Machine learning analysis of 5,059 patients with pulmonary arterial hypertension identified six distinct treatment clusters, with 82.9% of patients initiating nitric oxide pathway monotherapy.
Cohort (n=5,059)
Yes
Machine learning-based analysis of real-world claims data identified six distinct PAH treatment sequence clusters, highlighting heterogeneity in clinical practice and potential under-treatment compared to guideline recommendations.
Abstract Rationale Pulmonary arterial hypertension (PAH) is a progressive vascular disease leading to right heart failure. Although clinical guidelines recommend combination therapy through a defined algorithm, real-world practice can be heterogeneous and diverge from recommendations. Leveraging real-world data, this study aims to identify treatment sequence patterns in PAH in the United States (US) and describe patient profiles across treatment clusters. Methods This retrospective cohort study utilized claims data from Komodo Research Data, covering 130 million patient lives from January 1, 2016, to November 30, 2024. Eligible adults with newly diagnosed PAH were identified based on diagnostic criteria and claims for PAH-related medications. The index date was defined as the first claim for PAH-targeted therapy after diagnosis. The baseline period spanned 6 months before the index date, and follow-up extended until loss of eligibility, data cutoff, or death. Unsupervised hierarchical clustering identified patient clusters with similar PAH treatment sequence patterns. The number of clusters was selected based on statistical robustness and parsimony. Baseline characteristics and healthcare resource use were summarized and compared across clusters. Results A total of 5,059 patients met the eligibility criteria. The mean age at treatment initiation was 60.5 years, 54.6% were female, and the mean Modified Quan-Charlson Comorbidity Index (CCI) was 2.7. The median follow-up duration was 1.9 years. Overall, most patients initiated therapy with monotherapy (82.9% nitric oxide NO pathway, 9.3% endothelin receptor antagonist ERA). Unsupervised clustering identified six clusters, or similar treatment patterns (Figure 1). Cluster 1 patients had the shortest follow-up duration, while clusters 4 and 5 were characterized by early treatment discontinuation and early mortality, respectively. In cluster 2, 80.9% escalated to NO + ERA by month 6, while most (≥92%) patients in clusters 3, 4, and 6 remained on NO or ERA monotherapy at month 24. At baseline, cluster 2 patients were younger (mean 54.7 years), more often female (71.2%), and more frequently underwent right-heart catheterization (53.7%). Cluster 4 patients were more often male (60.2%). Cluster 6 patients had the lowest comorbidity burden (mean CCI 1.7) and hospitalization rate (14.3%), while clusters 1 and 5 had the highest hospitalization rates (45.5% and 45.3%, respectively). Conclusions Machine learning-based sequence analysis revealed six real-world treatment clusters with different escalation trajectories, each associated with distinct clinical and demographic profiles. These findings enhance understanding of current practice, and generate actionable insights for medical community to identify opportunities for optimized treatment (e.g., patient profiles with insufficient therapy). This abstract is funded by: Merck & Co., Inc
Wang et al. (Fri,) conducted a cohort in Pulmonary arterial hypertension (n=5,059). PAH-targeted therapy was evaluated on Patient clusters with similar PAH treatment sequence patterns. Machine learning analysis of 5,059 patients with pulmonary arterial hypertension identified six distinct treatment clusters, with 82.9% of patients initiating nitric oxide pathway monotherapy.