Artificial intelligence for predicting surgical difficulty in laparoscopic cholecystectomy: a systematic review and meta-analysis.

backgroundAccurately predicting operative difficulty in laparoscopic cholecystectomy (LC) is foundational to personalized surgical planning and patient safety assurance. However, the reliability, generalizability, and true clinical utility of current Artificial Intelligence (AI) models are currently unsubstantiated. This review aimed to evaluate the predictive performance and methodological quality of AI models designed to predict LC surgical difficulty.

methodsPubMed, Embase, Web of Science, and the Cochrane Library were searched from inception to March 2, 2026. The Prediction Model Risk of Bias Assessment Tool was used to assess the risk of bias and applicability. The areas under the curve (AUC) with 95% confidence intervals were pooled using random-effects meta-analysis. The overall certainty of evidence was evaluated using the GRADE framework. The study followed PRISMA guidelines and was registered with PROSPERO (CRD420251267805).

resultsA total of 18 studies were included in this review. Sixteen studies were at high risk of bias. The pooled AUC for 27 training models was 0.848 (95% CI, 0.829-0.868). For 32 validation models, the pooled AUC was 0.818 (95% CI, 0.797-0.840). Ensemble models achieved the highest pooled AUCs (0.889 and 0.861) in both training and validation set. Multimodal integration of clinical features, imaging, and intraoperative video also yielded superior performance.

conclusionCurrent research showed significant methodological flaws. AI models based on ensemble architecture and multimodal approaches warrant further exploration. Most studies carry a high risk of bias and rarely undergo external validation, which limits clinical translation. Before clinical implementation, these models still require strict methodological evaluation, prospective multicenter testing, and proper calibration.