Characterization of injuries related to bicycle crashes in urban environments

To explore whether an uncertainty-aware fuzzy clustering framework can organize demographic, temporal, collision-related, and clinical patterns among bicycle-related emergency department cases.

The 698 cases included in the study were individual patients treated in the emergency department in Istanbul, Türkiye, for bicycle-related injuries. Exploratory profiles were generated using a PSO-guided, FCM-fuzzy membership architecture with Pearson correlation distance. A Random Forest surrogate and SHAP were used only for post-hoc description of maximum-membership label recoverability and conditional attribution within the same feature space.

The analysis produced four overlapping, model-derived case profiles. Cluster 1 (n = 215) included evening/night cases of adolescents (mean age 14) and had the highest proportion of admission to the ICU (21/215, 9.8%; 95% CI: 6.5-14.5). Cluster 2 (n = 271) included school-age afternoon cases (mean age 11) and had a 96.3% proportion of immediate discharge (261/271; 95% CI: 93.3-98.0). Cluster 3 (n = 54) included morning school-age cases (mean age 11.5) with gender parity and low observed severity. Cluster 4 (n = 158) included young adult cases (mean age 24) and had the highest fracture (41/158, 25.9%; 95% CI 19.7-33.3) and non-immediate-discharge proportions (51/158, 32.3%; 95% CI: 25.5-39.9).

The findings show that bicycle-related emergency department cases are heterogeneous, reflecting distinct demographic, temporal, collision-related, and clinical patterns. The PSO-Fuzzy-Pearson framework identifies interpretable injury-case phenotypes that may support post-crash triage, injury surveillance, and emergency resource prioritization. Patterns also provide useful safety information for targeted prevention, including visibility, traffic calming, and protected cycling measures, while requiring future validation with exposure, infrastructure, route choice, and police crash data.