Effect of AI-assisted radiographic fracture detection on paediatric emergency department length of stay through fracture exclusion: A before-after cohort study.

BACKGROUND: Artificial intelligence-assisted radiographic fracture detection tools (AI-RFDT) have the potential to support confident fracture exclusion by physicians in the paediatric emergency department (ED), where presentations for suspected fractures substantially outnumber confirmed fractures. This study aimed to evaluate whether implementation of AI-RFDT in a paediatric ED is associated with reduced length of stay (LOS) among patients presenting with suspected fractures, and to determine whether any observed effect is specifically mediated through the fracture-exclusion pathway.

PATIENTS AND METHODS: This retrospective before-after cohort study was conducted at a single tertiary paediatric ED. AI-RFDT (BoneView, Gleamer, Paris, France) was deployed on 1 October 2023 as a passive, second-read decision-support tool. The study population comprised 4172 paediatric patients with rule-out fracture diagnoses (fracture-excluded; F-E) during standard working hours (08:00-15:59; pre-AI n = 2514; post-AI n = 1658). Fracture-confirmed (F-C) patients served as a pre-specified internal control (n = 1155). The primary outcome was ED LOS; secondary outcomes were orthopaedic consultation rate, LOS by consultation status, anatomical subgroup LOS, and 30-day re-admission.

RESULTS: Orthopaedic consultation rates in the paediatric ED decreased from 73% to 68.3% (Δ -4.7 %age points, p = 0.002). The greatest reduction was observed in patients discharged without orthopaedic consultation (2.25→1.78 h, Δ -28 min, p < 0.001), while those with confirmed fractures demonstrated no significant change in LOS (Δ+3 min, p = 0.51). 30-day re-admission rates remained unchanged (6.9% vs 6.9%, p = 1.00). Wrist and forearm injuries demonstrated the greatest LOS reduction (Δ -24 min, p = 0.004).

CONCLUSIONS: Deployment of AI-RFDT in a paediatric ED was associated with a reduction in LOS that was concentrated among patients discharged without orthopaedic consultation. The absence of LOS change in confirmed-fracture controls supports a mechanism specific to fracture exclusion rather than generalised care acceleration.