead ECG aimed to detect PH in two large cohorts of patients from different institutions. This algorhytm showed overall high performance at both medical centers, with a sensitivity of 85% and 82% and a specificity of 84% and 77% at Mayo Clinic and Vanderbilt University Medical Center, respectively. However, it should be noted that sensitivity was markedly reduced in specific subgroups (i.e. pre‐capillary PH and isolated post‐capillary PH). Faced with the intrinsic limits of diagnostic sensitivity of standard ECG, ECHO is regarded as the first line imaging modality of non‐invasive PH assessment, providing important integrated information about function, hemodynamics and structure of the right heart [9]. Given the complexity of the structural and functional alterations associated with PH, a diagnostic approach based on multiple complementary ECHO parameters is recommended by the guidelines. Changes in RV size, wall thickness and function reflect the response of this chamber to elevated afterload. RV enlargement, in turn, can lead to tricuspid annular dilation and functional tricuspid regurgitation (TR), triggering RV volume overload and impairing LV filling via ventricular interdependence. RV systolic function can be assessed through global and longitudinal measures. RV fractional area change it is a conventionally accepted parameter for estimation of RV contractile performance. RV longitudinal function can be captured by tricuspid annular plane systolic excursion (TAPSE). Although pulmonary artery hemodynamics can be assessed through numerous indices, the assessment of RV systolic pressure, which approximates PASP in the absence of pulmonic stenosis, play a central role for detection and monitoring of PH in clinical practice.

The ESC/ERS guidelines recommend classifying patients as low, intermediate or high probability of PH according to their peak TR velocity measurement (≤ 2.8 m/s or not measurable; 2.9–3.4 m/s; > 3.4 m/s, respectively), and whether they have other echocardiographic signs of PH [4]. More recently, the echocardiographic pulmonary left atrial ratio (ePLAR), which reflects the trans‐pulmonary gradient adjusted for LV filling pressure, has been investigated in studies that have invasively measured PVR revealing a high discriminatory capacity for PH [10, 11]. A meta‐analysis of 27 studies, including 4386 patients undergoing ECHO and right heart catheterization, revealed that ECHO had a pooled sensitivity of 85%, a pooled specificity of 74%. A lower sensitivity (81%) and specificity (61%) was found in patients with definite lung diseases and in those with greater time interval between ECHO and RHC [12].

The study by Fuchigami [5], exploring the relationship between machine‐generated ECG findings suggestive of right‐heart involvement and ECHO PH phenotypes, has the merit of adding a new piece of information on the clinical value of routinely generated machine‐reported ECG findings. The entire cohort analyzed consisted of 42 078 patients (mean age 68 years, 46% women) of whom approximately one fifth were admitted to intensive care units at the time of ECHO. The most common comorbidities were in ranking order: hypertension (60.0%), coronary artery disease (37.9%), heart failure (36.1%), atrial fibrillation (30.2%), chronic kidney disease (23.5%), and chronic obstructive pulmonary disease (12.0%). The non‐invasive metric to identify echocardiographic PH phenotypes was a composite of ECHO findings, including TR velocity greater than 3.4 m/s, structured report evidence of moderate or severe PH, or RV pressure overload. In the whole population, machine‐reported RAD/RVH, found in 4% of all ECG recordings, revealed low sensitivity (7.5%) but high specificity (96.7%) for ECHO outcomes. The corresponding sensitivity values for the individual ECHO components were: 9.7% for RV dysfunction, 8.7% for RV dilation, 7.3% for moderate/severe TR and 4.9% for RA enlargement, respectively. These data highlight that detection of a complex condition such as PH by capturing machine‐reported ECG markers of right‐heart involvement remains challenging although it cannot be excluded that the low performance of this diagnostic approach was in turn affected by the limited precision of the ECHO criteria in identifying PH phenotypes. Among the limitations of the echocardiographic metric, it is worth mentioning the impossibility of measuring the TR velocity in a large fraction of patients. Increasing diagnostic sensitivity through multi‐parametric algorithms based on easily acquired variables could be the turning point to facilitate and improve the diagnostic process of PH. In this regard, a recent study by McLean et al. [13] deserves to be mentioned. The authors validated an algorithm based on the orthogonal ECG voltage gradient and photoplethysmogram data to assess mPAP elevation in 315 patients with elevated mPAP from RHC (positive cohort) and 147 subjects with low probability of PH. This machine‐learned algorithm, which incorporated three families of features—conduction, repolarization, and respiration, along with patient age, sex, weight, and height, showed a diagnostic sensitivity of 82% with a specificity of 0.92, comparable, or possibly superior to ECHO, given that the TR velocity was not measurable in up to 41% of the cases.

Therefore, AI could soon play a crucial role in identifying patients with PH using machine‐learned algorithms that simultaneously include clinical, ECG, plethysmographic, and ECHO variables easily obtainable in clinical practice, overcoming the limitations related to the predictive value of a single diagnostic technique [14, 15].

Funding

The authors have nothing to report.

Conflicts of Interest

The authors report no conflicts of interest.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.