International Conference on Computer and Knowledge Engineering

Home / 15th International Conference on Computer and Knowledge Engineering

A Comparative Analysis of Clinical Note Categories for Mortality Prediction in ICU Patients

Authors :

Maryam Karrabi¹ Mohsen Kahani² Mina Afzali³ Nadieh Armin⁴

1- Computer Engineering Dept. Ferdowsi University of Mashhad Mashhad, Iran 2- Computer Engineering Dept. Ferdowsi University of Mashhad Mashhad, Iran 3- Computer Engineering Dept. Ferdowsi University of Mashhad Mashhad, Iran 4- Computer Engineering Dept. Ferdowsi University of Mashhad Mashhad, Iran

Keywords :

Electronic Health Record،Mortality Prediction،Clinical notes،Note embedding،MIMIC III

Abstract :

Mortality prediction in intensive care unit (ICU) patients is crucial for improving patient outcomes and optimizing the use of resources. The accessibility of electronic health records (EHRs) has enabled the use of data-driven predictive modeling through machine learning. In the medical field, natural language processing (NLP) techniques have demonstrated their ability to extract valuable insights from EHRs. Contextualized word embedding-based models, along with preprocessing approaches, are key to better representing unstructured clinical data. In this work, we propose a comparative analysis of different clinical note categories for predicting in-hospital mortality among ICU patients within the first 24 hours of admission. Results indicate that nursing/other and nursing notes are the most informative when used individually, while combining multiple note categories improves predictive performance. These findings highlight the importance of note category selection in developing effective clinical note-based mortality prediction models.