Large Language Modelsfor Injury Surveillance: Evidence from E-Bike Incidents
Organizations: Civil and Environmental Engineering, University of Maryland · Computer and Information Science, University of Pennsylvania
Abstract
This study analyzes e-bike-related injury incidents using two complementary data sources: the Consumer Product Safety Risk Management System (CPSRMS) and the National Electronic Injury Surveillance System (NEISS). We develop an LLM-assisted framework that converts unstructured incident narratives into structured variables, notably human-, equipment-, and environment-related contributing factors, for use alongside demographic, locational, temporal, and weather-derived predictors. To improve cross-source comparability, injury severity is harmonized into three ordered categories. Separate ordered logit models are estimated for CPSRMS and NEISS, together with two supplementary joint scaled specifications. In CPSRMS, human-related factors are associated with higher injury severity, whereas equipment- and environment-related factors, riders under age 18, spring incidents, and recreation/sport locations are associated with lower severity categories. In NEISS, riders under age 18 and female riders are associated with lower treatment severity, whereas riders above age 59 are associated with higher severity. Likelihood-ratio tests support selected coefficient-sharing restrictions but reject equality of the source thresholds. The findings demonstrate that narrative-rich incident reports and standardized emergency-department surveillance data capture complementary rather than directly interchangeable dimensions of e-bike safety. The proposed framework illustrates how LLM-derived variables can be integrated with econometric models to support injury surveillance, safety intervention design, and evidence-based micromobility policy.
Figures & tables
| Mechanical/Visibility Component | Keyword(s) |
|---|---|
| Brake System | Brake, Pads, Cable |
| Steering System | Handlebar, Stem, Cockpit |
| Pedals | Pedal |
| Drive System | Battery, battery pack, charger/charging system, motor, controller, chain, belt, sprocket, gear, derailleur, drivetrain |
| Wheel/Tire | Wheel or Tire, Spokes |
| Front Fork | Front Fork |
| Variable | Definition and Coding | Reference Category | Information Source |
|---|---|---|---|
| Outcome | |||
| Injury severity | Harmonized ordered outcome (1–3), with higher values indicating greater severity. Original CPSRMS and NEISS classifications are mapped to three levels. | – | Source-specific structured severity/treatment fields |
| Demographic variables | |||
| Age under 18 | 1 if rider age is below 18; 0 otherwise. | Age 25–59 | Structured age field; narrative-assisted completion when needed |
| Age 18–24 | 1 if rider age is 18–24; 0 otherwise. | Age 25–59 | Structured age field; narrative-assisted completion when needed |
| Age 45–59 | 1 if rider age is 45–59; 0 otherwise. Considered but not retained in the final models. | Age 25–44 (candidate coding) | Structured age field; narrative-assisted completion when needed |
| CPSRMS ( ) | NEISS ( ) | |||
|---|---|---|---|---|
| Variable | Missing | Missing (%) | Missing | Missing (%) |
| Injury severity | 0 | 0.0 | 0 | 0.0 |
| Age indicators | 314 | 44.2 | 0 | 0.0 |
| Female | 257 | 36.2 | 2 | 0.1 |
| Wet road | 203 | 28.6 | 584 | 24.6 |
| Street/highway location | 230 | 32.4 | 814 | 34.3 |
| Task | Example Prompt | Expected Output |
|---|---|---|
| Transportation Mode | Please retrieve the transportation mode in the incident: “A 66-year-old male riding an electric bicycle eastbound on a road was seriously injured after colliding with a southbound vehicle at the intersection.” | electric bicycle |
| Demographic Information | Please retrieve the demographic information in the incident: “42YOF PT RIDING AN E BIKE, LOST CONTROL & CRASHED BIKE. DX: HEAD CONTUSION, CHI.” | Female Age 42 |
| Contributing Factors | Determine whether human-related, equipment-related, and environment-related factors contributed to the incident: “9 yom riding an electric bike down a hill, lost control and hit a car. dx orbital fx.” | Human = Yes Equipment = No Environment = No |
| Incident–Component Link | Identify the e-bike component related to the incident: “PEDAL BROKE” | Pedals = Yes |
| Predicted Positive | Predicted Negative | |
|---|---|---|
| Actual Positive | True Positive (TP) | False Negative (FN) |
| Actual Negative | False Positive (FP) | True Negative (TN) |
| Model 1 | Model 2 | Model 3 | Model 4 | |||||
| CPSRMS only | NEISS only | Joint: separate + relative scale | Joint: source allocated/ partially unified + relative scale | |||||
| Model information | Coeff. | t-stat | Coeff. | t-stat | Coeff. | t-stat | Coeff. | t-stat |
| Female | -0.403** | -2.05 | N: -0.404** | N: -2.06 | N: -0.402** | N: -2.05 | ||
| Rural | 0.325 | 0.85 | N: 0.326 | N: 0.85 | N: 0.330 | N: 0.86 | ||
| Spring | -0.978** | -2.13 | C: -0.516 | C: -1.49 | C: -0.276 | C: -1.34 | ||
| Weekend | 0.552 | 1.23 | C: 0.294 | C: 1.18 | C: 0.161 | C: 1.03 | ||