From dusk till dawn?-Night shift-induced fatigue, diagnostic performance, and AI assistance in chest radiography.
文献摘要
Objectives To evaluate the impact of night shift-induced fatigue on radiology residents' diagnostic performance and to assess whether AI assistance offsets potential performance loss. Materials and methods In this prospective intra-individual reader study, ten radiology residents (mean experience: 35 months; three women) interpreted three sets of 33 chest radiographs before and after a first-call on-site night shift. Sessions included two pre-shift readings (morning/afternoon, without AI) and two post-shift readings (morning, with/without AI). Eye tracking captured fixation duration and saccade count, and fatigue was defined as a ≥ 5% reduction in the slope of the saccadic main sequence between pre- and post-shift sessions. Diagnostic accuracy (Cohen's κ vs. expert majority vote), efficiency (mean reporting time per radiograph), eye-tracking metrics, and user surveys were analyzed with generalized linear mixed models and post-hoc Tukey-adjusted pairwise comparisons. Results Six of ten radiology residents were classified as fatigued, among whom diagnostic accuracy declined significantly post-shift (from κ = 0.71 ± 0.05 to κ = 0.61 ± 0.10; p Conclusion Night-shift-induced fatigue impaired diagnostic accuracy without slowing reporting speed, raising patient safety concerns. The utilized AI model did not restore pre-shift performance. Key points Question Night-shift-induced fatigue may reduce diagnostic accuracy of radiology residents in chest radiography, yet its objective quantification and the compensatory potential of AI assistance remain unclear. Findings Fatigue, based on eye-tracking measurements of oculomotor efficiency, reduced diagnostic accuracy after night shifts without slowing reporting speed; AI assistance did not significantly improve accuracy. Clinical relevance Night-shift-induced fatigue can lower diagnostic accuracy in chest radiography despite unchanged reporting speed, posing patient-safety risks. The AI model used in this study cannot fully compensate for fatigue-related performance decline.