Endotypes of Vascular Health Predict Transplantation‐Free Survival in Pulmonary Hypertension

Abstract

Background

The current classification of pulmonary hypertension (PH), based largely on expert opinion, has limitations in prognostication and guiding therapies. We hypothesize that novel PH clusters that predict survival will reveal mechanistic phenotypes associated with biomarkers of vascular health across all PH groups.

Methods

We first identify novel PH clinical clusters by performing unsupervised clustering analysis on the CC‐PH (Cleveland Clinic PH) registry (N=1529). We develop classification models to predict the new PH clusters and then apply them to the multicenter PVDOMICS (Pulmonary Vascular Diseases Phenomics) cohort (N=853) for validation. We compare transplantation‐free survival across the new PH clusters. We quantify metabolites of the arginine‐nitric oxide pathway and D‐dimer levels and calculate global arginine bioavailability (arginine/[ornithine+citrulline]) to assess endothelial function and activation in the new clusters and link these biomarkers to clinical outcomes.

Results

Clustering analysis identify 3 clear clusters in CC‐PH that are validated in PVDOMICS and outperform conventional classifications in predicting transplantation‐free survival. The phenotype associated with the worst survival is characterized by reduced lung diffusion capacity, decreased arginine bioavailability and nitrate levels, and elevated D‐dimer levels, consistent with loss of pulmonary microcirculation and endothelial dysfunction.

Conclusions

We identify new informative PH phenotypes associated with mortality and defined by biomarkers of endothelial function and activation. Loss of endothelial health and pronounced pulmonary vascular rarefication contribute more substantially to mortality across the spectrum of PH than right heart function.

REGISTRATION: URL: https://www.clinicaltrials.gov; Unique identifier: NCT02980887.

Nonstandard Abbreviations and Acronyms

6MWD

6‐minute walk distance

ADMA

asymmetric dimethylarginine

CC‐PH

Cleveland Clinic Pulmonary Hypertension registry

DLCO

lung diffusing capacity for carbon monoxide

GABR

Global arginine bioavailability

mPAP

mean pulmonary arterial pressure

NO

nitric oxide

PAH

pulmonary arterial hypertension

PCWP

pulmonary capillary wedge pressure

PH

pulmonary hypertension

PVDOMICS

Pulmonary Vascular Diseases Phenomics

WSPH

World Symposium on Pulmonary Hypertension

Clinical Perspective.

What Is New?

• Three pulmonary hypertension clusters associated with transplantation‐free survival were identified in a discovery cohort and validated in an independent cohort, revealing biologically distinct phenotypes linked to endothelial health as reflected by global arginine bioavailability and D‐dimer levels.

What Are the Clinical Implications?

• Transplantation‐free survival in pulmonary hypertension depends on preservation of vascular endothelial health, underscoring the need for targeted interventions to restore endothelial function, particularly in high‐risk phenotypes.

Pulmonary hypertension (PH) is defined hemodynamically by a mean pulmonary arterial pressure (mPAP) of >20 mm Hg¹ and is classified by the World Symposium on Pulmonary Hypertension (WSPH) into 5 groups based on hemodynamics, clinical features, and underlying cause. ¹ Despite marked changes over the past 5 decades in an effort to capture the diverse phenotypes of this complex disease, the current clinical classification remains limited. It does not fully capture patient heterogeneity and group overlap, hindering granular biological distinction, prognostication, and potential for biologically informed individualized treatment. Notably, the demographic profile of patients with WSPH 1 pulmonary arterial hypertension (PAH) has changed in recent decades, with increasing age and more comorbidities. ² , ³ , ⁴ , ⁵ These observations raise critical questions about the limits of current classification schemes and highlight the need for mechanistically informed groupings.

The pathogenesis of PAH is multifactorial, with endothelial dysfunction playing a central role. ⁶ , ⁷ , ⁸ The pulmonary endothelium has several functions including preventing leakage, maintenance of vascular tone, leukocyte trafficking, production of growth factors and cell signals and prevention of blood clotting. In PAH, abnormal endothelial proliferation and angiogenesis result in plexiform lesions, a hallmark feature of the disease. ⁹ Disrupted production of vasoactive mediators, primarily nitric oxide (NO), endothelin‐1, and prostacyclin, leads to vasoconstriction, smooth muscle proliferation, and vascular remodeling. ⁸ In addition, reduced levels of prostacyclin and NO, both antithrombotic, along with endothelial injury and sluggish pulmonary blood flow create a thrombogenic environment with impaired fibrinolysis and microthrombus formation. ⁸ Therapies targeting NO, endothelin‐1, and prostacyclin are the mainstay of PAH treatment and are used in other WSPH groups despite less established mechanistic basis. ¹⁰ NO is produced by endothelial NO synthase by converting arginine to NO and citrulline. Functional NO deficiency is a key pathogenic mechanism in PAH. ¹¹ , ¹² , ¹³ Global arginine bioavailability (GABR) is a better measure of NO production than arginine levels and is a strong predictor of cardiovascular health. ¹⁴ , ¹⁵ , ¹⁶ We recently investigated NO‐arginine metabolism in PH and showed reduced GABR across all WSPH groups suggesting shared endothelial dysfunction. However, marked heterogeneity in metabolites among and within the PH groups supports the presence of distinct pulmonary vascular endotypes. ¹⁷ Conversely, D‐dimer, a marker of endothelial activation and intravascular thrombosis has not been adequately studied in PH. One study reported elevated D‐dimer levels in PAH associated with congenital heart disease, whereas another found no difference in idiopathic and PAH associated with systemic sclerosis after adjusting for age. ¹⁸ , ¹⁹ Classifying patients with PH based on endothelial health using plasma measures of NO‐arginine metabolites and D‐dimer levels has not been explored in PH.

We hypothesize that grouping based on hard clinical end points like death or transplantation, coupled with deep phenotyping, can reveal novel subgroups with shared biological mechanisms, enabling more precise therapeutic interventions. Using a CC‐PH (Cleveland Clinic Pulmonary Hypertension) cohort for discovery, we identify clinical clusters and validate our findings in the multicenter National Heart, Lung, and Blood Institute PVDOMICS (Pulmonary Vascular Diseases Phenomics) cohort. The deeply phenotyped PVDOMICS cohort is used to test whether the PH phenotype linked to long‐term survival is associated with biomarkers of endothelial health, including high GABR and nitrates, low D‐dimer, and preserved diffusing capacity of the lung for carbon monoxide (DLCO). The findings provide proof of concept that transplantation‐free survival in PH relies on preservation of vascular endothelial health and highlight the need to develop targeted interventions that restore endothelial function, especially in high‐risk phenotypes.

METHODS

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

Study Design

The overall study design is illustrated in Figure 1. We used the clinical parameters detailed in Table S1 for clustering analysis. We identified new PH clusters through clustering of the CC‐PH registry patients. Prediction models for the new PH clusters were trained and then applied to the PVDOMICS cohort. To assess prognostic validity, we estimated transplantation‐free survival across the newly identified clusters. The PVDOMICS cohort enabled investigation of endothelial biology and its association with clinical outcomes. Endothelial function and activation were evaluated by measuring metabolites of the arginine–NO pathway and D‐dimer levels and by calculating global arginine bioavailability (arginine / [ornithine + citrulline]) within each cluster.

Figure 1. Overview of study analyses.

CC‐PH indicates Cleveland Clinic Pulmonary Hypertension registry; PVDOMICS, Pulmonary Vascular Diseases Phenomics; and WSPH, World Symposium on Pulmonary Hypertension.

Cleveland Clinic Pulmonary Hypertension Registry

The CC‐PH registry is an institutional review board‐approved registry that was established in May 2005. It enrolled consecutive patients seen by the Pulmonary Vascular Group at the Cleveland Clinic. Those seen before May 2005 were enrolled retrospectively, and those seen after that date were enrolled prospectively. For this study, the registry was queried in December 2013 and yielded 2010 patients who underwent right heart catheterization for suspected PH from as early as August 1980. Data collected from the study participants include demographics, comorbidities, body mass index, right heart catheterization, Doppler echocardiography, 6‐minute walk distance (6MWD), pulmonary function testing, BNP (B‐type natriuretic peptide) levels, complete blood counts and chemistries, and PAH‐targeted therapies at baseline and on follow‐up. A group of expert heart failure cardiologists performed all right heart catheterization. Mortality and lung transplantation were determined in these participants by manual and electronic review of the medical records and the Social Security Death Index. Entry point into the study was the date of the diagnostic right heart catheterization. All baseline variables were obtained closest to the time of the right heart catheterization. The median follow‐up time was 3.9 years with an interquartile range of 1.4 to 7.2.

Pulmonary Vascular Diseases Phenomics

The PVDOMICS consortium, funded by the National Institutes of Health, consisted of 7 clinical centers across the United States and a Data Coordinating Center at Cleveland Clinic. It is an observational prospective longitudinal cohort (clinicaltrials.gov NCT02980887). The study protocol and a Strengthening the Reporting of Observational Studies in Epidemiology diagram depicting enrollment and patient classification have been previously published. ²⁰ A total of 1193 participants were enrolled from November 2016 to October 2019. All participants provided written consent. The inclusion/exclusion criteria and the study protocol have been previously published. ²¹ In summary, adult patients with known or suspected pulmonary vascular disease, matched disease comparators and healthy controls were recruited at the consortium clinical centers. Each participant underwent comprehensive clinical data collection, including demographics, laboratory variables, medical history, 6MWD, pulmonary function testing chest imaging, echocardiogram and cardiac magnetic resonance imaging. ²¹ , ²² Furthermore, fasting blood samples were drawn from the study participants for comprehensive omics profiling, including genome, transcriptome, proteome, metabolome, coagulome, and cell biome. ²⁰ Arginine pathway metabolites and urine nitrate were measured using high‐performance liquid chromatography as previously described. ¹⁷ Nontargeted metabolomics analysis were performed at Metabolon, Inc (Durham, NC). The details of analytical platform and data curation have been described in detail. ¹⁷ , ²³ Log2 transformed, normalized. and imputed metabolite values were used for the analysis. The Data Coordinating Center at Cleveland Clinic provided centralized reading for echocardiography, cardiac magnetic resonance imaging, cardiopulmonary exercise testing, electrocardiogram, chest computerized tomography scan, overnight sleep study, pulmonary function testing, and right heart catheterization. Information about death and transplantation was prospectively collected. The follow‐up was up to 7.0 years (median=4.8, [interquartile range, 2.8–6.0]).

Hemodynamic Criteria Classifications

All participants from the CC‐PH registry and PVDOMICS were classified based on the seventh WSPH hemodynamic criteria ¹ as no PH (mPAP ≤20 mm Hg); precapillary PH (mPAP >20 mm Hg, pulmonary vascular resistance (PVR) >2 Wood units; pulmonary capillary wedge pressure (PCWP) ≤ 15 mm Hg), postcapillary PH (mPAP >20 mm Hg, PVR ≤2 Wood units, PCWP >15 mm Hg); combined pre‐ and postcapillary PH (mPAP >20 mm Hg, PVR >2 Wood units, PCWP >15 mm Hg), and other PH (mPAP >20 mm Hg and rest of hemodynamics missing or not fitting any of the previous categories) (Table 1). For this study, we excluded participants with no PH (mPAP ≤20 mm Hg) or younger than 18 years old. The final cohort included 1529 adult participants with PH from the CC‐PH registry and 853 from PVDOMICS. The hemodynamic criteria used in this study differ from those applied in PVDOMICS, ²⁰ , ²¹ which accounts for discrepancies in numbers and characteristics of participants with PH compared with prior publications.

Table 1.

CC‐PH Registry and PVDOMICS Participants by WSPH and Hemodynamic Pulmonary Hypertension Classification

Classification	CC‐PH registry, N=1529	PVDOMICS, N=853
WSPH clinical classification
Group 1: Pulmonary arterial hypertension	711 (46.5)	328 (38.5)
Group 2: PH associated with left heart disease	406 (26.6)	217 (25.4)
Group 3: PH associated with lung disease or hypoxia	180 (11.8)	219 (25.7)
Group 4: PH associated with pulmonary artery occlusions	119 (7.8)	57 (6.7)
Group 5: PH with unclear or multifactorial mechanisms	113 (7.4)	32 (3.8)
Hemodynamic classification
Isolated postcapillary PH	122 (8.0)	69 (8.1)
Precapillary PH	746 (48.8)	508 (59.6)
Combined post‐ and precapillary PH	455 (29.8)	200 (23.4)
Unclassified PH	206 (13.5)	76 (8.9)

N (%) presented.

CC‐PH indicates Cleveland Clinic Pulmonary Hypertension registry; PH, pulmonary hypertension; PVDOMICS, Pulmonary Vascular Diseases Phenomics; and WSPH, World Symposium on Pulmonary Hypertension.

Data Harmonization

Data dictionaries of the 2 studies were thoroughly examined to identify common variables and discrepancies. Study protocols were also compared to identify possible differences in data collection methods. The common variables were then aligned across the 2 data sets. The measurement units were compared. For the variables measuring the same parameter with different units or scales, appropriate conversions were performed. BNP was estimated from N‐terminal‐proBNP in PVDOMICS. ²⁴ Standard terminologies and codes (eg, New York Heart Classification class) were applied to ensure consistency. Range and logistic checks were also conducted.

Statistical Analysis

Participant characteristics were summarized as means±SD, medians, and interquartile ranges or frequencies and percentages by study cohort or PH group. ANOVA or the Kruskal–Wallis test was used to compare continuous variables, and the chi‐squared test was used to compare categorical variables among different groups. Kaplan–Meier curves were used to estimate the transplant‐free survival probabilities over time for different PH groups or clusters. Log rank test and Cox proportional hazard model were used to compare the survival curves. All analyses were conducted using R (R core team, Vienna, Austria). Statistical significance was established at 2‐sided P values <0.05.

Clustering Variables and Missing Data

Eighty‐five common variables were identified in both cohorts. To reduce collinearity, pairwise Pearson correlation coefficients were calculated for continuous variables. For any pair of continuous variables with a correlation >0.9 (eg, body surface area and weight), the variable with fewer missing data was kept. A similar filtering step was applied to any pair of categorical variables with a Cramer's V >0.5. Furthermore, the variables with at least 80% missing data were excluded. These eventually led to 56 variables for clustering analysis (Table S1). The chained equation approach was applied to impute the missing data in these 56 variables. ²⁵ Five complete data sets were generated for the CC‐PH and PVDOMICS cohorts, respectively.

Clustering Analysis

To identify the new PH clusters, the K‐means clustering algorithm was applied to regroup the CC‐PH participants based on the similarities of their characteristics in terms of the 56 clustering variables. The K‐means method was applied to each imputed data set, and the final clustering results were pooled across the analyses of the 5 imputed data sets using nonnegative matrix factorization. ²⁶ The number of clusters was determined using the Bayesian information criterion.

PH Cluster Prediction

Considering the new cluster labels as the ground truth, a Gradient Boosting model was trained with the clustering variables as predictors. The training process was performed using the XGBoost package in R. One model was trained for each imputed CC‐PH data set. Feature importance was determined by averaging the gains of each feature across the 5 boosting models. The trained boosting models were then applied to predict the new PH clusters for the PVDOMICS subjects. Because there were 5 five imputed PVDOMICS data sets, each PVDOMICS participant had a total of 25 predictions, and the final PH cluster was determined by majority voting.

To link the new PH clusters with arginine‐NO metabolites and D‐dimer, a multinomial logistic regression model was further developed to predict the new PH clusters using GABR and D‐dimer instead of 56 clinical variables. This model was developed using the PVDOMICS cohort only because GABR and D‐dimer were not available in the CC‐PH registry. Micro‐averaged receiver operating characteristic curve was used to assess the performance of the model.

RESULTS

Among CC‐PH participants, 46.5% were WSPH 1, 26.6% WSPH 2, and 11.8% WSPH 3. (Table 1). In PVDOMICS, most participants were also classified within WSPH Groups 1 to 3. Based on hemodynamic criteria, 48.8% of CC‐PH participants had precapillary PH, 29.8% had combined pre‐ and postcapillary PH, and 8% isolated postcapillary PH; corresponding proportions in PVDOMICS were 59.6%, 23.4%, and 8.1% respectively. Demographics and clinical characteristics of both cohorts are summarized in Table 2. The mean age was 59.6 years in CC‐PH, 34.8% were male, and 78.6% were White. PVDOMICS participants had similar demographic profiles, with a mean age of 59.5 years, 39.4% male, and 78.7% White. CC‐PH participants had more severe hemodynamics with higher mPAP, PVR, and PCWP along with lower stroke volume. The 6MWD was shorter in CC‐PH (293 m versus 338 m) and pulmonary functions were lower (forced vital capacity: 2.4 L versus 2.7 L; forced expiratory volume in 1 second [FEV1]: 1.7 L versus 1.9 L; DLCO: 12.6 versus 13.6 mL/min/mm Hg). BNP levels were comparable between the 2 cohorts. A higher proportion of PVDOMICS participants exhibited right atrial dilation. A greater proportion of PVDOMICS participants were receiving PH medications (46.1% versus 37.3%).

Table 2.

Demographic and Clinical Characteristics of CC‐PH and PVDOMICS Participants

	CC‐PH registry, N=1529	PVDOMICS, N=853	P value
Age, y	59.6±14.8	59.5±14.2	0.85
Male sex, (%)	532 (34.8)	336 (39.4)	0.03
White race (%)	1195 (78.6)	671 (78.7)	1.00
Body mass index, kg/m2	30.9±13.7	30.8±8.1	0.87
6‐min walk distance, m	292.9±119.2	338.3±129.0	<0.001
FVC, L	2.4 (1.8–3.1)	2.7 (2.1–3.4)	<0.001
FEV1, L	1.7 (1.3–23)	1.9 (1.4–2.5)	<0.001
FEV1/FVC	73.1±12.4	72.5±12.5	0.38
Lung diffusing capacity for carbon monoxide, mL/min/mm Hg	12.6±6.3	13.6±6.6	0.001
Echocardiography
Left ventricular ejection fraction (%)	55.5 (9.1)	58.5 (10.0)	<0.001
Tricuspid regurgitant, cm/s	397.9±77.0	355.4±71.8	<0.001
Right atrium mean pressure, mm Hg	11.6±6.9	8.4±4.9	<0.001
Right ventricular dilation (%)	711 (69.0)	554 (70.3)	0.57
Right atrium dilation (%)	575 (56.2)	619 (80.7)	<0.001
Systolic blood pressure, mm Hg	131.5±24.7	130.0±24.7	0.15
Mean blood pressure, mm Hg	95.5±15.9	92.7±15.5	<0.001
Heart rate, beats/min	79.9±15.7	74.8±14.0	<0.001
Hemodynamics
Mean pulmonary artery pressure, mm Hg	45.5±14.2	38.7±13.2	<0.001
Pulmonary wedge pressure, mm Hg	15.7±8.7	13.5±6.3	<0.001
Pulmonary vascular resistance, Wood units	7.4±5.7	5.4±4.2	<0.001
Thermodilution cardiac index, L/min	2.6±0.9	2.7±0.9	0.12
Fick cardiac index, L/min	2.6±0.9	2.5±0.8	0.16
Fick cardiac output, L/min	4.9±1.9	4.9±1.6	0.65
Stroke volume (ml)	63.3±25.5	73.4±27.1	<0.001
Mixed venous oxygenation (%)	64.2±9.4	64.8±9.2	0.22
B‐type natriuretic peptide, pg/ml	213.5 (72.0–523.0)	195.0 (69.0–686.0)	0.35
Serum urea nitrogen, mg/dl	18.0 (14.0–26.0)	17.50 (13.0–23.0)	0.01
Creatinine, mg/dl	0.94 (0.80–1.20)	0.95 (0.80–1.18)	0.69
White blood cell count, cells/μL	8.8±5.6	7.2±2.6	<0.001
Red blood cell count, cells/μL	4.3±0.9	4.6±0.8	<0.001
Hemoglobin, g/dl	12.1±2.5	13.4±2.2	<0.001
New York Heart Association Class N (%)			<0.001
I	18 (2.4)	49 (6.0)
II	196 (26.4)	277 (33.8)
III	386 (52.0)	450 (54.9)
IV	143 (19.2)	43 (5.3)
Smoker N (%)			<0.001
No	455 (29.8)	419 (49.1)
Yes	418 (27.3)	428 (50.2)
Unknown	656 (42.9)	6 (0.7)
Pulmonary hypertension vasodilators N (%)	532 (37.3)	392 (46.1)	<0.001
Prostacyclin		167 (19.6)
Phosphodiesterase‐5 inhibitor		302 (35.5)
Soluble guanylate cyclase stimulator		38 (4.5)
Endothelin receptor antagonist		216 (25.4)
Monotherapy		145 (17.1)
Dual therapy		160 (18.8)
Triple therapy		83 (9.8)

Mean±SD or median [interquartile range] presented for continuous variables, N (%) for categorical variables.

CC‐PH indicates Cleveland Clinic Pulmonary Hypertension registry; FEV1, forced expiratory volume in 1 s; FVC, forced vital capacity; and PVDOMICS, Pulmonary Vascular Diseases Phenomics.

In CC‐PH, 39.5% of participants died, and 42.0% either died or underwent lung transplantation during follow‐up. The median transplantation‐free survival was 9.5 years. Transplantation‐free survival differed across WSPH groups (Figure 2A) and hemodynamic classifications (Figure 2B); however, not all group comparisons reached statistical significance, and some showed overlapping outcomes. For example, there was no significant difference in transplantation‐free survival between WSPH 1 and 2 or between WSPH 1 and 5. In the PVDOMICS cohort, the mortality rate was 29.2%, and 38.2% of participants either died or underwent lung transplantation. The median transplantation‐free survival was 6.3 years. Similar to CC‐PH, transplantation‐free survival varied across WSPH groups (Figure 2C) and hemodynamic classifications (Figure 2D), with some overlap observed. For instance, WSPH 1 did not differ significantly from Groups 4 or 5, and precapillary PH showed similar survival to isolated postcapillary PH and other PH subtypes.

Figure 2. Kaplan–Meier transplantation‐free survival curves by WSPH groups and hemodynamic classifications in both cohorts.

A, Kaplan–Meier curves by WSPH groups in CC‐PH registry; B, Kaplan–Meier curves by hemodynamic classifications in CC‐PH registry; C, Kaplan–Meier curves by WSPH groups in PVDOMICS; D, Kaplan–Meier curves by hemodynamic classifications in PVDOMICS. CC‐PH indicates Cleveland Clinic Pulmonary Hypertension; PVDOMICS, Pulmonary Vascular Diseases Phenomics; and WSPH, World Symposium on Pulmonary Hypertension. *Two patients in the CC‐PH registry did not have survival times and were not included in the survival analysis.

New Identified PH Clusters

The K‐means algorithm uncovered th3ree distinct clinical clusters in the CC‐PH cohort. The optimal number of clusters (K=3) was determined using the Bayesian information criterion (Figure S1). Of the 1529 participants, 574 (38%) were assigned to Cluster 1, 470 (31%) to Cluster 2, and 485 (32%) to Cluster 3 (Table 3). The 3 clusters demonstrated significantly different transplantation‐free survival trajectories (Figure 3A, P<0.001). Median transplantation‐free survival time was 23.2 years for Cluster 1, 7.2 years for Cluster 2 and 4.3 years for Cluster 3. Lung transplantation occurred in 2.1% of patients in Cluster 3 compared with 3.1% in Cluster 1 and 5.3% in Cluster 2. Compared with Cluster 1, the hazard ratio for death or transplantation was 2.0 (95% CI, 1.6–2.4) for Cluster 2 and 2.7 (95% CI, 2.3–3.4) for Cluster 3, showing an increasingly worse prognosis from Cluster 1 to Cluster 3 (P<0.001).

Table 3.

Characteristics of CC‐PH and PVDOMICS Participants by New PH Clusters

CC‐PH cohort	Cluster 1 (N=574)	Cluster 2 (N=470)	Cluster 3 (N=485)	P‐ value
Age, y	57.8±13.5	54.1±15.3	67.0±12.9	<0.001
Male sex (%)	225 (39.2)	164 (34.9)	143 (29.5)	0.004
White race (%)	473 (83.1)	371 (79.1)	351 (72.8)	<0.001
BMI, kg/m2	32.1±9.2	28.2±7.2	32.2±20.8	<0.001
6‐min walk distance, m	339.6±108.4	303.5±112.0	211.7±99.7	<0.001
FVC, L	2.8 (2.2–3.5)	2.7 (2.2–3.4)	1.7 (1.4–2.1)	<0.001
FEV1, L	2.0 (1.5–2.5)	2.0 (1.7–2.6)	1.2 (1.0–1.5)	<0.001
FEV1/FVC	72.4±12.8	74.2±10.8	72.7±13.4	0.13
DLCO, mL/min/mm Hg	14.7±6.2	13.7±6.3	8.7±4.1	<0.001
Echocardiography
LV ejection fraction (%)	57.4 (7.9)	54.5 (9.1)	54.3 (10.0)	<0.001
Tricuspid regurgitant, cm/s	354.4±68.2	434.3±64.7	404.0±75.8	<0.001
Right atrium mean pressure, mm Hg	8.4±4.8	12.5±7.1	14.5±7.2	<0.001
Right ventricular dilation–yes N (%)	164 (42.3)	336 (92.8)	211 (75.1)	<0.001
Right atrium dilation–yes N (%)	105 (27.0)	288 (80.9)	182 (65.5)	<0.001
Systolic blood pressure, mm Hg	133.1±23.2	122.0±20.9	139.2±26.7	<0.001
Mean blood pressure, mm Hg	95.9±15.2	93.1±15.4	97.5±16.9	<0.001
Heart rate, beats/min	77.1±14.8	83.3±15.6	79.9±16.0	<0.001
Hemodynamics
Mean pulmonary artery pressure, mm Hg	37.3±11.4	55.0±13.2	46.1±12.1	<0.001
Pulmonary wedge pressure, mm Hg	14.1±6.5	12.9±7.9	20.3±9.8	<0.001
Pulmonary vascular resistance, WU	4.3±2.7	12.4±6.1	6.4±4.7	<0.001
Thermodilution cardiac index, L/min	3.2±1.0	2.1±0.6	2.5±0.7	<0.001
Fick cardiac index, L/min	3.1±1.0	2.0±0.5	2.6±0.8	<0.001
Fick cardiac output, L/min	6.1±2.1	3.7±1.0	4.8±1.5	<0.001
Stroke volume, ml	79.8±25.9	46.3±16.3	59.5±19.0	<0.001
Mixed venous oxygenation (%)	69.9 (6.6)	58.3 (9.5)	63.2 (8.2)	<0.001
B‐type natriuretic peptide, pg/ml	69.0 (30.0–160.0)	330.5 (143.0–795.0)	350.0 (178.0–715.0)	<0.001
White blood cell count, cells/μL	8.5±5.4	8.4±4.9	9.4±6.3	0.02
Red blood cell count, cells/μL	4.4±0.8	4.4±0.9	4.0±0.8	<0.001
Hemoglobin, g/dl	12.8±2.4	12.5±2.7	11.0±2.2	<0.001
Serum urea nitrogen, mg/dl	16.0 (12.0–21.0)	17.0 (13.0–23.0)	24.0 (17.0–37.0)	<0.001
Creatinine, mg/dl	0.9 (0.7–1.0)	0.9 (0.8–1.2)	1.1 (0.8–1.4)	<0.001
NYHA Class N (%)				<0.001
I	13 (5.1)	3 (1.1)	2 (1.0)
II	95 (37.4)	71 (25.4)	30 (14.4)
III	126 (49.6)	144 (51.4)	116 (55.5)
IV	20 (7.9)	62 (22.1)	61 (29.2)
Smoker N (%)				<0.001
No	170 (29.6)	159 (33.8)	126 (26.0)
Yes	181 (31.5)	122 (26.0)	115 (23.7)
Unknown	223 (38.9)	189 (40.2)	244 (50.3)
PH medication N (%)	162 (30.5)	198 (44.5)	172 (38.3)	<0.001

PVDOMICS cohort	Cluster 1 (N=503)	Cluster 2 (N=179)	Cluster 3 (N=171)	P value
Age, y	57.9±13.5	56.3±15.9	67.2±11.1	<0.001
Male sex (%)	211 (41.9)	76 (42.5)	49 (28.7)	0.01
White race (%)	399 (79.3)	137 (76.5)	135 (78.9)	0.73
BMI, kg/m2	31.5±8.0	27.6±6.3	32.1±9.2	<0.001
6‐min walk distance, m	365.5±121.3	337.1±131.3	238.7±103.0	<0.001
FVC, L	2.9 (2.3–3.6)	3.0 (2.5–3.6)	1.8 (1.5–2.2)	<0.001
FEV1, L	2.1 (1.6–2.6)	2.2 (1.8–2.5)	1.3 (1.1–1.5)	<0.001
FEV1/FVC	72.6±13.1	73.6±9.4	71.3±13.1	0.29
DLCO, mL/min/mm Hg	14.9±6.9	13.6±6.4	9.4±3.8	<0.001
Echocardiography
LV ejection fraction (%)	59.6 (8.0)	57.3 (12.8)	56.3 (11.4)	<0.001
Tricuspid regurgitant, cm/s	327.9±61.3	410.4±65.2	367.4±66.8	<0.001
Right atrium mean pressure, mm Hg	7.0±4.0	9.5±5.0	11.2±5.8	<0.001
Right ventricular dilation–yes N (%)	264 (56.4)	165 (98.2)	125 (82.2)	<0.001
Right atrium dilation–yes N (%)	321 (71.3)	161 (95.8)	137 (91.9)	<0.001
Tricuspid annular plane systolic excursion/pulmonary arterial systolic pressure, mm/mm Hg	0.41 (0.31–0.55)	0.20 (0.16–0.25)	0.28 (0.21–037)	<0.001
Systolic blood pressure, mm Hg	131.7±24.7	119.0±21.3	136.4±24.7	<0.001
Mean blood pressure, mm Hg	93.2±15.2	88.4±14.9	96.0±16.0	<0.001
Heart rate, beats/min	72.4±13.0	78.6±14.1	77.8±15.1	<0.001
Hemodynamics
Mean pulmonary artery pressure, mm Hg	32.8±10.0	51.2±12.2	42.8±11.6	<0.001
Pulmonary wedge pressure, mm Hg	12.4±5.5	12.5±6.9	17.4±6.6	<0.001
Pulmonary vascular resistance, WU	3.6±2.0	10.6±5.3	5.5±2.9	<0.001
Thermodilution cardiac index, L/min	3.0±0.9	2.1±0.6	2.6±0.8	<0.001
Fick cardiac index, L/min	2.8±0.7	2.0±0.5	2.2±1.0	<0.001
Fick cardiac output, L/min	5.6±1.5	3.7±0.9	4.1±1.3	<0.001
Stroke volume, ml	83.8±26.5	51.3±14.9	65.7±21.8	<0.001
Mixed venous oxygenation (%)	68.3 (7.2)	59.9 (10.9)	59.7 (7.9)	<0.001
B‐type natriuretic peptide, pg/ml	89.0 (45.0–208.0)	694.0 (344.0–1482.0)	700.0 (252.0–1198.0)	<0.001
White blood cell count, cells/μL	7.1±2.5	7.2±2.4	7.7±3.1	0.03
Red blood cell count, cells/μL	4.6±0.8	4.8±0.7	4.3±0.8	<0.001
Hemoglobin, g/dl	13.6±2.0	14.2±2.0	12.2±2.2	<0.001
Serum urea nitrogen, mg/dl	16.0 (13.0–21.0)	18.0 (14.0–24.0)	23.0 (16.0–33.0)	<0.001
Creatinine, mg/dl	0.9 (0.8–1.1)	1.0 (0.8–1.2)	1.10 (0.9–1.5)	<0.001
NYHA Class N (%)				<0.001
I	38 (7.8)	9 (5.2)	2 (1.3)
II	193 (39.7)	53 (30.5)	31 (19.5)
III	246 (50.6)	90 (51.7)	114 (71.7)
IV	9 (1.9)	22 (12.6)	12 (7.5)
Smoker N (%)				<0.001
No	258 (51.3)	87 (48.6)	74 (43.3)
Yes	242 (48.1)	90 (50.3)	96 (56.1)
Unknown	3 (0.6)	2 (1.1)	1 (0.6)
PH medication N (%)	227 (45.2)	103 (57.9)	62 (36.5)	<0.001

Mean±SD or median [interquartile range] presented for continuous variables, N (%) for categorical variables.

BMI indicates body mass index; CC‐PH, Cleveland Clinic Pulmonary Hypertension registry; DLCO, lung diffusing capacity for carbon monoxide; FEV1, forced expiratory volume in 1 second; FVC, forced vital capacity; LV, left ventricular; NYHA, New York Heart Association; PH, pulmonary hypertension; PVDOMICS, Pulmonary Vascular Diseases Phenomics; and WU, Wood unit.

Figure 3. New clusters identified in the discovery cohort and then validated in an independent cohort uncover phenotypes associated with transplantation‐free survival and linked to underlying biological mechanisms‐ endothelial health.

A, Kaplan–Meier transplantation‐free survival curves by new (PH) cluster in the CC‐PH registry; B, Kaplan–Meier curves by the predicted new PH clusters in PVDOMICS; C, The cluster with the worse outcome has the lowest global arginine bioavailability and highest D‐dimer and elevated BNP. BNP indicates B‐type natriuretic peptide; CC‐PH, Cleveland Clinic Pulmonary Hypertension; PH, pulmonary hypertension; and PVDOMICS, Pulmonary Vascular Diseases Phenomics. *Two patients in the CC‐PH registry did not have survival times and were not included in the survival analysis.

The top 15 most important variables contributing to cluster assignment, in descending order of importance included: PVR, FEV1, BNP, stroke volume, forced vital capacity, mixed venous oxygen saturation, 6MWD, mPAP, mean right atrial pressure, DLCO, tricuspid regurgitant jet velocity, serum urea nitrogen, PCWP, indirect Fick cardiac output, and hemoglobin (Figure S2).

When comparing the clinical characteristics across the 3 new clusters, patients in Cluster 3 were older (mean age 67 years versus 57.8 in Cluster 1 and 54.1 in Cluster 2) and more likely to be female (70% versus 60.8% in Cluster 1 and 65.1% in Cluster 2) (Table 3). The 6MWD decreased from 339.6 m in Cluster 1 to 303.5 m in Cluster 2 and 211.7 m in Cluster 3. Other markers of disease severity worsened from Cluster 1 to Cluster 3 including right arterial pressure, BNP, and serum urea nitrogen. Cluster 3 exhibited the most impaired pulmonary function (lowest FEV1, forced vital capacity, and DLCO) compared with Clusters 1 and 2. However, Cluster 3 had less right arterial and right ventricular dilation by echocardiography compared with Cluster 2. Among the 3 clusters, Cluster 2 had the worst hemodynamics with the highest mPAP and PVR and the lowest cardiac output and stroke volume (Table 3). Cluster 1 had the least severe hemodynamics. Cluster 3 exhibited intermediate values for most hemodynamic measures but had the highest PCWP (20.3 mm Hg), compared with 14.1 mm Hg in Cluster 1 and 12.9 mm Hg in Cluster 2. Use of PH‐specific medications was more common in Cluster 2 (44.5%), followed by Cluster 3 (38.3%) and Cluster 1 (30.5%).

The new PH clusters spanned multiple WSPH groups and hemodynamic profiles (Figure 4; Table S2). 52.2% of WSPH 2 patients and 51% of those with combined pre‐ and postcapillary PH were in Cluster 3. Most patients with WSPH 4 were in Clusters 1 and 2 (42% and 43.7%, respectively). Only 2.5% of patients with isolated postcapillary PH were in Cluster 2. Cluster 2 was predominantly WSPH 1 (69.1%). Clusters 1 and 3 had a similar proportion of WSPH 3 (~15%); however, Cluster 3 had more WSPH 2 (43.7%) and less WSPH 1 (29.1%).

Figure 4. Distribution of WSPH groups within the new clusters in the CC‐PH and PVDOMICS cohorts.

The new pulmonary hypertension clusters include a mix of WSPH groups with Cluster 2 enriched in WSPH 1 patients and Cluster 3 in WSPH 2 patients. CC‐PH indicates Cleveland Clinic Pulmonary Hypertension registry; PVDOMICS, Pulmonary Vascular Diseases Phenomics; and WSPH, World Symposium on Pulmonary Hypertension.

Validating New PH Clusters in PVDOMICS

Based on the classification model derived from the CC‐PH cohort, 503 PVDOMICS participants (59%) were assigned to the new PH Cluster 1, 179 (21%) to Cluster 2 and 171 (20%) to Cluster 3 (Table 3). The higher proportion of PVDOMICS patients in Cluster 1 is consistent with the cohort's less severe disease compared with the CC‐PH. The predicted clusters in PVDOMICS showed clear and statistically significant separation in transplantation‐free survival (Figure 3B, P<0.001), with a median transplantation‐free survival of 3.6 years in Cluster 3. Transplantation rates were 3.5% in Cluster 3, compared with 9.1% in Cluster 1 and 11.2% in Cluster 2. compared with Cluster 1, the hazard ratio for transplantation‐free survival was 1.8 (95% CI, 1.4–2.4) for Cluster 2 and 2.9 (95% CI, 2.3–3.8) for Cluster 3. PVR, FEV1, and BNP were the strongest predictors for the new clusters. Considering the cluster membership predicted from the full model as the reference standard, the prediction accuracy of a reduced model with PVR, FEV1, and BNP was 79.5%.

The clinical characteristics of the predicted PH clusters in PVDOMICS showed similar patterns to those observed in the CC‐PH clusters (Table 3). Participants in Cluster 3 were older (mean age, 67.2 years) with the highest proportion of female participants (71.3%). 6MWD decreased across the clusters, from 365.5 m in Cluster 1 to 337.1 m in Cluster 2 and 238.7 m in Cluster 3. Lung function measures were lowest in Cluster 3.

Cluster 2 had the worst hemodynamics with Cluster 3 showing intermediate values but the highest PCWP. Echocardiographic findings paralleled the hemodynamic trends. Right ventricular systolic pressure was highest in Cluster 2 and lowest in Cluster 1 (P<0.001; Figure S3A). Both right ventricular fraction area change and tricuspid annular plane systolic excursion were highest in Cluster 1 and lowest in Cluster 2 (P<0.001; Figure S3B and S3C). The right ventricular–pulmonary arterial coupling, measured by tricuspid annular plane systolic excursion/pulmonary arterial systolic pressure was lowest in Cluster 2 (Table 3). As observed in the CC‐PH cohort, the newly identified clusters in PVDOMICS included participants across multiple hemodynamic profiles and WSPH groups (Figure 4; Table S2).

NO‐Arginine Metabolism and D‐Dimer Levels Characterize the New PH Phenotypes

NO‐arginine metabolites were available only in the PVDOMICS cohort. Arginine levels were highest in Cluster 1 (106.9 μM), intermediate in Cluster 2 (94.4 μM) and lowest in Cluster 3 (92.1 μM) (P<0.001) (Table 4). Ornithine and citrulline levels were highest in Cluster 3 followed by Cluster 2 and the lowest levels in Cluster 1 (P<0.001). GABR was reduced in Cluster 3 (0.71) compared with Cluster 1 (0.95) and Cluster 2 (0.79) (P<0.001) (Figure 3C). To mechanistically delineate the alterations in NO‐arginine metabolism and indirectly estimate arginase and NO synthase activity, we calculated the ornithine‐to‐arginine and citrulline‐to‐arginine ratios. Cluster 3 demonstrated higher ornithine‐to‐arginine and citrulline‐to‐arginine ratios compared with Clusters 1 and 2 (both P<0.001) consistent with reduced arginine bioavailability due to increased enzymatic conversion (Table 4). Urinary nitrate‐to‐creatinine ratio was lowest in Cluster 3 (44.38 versus 58.56 and 57.46 nmol/μmol in Clusters 1 and 2 respectively) though not statistically significant (P=0.06). We used nontargeted metabolomic data to determine differences in asymmetric dimethylarginine (ADMA), an inhibitor of NO synthetase, across clusters. ADMA was quantified as the combined signal of ADMA and symmetric dimethylarginine. ADMA+ symmetric dimethylarginine levels were significantly increased in Cluster 3 (0.38) compared with Clusters 1 (0.05) and 2 (0.23) (P<0.001). We also examined arginine, ornithine, and citrulline levels. Consistent with the targeted metabolite measurements, Cluster 3 had higher ornithine‐to‐arginine and citrulline‐to‐arginine ratios, along with lower GABR (all P<0.001). D‐dimer levels were significantly higher in Cluster 3 (990 ng/mL) compared with Cluster 2 (635.3 ng/mL) and Cluster 1 (520.6 ng/mL) (P<0.001) (Figure 3C).

Table 4.

Distribution of Arginine‐Nitric Oxide Pathway Metabolites and D‐Dimer Levels by New PH Clusters in PVDOMICS Cohort

	Overall (N=853)	Cluster 1 (N=503)	Cluster 2 (N=179)	Cluster 3 (N=171)	P value
Arginine, μM	101.3±30.6	106.9±30.3	94.4±29.8	92.1±28.5	<0.001
Ornithine, μM	81.5±28.5	77.6±25.6	84.5±30.8	90.1±32.2	<0.001
Citrulline, μM	42.4±15.8	40.3±13.7	41.9±15.9	49.3±19.3	<0.001
Ornithine/arginine	0.77 [0.60–1.01]	0.70 [0.56–0.89]	0.86 [0.66–1.19]	0.94 [0.69–1.30]	<0.001
Citrulline/arginine	0.40 [0.32–0.52]	0.36 [0.30–0.46]	0.42 [0.34–0.55]	0.53 [0.40–0.70]	<0.001
GABR	0.87±0.30	0.95±0.29	0.79±0.27	0.71±0.27	<0.001
D‐dimer, ng/ml	636.7±870.6	520.6±501.3	635.3±757.3	990.0±1530.2	<0.001
Urinary nitrate/creatinine, nmol/μmol	56.27 [36.01–88.23]	58.56 [39.33–89.23]	57.46 [34.37–89.15]	44.38 [31.66–84.09]	0.06

Mean±SD or median [interquartile range] presented, GABR=arginine/(ornithine + citrulline). GABR indicates global arginine bioavailability ratio; PH, pulmonary hypertension; and PVDOMICS, Pulmonary Vascular Diseases Phenomics.

Placing subjects into 1 of the 3 new PH clusters could be efficiently determined using GABR and D‐dimer (Table S3). Elevated GABR was associated with lower odds of being in Cluster 2 or 3. Compared with the first quartile of GABR, odds ratio decreases progressively across quartiles. For quartile 2, the odds ratio was 0.51 (95% CI, 0.31–0.84) and 0.44 (95% CI, 0.27–0.72) for Clusters 2 and 3 respectively. The odds ratios were 0.33 and 0.14, respectively, for GABR quartile 3, and were 0.23 and 0.15 for GABR quartile 4. Higher D‐dimer was associated with greater probabilities of being in Cluster 3. Compared with the first quartile, the odds ratios were 2.69 (95% CI, 1.26–5.33), 3.65 (95% CI, 1.90–7.00) and 5.01 (95% CI, 2.63–9.54), respectively, across quartiles 2 to 4. The area under the micro‐average receiver operating characteristic curve of the multinomial model was 0.71 (Figure S4A). The transplantation‐free survival curves by PH clusters predicted using GABR and D‐dimer exhibited clear separations (Figure S4B, P<0.001). The hazard ratio for a 0.1 increase in GABR was 0.90 (95% CI, 0.86–0.94, P<0.001); and the hazard ratio for doubling of D‐dimer was 1.23 (95% CI, 1.12–1.35, P<0.001).The clinical characteristics of the predicted clusters showed similar trends to the clusters identified using the clinical variables (Table S4).

DISCUSSION

This study identifies clinical PH phenotypes associated with survival and linked to disease‐specific biological mechanisms (endotypes), enabling a more precise framework for patient classification that could inform targeted therapies across a wide range of patients with PH. Vascular health emerged as a strong predictor of survival across all WSPH groups. Using the CC‐PH registry, we derived 3 distinct clinical clusters with differential transplantation‐free survival, which were validated in the multicenter PVDOMICS cohort. These clusters are strongly associated with endothelial health, and, to our knowledge, this is the first study to identify clinical clusters across all PH groups and directly link these phenotypes to underlying biological mechanisms. ²⁷ , ²⁸ , ²⁹ , ³⁰ The high‐risk phenotype is characterized by loss of pulmonary microcirculation, supported by severely reduced DLCO, low GABR and urinary nitrates, and elevated D‐dimer levels.

Although several hemodynamic variables that overlap with traditional classification systems informed the new clusters, the approach was data driven and accounted for interactions among multiple variables generating multidimensional phenotypes rather than categories defined solely by individual hemodynamic thresholds. Among the newly identified clusters, Cluster 1 had the best survival, less severe hemodynamics, higher 6MWD, lower BNP, and better functional class compared with the other 2 clusters. Cluster 1 also included a higher proportion of WSPH 1 patients than Cluster 3 but a greater proportion of WSPH 2 and 3 patients than Cluster 2. Patients in Cluster 3 were older with a higher proportion of women compared with the other 2 clusters. They had a higher burden of cardiac and pulmonary comorbidities with higher PCWP, lower pulmonary function tests, lower hemoglobin levels, and worse renal function. They had the lowest DLCO, lowest walk distance, and highest BNP; however, their hemodynamics except for higher PCWP were better than Cluster 2 with lower PVR, mPAP, and higher CI. Cluster 3 had more WSPH 2 patients (43.7%), with 47.8% of patients having combined pre‐ and postcapillary PH and 30.9% precapillary PAH. Despite having less severe hemodynamics than Cluster 2, Cluster 3 had the worst survival. The findings suggest that patients in Cluster 3 had marked pulmonary vascular loss compared with the other 2 clusters. Most studies have shown that cardiac output or cardiac index and right atrial pressure at baseline predict prognosis in PH, whereas other variables such as mPAP and PVR have not shown consistent associations with survival. ³¹ , ³² , ³³ , ³⁴ Weatherald et al. showed that no baseline hemodynamic variables independently predict prognosis in a multivariate model. ³⁵ In another study, incremental modeling analyses using the c‐index demonstrated that the prediction of mortality in patients with PAH was incrementally improved by incorporating clinical and noninvasive data with minimal increase when hemodynamic measurements are added to the model. ³¹ Taken together, Cluster 3 represents a PH phenotype characterized by pulmonary vascular rarefication that leads to a greater risk of death.

Recent evidence has emerged supporting a lung phenotype in Group 1 PAH that is characterized by a markedly reduced DLCO and associated with worse survival. ² Patients with this phenotype are older, have a smoking history, and behave like those in Group 3 PH due to lung disease rather than classical PAH. ²⁷ Pathologically, this subgroup does not exhibit the pathological plexiform lesions of PAH but has more arteriolar and venular microvascular and capillary pathology. ³⁶ Our study aligns with and builds up on these findings by including all WSPH groups and linking clinical phenotypes to endothelial health. The 3 identified phenotypes show significant differences in NO‐arginine metabolism and D‐dimer levels, supporting the notion that we have identified phenotypes that correlate to underlying endothelial biology. Patients in Cluster 3 with reduced DLCO had the lowest GABR, lowest urinary nitrates, and highest D‐dimer levels, consistent with an endotype associated with NO deficiency, impaired vasodilation, a prothrombotic state, and compromised gas exchange. Interestingly, Cluster 3 had lower PVR and higher CI compared with Cluster 2 indicating that endothelial loss and alterations in NO‐arginine metabolism drive mortality outcomes in PH independent of hemodynamics. Furthermore, using only GABR and D‐dimer levels, we can stratify patients into endotypes that strongly predict mortality, highlighting the concept that preserving endothelial health is crucial for survival in PH. These findings align with prior mechanistic studies implicating NO deficiency and dysregulated arginine metabolism in disease pathogenesis, supporting the biological relevance of these markers beyond reflecting disease severity and suggesting that the identified clusters represent mechanistically informative phenotypes.

Importantly, both targeted and nontargeted metabolomic analyses of arginine metabolites demonstrated reduced functional arginine bioavailability in Cluster 3. Elevated ADMA levels in this cluster indicate NO synthetase inhibition, whereas higher ornithine‐to‐arginine ratios suggest enhanced arginase conversion of arginine away from NO synthesis. The concomitant increase in citrulline‐to‐arginine ratio further supports the heightened arginine metabolic flux. Together with reduced GABR, these findings suggest impaired NO availability in Cluster 3 driven by both increased metabolic demand and competitive NO synthase inhibition.

This study has several limitations. The 2 cohorts were recruited over different time periods, during which the hemodynamic definition of pulmonary hypertension evolved, and new PH therapies, including upfront combination therapy were introduced, potentially resulting in time‐lag bias. Additionally, there were significant differences in disease severity between the cohorts. Despite this, the newly identified clusters in the CC‐PH registry were successfully validated in PVDOMICS, and their clinical characteristics were largely consistent supporting the robustness of our findings and suggesting minimal influence from temporal factors. Markers of endothelial health were associated with cluster membership in both cohorts despite differing therapeutic strategies. The results underscore the limitations of current treatment approaches and emphasize the need for precision medicine. Given that new treatments became available over the course of data collection, the findings reflect real‐world clinical settings where practice evolves over time. Another limitation is that the clustering analysis relied on imputed missing data, with 24.5% of the clustering variable data missing and imputed in CC‐PH registry and 15.8% missing in PVDOMICS. In addition, the analyses did not account for comorbidities or the use of pulmonary vasodilators, both of which may influence disease progression and treatment response. Finally, it is important to acknowledge that different approaches can be explored to reclassify patients with pulmonary vascular diseases, especially in a deeply phenotyped cohort like PVDOMICS, with ongoing efforts investigating clinical clustering and hemodynamic‐based reclassification, particularly in areas of PH group overlap. ³⁷ These strategies are complementary and will enhance our understanding of patient heterogeneity beyond current classification. These complementary strategies enhance our understanding of patient heterogeneity and, although each may serve distinct clinical roles, they may ultimately be integrated into a unified reclassification framework.

CONCLUSIONS

In summary, we identify PH phenotypes associated with mortality, first in a discovery cohort and then validate them in an independent cohort. These phenotypes span multiple hemodynamic profiles, WSPH groups, and comorbidities. Importantly, they are linked to endothelial health with GABR and D‐dimer emerging as strong predictors of the newly identified phenotypes. The high‐risk phenotype is characterized by loss of endothelial vasodilation, in situ thrombosis and pulmonary vascular rarefication leading to markedly reduced gas exchange.

Sources of Funding

The study was supported by grants U01 HL125218 (principal investigator [PI]: E. B. Rosenzweig), U01 HL125205 (PI: R. P. Frantz), U01 HL125212 (PI: A. R. Hemnes), U01 HL125208 (PI: F. P. Rischard), U01 HL125175 (PI: P. M.Hassoun), U01 HL125215 (PI: J. A. Leopold), U01 HL125177 (PI: G. J. Beck), RO1 HL060917 (PI: S. Erzurum) and the Pulmonary Hypertension Association.

Disclosures

Dr Heresi participates in Bayer and Merck Speakers Bureau; nonpromotional, nonbranded, advisory boards for Merck and United Therapeutics; has Bayer research funds for investigator‐initiated study; and is on the steering committee for Johnson and Johnson. Dr Tonelli participated in advisory boards of United Therapeutics, Merck, and Janssen. Dr Tang has served as consultant for Cardiol Therapeutics, CardiaTec Biosciences, Astra Zeneca, Alleviant Medical, Salubris Biotherapeutics, BioCardia, BridgeBio Pharma, Tenax Therapeutics, and Vasa Therapeutics and has received honoraria from Springer Nature and Belvoir Media Group. Dr Hemnes serves as a consultant for Merk, UT, Janssen, GossamerBio, and Johnson and Johnson;she holds stock in Tenax Therapeutics and has received grants from the National Institutes of Health. Dr Horn has served on the Data and Safety Monitoring Board of Trisol and the steering committee for Pulnovodmed; and has served as a consultant for Biotronik and 35Pharma. Dr Leopold receives salary support from the Massachusetts Medical Society for her role as Deputy Editor, NEJM. Dr Frantz has consulting, steering committee, and advisory board relationships with Gossamer Bio, Liquidia, Merck, Insmed, Inhibikase, and UpToDate. Dr Hassoun reports participation on a data safety monitoring board or advisory board with ARIA SAB, serves on a scientific steering committee for Merck, and is an Associate Editor of the European Respiratory Journal. Dr Rischard has consulting relationships with Acceleron and United Therapeutics; is on a Steering Committee for Acceleron; and receives research support from Ismed, United Therapeutics, Bayer, and Tenax. Dr Hill has received research grants to Tufts Medical Center from Gossamer, Merck, Insmed, and United Therapeutics and has served on advisory boards for Gossamer, Liquida, Merck, and Insmed. Dr Rosenzweig has relationships with Janssen Pharmaceuticals. Dr DuBrock has consulting relationships with Merck, Janssen Global Services, and United Therapeutics and has Beyer research funding. The remaining authors have no disclosures to report.

Supporting information

Tables S1–S4

Figures S1–S4

PVDOMICS Group Investigators

STROBE Statement