Machine learning predictive models for acute pancreatitis: A systematic review.

Area of Science:

• Medical Informatics
• Computational Biology
• Clinical Decision Support

Background:

• Acute pancreatitis (AP) is a prevalent pancreatic disease with variable patient outcomes.
• Machine learning (ML) offers algorithmic advantages for disease prediction and clinical decision support.

Purpose of the Study:

• To systematically review studies utilizing ML for predictive tool development in acute pancreatitis.
• To assess the quality and identify limitations of existing ML-based AP prediction models.

Main Methods:

• A systematic review of PubMed, Web of Science, Scopus, and Embase databases was performed.
• Studies published up to May 2021 using ML for AP prediction were included and quality assessed using the IJMEDI checklist.

Main Results:

• 24 studies involving 8,327 patients and 47 models were included, focusing on severity, complication, mortality, recurrence, and surgery timing prediction.
• ML demonstrated high accuracy in various AP prediction tasks, but most studies were retrospective, single-center, and lacked external validation.
• Only two studies met high-quality criteria, with most exhibiting bias in data preparation, validation, and deployment.

Conclusions:

• ML holds significant potential for improving AP prediction and aiding clinical decision-making.
• Existing ML models for AP require optimization in construction, validation, and comparability to ensure clinical applicability.