Identification of Clinically Meaningful Plasma Transfusion Subgroups Using Unsupervised Random Forest Clustering

Area of Science:

• Biomedical Informatics
• Clinical Research Methodology
• Translational Medicine

Background:

• Comparative effectiveness research often relies on statistical methods like propensity score matching to adjust for confounders.
• Current methods yield an "average" treatment effect, which is limited due to significant patient heterogeneity in treatment response.
• Individual patient variability from the population average necessitates more personalized analytical approaches.

Purpose of the Study:

• To develop a framework for discovering clinically meaningful homogeneous subgroups with differential treatment effects.
• To apply unsupervised random forest clustering for identifying these subgroups in blood transfusion research.
• To enable customized blood transfusion therapy by accounting for individual patient variability.

Main Methods:

• Utilized unsupervised random forest clustering to identify patient subgroups.
• Performed subgroup analysis on two distinct blood transfusion datasets.
• Compared treatment effects and risk factors between identified subgroups and the general population.

Main Results:

• Identified considerable variability in plasma transfusion effects on bleeding and mortality across different patient subgroups.
• Discovered distinct risk factors for bleeding and mortality within these subgroups.
• Demonstrated that subgroup effects can significantly differ from population-level averages.

Conclusions:

• The developed framework effectively identifies patient subgroups with differential responses to plasma transfusion.
• Subgroup analysis reveals significant heterogeneity in treatment outcomes and risk factors.
• Findings support the customization of blood transfusion therapy for individual patients based on subgroup characteristics.