Machine Learning and Antibiotic Management.

Area of Science:

• * Clinical Informatics
• * Artificial Intelligence in Medicine
• * Critical Care Medicine

Background:

• * Intensive care units (ICUs) face increasing complexity in patient monitoring and management.
• * Experienced clinicians rely on an intuitive 'gestalt' for decision-making, which is difficult to quantify.
• * There is a need for objective methods to aid clinical judgment in complex ICU settings.

Purpose of the Study:

• * To apply machine learning and cluster analysis to evaluate patient vitals and antimicrobial therapy in the ICU.
• * To develop a quantifiable measure of clinical experience and decision-making processes.
• * To provide a tool for assessing the progress of complex ICU patients.

Main Methods:

• * Retrospective analysis of single-center ICU data.
• * Machine learning and cluster analysis applied to daily records of antibiotic therapy (duration, type, combinations).
• * Fuzzy logic normalization used to categorize parameters into descriptive linguistic terms (e.g., 'very low' to 'very high').
• * Hourly bedside monitoring data also analyzed using the same methodology.
• * Patient conditions assessed by monitoring shifts between clusters.

Main Results:

• * Clustering of normalized therapy and monitoring data created distinct patient/day clusters.
• * Changes in patient condition correlated with shifts between these clusters.
• * Fuzzy logic normalization provided a human-interpretable representation of complex clinical data.

Conclusions:

• * Machine learning and fuzzy logic offer a valuable aid for clinical management in ICUs.
• * The proposed method provides a proxy for experienced clinician judgment, aiding in decision-making.
• * This approach can serve as an additional tool for monitoring and assessing the progress of complex ICU patients.