Safety in one anastomosis gastric bypass (OAGB) is judged by outcomes, but it seems reasonable to utilize best practices for safety, whose performance can be evaluated and therefore improved. We aimed to test an artificial intelligence-based model in real world for the evaluation of adherence to best practices in OAGB.Please check and confirm that the authors and their respective affiliations have been correctly identified and amend if necessary. A retrospective single-center study of 89 consecutive OAGB videos was captured and analyzed by an artificial intelligence platform (10/2020-12/2023). The platform currently provides assessment of four elements, including bougie insertion, full division of pouch, view of Treitz ligament, and leak test performed. Two bariatric surgeons viewed all videos, categorizing these elements into Yes/No adherence. Intra-rater and inter-rater agreements were computed. The estimates found in greatest consensus were used to determine the model's performance. Clinical data retrieval was performed. Videos included primary (71.9%) and conversion (28.1%) OAGB. Patients' age was 41.5 ± 13.6y and body mass index 42.0 ± 5.7kg/m2. Anastomosis width was 40mm (IQR, 30-45), and biliopancreatic limb length was 200cm (IQR, 180-200). Operative duration was 69.1min (IQR 55.3-97.4), mainly spent on gastric transection (26%) and anastomosis (45%). Surgeons' intra-rater overall agreements ranged 93-100% (kappa 0.57-1). Inter-rater overall agreements increased to 99-100% (kappa 0.95-1) in the second review, set as reference point to the model. The model's overall accuracy ranged 82-98%, sensitivity 91-94%, and positive predictive value 88-99%. Specificity ranged 17-92% and negative predictive value 20-68%. The model appears to have high accuracy, sensitivity, and positive predictive value for evaluating adherence to best practices for safety in OAGB. Considering the paucity of negative estimates in our study, more low-performance cases are needed to reliably define the model's specificity and negative predictive value. Adding more best practices, tested in multi-center studies will enable cross-border standardization of the procedure.
Read full abstract