Bayesian Outcome Prediction After Resuscitation From Cardiac Arrest.

Area of Science:

• Critical Care Medicine
• Biostatistics
• Neurology

Background:

• Post-cardiac arrest prognostication often overlooks sequential data acquisition, leading to uncertain estimates.
• Bayesian approaches are well-suited for prognostication due to their ability to incorporate prior knowledge and handle sequential data.

Purpose of the Study:

• To explore the utility of sequential prognostic indicators using Bayesian regression.
• To compare Bayesian prognostication with a guideline-concordant algorithm for post-cardiac arrest outcomes.

Main Methods:

• Bayesian hierarchical generalized linear multivariate models were used to predict Cerebral Performance Category (CPC) scores.
• Prospective data from 2,692 post-cardiac arrest patients were analyzed, incorporating demographic, clinical, laboratory, and EEG data sequentially.
• The 2021 European Resuscitation Council and European Society of Intensive Care Medicine (ERC/ESICM) guidelines were used as a comparator.

Main Results:

• Bayesian models demonstrated progressively narrower outcome probability distributions as sequential data were added.
• The most comprehensive Bayesian model achieved 76% sensitivity for predicting poor outcomes (CPC 4-5) with a 0.6% false-positive rate.
• The ERC/ESICM algorithm showed 36% sensitivity with 0% false-positive rate in a subset of patients.

Conclusions:

• Bayesian models offer a robust framework for accurate neurologic prognostication in post-cardiac arrest patients.
• Accurate prognostication is achievable before 72 hours post-arrest, informing clinical decision-making and clinical trials.
• While cautioning against premature withdrawal of care, rapid outcome prediction enhances patient management strategies.