IntraCranial pressure prediction AlgoRithm using machinE learning (I-CARE): Training and Validation Study.

Area of Science:

• Neurology
• Critical Care Medicine
• Data Science

Background:

• Elevated intracranial pressure (ICP) is a critical complication following neurological injury.
• Timely intervention is crucial for managing ICP and preventing adverse outcomes.

Purpose of the Study:

• To develop and validate an ensemble machine learning model for predicting intracranial pressure (ICP) 30 minutes in advance.
• To assist clinicians in proactive treatment adjustments for patients with neurological injuries.

Main Methods:

• Retrospective analysis of 335 ICUs across 208 US hospitals using the eICU Collaborative Research Database.
• An ensemble machine learning model was trained on patient data, including demographics, labs, medications, vitals, and ICP history.
• Model performance was evaluated on a held-out eICU test set and externally validated on the MIMIC-III database.

Main Results:

• The model utilized predictors such as age, GCS, temperature, creatinine, and historical ICP and hemodynamic data.
• The ensemble model achieved a root mean squared error of 4.51 mm Hg in the eICU test set and 3.56 mm Hg in the MIMIC-III dataset.
• Key predictors for ICP included prior ICP values, patient temperature, weight, serum creatinine, age, GCS, and hemodynamic parameters.

Conclusions:

• The developed IntraCranial pressure prediction AlgoRithm using machinE learning (ICP-RAM) demonstrates promising predictive performance for future ICP.
• External validation confirmed the model's generalizability, suggesting its potential utility in clinical settings.
• This ICP prediction tool can support clinical decision-making and potentially mitigate complications associated with elevated intracranial pressure.