BACKGROUND: Several recent studies have found that the efficacy of computer-aided polyp detection (CADe) on the adenoma detection rate (ADR) diminished in real-world settings. The role of unmeasured factors in AI-human interaction, such as monitor approaches, remains unknown. This study aimed to validate the effectiveness of CADe in the real world and assess the impact of monitor approaches. METHODS: A retrospective propensity score-matched cohort study was conducted using routine data from a tertiary endoscopy center in China before and after the implementation of CADe. Four propensity score-matched cohorts were established: Cohort 1: pre-CADe matched with dual-monitor CADe-assisted group
Cohort 2: dual-monitor CADe-assisted with single-monitor CADe-assisted group
Cohort 3: pre-CADe with single-monitor CADe-assisted group
and Cohort 4: pre-CADe with CADe period. ADR was set as the primary outcome. RESULTS: There were 5390, 6083, and 6131 eligible patients in the pre-CADe group, dual-monitor group, and single-monitor group, respectively. In the matched analysis, results indicated that regardless of the monitor setup, CADe-assisted groups showed a trend of increased ADR compared with the pre-CADe period (CADe period: OR 1.141, 95% CI 1.047-1.243
p = 0.003
dual-monitor: OR 1.178, 95% CI 1.069-1.299, p = 0.002
single-monitor: OR 1.094, 95% CI 0.998-1.200, p = 0.056). Moreover, no significant difference between different monitor approaches was observed, although dual-monitor setup showed an increasing tendency on ADR compared with single-monitor setup (OR 1.069, 95% CI 0.985-1.161, p = 0.109). CONCLUSION: CADe shows great potential to improve ADR during colonoscopy in the real world. Meanwhile, changes in monitor setup do not significantly impact the assistance capability of CADe. Further research dedicated to evaluating the unmeasured elements in the AI-clinician hybrid for better implementation of CADe would be beneficial.