Machine Learning Uncovers New Pulmonary Embolism Disease Clusters

Researchers recently used precision hierarchical clustering to identify disease subtypes among patients with intermediate- to high-risk pulmonary embolism at admission, a new study showed. “Prognostic risk assessment and stratification of patients with PE are of great significance for guiding the diagnosis and treatment strategy of PE to reduce the mortality of PE,” study researchers wrote. “However, current intermediate-high-risk PE encompasses heterogeneous subgroups with varied prognoses, leading to thrombolysis controversies.” A retrospective study was designed to apply machine learning algorithms to patients with intermediate- to high-risk pulmonary embolism to further risk-stratify. The study analyzed data from 79 patients at two clinical centers. The cluster analysis was based on 10 continuous variables, including age, sex, chronic lung disease, Pulmonary Embolism Severity Index (PESI) score, and more. Using this method, three clusters were identified: a dominant cluster that included 67 cases (cluster 2), and two additional clusters that included six cases each (clusters 1 and 3). There were statistically significant differences between the clusters for age, chronic lung disease, chronic heart disease, and diabetes. Specifically, cluster 3 (6 cases) “was characterized by advanced age, chronic heart disease, and diabetes.” Importantly, significant differences in the 1-year and 3-year PE mortality were also found between the groups. For example, the 1-year PE mortality was 100% in cluster 1 (6 cases), 97.0% in the dominant cluster, but 50% in cluster 3. Cluster 3 also had significantly longer ICU and hospital stays and higher total expenses than the other two groups. “In this study, hierarchical clustering discovered a subgroup with a poor prognosis in intermediate-high-risk PE patients at admission,” the researchers wrote. “Therefore, further research is needed to be certified for a large sample size by a prospective study and verify our results on an external dataset.”