Artificial intelligence assessment of Parkland’s grading scale in laparoscopic cholecystectomy: a step toward real-world outcome prediction

Abstract Routine real-world use of a surgical artificial-intelligence (AI) platform has led us to adopt Parkland’s Grading Scale (PGS) for disease-severity in laparoscopic cholecystectomy (LC). We evaluated model performance in estimating PGS and explored correlations with surgical outcomes. The platform routinely video-captured LCs and automatically assigned PGS scores across 249 consecutive LCs that were classified into Low (PGS 1–2; n = 78, 31.3%) and High (PGS 3–5; n = 171, 68.7%) groups. Surgical outcomes were compared. Case-control matching (CCM; n = 84) was performed to balance possible confounders. Two surgeons independently reviewed a video sample ( n = 75) twice to establish ground-truth for F1-score calculations. Model discrimination (AUC) and calibration were evaluated separately against each rater. High-severity cases presented with significantly older age (63.2y vs. 47.1y), higher ASA scores ( ≥ 3: 29.8% vs. 9%), cholecystitis (47.4% vs. 10.3%), and urgent surgery (12.3% vs. 0) (all p < 0.001). The High group experienced longer operative durations (57.6 vs. 35.3 min; p < 0.001), more intraoperative events (80.1% vs. 61.5%; p = 0.003) and bailouts (10.5% vs. 0; p < 0.001). Hospitalizations were longer (2d vs. 1d; p < 0.001), with more 90-day major complications and readmissions (10.5% vs. 2.6%; p = 0.04). After CCM, operative durations (mean rank 49.51 vs. 35.49; p = 0.008) and hemorrhage-related events (mean rank 46.0 vs. 39.0; p = 0.047) maintained significance. The AI model achieved high F1 scores (High = 0.96, Low = 0.93), strong discrimination (AUC: 0.932 and 0.896), and robust calibration- peaking at PGS = 3. The AI model shows promise for surgical outcome prediction in LC and, with further validation, could be integrated into routine clinical frameworks.