Friedman's Gradient-Boosting Algorithm Predicts Lactate-Pyruvate Ratio Trends in Cases of Intracerebral Hemorrhages

The local effects of an intracerebral hemorrhage (ICH) on surrounding brain tissue can be detected bedside using multimodal brain monitoring techniques. The aim of this study is to design a gradient boosting regression model using the R package boostmtree with the ability to predict lactate-pyruvate ratio measurements in ICH.

We performed a retrospective analysis of 6 spontaneous ICH and 6 traumatic ICH patients who underwent surgical removal of the clot with microdialysis catheters placed in the perihematomal zone. Predictors of glucose, lactate, pyruvate, age, sex, diagnosis, and operation status were used to design our model.

In a holdout analysis, the model forecasted lactate-pyruvate ratio trends in a representative in-sample testing set. We anticipate that boostmtree could be applied to designs of similar regression models to analyze trends in other multimodal monitoring features across other types of acute brain injury.

The model successfully predicted hourly lactate-pyruvate ratios in spontaneous ICH and traumatic ICH cases after the hemorrhage evacuation and displayed significantly better performance than linear models. Our results suggest that boostmtree may be a powerful tool in developing more advanced mathematical models to assess other multimodal monitoring parameters for cases in which the perihematomal environment is monitored.