Micromobility-Related Injuries: Novel Methodology for Surveillance and Granular Data Extraction

Katia C. Genadry, MD
Amir A. Kimia, MD
James M. Chamberlain, MD
Alek Jorge, BA
Jennifer E. Welcher, MBA
Assaf Landschaft, MS
Kavya A. John, BS
Nina D. Kosciuszek, DO, MS
Fahd A. Ahmad, MD, MSCI
Lois K. Lee, MD, MPH

Micromobility device injuries, including e-bikes and e-scooters, are increasing. Injury surveillance commonly relies on administrative data (e.g. ICD-10 codes). However, their accuracy for identifying micromobility injuries is unknown, and relevant clinical details are often lacking. The objectives of this study were to: (1) analyze micromobility injury case identification when ICD-10 codes are augmented with natural language processing (NLP) of emergency department (ED) clinical narratives, and (2) examine detailed characteristics of these injuries and trends over time.

We conducted a multicenter cross-sectional study of patients 0-18 years old presenting with e-bike and e-scooter injuries to four tertiary-care pediatric EDs. Two centers contributed data from January 2022–December 2025, while the other two contributed data from January–December 2025. ED narratives were annotated using the NLP platform, Document Review Tools (DrT), which uses regular expressions to highlight relevant specific text. Two mutually exclusive approaches were used to identify micromobility-related injuries: (1) NLP analysis of ED electronic health record (EHR) narratives, and (2) ICD-10 codes. We a priori established a sensitivity ?95% for the models. The final NLP models were applied across all 4 site EHR narratives and aggregated results were analyzed. Data from all 4 centers was used to analyze characteristics of the injuries. Data from 2 centers covering 4 years were used to evaluate trends. We performed linear regression to analyze temporal trends in injury rates over time.

There were a total of 719 micromobility-related injuries identified using DrT from the study sample including all 4 sites: 303 (42.1%) e-bikes and 416 (57.8%) e-scooters. The performance metrics for the NLP models were: sensitivity 97.1% (95% CI 92-100%), specificity 84% (95% CI 79-89%), accuracy 91%, and F1-score 0.86. ICD-10 codes identified only 198 cases (27%) of all confirmed micromobility injuries. The median age of patients was 13 years [IQR 10.5-15.0 years]. Overall 29% of injures were in females, with differences by device: 18% females for e-bikes and 36% females for e-scooter. For e-bikes the leading mechanisms of injuries were: falls (32%), motor vehicle collisions (MVCs) (20%), and collisions with stationary object (13%). For e-scooters the leading mechanisms of injuries were: falls (48%), MVCs (20%), and due to road bump/pothole (9%). Helmet use was only documented as being used for 22% of e-bike injuries and 10% of e-scooter injuries. Time trends demonstrated statistically significant increases in micromobility injuries from 2022-2025 (p<0.01) (Figure 1A), as well as the proportion of micromobility injuries from MVCs (p<0.01) (Figure 1B).

Micromobility injuries from e-bikes and e-scooters are increasing, including from collisions with motor vehicles. Reliance on ICD-10 codes alone underestimates the incidence and limits surveillance. NLP of ED narratives improves injury identification and captures granular clinical details unavailable in administrative coding, supporting more effective surveillance and injury prevention research.

1- Describe the increasing burden of micromobility injuries (e-bikes, e-scooters) including injuries from motor vehicle collisions.
2- Recognize the limitations of relying solely on ICD-10 coding for identifying and surveilling micromobility-related injuries.
3-Explain how natural language processing of emergency department narratives can improve injury identification and provide detailed clinical information to support surveillance and injury prevention research.