Cluster analysis identifies novel real-world lung disease-pulmonary hypertension subphenotypes: implications for treatment response

Shelsey W Johnson; Rui-Sheng Wang; Michael R Winter; Kari R Gillmeyer; Katarina Zeder; Elizabeth S Klings; Ronald H Goldstein; Renda Soylemez Wiener; Bradley A Maron

doi:10.1183/23120541.00959-2023

Cluster analysis identifies novel real-world lung disease-pulmonary hypertension subphenotypes: implications for treatment response

ERJ Open Res. 2024 May 20;10(3):00959-2023. doi: 10.1183/23120541.00959-2023. eCollection 2024 May.

Authors

Shelsey W Johnson^{1

2

3}, Rui-Sheng Wang⁴, Michael R Winter⁵, Kari R Gillmeyer^{1

2

6}, Katarina Zeder^{7

8}, Elizabeth S Klings², Ronald H Goldstein¹, Renda Soylemez Wiener^{1

2

6}, Bradley A Maron^{1

7

8}

Affiliations

¹ VA Boston Healthcare System, Boston, MA, USA.
² The Pulmonary Center, Division of Pulmonary, Allergy, Sleep and Critical Care, and Boston University School of Medicine, Boston, MA, USA.
³ Department of Pulmonary and Critical Care Medicine, Brigham and Women's Hospital, Boston, MA, USA.
⁴ Department of Medicine, Brigham and Women's Hospital, Boston, MA, USA.
⁵ Boston University School of Public Health, Biostatistics and Epidemiology Data Analytics Center, Boston, MA, USA.
⁶ Center for Healthcare Organization and Implementation Research, VA Bedford Healthcare System, Bedford, MA, USA.
⁷ Department of Medicine, University of Maryland School of Medicine, Baltimore, MD, USA.
⁸ The University of Maryland-Institute for Health Computing, Bethesda, MD, USA.

Abstract

Background: Clinical trials repurposing pulmonary arterial hypertension (PAH) therapies to patients with lung disease- or hypoxia-pulmonary hypertension (PH) (classified as World Health Organization Group 3 PH) have failed to show a consistent benefit. However, Group 3 PH clinical heterogeneity suggests robust phenotyping may inform detection of treatment-responsive subgroups. We hypothesised that cluster analysis would identify subphenotypes with differential responses to oral PAH therapy.

Methods: Two k-means analyses were performed on a national cohort of US veterans with Group 3 PH; an inclusive model (I) of all treated patients (n=196) and a haemodynamic model (H) limited to patients with right heart catheterisations (n=112). The primary outcome was organ failure or all-cause mortality by cluster. An exploratory analysis evaluated within-cluster treatment effects.

Results: Three distinct clusters of Group 3 PH patients were identified. In the inclusive model (C1^I n=43, 21.9%; C2^I n=102, 52.0%; C3^I n=51, 26.0%), lung disease and spirometry drove cluster assignment. By contrast, in the haemodynamic model (C1^H n=44, 39.3%; C2^H n=43, 38.4%; C3^H n=25, 22.3%), right heart catheterisation data surpassed the importance of lung disease and spirometry. In the haemodynamic model, compared to C3^H, C1^H experienced the greatest hazard for respiratory failure or death (HR 6.1, 95% CI 3.2-11.8). In an exploratory analysis, cluster determined treatment response (p=0.006). Conclusions regarding within-cluster treatment responses were limited by significant differences between select variables in the treated and untreated groups.

Conclusions: Cluster analysis identifies novel real-world subphenotypes of Group 3 PH patients with distinct clinical trajectories. Future studies may consider this methodological approach to identify subgroups of heterogeneous patients that may be responsive to existing pulmonary vasodilatory therapies.