SP 5.07 Machine Learning for Acute Cholecystitis Management: A Pilot Study Comparing an NLP-Enhanced Decision-Support Model with Clinical Judgement
Raghav Anirudh, Alexandra Whitworth, Marie Laanani, Christie Swaminathan, Alex Bull, Muhammad S SajidAbstract
Aims
Determining the need for emergency admission for acute cholecystitis (AC) versus outpatient management with elective laparoscopic cholecystectomy remains variable. This pilot study evaluates a machine learning (ML)–based decision-support model designed to standardize acute admission decisions by integrating clinical, biochemical, and free-text data.
Methods
Data from patients admitted with AC over eight consecutive weeks at a tertiary NHS hospital were retrospectively analysed. A hybrid system was developed using natural language processing (NLP) with regex-based extraction of 27 clinical keywords from emergency triage notes. Feature engineering included calculation of inflammatory markers, such as neutrophil-to-lymphocyte ratio (NLR), and severity grading aligned with Tokyo Guidelines 2018 (TG18). A logistic regression classifier with balanced class weights was trained using an 80/20 train–test split, incorporating age, ASA grade, NLR, and TG18-defined organ dysfunction parameters. Model recommendations were compared with actual clinical outcomes.
Results
Thirty-eight patients met inclusion criteria. The ML model demonstrated 76.3% agreement with clinical decision-making and achieved a test-set ROC AUC of 1.0. The model identified 10.5% of patients as suitable for outpatient management despite their admission; all were Tokyo Grade I with low inflammatory markers (CRP ≤8 mg/L). Conversely, five discharged patients were flagged for admission due to high-risk physiological or radiological features.
Conclusion
This pilot study demonstrates that ML models integrating NLP and biochemical data can provide a reliable second opinion for AC triage. Such tools may reduce unnecessary emergency admissions, improve identification of high-risk patients, enhance safety, and contribute to cost savings and optimized hospital bed utilization.