From Clusters to Outcomes: Machine Learning-Based Phenotyping in Intermediate-High-Risk Acute Pulmonary Embolism

Area of Science:

• Cardiology
• Data Science
• Pulmonary Medicine

Background:

• Intermediate-high-risk pulmonary embolism (PE) is heterogeneous, and current guidelines may not fully capture risks.
• Data-driven phenotyping can enhance risk stratification for PE patients.
• Individualized treatment decisions, including reperfusion therapy, are crucial for managing PE.

Purpose of the Study:

• To apply unsupervised machine learning to identify distinct phenotypes within intermediate-high-risk PE.
• To compare clinical characteristics, treatment patterns, and outcomes across identified PE phenotypes.
• To assess the utility of phenotype-based stratification for guiding reperfusion therapy decisions.

Main Methods:

• Retrospective analysis of 553 guideline-defined intermediate-high-risk PE patients (2012-2025).
• Unsupervised machine learning (k-prototypes algorithm) applied to 36 variables for patient clustering.
• Comparison of clinical, imaging, treatment, and mortality outcomes between identified phenotypes.

Main Results:

• Two phenotypes were identified: RV-failure (n=360) and comorbidity-dominant (n=193).
• The comorbidity-dominant phenotype had lower in-hospital mortality (3.6% vs. 7.2%) and was independently associated with reduced early mortality.
• Both phenotypes showed RV function improvement post-reperfusion, with greater gains in the RV-failure phenotype.

Conclusions:

• Unsupervised machine learning successfully identified two clinically relevant intermediate-high-risk PE phenotypes.
• These phenotypes exhibit different early mortality risks but similar long-term outcomes.
• Phenotype-based assessment can refine risk stratification and inform individualized reperfusion strategies for acute PE.