Machine Learning Classification Algorithms to Predict aGvHD following Allo-HSCT: A Systematic Review.

Area of Science:

• Hematopoietic stem cell transplantation research
• Medical informatics and machine learning

Background:

• Acute graft-versus-host disease (aGvHD) is a primary cause of mortality post-allogeneic hematopoietic stem cell transplantation.
• Late diagnosis of aGvHD is common due to its occurrence during severe tissue damage.
• Machine learning (ML) offers potential for real-time aGvHD prediction models.

Purpose of the Study:

• To systematically review and synthesize literature on ML classification algorithms for predicting aGvHD.
• To identify commonly used ML algorithms and significant predictor variables in aGvHD prediction.

Main Methods:

• A systematic review adhering to PRISMA guidelines was conducted.
• Searches were performed across major scientific databases (PubMed, Embase, Web of Science, Scopus, Springer, IEEE Xplore) up to April 2019.
• Studies focusing on ML classification algorithms for aGvHD prediction were included.

Main Results:

• Fourteen studies met the inclusion criteria.
• Artificial Neural Network (79%) was the most frequent algorithm, followed by Support Vector Machine (50%) and Naive Bayes (43%).
• Predictor variables were categorized into biomarkers, demographics, infections, clinical factors, genes, transplant details, and drugs.

Conclusions:

• Various ML algorithms possess distinct characteristics and utilize diverse predictors for aGvHD.
• Correctly implemented ML modeling holds significant potential for accurate aGvHD prediction.
• Further research into optimal algorithm selection and predictor variable integration is warranted.