Uncovering distinct clinical phenotypes in disseminated intravascular coagulation through machine learning-enabled...

Area of Science:

• Critical Care Medicine
• Computational Biology
• Hematology

Background:

• Disseminated intravascular coagulation (DIC) is a life-threatening ICU condition with varied causes and outcomes.
• Differentiating DIC patient phenotypes is challenging, hindering personalized treatment strategies.

Purpose of the Study:

• To apply unsupervised machine learning (ML) for stratifying DIC patients into distinct phenotypes.
• To enable more personalized treatment approaches for DIC based on identified subtypes.

Main Methods:

• Retrospective analysis of 134 DIC patients admitted to the ICU.
• Unsupervised machine learning (consensus clustering) using key variables: Thrombin-Antithrombin Complex (TAT), Plasmin-α2-Plasmin Inhibitor Complex (PIC), tissue plasminogen activator-inhibitor complex (tPAIC), and thrombomodulin (TM).
• Logistic regression to assess the association between phenotypes and clinical endpoints.

Main Results:

• Two distinct DIC subtypes were identified: mild (n=79) and severe (n=55).
• Significant differences were observed in coagulation markers (TAT) and clinical parameters (HR, SBP) between subtypes.
• The severe subtype was associated with increased 7-day (OR 4.71) and 28-day (OR 2.29) mortality risk.

Conclusions:

• Unsupervised ML successfully stratified DIC patients into two distinct phenotypes.
• These phenotypes exhibit different laboratory profiles and associated mortality risks.
• The findings support personalized treatment strategies for DIC patients based on identified subtypes.