ABSTRACT Purpose Laparoscopic subtotal cholecystectomy (LSTC) serves as a bailout procedure when the critical view of safety cannot be achieved during laparoscopic cholecystectomy (LC) for acute cholecystitis (AC). Previous prediction models rely solely on baseline variables without considering temporal changes. We aimed to develop prediction models incorporating both static clinical factors and dynamic laboratory changes using machine learning (ML). Methods We retrospectively analyzed 104 patients who underwent LC for AC between January 2020 and October 2022. Three models were developed: Model 1 using logistic regression with baseline variables, Model 2 incorporating temporal laboratory changes, and Model 3 employing ML algorithms. Model performance was evaluated using the area under the receiver operating characteristic curve (AUROC). Results LSTC was performed in 19 patients (18.3%). In Model 1, severity grade II–III was the strongest predictor with an AUROC of 0.735. Model 2 improved performance to an AUROC of 0.782 by incorporating the ALT change rate. Among seven ML algorithms, elastic net achieved the highest AUROC of 0.821. Conclusions The ML‐based model enabled preoperative identification of patients more likely to require LSTC. The model may support preoperative discussions in settings where operative strategies vary.
SHINOZUKA et al. (Thu,) studied this question.