Machine learning-based predictive model for immune checkpoint inhibitors response in gastrointestinal cancers

Area of Science:

• Oncology
• Bioinformatics
• Machine Learning

Background:

• Gastrointestinal (GI) cancers have poor prognoses, with limited predictors for immunotherapy response.
• Accurate prediction models for immunotherapy response are crucial for effective clinical decision-making.

Purpose of the Study:

• To develop and validate a machine learning model for predicting immunotherapy response rates in GI cancers.
• To identify key clinical and molecular features influencing treatment outcomes.

Main Methods:

• A multicentre retrospective study of 506 GI cancer patients.
• Utilized 14 features (demographic, clinical, biochemical, genetic) as input for eight machine learning algorithms.
• Employed eXtreme Gradient Boosting (XGBoost) for its superior predictive performance, validated with metrics like AUC, accuracy, and sensitivity.
• Shapley Additive explanations (SHAP) analysis was used for model interpretability.

Main Results:

• The XGBoost model achieved an AUC of 0.829, accuracy of 78.43%, and sensitivity of 86.67%.
• Chemotherapy (28% contribution) and tumor stage were the most significant predictors.
• Factors like hemoglobin, BMI, age, and lower neutrophil-to-lymphocyte ratio (NLR) also positively influenced predictions.

Conclusions:

• Interpretable XGBoost models accurately predict immunotherapy response in GI cancers.
• Personalized predictive models, informed by factors like chemotherapy and tumor stage, can enhance clinical decision-making.
• This model offers precise support for personalized treatment strategies and risk management in GI oncology.