Clustering and Subgroup Analysis of Traumatic Subdural Hemorrhage Patients Using Machine Learning

Introduction: Traumatic subdural hemorrhage is a severe brain injury that requires prompt treatment to prevent significant morbidity and mortality, primarily affecting the elderly. This study employs machine learning-based clustering algorithms to identify unique clinical profiles in patients with traumatic subdural hemorrhage (SDH), providing new insights into the stratification of risk factors and patient outcomes.

Methods: The 2015-2019 National Inpatient Sample was queried using ICD-10 CM/PCS codes identifying patients with traumatic SDH. Machine learning clustering analysis evaluated the population based on 49 comorbidities, complications and clinical covariates. Optimal number of clusters was determined using Davies-Bouldin Index (DBI) and Calinski-Harabasz Index (CHI). Between-cluster multivariate logistic regression analysis was performed to assess risk of mortality and non-routine discharge. Kruskal-Wallis H-Testing was performed to assess variance in length-of-stay between clusters. Statistical analysis was performed using Python.

Results: A total of 35,011 patients were included in this study. DBI-CHI composite determined the optimal number of clusters at 5, with sizes ranging from 651 to 27573 patients. Mortality ranged from 6.60% in Cluster 1 to 33.49% in Cluster 5. Cluster 5 had the highest prevalence of anemia, coagulopathy, sepsis, and pneumonia. Cluster 5 was also associated with a seven-fold increased risk of mortality relative to Cluster 1 [OR= 7.01, p< 0.001]. Kruskall-Wallis H-testing and post-hoc pairwise testing of length-of-stay distributions showed significant (p < 0.001) differences between all clusters, with the greatest difference in medians occurring when comparing clusters 1 and 4.

Conclusion : Clustering analysis of patients with traumatic SDH identified five distinct groups with unique comorbidity profiles. The cluster with the highest prevalence of anemia, coagulopathy, sepsis and pneumonia was also associated with the highest mortality rate. This clustering machine learning approach enables a nuanced perspective of previously imperceptible comorbidity interactions, providing a basis for more personalized clinical decision-making and treatment interventions.