AI for reducing claims and compensation payments from missed fractures on radiographs.
Authors
Affiliations (14)
Affiliations (14)
- Department of Radiology, Hotel Dieu Hospital/Cochin Hospital, APHP, Paris, France. [email protected].
- Department of Radiology, Raymond Poincare Hospital, APHP, Garches, France. [email protected].
- Biomedical Engineering and Imaging Institute, Icahn School of Medicine at Mount Sinai, New York, NY, USA. [email protected].
- Department of Radiology, Hotel Dieu Hospital/Cochin Hospital, APHP, Paris, France.
- Department of Radiology, Ambroise Pare Hospital, APHP, Paris, France.
- Gleamer, Paris, France.
- Biomedical Engineering and Imaging Institute, Icahn School of Medicine at Mount Sinai, New York, NY, USA.
- Réseau Imagerie Sud Francilien, Lieusaint, France.
- Ramsay Santé, Clinique du Mousseau, Evry, France.
- Department of Radiology, Raymond Poincare Hospital, APHP, Garches, France.
- Department of Traumatology and Orthopaedic Surgery, APHP, Raymond Poincaré Hospital, Garches, France.
- Boston University School of Medicine, Boston, MA, USA.
- Service de Radiologie, Institut Mutualiste Montsouris, Paris, France.
- Université Paris-Cité, Paris, France.
Abstract
To evaluate the potential of BoneView, an artificial intelligence (AI) tool for fracture detection on radiographs in claims cases related to missed fractures that resulted in financial compensation. This retrospective, multicenter study included all patients who submitted a claim and received financial compensation for missed fractures on radiographs between January 2013 and December 2019 across the 38 university hospitals of the Greater Paris area (AP-HP-France). For each claimant, 5 control patients were selected from a contemporary single-center cohort of patients who underwent radiographs of the same anatomical regions (with or without fracture) performed in the emergency department of a university hospital, irrespective of fracture status. Three radiologists (musculoskeletal fellow, 5-year fellowship-trained radiologist, and > 20-year expert) independently performed blinded, unassisted, and AI-assisted interpretations following a 3-month washout. Sensitivity, specificity, and accuracy were calculated with and without AI and for standalone AI. Compensation amounts were used to estimate potentially avoidable payouts. 156 patients (mean age, 42.2 years; 113 (72.4%) males), including 26 claimants with missed fractures on radiographs and 130 controls, were evaluated. 27/130 (20.8%) of the control cohort had fractures. Standalone AI sensitivity was 80.8% (21/26; 95% CI [60.7-93.5]) for claimants and 96.3% (26/27; 95% CI [81.0-99.9]) for the control cohort (p = 0.10). AI assistance did not significantly improve sensitivity, with expert reader sensitivity increasing from 81.1% to 92.5% (p = 0.10). Total compensation was €282,850 (€10,878 per case). AI-assisted detection could have potentially mitigated €265,314, representing a 93.8% theoretical reduction in institutional financial liability. AI-assisted fracture detection identified most missed fractures on radiographs leading to compensation and may help reduce medicolegal risk and compensation-related financial burden. Question Can AI-assisted fracture detection identify fractures missed by radiologists on radiographs that led to compensated medicolegal claims and institutional financial liability? Findings Standalone AI detected 21/26 (80.8%) of missed fractures in patients with compensated claims, corresponding to €265,314/€282,850 (93.8%) of total compensation in this cohort. Clinical relevance AI-assisted fracture detection can potentially increase radiologists' diagnostic accuracy and reduce medicolegal risk and compensation-related financial burden. Implementing AI tools in emergency settings may significantly mitigate patient harm and institutional costs associated with missed fractures.