Predicting the efficiency of chidamide in patients with angioimmunoblastic T-cell lymphoma using machine learning...

Area of Science:

• Oncology
• Hematology
• Computational Biology

Background:

• Angioimmunoblastic T-cell lymphoma (AITL) is a rare non-Hodgkin lymphoma with challenging prognostication.
• Chidamide, a subtype-selective histone deacetylase (HDAC) inhibitor, has shown potential in clinical trials for AITL.
• Real-world data on chidamide's impact on overall survival (OS) in AITL patients remain contradictory.

Purpose of the Study:

• To develop an interpretable machine learning (ML) model for predicting 2-year OS in AITL patients.
• To assess the predictive value of chidamide usage and baseline features for AITL patient survival.
• To clarify the role of chidamide in improving OS for AITL patients in real-world settings.

Main Methods:

• Utilized a dataset of 183 AITL patients, randomly divided into training and testing sets.
• Employed five ML algorithms, with recursive feature elimination (RFE) for feature selection.
• Interpreted model relevance using Shapley additive explanations (SHAP) and local interpretable model-agnostic explanations.

Main Results:

• The Catboost model achieved the best predictive performance (AUC = 0.8651) with 12 selected features.
• Chidamide usage was identified as the third most important variable correlating with 2-year OS.
• Seventy-one out of 183 patients died within 2 years, indicating a significant survival challenge.

Conclusions:

• The developed Catboost model effectively predicts 2-year OS in AITL patients.
• Incorporating chidamide into treatment regimens is positively associated with improved OS for AITL patients.
• This study provides valuable insights for personalized treatment strategies in AITL.