Guiding Efficient, Effective, and Patient-Oriented Electrolyte Replacement in Critical Care: An Artificial...

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Decision Support Systems
• Intensive Care Unit (ICU) Management

Background:

• Electrolyte replacement in ICUs is often over-prescribed due to provider- and protocol-driven approaches.
• Current methods may lead to suboptimal patient outcomes and increased healthcare costs.

Purpose of the Study:

• To develop and validate a data-driven clinical decision support tool using reinforcement learning (RL) for personalized electrolyte replacement in ICU patients.
• To reduce the frequency and improve the safety, precision, efficacy, and cost of electrolyte repletion.

Main Methods:

• Development of an RL-based clinical decision support tool utilizing electronic health records (EHR) data from two institutions.
• The tool integrates patient characteristics, physiological and pharmacological states, repletion actions, and clinical goals.
• Retrospective validation of the RL tool's recommendations on patient data.

Main Results:

• RL-driven electrolyte repletion significantly reduced magnesium and potassium replacements (up to 60%).
• Intervention timing for potassium, magnesium, and phosphate was optimized.
• A shift towards oral repletion over intravenous administration was recommended, limiting over-repletion risks.
• Monetary savings were implied through optimized repletion strategies.

Conclusions:

• The RL-driven tool enhances the safety, precision, efficacy, and cost-effectiveness of electrolyte repletion in ICUs.
• This approach offers robust performance across diverse patient cohorts and hospital systems.
• Personalized, data-driven recommendations can mitigate the risks associated with over-prescription of electrolytes.