Hierarchal clustering for the identification of distinct rib fracture patterns

Background: Treatment and prognosis of rib injury is suspected to be dependent on patterning of rib fractures. Pattern data is limited, restricted to number of individual ribs fractured, segmentality, or overall fracture number. Analyzing fracture patterns may reveal other associations, but a method for analysis is needed that maintains geospatial fidelity. We hypothesize that clustering methodology can detect clinically important fracture patterns between patients seen following motor vehicle collisions (MVCs) and motorcycle collisions (MCCs), and patients who do and do not require tracheostomy. Methods: A prospectively kept anatomic database was reviewed for patients with multiple rib fractures. Fracture coordinate locations were described by rib number and anterior to posterior locations. Fractures incidence was mapped to a corresponding 5×12 matrix and organized by paired conditions with non-overlapping cohorts. Hierarchical clustering was optimized and utilized to detect associations between locations of rib fractures. Rib fracture location and clustering for cohorts were compared. Results were confirmed with traditional regression analysis. Results: Clustering outcomes were displayed as hierarchical dendrograms demonstrating the most highly associated rib fractures for each condition. Compared to MVCs, MCCs produced fractures that clustered more significantly posteriorly and on the left versus laterally [Baker's gamma index (BGI) P<0.01 (right) and P=0.03 (left)]. Compared to patients without tracheostomy, patients who underwent tracheostomy had high density of fractures clustering around very superior and inferior ribs with the greatest density at rib 3. This was confirmed on regression analysis. Conclusions: Using high-level rib fracture clustering, motorcycle injuries appear to display a pattern of more posteriorly located rib fractures versus MVCs. Similarly, posterior-lateral fracture patterns centered around rib 3 appear to associate with need for tracheostomy. Further work is needed for improved contextualization. Hierarchical clustering appears useful in hypothesis generation and evaluating large rib fracture data sets to for relevant patterns.