Linking preoperative and early intensive care unit data for prolonged intubation prediction

Area of Science:

• Cardiovascular Surgery
• Intensive Care Medicine
• Respiratory Therapy

Background:

• Prolonged intubation (PI) is a severe complication following cardiac surgery (CS).
• Predictive models solely using preoperative data may lack accuracy due to intraoperative and early postoperative factors.
• Accurate prediction is crucial for timely intervention and improved patient outcomes.

Purpose of the Study:

• To develop and validate a prediction model for PI in CS patients.
• To integrate both preoperative and early intensive care unit (ICU) data for enhanced predictive accuracy.
• To facilitate early prophylaxis strategies for PI.

Main Methods:

• Retrospective cohort study using data from two public ICU databases (MIMIC and eICU).
• Multivariable logistic regression was employed to construct the predictive model.
• The model underwent external validation to assess its predictive performance.

Main Results:

• The model identified 11 significant predictors of PI, including plateau pressure, lactate level, and SOFA score.
• The incidence of PI was 11.8% (training), 12.1% (testing), and 17.5% (external validation).
• The model demonstrated strong discriminative capability with an AUROC of 0.853 (training) and 0.704 (external validation).

Conclusions:

• The developed prediction model, integrating preoperative and early ICU data, shows excellent discriminative ability for PI risk in CS patients.
• This model enables accurate risk assessment in the early post-CS period.
• Timely interventions based on this model can help mitigate adverse outcomes associated with PI.