Applying Machine Learning to the ANZELA-QI Database to Predict Adverse Outcomes for Patients Undergoing Emergency...

Area of Science:

• Surgical Outcomes
• Machine Learning in Medicine
• Health Informatics

Background:

• Emergency laparotomy carries significant risks of morbidity and mortality.
• Accurate risk prediction is crucial for improving patient management and decision-making.
• The Australian and New Zealand Emergency Laparotomy Audit (ANZELA-QI) database provides a valuable resource for such studies.

Purpose of the Study:

• To apply novel machine learning models for risk stratification in emergency laparotomy.
• To predict adverse outcomes including mortality, ICU admission, non-return home, and prolonged hospitalization.
• To enhance patient flow and shared decision-making through accurate risk assessment.

Main Methods:

• Utilized the ANZELA-QI database, extracting data from 8615 cases.
• Employed logistic regression, XGBoost, and random forest machine learning techniques.
• Trained and tested models using clinical and demographic predictors for adverse outcomes.

Main Results:

• Machine learning models showed high accuracy in predicting intensive care unit (ICU) admission and prolonged hospital stays.
• For ICU admission, sensitivity was 0.7 and specificity was 0.74.
• The models achieved 75% accuracy in predicting hospital admissions longer than one week.

Conclusions:

• Novel machine learning models were successfully developed using the ANZELA-QI database.
• These models demonstrate high accuracy in stratifying risk for prolonged hospital stays and post-operative ICU admission.
• The findings support the use of machine learning for improved risk prediction in emergency laparotomy patients.