icuARM-An ICU Clinical Decision Support System Using Association Rule Mining

Area of Science:

• Biomedical Engineering
• Medical Informatics
• Data Science

Background:

• Intensive care units (ICUs) generate vast multimodal patient data, posing real-time processing challenges.
• Developing effective data analysis tools is crucial for improving patient care in critical settings.

Purpose of the Study:

• To design and validate icuARM, an ICU clinical decision support system utilizing associate rule mining (ARM).
• To leverage the MIMIC-II database for real-time data mining and clinical insights.

Main Methods:

• Developed icuARM, integrating multiple association rules and a graphical user interface (GUI).
• Utilized the Multi-parameter Intelligent Monitoring in Intensive Care II (MIMIC-II) database containing over 40,000 ICU records.
• Investigated associations between patient characteristics (comorbidities, demographics, medications) and ICU outcomes (length of stay).

Main Results:

• Coagulopathy identified as the most dangerous comorbidity, associated with a 54.1% possibility of prolonged ICU stay.
• Older women (over 50) showed the highest possibility (38.8%) of prolonged ICU stay.
• icuARM demonstrated potential for optimizing medication choices based on patient-specific factors.

Conclusions:

• icuARM provides valuable real-time insights for ICU physicians.
• The system aids in tailoring patient treatment based on clinical status and identified risk factors.
• Associate rule mining offers a powerful approach for clinical decision support in ICUs.