Reinforcement learning model for optimizing dexmedetomidine dosing to prevent delirium in critically ill patients

Area of Science:

• Critical Care Medicine
• Artificial Intelligence in Healthcare
• Pharmacology

Background:

• Delirium in intensive care unit (ICU) patients is associated with adverse outcomes, including prolonged hospital stays and increased mortality.
• Dexmedetomidine is utilized for delirium prevention in ICUs, but achieving optimal dosing remains a clinical challenge.

Purpose of the Study:

• To develop and validate an Artificial Intelligence (AI) model, termed AID (AI model for Delirium prevention), to optimize dexmedetomidine dosing for delirium prevention in ICU patients.
• To compare the performance of the AID model against clinician dosing strategies.

Main Methods:

• A reinforcement learning-based AI model (AID) was developed.
• The model was trained and internally validated on a cohort of 2416 patients (2531 ICU admissions).
• External validation was performed on an independent cohort of 270 patients (274 ICU admissions).

Main Results:

• The AID model demonstrated a significantly higher estimated performance return compared to the clinicians' policy in both the derivation cohort (0.390 vs. -0.051) and the external validation cohort (0.186 vs. -0.436).
• These findings suggest the AID policy is more effective in managing dexmedetomidine dosing for delirium prevention.

Conclusions:

• The AID model shows potential as a supportive tool for clinicians in optimizing dexmedetomidine dosing to prevent delirium in ICU patients.
• Further off-policy evaluation is necessary to confirm the real-world efficacy and safety of the AID model.