Introduction
Pulmonary hypertension (PH) is a progressive cardiopulmonary disorder characterized by elevated pulmonary arterial pressure and pulmonary vascular resistance. These pathophysiological changes cause right ventricular dysfunction and mortality [1]. Although advances in targeted therapies have improved outcomes, delayed diagnosis and treatment limit prognosis.
Early detection and intervention are critical to improving survival and reducing burden [2].
Right heart catheterization (RHC) is the reference standard for diagnosing PH and assessing severity. This technique provides direct and accurate measurements of pulmonary arterial pressure, vascular resistance, and cardiac output, which are essential for disease classification and decision-making. According to the updated international consensus criteria, PH is defined as a mean pulmonary arterial pressure (mPAP) greater than 20 mmHg at rest, measured by RHC [3, 4].
Despite its diagnostic precision, RHC is invasive and requires specialized equipment and expertise. Its risks, resource requirements, and limited accessibility limit its use, especially in general clinical settings [5]. Consequently, there is a requirement for reliable, non-invasive approaches to identify individuals at high risk of PH who would benefit from confirmatory RHC.
Transthoracic echocardiography(ECHO) is the most commonly used non-invasive modality for evaluating patients with suspected PH. Echocardiographic parameters, including tricuspid regurgitation velocity (TR Vmax), tricuspid regurgitation pressure gradient (TR-PG), and estimated pulmonary artery systolic pressure (PASP ECHO), provide indirect assessments of pulmonary hemodynamics and right ventricular pressure overload [6]. Current guidelines integrate these indices to estimate the probability of PH and to guide further evaluation [4].
Previous studies have reported moderate-to-strong correlations between echocardiographic measurements and hemodynamic parameters obtained by RHC [7]. However, echocardiography has important limitations. Measurement accuracy depends on Doppler signal quality, operator expertise, and anatomical factors.
Moreover, reliance on a single echocardiographic parameter may lead to an incorrect characterization of PH probability [8].
Beyond imaging, biochemical and electrocardiographic markers provide complementary information on the pathophysiological effects of pulmonary vascular disease. Natriuretic peptides, such as B-type natriuretic peptide (BNP) and N-terminal pro-BNP (NT-proBNP), are biomarkers of myocardial strain and right ventricular dysfunction in PH [9]. Elevated marker levels correlate with disease severity, impairment, and adverse outcomes [10].
Similarly, electrocardiographic abnormalities, including right ventricular hypertrophy, right atrial enlargement, and right ventricular strain, reflect chronic right ventricular pressure overload [11]. Because electrocardiography (ECG) and natriuretic peptide testing are inexpensive, widely available, and easily performed, integrating these markers with echocardiographic parameters may enhance the performance of non-invasive screening strategies.
Despite these potential advantages, few studies have assessed the combined diagnostic value of echocardiographic indices, electrocardiographic findings, and natriuretic peptide levels in a unified screening framework compared to invasive hemodynamic measurements. Evidence remains limited regarding the incremental clinical benefit of such integrated approaches in hospitalized patients with a high clinical suspicion of PH. Developing a practical and reliable non-invasive screening strategy could improve patient stratification and the use of invasive diagnostic procedures.
In this study, we evaluated whether routinely available ECG and BNP/NT-proBNP positivity provide incremental information beyond PASP ECHO in an RHC-referred cohort of patients with suspected PH. Using RHC as the reference standard, we assessed discrimination, calibration, temporal validation, and decision curve analysis. We aimed to determine whether this multimodal approach could provide complementary information for triage, while recognizing that it is not intended to replace ESC/ERS probability-based assessment or confirmatory RHC.
Methods
Study design and participants
This single-center retrospective study was performed at the First Affiliated Hospital of Fujian Medical University. Hospitalized patients with suspected PH who were admitted between January 1, 2023, and January 31, 2025, were screened for inclusion in the derivation cohort. After applying the eligibility criteria, 130 patients were included in the derivation cohort.
A temporal validation cohort consisted of consecutive eligible patients admitted between March 2025 and February 2026 (n = 42). The final model developed in the derivation cohort was applied to this temporal validation cohort without refitting.
Transthoracic Doppler echocardiography, 12-lead ECG, BNP or NT-proBNP testing, and RHC were performed during the same hospital admission.
Inclusion criteria
Patients were eligible if they met all of the following criteria: (1) echocardiography-estimated pulmonary artery systolic pressure (PASP ECHO) > 30 mmHg or clinical features prompting further evaluation for suspected PH; (2) RHC with available hemodynamic measurements; (3) available transthoracic echocardiography, ECG, and BNP or NT-proBNP data; and (4) PH status classifiable according to RHC-measured mean pulmonary arterial pressure (mPAP). PH was defined as mPAP > 20 mmHg on RHC, consistent with the 2022 ESC/ERS hemodynamic definition.
The inclusion criterion was not intended to reproduce the ESC/ERS high echocardiographic PH probability definition. Therefore, ESC/ERS echocardiographic PH probability was classified separately using available TR Vmax and additional echocardiographic signs and is reported in Tables 2 and 3.
Exclusion criteria
Patients were excluded if RHC was not performed, if key echocardiographic, ECG, or BNP/NT-proBNP data were missing, if duplicate records were identified, or if RHC was contraindicated. Contraindications included acute infectious disease; prior mechanical tricuspid or pulmonary valve replacement; floating thrombus in the right heart chambers or main pulmonary artery; severe electrolyte disturbance; severe malignant arrhythmia; complete left bundle branch block unless protected by temporary pacing; decompensated right heart failure in severe PH; severe coagulation disorder; recent pacemaker or defibrillator implantation; severe hepatic or renal dysfunction; pregnancy; and refusal of or inability to cooperate with RHC.
Grouping
Patients were classified into the PH group (mPAP > 20 mmHg) and the non-PH group (mPAP ≤ 20 mmHg) according to RHC results. Because this study used hemodynamic PH status rather than etiologic classification, the term PH is used throughout instead of PAH.
Clinical data
Baseline demographic and clinical data were extracted from the electronic medical record, including sex, age, height, weight, body mass index (BMI), medical history, and comorbidities. Comorbidities included hypertension, diabetes mellitus, hyperlipidemia, atrial fibrillation, coronary artery disease, heart failure, and prior cardiovascular events.
Biomarker measurement and NT-proBNP/BNP positivity
Venous blood samples were obtained at admission, and serum BNP or NT-proBNP levels were measured using an automated analyzer according to the manufacturer’s instructions. NT-proBNP positivity was defined as NT-proBNP ≥ 450 pg/mL, and BNP positivity was defined as BNP ≥ 125 pg/mL. The combined biomarker variable, BNP/NT-proBNP positivity, was considered positive if either threshold was met.
Echocardiography protocol
Transthoracic Doppler echocardiography was performed at rest in the supine or left lateral decubitus position using a Philips Affiniti 50 system. Formal reports were generated by experienced sonographers. Standard measurements included aortic diameter, left atrial diameter, left ventricular end-diastolic and end-systolic diameters, left ventricular volumes, mitral inflow velocity (E), early diastolic mitral annular velocity (e′), E/e′ratio, heart rate, cardiac output, ejection fraction, and maximal and minimal inferior vena cava diameters.
TR Doppler spectra were acquired to obtain peak TR velocity (TR Vmax) and tricuspid regurgitation pressure gradient (TR-PG). PASP ECHO was estimated using the modified Bernoulli equation, with right atrial pressure estimated from inferior vena cava size and respiratory variation. Each measurement was obtained three times when feasible, and the average value was used for analysis.
ESC/ERS echocardiographic PH probability was classified according to TR Vmax and additional echocardiographic signs. Available additional signs were mapped to the ESC/ERS categories and summarized as category-level positivity. The distribution of additional signs and low, intermediate, and high echocardiographic PH probability is presented in Tables 2 and 3.
Electrocardiography
Standard 12-lead ECGs were obtained using a GE MAC1200 system with patients in the supine position, standard limb and precordial lead placement, a paper speed of 25 mm/s, and a calibration of 10 mm/mV. ECG positivity was defined as the presence of any of the following prespecified right-heart strain criteria: (1) right axis deviation ≥ 90°; (2) R/S ratio ≥ 1 in lead V1 with RV1 > 0.5 mV; (3) RV1 + SV5 > 1.0 mV; (4) complete or incomplete right bundle branch block; or (5) P pulmonale. The frequency of each ECG criterion is reported in Table 4.
Right heart catheterization
RHC was performed using a 7-F Swan-Ganz catheter under fluoroscopic guidance with a Philips Allura Xper FD20 angiography system. After local anesthesia, vascular access was obtained via an antecubital vein. The catheter was advanced through the vena cava, right atrium, right ventricle, main pulmonary artery, and branch pulmonary arteries.
Hemodynamic parameters included right atrial pressure (RAP), right ventricular pressure (RVP), systolic pulmonary artery pressure measured by RHC (sPAP RHC), diastolic pulmonary artery pressure (dPAP), and mPAP. Measurements were repeated three times, and the mean value was used for analysis.
Pulmonary artery wedge pressure (PAWP) was measured by inflating the balloon and advancing the catheter to a distal pulmonary artery branch until wedged. The mean of 3–5 end-expiratory measurements during normal breathing, without breath-holding or Valsalva maneuver, was recorded. Because pulmonary vascular resistance was not available for all patients, complete hemodynamic subtype classification was not performed.
Model development
Candidate predictors were selected a priori according to clinical relevance, routine availability, and the need to minimize collinearity. The final multimodal model included PASP ECHO, ECG positivity, and BNP/NT-proBNP positivity. Because PASP ECHO, TR Vmax, and TR-PG are physiologically and mathematically correlated, TR Vmax and TR-PG were not entered simultaneously with PASP ECHO in the final model.
The model was fitted using Firth penalized logistic regression to reduce small-sample bias and potential separation caused by the limited number of non-PH patients. No automated stepwise variable selection was used. Predicted probabilities from the final model were used for ROC analysis, calibration assessment, threshold-dependent diagnostic indices, and decision curve analysis.
Model evaluation, calibration, and decision curve analysis
Diagnostic discrimination was assessed using receiver operating characteristic (ROC) analysis. AUCs were compared using DeLong’s test [16]. Threshold-dependent diagnostic performance was reported for the Youden threshold and for a sensitivity-oriented threshold.
The Youden threshold was considered a statistical operating point, whereas the sensitivity-oriented threshold was selected to achieve sensitivity of at least 90% while maximizing specificity because high sensitivity is clinically desirable in a screening-oriented setting.
Calibration was assessed using calibration plots, Brier score, calibration intercept, and calibration slope. Decision curve analysis compared the net benefit of the final model, PASP ECHO alone, treat-all strategy, and treat-none strategy across threshold probabilities from 0.05 to 0.80.
Internal and temporal validation
Internal validation was performed using bootstrap resampling with 1,000 replicates to estimate optimism-corrected discrimination [20]. Ten-fold cross-validation was not retained as the primary internal validation approach because of the small number of non-PH patients and the instability of fold-level estimates.
For temporal validation, the final derivation model and fixed thresholds were applied to the temporal validation cohort without refitting. Discrimination, calibration, and threshold-dependent diagnostic indices in the temporal validation cohort are reported in Table 12 and Figs. 2–3.
Agreement analysis
Spearman correlation analysis was used to assess the association between PASP ECHO and sPAP RHC. Because correlation does not measure agreement, Bland–Altman analysis was additionally performed to evaluate agreement between PASP ECHO and sPAP RHC [15].
Statistical analysis
Continuous variables were summarized as mean ± standard deviation or median with interquartile range, depending on distribution. Categorical variables were summarized as counts and percentages. Between-group comparisons were performed using the Wilcoxon rank-sum test for continuous variables and the chi-square test or Fisher’s exact test for categorical variables, as appropriate.
A two-sided P value < 0.05 was considered statistically significant.
All analyses were performed using R version 4.4.2 [2024–10–31] with the packages logistf, pROC, ggplot2, and related packages, together with SPSS Statistics 27.0.1, GraphPad Prism 10.1.2, and MedCalc 23.1.6 where applicable.
Ethics approval
The study protocol was reviewed and approved by the Institutional Review Board of the First Affiliated Hospital of Fujian Medical University (reference No. 闽医大附一伦理医研[2023]613号). Owing to the retrospective nature of the study, the requirement for informed consent was waived by the Institutional Review Board.
Results
Study population and baseline characteristics
A total of 333 patients were initially screened for the derivation cohort. After excluding patients without RHC, patients with missing echocardiographic, ECG, or BNP/NT-proBNP data, duplicate records, and other prespecified exclusions, 130 patients were included in the derivation cohort. Among them, 111 patients were classified as having PH and 19 as non-PH according to RHC-defined mPAP > 20 mmHg.
During the temporal validation period, 128 patients were initially screened, and 42 consecutive eligible patients were included in the temporal validation cohort (Fig. 1).Fig. 1Study flow diagram.
Flow diagram showing patient screening and inclusion in the derivation and temporal validation cohorts. RHC, right heart catheterization; ECG, electrocardiography; BNP, B-type natriuretic peptide; NT-proBNP, N-terminal pro-B-type natriuretic peptide
Baseline characteristics of the derivation cohort are summarized in Table 1. The derivation cohort had a high prevalence of RHC-confirmed PH, reflecting a clinically selected population referred for invasive hemodynamic assessment rather than a general screening population. Baseline demographic characteristics and comorbidities were compared between PH and non-PH patients.Table 1Baseline characteristics by PH statusVariableOverall, N = 130non-PH, N = 19PH, N = 111P value2Age, years69.0 (59.0, 75.0)68.0 (54.0, 75.0)69.0 (60.0, 75.0)0.50Sex0.13 Male71 (54.6%)7 (36.8%)64 (57.7%) Female59 (45.4%)12 (63.2%)47 (42.3%)Heart rate, bpm72.0 (62.0, 81.0)68.0 (60.0, 84.0)72.0 (64.0, 81.0)0.80Weight, kg64.0 (57.5, 72.0)62.5 (59.5, 70.0)65.0 (57.0, 73.0) > 0.90Height, cm163.0 (156.0, 168.0)160.0 (157.0, 170.0)163.0 (156.0, 168.0) > 0.90Body mass index, kg/m224.4 (22.3, 27.0)24.5 (22.4, 27.3)24.4 (22.3, 27.0)0.90Suspected PH etiology0.90 Group 1: CHD-shunt related3 (2.3%)0 (0.0%)3 (2.7%) Group 2: LHD-related75 (57.7%)13 (68.4%)62 (55.9%) Group 3: Lung/hypoxia-related13 (10.0%)2 (10.5%)11 (9.9%) Group 4: CTEPH/PE-related1 (0.8%)0 (0.0%)1 (0.9%)Multiple suspected categories17 (13.1%)1(5.3%)16 (14.4%)Other/Unclear21 (16.2%)3 (15.8%)18 (16.2%)Values are presented as median (interquartile range), mean ± standard deviation, or n (%), as appropriate.
PH was defined as mPAP > 20 mmHg measured by right heart catheterization. Between-group comparisons were performed using the Wilcoxon rank-sum test, Student’s t-test, chi-square test, or Fisher’s exact test, as appropriateBMI Body mass index, CAD Coronary artery disease, PH Pulmonary hypertension, RHC Right heart catheterization, mPAP Mean pulmonary arterial pressure1Median (Q1, Q3); n (%)2Wilcoxon rank sum test; Fisher's exact test
ESC/ERS echocardiographic PH probability
Because the inclusion criteria were based on clinical suspicion and referral for RHC rather than restriction to the ESC/ERS high echocardiographic probability category, ESC/ERS echocardiographic PH probability was analyzed separately. Additional echocardiographic signs were summarized according to the 2022 ESC/ERS categories. Additional signs in at least two ESC/ERS categories were more frequent in patients with PH than in non-PH patients (Table 2).Table 2ESC/ERS additional echocardiographic signs by PH statusVariableOverall, N = 130non-PH, N = 19PH, N = 111P valueA category: ventricular signs51/130 (39.2%)5/19 (26.3%)46/111 (41.4%)0.310B category: pulmonary artery signs17/130 (13.1%)0/19 (0.0%)17/111 (15.3%)0.075C category: IVC/right atrial signs18/130 (13.8%)0/19 (0.0%)18/111 (16.2%)0.073Additional echocardiographic signs present, ≥ 2 categories positive20/130 (15.4%)0/19 (0.0%)20/111 (18.0%)0.043Additional echocardiographic signs were classified according to the 2022 ESC/ERS pulmonary hypertension guideline categories.
The categories included signs related to right ventricular morphology/function, pulmonary artery findings, and inferior vena cava/right atrial findings. The item “ ≥ 2 categories positive”indicates the presence of additional echocardiographic signs in at least two categories. PH was defined as mPAP > 20 mmHg measured by right heart catheterization.
Between-group comparisons were performed using Fisher’s exact testESC/ERS European Society of Cardiology/European Respiratory Society, PH Pulmonary hypertension, mPAP Mean pulmonary arterial pressure
The distribution of low, intermediate, and high echocardiographic PH probability also differed between groups (overall P = 0.011; Table 3).Table 3ESC/ERS echocardiographic PH probability by PH statusEchocardiographic PH probabilityOverall, N = 130non-PH, N = 19PH, N = 111P valueLow31/130 (23.8%)8/19 (42.1%)23/111 (20.7%)Intermediate59/130 (45.4%)10/19 (52.6%)49/111 (44.1%)High40/130 (30.8%)1/19 (5.3%)39/111 (35.1%)0.011Echocardiographic PH probability was classified as low, intermediate, or high according to the 2022 ESC/ERS approach using TR Vmax and additional echocardiographic signs. The P value refers to the overall distribution across the three probability categories. PH was defined as mPAP > 20 mmHg measured by right heart catheterizationESC/ERS European Society of Cardiology/European Respiratory Society, PH Pulmonary hypertension, TR Vmax Peak tricuspid regurgitation velocity, mPAP Mean pulmonary arterial pressure
These findings indicate that the study population included patients across different echocardiographic probability categories and was not limited to those with high echocardiographic PH probability.
ECG criteria
The distribution of prespecified ECG right-heart strain criteria is shown in Table 4. Any ECG positivity was more frequent in patients with PH than in those without PH. Among the individual ECG criteria, RV1 + SV5 > 1.0 mV was significantly more common in the PH group.
P pulmonale was not observed in either group. These findings suggest that ECG abnormalities may provide complementary information beyond echocardiographic pressure estimates in selected patients with suspected PH.Table 4ECG criteria by PH statusECG criterionOverall, N = 130non-PH, N = 19PH, N = 111P valueRight axis deviation ≥ 90°18/130 (13.8%)1/19 (5.3%)17/111 (15.3%)0.470R/S ratio ≥ 1 in V1 with RV1 > 0.5 mV17/130 (13.1%)1/19 (5.3%)16/111 (14.4%)0.465RV1 + SV5 > 1.0 mV37/130 (28.5%)1/19 (5.3%)36/111 (32.4%)0.013Complete/incomplete RBBB16/130 (12.3%)1/19 (5.3%)15/111 (13.5%)0.465P pulmonale0/130 (0.0%)0/19 (0.0%)0/111 (0.0%)-Any ECG positivity43/130 (33.1%)1/19 (5.3%)42/111 (37.8%)0.004ECG positivity was defined as the presence of any prespecified right-heart strain criterion, including right axis deviation ≥ 90°, R/S ratio ≥ 1 in V1 with RV1 > 0.5 mV, RV1 + SV5 > 1.0 mV, complete or incomplete right bundle branch block, or P pulmonale. PH was defined as mPAP > 20 mmHg measured by right heart catheterization.
Between-group comparisons were performed using Fisher’s exact test. P values were not estimated for criteria with no events in either groupECG Electrocardiography, PH Pulmonary hypertension, RBBB Right bundle branch block, mPAP Mean pulmonary arterial pressure
Echocardiographic and invasive hemodynamic findings
Echocardiographic parameters are summarized in Table 5. Patients with PH showed higher values of TR Vmax, TR-PG, and PASP ECHO than non-PH patients, consistent with elevated pulmonary artery pressure. RHC-derived hemodynamic measurements are summarized in Table 6.
As expected, invasive pulmonary pressure parameters were higher in the PH group than in the non-PH group. PAWP category distribution is shown in Table 7. Because pulmonary vascular resistance was not available, complete hemodynamic subtype classification was not performed.Table 5Echocardiographic parameters by PH statusVariableOverall, N = 130non-PH, N = 19PH, N = 111P valueAOD, cm3.33 (3.07, 3.66)3.19 (2.89, 3.60)3.37 (3.08, 3.67)0.272CO, L/min4.63 (3.73, 5.73)4.68 (3.62, 5.48)4.63 (3.74, 5.73)0.869LAD, cm4.37 (3.90, 4.94)4.05 (3.47, 4.60)4.47 (3.92, 5.02)0.046LVDS, cm3.14 (2.90, 3.89)3.00 (2.81, 3.32)3.20 (2.92, 3.93)0.130LVDD, cm4.94 (4.52, 5.42)4.79 (4.39, 5.17)4.99 (4.53, 5.47)0.346EF, %62.81 (53.27, 66.56)66.12 (64.12, 68.20)62.22 (51.75, 66.42)0.036E/e'11.31 (8.49, 16.86)8.75 (6.53, 11.32)12.13 (9.07, 17.33)0.012LVVd, mL114.16 (93.41, 141.00)107.00 (87.22, 127.75)116.05 (93.66, 143.74)0.363LVVs, mL39.11 (32.27, 63.71)35.00 (29.80, 44.79)40.95 (32.75, 64.71)0.141TR Vmax, m/s3.09 (2.82, 3.44)2.86 (2.75, 3.12)3.13 (2.85, 3.51)0.008TR-PG, mmHg38.00 (32.00, 46.75)32.00 (30.50, 38.50)39.00 (33.00, 50.50)0.006Values are presented as median (interquartile range), mean ± standard deviation, or n (%), as appropriate.
PASP ECHO was estimated from TR velocity using the modified Bernoulli equation plus estimated right atrial pressure derived from inferior vena cava size and respiratory variation. PH was defined as mPAP > 20 mmHg measured by right heart catheterization. Between-group comparisons were performed using the Wilcoxon rank-sum test, Student’s t-test, chi-square test, or Fisher’s exact test, as appropriatePASP ECHO Pulmonary artery systolic pressure estimated by echocardiography, TR Vmax Peak tricuspid regurgitation velocity, TR-PG Tricuspid regurgitation pressure gradient, LAD Left atrial diameter, LVDD Left ventricular end-diastolic diameter, LVDS Left ventricular end-systolic diameter, EF Ejection fraction, CO Cardiac output, IVC Inferior vena cava, PH Pulmonary hypertensionTable 6RHC-derived hemodynamics by PH statusVariableOverall, N = 130non-PH, N = 19PH, N = 111P valuemPAP, mmHg27.00 (22.25, 33.00)17.00 (14.00, 17.50)28.00 (24.00, 34.00) < 0.001RVP, mmHg18.00 (15.00, 23.75)14.00 (9.50, 16.50)20.00 (15.00, 24.00) < 0.001RAP, mmHg9.00 (6.00, 12.00)7.00 (5.50, 9.00)9.00 (7.00, 13.00)0.031SVCP, mmHg8.00 (5.00, 9.75)5.00 (3.50, 7.50)8.00 (5.00, 10.00)0.010PAWP, mmHg18.00 (11.00, 24.75)11.00 (8.50, 15.50)19.00 (12.00, 26.00) < 0.001Values are presented as median (interquartile range), mean ± standard deviation, or n (%), as appropriate.
PH was defined as mPAP > 20 mmHg measured by right heart catheterization. Between-group comparisons were performed using the Wilcoxon rank-sum test, Student’s t-test, chi-square test, or Fisher’s exact test, as appropriateRHC Right heart catheterization, RAP Right atrial pressure, RVP Right ventricular pressure, sPAP RHC Systolic pulmonary artery pressure measured by right heart catheterization, dPAP Diastolic pulmonary artery pressure, mPAP Mean pulmonary arterial pressure, PAWP Pulmonary artery wedge pressure, PH Pulmonary hypertensionTable 7PAWP category by PH statusPAWP categoryOverall, N = 130non-PH, N = 19PH, N = 111P valuePAWP ≤ 15 mmHg49/130 (37.7%)14/19 (73.7%)35/111 (31.5%)PAWP > 15 mmHg81/130 (62.3%)5/19 (26.3%)76/111 (68.5%) < 0.001PAWP was categorized as ≤ 15 mmHg or > 15 mmHg. PH was defined as mPAP > 20 mmHg measured by right heart catheterization.
The P value refers to the overall distribution of PAWP categories between PH and non-PH patients. Because pulmonary vascular resistance was not available, complete hemodynamic subtype classification was not performedPAWP Pulmonary artery wedge pressure, PH Pulmonary hypertension, mPAP Mean pulmonary arterial pressure
Diagnostic performance in the derivation cohort
Diagnostic performance of PASP ECHO alone and the final multimodal model is shown in Tables 8, 9, 10, 11 and Fig. 2A. PASP ECHO alone showed an AUC of 0.730 (95% CI, 0.620–0.841).
The final multimodal model, which included PASP ECHO, ECG positivity, and BNP/NT-proBNP positivity, showed a higher AUC of 0.825 (95% CI, 0.722–0.928). The improvement over PASP ECHO alone was statistically significant by DeLong’s test (P = 0.008; Table 8).Table 8Diagnostic discrimination and AUC comparison in the derivation cohortModelAUC95% CIΔAUC vs PASP ECHOP value by DeLong testPASP ECHO alone0.7300.620–0.841ReferenceTR Vmax alone0.6910.574–0.808−0.0390.032TR-PG alone0.6990.582–0.817−0.0310.064BNP/NT-proBNP positivity alone0.6520.540–0.764−0.0780.158ECG positivity alone0.6630.594–0.732−0.0680.230Final multimodal model0.8250.722–0.9280.0950.008The final multimodal model included PASP ECHO, ECG positivity, and BNP/NT-proBNP positivity. AUCs were compared with PASP ECHO alone using DeLong’s test.
PH was defined as mPAP > 20 mmHg measured by right heart catheterizationAUC Area under the receiver operating characteristic curve, CI Confidence interval, PASP ECHO Pulmonary artery systolic pressure estimated by echocardiography, ECG Electrocardiography, BNP B-type natriuretic peptide, NT-proBNP N-terminal pro-B-type natriuretic peptide, PH Pulmonary hypertensionTable 9Threshold-dependent diagnostic performance in the derivation cohortModelThreshold typeThresholdTPFPTNFNSensitivitySpecificityPPVNPVPASP ECHOYouden threshold36.500949101784.7%52.6%91.3%37.0%Final modelYouden threshold0.891671184460.4%94.7%98.5%29.0%Final modelSensitivity-oriented threshold0.6761019101091.0%52.6%91.8%50.0%Threshold-dependent indices were calculated for PASP ECHO alone and the final multimodal model. The Youden threshold was selected by maximizing sensitivity + specificity − 1. The sensitivity-oriented threshold was selected to achieve sensitivity of at least 90% while maximizing specificity, reflecting a screening-oriented operating point.
PH was defined as mPAP > 20 mmHg measured by right heart catheterizationPPV Positive predictive value, NPV Negative predictive value, PASP ECHO Pulmonary artery systolic pressure estimated by echocardiographyTable 10Calibration metrics of the final multimodal model in the derivation cohortMetricFinal multimodal modelBrier score0.103Calibration intercept0.091Calibration slope1.144Calibration was assessed using the Brier score, calibration intercept, and calibration slope. A calibration intercept close to 0 and calibration slope close to 1 indicate better calibration. The final multimodal model included PASP ECHO, ECG positivity, and BNP/NT-proBNP positivity.
PH was defined as mPAP > 20 mmHg measured by right heart catheterizationTable 11Bootstrap internal validation of the final multimodal modelModelApparent AUCPlanned BValid BOptimismCorrected AUCFinal multimodal model0.8251,0001,0000.0140.811Bootstrap internal validation was performed using 1,000 resampling replicates to estimate optimism-corrected discrimination. The final multimodal model included PASP ECHO, ECG positivity, and BNP/NT-proBNP positivityAUC Area under the receiver operating characteristic curve, CI Confidence interval, PASP ECHO Pulmonary artery systolic pressure estimated by echocardiographyFig. 2ROC curves for PASP ECHO and the final multimodal model.
A Derivation cohort. B Temporal validation cohort. The final multimodal model included PASP ECHO, ECG positivity, and BNP/NT-proBNP positivity.
In the temporal validation cohort, the derivation model was applied without refitting. ROC, receiver operating characteristic; PASP ECHO, pulmonary artery systolic pressure estimated by echocardiography; BNP, B-type natriuretic peptide; NT-proBNP, N-terminal pro-B-type natriuretic peptide
Threshold-dependent diagnostic performance is shown in Table 9. At the Youden threshold of 0.891, the final multimodal model showed high specificity but limited sensitivity. Because this sensitivity may be insufficient for screening-oriented use, a sensitivity-oriented threshold of 0.676 was additionally evaluated.
This threshold achieved higher sensitivity at the expense of specificity, consistent with a screening-oriented trade-off. Calibration metrics are summarized in Table 10. Bootstrap internal validation yielded an optimism-corrected AUC of 0.811, suggesting acceptable internal discrimination after correction for optimism despite the small number of non-PH patients (Table 11).
Temporal validation
The final derivation model was applied to the temporal validation cohort without refitting. In this cohort, the final multimodal model showed an AUC of 0.792, whereas PASP ECHO alone showed an AUC of 0.749. The difference was not statistically significant by DeLong’s test (P = 0.442; Table 12 and Fig.
2B). At the fixed sensitivity-oriented threshold derived from the derivation cohort, At the fixed sensitivity-oriented threshold, the model achieved 100.0% sensitivity and 100.0% negative predictive value in the temporal validation cohort, although specificity was 50.0% in the temporal validation cohort, although specificity was lower. Calibration in the temporal validation cohort was less satisfactory than in the derivation cohort, with a negative calibration intercept and reduced calibration slope (Fig.
3 and Table 12). These findings suggest that further external validation and possible recalibration are needed before clinical implementation.Table 12Temporal validation of the final multimodal modelMetricDerivation cohort n = 130Temporal validation cohort n = 42PH prevalence111/130 (85.4%)30/42 (71.4%)Final model AUC0.8250.792Final model 95% CI0.722–0.9280.639–0.945PASP ECHO AUC0.7300.749PASP ECHO 95% CI0.620–0.8410.588–0.909DeLong P value: final model vs PASP ECHO0.0080.442Brier score0.1030.188Calibration intercept0.091−1.285Calibration slope1.1440.712Youden threshold0.8910.891Sensitivity at Youden threshold60.4%66.7%Specificity at Youden threshold94.7%66.7%PPV at Youden threshold98.5%83.3%NPV at Youden threshold29.0%44.4%Sensitivity-oriented threshold0.6760.676Sensitivity at sensitivity-oriented threshold91.0%100.0%Specificity at sensitivity-oriented threshold52.6%50.0%PPV at sensitivity-oriented threshold91.8%83.3%NPV at sensitivity-oriented threshold50.0%100.0%The final multimodal model was developed in the derivation cohort and applied to the temporal validation cohort without refitting. Thresholds were fixed from the derivation cohort and were not re-optimized in the validation cohort.
AUCs were compared with PASP ECHO alone using DeLong’s test. PH was defined as mPAP > 20 mmHg measured by right heart catheterization. Because the validation cohort was obtained from a later time period within the same institutional setting, it should be interpreted as temporal validation rather than definitive external validationAUC Area under the receiver operating characteristic curve, CI Confidence interval, PPV Positive predictive value, NPV Negative predictive value, PASP ECHO Pulmonary artery systolic pressure estimated by echocardiography, ECG Electrocardiography, BNP B-type natriuretic peptide, NT-proBNP N-terminal pro-B-type natriuretic peptideFig.
3Calibration plots of the final multimodal model. A Derivation cohort. B Temporal validation cohort.
The dashed diagonal line indicates perfect calibration. Points represent grouped observed PH proportions according to predicted probability, and the solid curve represents the fitted calibration curve. The final multimodal model was developed in the derivation cohort and applied to the temporal validation cohort without refitting.
PH, pulmonary hypertension
Agreement between PASP ECHO and sPAP RHC
The association and agreement between PASP ECHO and sPAP RHC are shown in Fig. 4. PASP ECHO showed a moderate positive correlation with sPAP RHC.
However, Bland–Altman analysis showed a mean bias of 5.69 mmHg and limits of agreement from − 21.07 to 32.44 mmHg. These results indicate that echocardiographic and invasive systolic pressure estimates were associated but not interchangeable at the individual-patient level.Fig. 4Agreement between PASP ECHO and sPAP RHC.
A Scatter plot showing the association between PASP ECHO and sPAP RHC. The dashed line represents the line of identity. B Bland–Altman plot showing the difference between PASP ECHO and sPAP RHC against their mean.
The solid horizontal line represents the mean bias, and the dashed horizontal lines represent the limits of agreement. PASP ECHO, pulmonary artery systolic pressure estimated by echocardiography; sPAP RHC, systolic pulmonary artery pressure measured by right heart catheterization; LoA, limits of agreement
Supplementary analyses
The coefficients of the final multimodal model are provided in Supplementary Table S1. Sensitivity analyses of candidate models and the collinearity assessment among echocardiographic variables are shown in Supplementary Table S2. PASP ECHO, TR Vmax, and TR-PG were highly correlated, supporting the decision not to include these echocardiographic variables simultaneously in the final model.
Single-marker ROC performance is summarized in Supplementary Table S3 and Supplementary Figure S2. Decision curve analysis is shown in Supplementary Figure S1. In the derivation cohort, the final multimodal model showed higher net benefit than PASP ECHO alone across part of the clinically relevant threshold probability range.
In the temporal validation cohort, the decision curves were less stable and should be interpreted cautiously because of the limited sample size.
Discussion
In this single-center RHC-referred cohort of patients with suspected PH, we evaluated whether routinely available ECG and natriuretic peptide information could complement echocardiographic assessment. Several findings are noteworthy. First, the final multimodal model, which included PASP ECHO, ECG positivity, and BNP/NT-proBNP positivity, showed higher discrimination than PASP ECHO alone in the derivation cohort.
Second, this improvement was statistically significant in the derivation cohort, but was not statistically significant in the temporal validation cohort. Third, the Youden threshold provided high specificity but limited sensitivity, whereas the sensitivity-oriented threshold increased sensitivity at the expense of specificity. Fourth, Bland–Altman analysis showed wide limits of agreement between PASP ECHO and sPAP RHC, indicating that echocardiographic and invasive pressure estimates are associated but not interchangeable at the individual-patient level [15].
Overall, these findings suggest that ECG and natriuretic peptide positivity may provide complementary information beyond PASP ECHO, but the proposed multimodal approach should be considered exploratory rather than ready for direct clinical implementation.
Relation to ESC/ERS echocardiographic probability assessment
Current ESC/ERS guidelines recommend a probability-based approach to suspected PH, integrating TR velocity, additional echocardiographic signs, and clinical context before confirmatory RHC [4]. RHC remains the reference standard for confirming PH and characterizing invasive hemodynamics [4, 5]. Our inclusion criteria were based on clinical suspicion and RHC referral rather than restriction to the ESC/ERS high echocardiographic probability category.
Therefore, the present cohort should be interpreted as an RHC-referred suspected PH cohort, not as a general screening cohort or a cohort limited to high echocardiographic probability. To address this issue, we summarized ESC/ERS additional echocardiographic signs and echocardiographic PH probability categories separately. The results showed that the study population included patients across low, intermediate, and high echocardiographic probability categories.
This distinction is important because the high PH prevalence in the derivation cohort reflects referral selection and limits generalizability to lower-prevalence populations.
Incremental value of ECG and natriuretic peptide positivity
Echocardiography remains the cornerstone of noninvasive assessment in suspected PH, but echocardiographic pressure estimates are affected by Doppler signal quality, right atrial pressure estimation, and operator-related factors [6–8, 12, 13]. In this study, PASP ECHO alone provided moderate discrimination. Adding ECG positivity and BNP/NT-proBNP positivity improved discrimination in the derivation cohort, suggesting that electrical right-heart strain and myocardial stress provide information not fully captured by PASP ECHO.
This is biologically plausible because ECG abnormalities may reflect right ventricular pressure overload or right-heart remodeling, whereas BNP and NT-proBNP reflect myocardial wall stress and have established prognostic relevance in PH [9–11]. Similar multimodal screening concepts have been used in selected high-risk populations, such as systemic sclerosis-associated pulmonary arterial hypertension, although the present cohort and model are different from those settings [14]. However, the temporal validation cohort showed only a numerically higher AUC for the final model compared with PASP ECHO alone, and the difference was not statistically significant.
Therefore, the incremental value of ECG and natriuretic peptide positivity should be interpreted cautiously and requires further confirmation in larger external cohorts.
Collinearity and overfitting considerations
An important revision of the present analysis was the use of a more parsimonious final model. PASP ECHO, TR Vmax, and TR-PG are physiologically and mathematically related, and including all three variables simultaneously may inflate variance and increase overfitting risk. This issue is particularly relevant because the derivation cohort included only 19 non-PH patients.
Therefore, the final model included only PASP ECHO as the echocardiographic pressure-related variable, together with ECG positivity and BNP/NT-proBNP positivity. Firth penalized logistic regression was used to reduce small-sample bias and potential separation [18]. Nevertheless, low events-per-variable conditions and marked outcome imbalance can still increase overfitting risk and limit model transportability [19, 20].
Bootstrap internal validation showed an optimism-corrected AUC of 0.811, but the small number of non-PH patients and high PH prevalence remain important limitations.
Threshold selection and clinical implications
The Youden threshold should be interpreted as a statistical operating point rather than a clinically preferred screening threshold. In the derivation cohort, the Youden threshold achieved high specificity but limited sensitivity, which may be inadequate for screening-oriented use because missing PH cases is clinically undesirable. Therefore, a sensitivity-oriented threshold was also evaluated.
This threshold increased sensitivity but reduced specificity, indicating a trade-off between case detection and false-positive referrals. In clinical practice, such a threshold may be more appropriate for triage or referral decisions than for confirming PH. Decision curve analysis was used to evaluate the potential net benefit of model-based decision-making across threshold probabilities [17].
However, the model should not be considered a replacement for the ESC/ERS diagnostic pathway or for RHC; rather, it may provide exploratory complementary information when echocardiography, ECG, and natriuretic peptide testing are already available [4, 5].
Calibration, temporal validation, and agreement with RHC
Temporal validation was performed by applying the final derivation model to a later cohort without refitting. The final model showed acceptable discrimination in the temporal validation cohort, but the difference from PASP ECHO alone was not statistically significant. Calibration was less satisfactory in the temporal validation cohort, as reflected by a negative calibration intercept and reduced calibration slope.
This finding suggests that predicted probabilities may require recalibration before use in different clinical settings or populations [20]. Because the validation cohort was obtained from the same institutional setting and was modest in size, it should be interpreted as temporal validation rather than definitive external validation.
Although PASP ECHO was correlated with sPAP RHC, Bland–Altman analysis demonstrated wide limits of agreement [15]. This finding reinforces that correlation does not imply interchangeability. Echocardiography is useful for screening, risk stratification, and determining whether further evaluation is needed, but invasive hemodynamic assessment remains necessary for definitive diagnosis and clinical classification [4–6].
This is particularly important when individual-level treatment decisions depend on accurate hemodynamic measurements.
Limitations
This study has several limitations. First, it was a single-center retrospective study with a relatively small sample size, particularly in the non-PH group. The high PH prevalence reflects an RHC-referred population with high pretest probability and limits generalizability to broader screening populations.
Second, the temporal validation cohort was modest in size and came from a later time period within the same institutional setting; therefore, it should not be considered definitive external validation. Third, complete etiologic classification and pulmonary vascular resistance were not available for all patients, so full hemodynamic subtype classification could not be performed. Model performance may differ across PH subtypes, especially in patients with left-heart-disease-related PH [1–4].
Fourth, echocardiographic measurements are operator dependent and may be affected by TR signal quality and estimation of right atrial pressure [6–8, 12]. Fifth, the biomarker variable combined BNP and NT-proBNP using predefined thresholds, which may not fully capture differences between assays or clinical contexts. Finally, although Firth penalized regression and bootstrap validation were used, the model remains exploratory because of the limited number of non-PH patients and potential overfitting risk [18–20].
Future directions
Future studies should validate this multimodal approach in larger, multicenter prospective cohorts with lower and more diverse PH prevalence. External validation should include detailed PH subtype classification, pulmonary vascular resistance, and standardized echocardiographic probability assessment. Further work should also evaluate whether recalibration or simplified risk categories can improve clinical usability and whether the model can safely support referral decisions without increasing unnecessary invasive testing [20].
Conclusion
In this single-center RHC-referred cohort of patients with suspected PH, ECG positivity and BNP/NT-proBNP positivity provided complementary information to PASP ECHO. The final multimodal model improved discrimination compared with PASP ECHO alone in the derivation cohort, but this improvement was not statistically significant in the temporal validation cohort. Calibration in the temporal validation cohort was less satisfactory, suggesting that the model may require further validation and recalibration before use in other clinical settings.
Therefore, the proposed multimodal approach should be considered exploratory and should not replace ESC/ERS probability-based assessment or confirmatory RHC. Larger prospective multicenter studies with detailed PH subtype classification, external validation, and model recalibration are needed before clinical implementation.
Supplementary Information
Supplementary Material 1: Supplementary Figure S1. Decision curve analysis of PASP ECHO and the final multimodal model. (A) Derivation cohort.
(B) Temporal validation cohort. Net benefit was calculated across threshold probabilities from 0.05 to 0.80. The final multimodal model included PASP ECHO, ECG positivity, and BNP/NT-proBNP positivity.
The model was developed in the derivation cohort and applied to the temporal validation cohort without refitting. PASP ECHO, pulmonary artery systolic pressure estimated by echocardiography.Supplementary Material 2: Supplementary Figure S2. Single-marker ROC curves in the derivation cohort.
(A) Echocardiographic markers, including PASP ECHO, TR Vmax, and TR-PG. (B) Adjunctive binary markers, including ECG positivity and BNP/NT-proBNP positivity. PH was defined as mPAP >20 mmHg measured by right heart catheterization.
ECG positivity and BNP/NT-proBNP positivity were analyzed as binary markers. ROC, receiver operating characteristic; PASP ECHO, pulmonary artery systolic pressure estimated by echocardiography; TR Vmax, peak tricuspid regurgitation velocity; TR-PG, tricuspid regurgitation pressure gradient; BNP, B-type natriuretic peptide; NT-proBNP, N-terminal pro-B-type natriuretic peptide.Supplementary Material 3: Supplementary Table S1. Coefficients of the final multimodal model.
The final multimodal model was fitted using Firth penalized logistic regression in the derivation cohort. PH was defined as mPAP >20 mmHg measured by right heart catheterization. PASP ECHO was modeled as a continuous variable per 1-mmHg increase.
ECG positivity and BNP/NT-proBNP positivity were modeled as binary variables. PASP ECHO, pulmonary artery systolic pressure estimated by echocardiography; ECG, electrocardiography; BNP, B-type natriuretic peptide; NT-proBNP, N-terminal pro-B-type natriuretic peptide; OR, odds ratio; CI, confidence interval; SE, standard error; PH, pulmonary hypertension.Supplementary Material 4: Supplementary Table S2. Sensitivity models and echo-variable collinearity.
Panel A. Incremental candidate model performance in the derivation cohort. Panel A shows the incremental performance of candidate models fitted using Firth penalized logistic regression in the derivation cohort.
Model 1, including PASP ECHO alone, was used as the reference model for DeLong comparisons. Panel B. Correlation among echocardiographic variables.
Panel B shows Spearman correlation coefficients among echocardiographic pressure-related variables. PASP ECHO, TR Vmax, and TR-PG were not included simultaneously in the final model because of substantial collinearity. PH was defined as mPAP >20 mmHg measured by right heart catheterization.
AUC, area under the receiver operating characteristic curve; CI, confidence interval; PASP ECHO, pulmonary artery systolic pressure estimated by echocardiography; TR Vmax, peak tricuspid regurgitation velocity; TR-PG, tricuspid regurgitation pressure gradient; ECG, electrocardiography; BNP, B-type natriuretic peptide; NT-proBNP, N-terminal pro-B-type natriuretic peptide.Supplementary Material 5: Supplementary Table S3. Single-marker ROC performance in the derivation cohort. Single-marker ROC analyses were performed in the derivation cohort.
PH was defined as mPAP >20 mmHg measured by right heart catheterization. ECG positivity and BNP/NT-proBNP positivity were analyzed as binary markers. PASP ECHO, TR Vmax, and TR-PG were analyzed as continuous echocardiographic markers.
ROC, receiver operating characteristic; AUC, area under the receiver operating characteristic curve; CI, confidence interval; PASP ECHO, pulmonary artery systolic pressure estimated by echocardiography; TR Vmax, peak tricuspid regurgitation velocity; TR-PG, tricuspid regurgitation pressure gradient; ECG, electrocardiography; BNP, B-type natriuretic peptide; NT-proBNP, N-terminal pro-B-type natriuretic peptide.