Predicting mortality in critically ill patients with diabetes using machine learning and clinical notes.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Decision Support Systems

Background:

• Diabetes mellitus is a common metabolic disease causing chronic hyperglycemia.
• Healthcare data is rapidly growing, driving the need for precision medicine.
• Artificial intelligence (AI) and machine learning (ML) are crucial for clinical decision-making, augmenting healthcare providers' capabilities.
• Few studies have explored predictive modeling for comorbidities in diabetic ICU patients.

Purpose of the Study:

• To predict the risk of mortality in intensive care unit (ICU) patients with diabetes.
• To leverage Unified Medical Language System (UMLS) resources, machine learning (ML), and natural language processing (NLP) for mortality risk prediction.
• To uncover associations between comorbidities in diabetic ICU patients and mortality risk.

Main Methods:

• Secondary analysis of the Medical Information Mart for Intensive Care III (MIMIC-III) database.
• Application of various ML modeling and NLP approaches, including knowledge-guided models.
• Utilized UMLS entity embeddings and convolutional neural networks (CNNs) with word embeddings for clinical text representation.
• Mortality classification based on a combination of knowledge-guided features and rules.

Main Results:

• The best ML model configurations achieved a high Area Under the Curve (AUC) of 0.97.
• ML models combined with NLP of clinical notes show promise in assisting healthcare providers.
• The knowledge-guided CNN model demonstrated effectiveness in learning hidden features for mortality prediction.

Conclusions:

• UMLS resources and clinical notes are valuable tools for predicting mortality in diabetic ICU patients.
• The knowledge-guided CNN model is effective, achieving an AUC of 0.97.
• AI-driven predictive modeling can significantly aid in managing critically ill diabetic patients.