Acute brain dysfunction clusters in COVID-19: a pilot machine learning-based analysis of the COVID-D cohort

Area of Science:

• Critical Care Medicine
• Neurology
• Infectious Diseases
• Machine Learning in Healthcare

Background:

• Acute brain dysfunction (ABD), encompassing delirium and coma, significantly increases morbidity in critically ill patients.
• The heterogeneity of ABD presents challenges in management and prognostication, especially in COVID-19 patients.
• Machine learning offers a potential approach for identifying subgroups within critically ill patients with ABD.

Purpose of the Study:

• To identify distinct patient clusters among critically ill COVID-19 patients with ABD using ICU admission data.
• To evaluate the association of these identified clusters with clinical outcomes, including neurological status and survival.

Main Methods:

• Retrospective analysis of an international multicenter database (COVID-D study) of critically ill adult COVID-19 patients with ABD during the first pandemic wave.
• Application of unsupervised machine learning (hierarchical clustering) with dimensionality reduction and bootstrap assessment.
• Analysis of clusters for differences in neurological outcomes, mechanical ventilation duration, and survival rates.

Main Results:

• Four reproducible clusters of critically ill COVID-19 patients with ABD were identified, each with distinct clinical and neurological profiles.
• Clusters varied in delirium and coma duration, with Cluster 4 showing the longest coma duration (11.2 days) and fewest delirium-free and coma-free (DFCF) days (4.74 days).
• Despite observed differences in ABD characteristics, no significant variations in 28-day mortality, ICU length of stay, or hospital length of stay were found among the clusters.

Conclusions:

• This pilot study suggests the existence of clinically distinct clusters among critically ill COVID-19 patients with acute brain dysfunction.
• While cluster differences in delirium and coma duration were noted, they did not translate to significant differences in 28-day survival.
• Further prospective studies are needed to confirm the clinical utility of these clusters for targeted delirium prevention strategies in modern ICUs.