Machine learning models for predicting response to epidermal growth factor receptor tyrosine kinase inhibitors in non-small cell lung cancer brain metastases: a systematic review and meta-analysis.
Predicting clinical and radiological outcomes of epidermal growth factor receptor tyrosine kinase inhibitors (EGFR-TKIs) in patients with non-small cell lung cancer (NSCLC) and brain metastases (BMs) is crucial for effective patient management. Machine learning (ML)-based models have increasingly been utilized to predict EGFR-TKI response in patients with lung cancer brain metastasis (LCBM). In this study, we aimed to evaluate the predictive performance of ML-based models for EGFR-TKI response prediction.
A comprehensive literature search was conducted using PubMed, Embase, Scopus, and Web of Science from database inception to April 25, 2025. Studies that developed ML-based models to predict EGFR-TKI response were included.
Eight studies involving 1322 LCBM patients were included. The included studies used logistic regression (LR), LR with least absolute shrinkage and selection operator (LASSO), decision tree (DT), and a Cox-based deep learning model (DL-Cox). The meta-analysis revealed a pooled area under the curve (AUC) of 0.84 (95% CI 0.78-0.91) and accuracy (ACC) of 0.75 (95% CI 0.62-0.88) with a sensitivity (SEN) of 0.82 (95% CI 0.77-0.87) and a specificity (SPE) of 0.73 (95% CI 0.66-0.80) for prediction of EGFR-TKI response. The meta-analysis of diagnostic odds ratios (DOR) exhibited a pooled DOR of 12.41 (95% CI 7.32-21.04).
ML-based models show promising ability to predict EGFR-TKI response in LCBM, supporting their potential to guide treatment selection. However, their use in clinical practice remains limited by small retrospective datasets and lack of external validation.
A comprehensive literature search was conducted using PubMed, Embase, Scopus, and Web of Science from database inception to April 25, 2025. Studies that developed ML-based models to predict EGFR-TKI response were included.
Eight studies involving 1322 LCBM patients were included. The included studies used logistic regression (LR), LR with least absolute shrinkage and selection operator (LASSO), decision tree (DT), and a Cox-based deep learning model (DL-Cox). The meta-analysis revealed a pooled area under the curve (AUC) of 0.84 (95% CI 0.78-0.91) and accuracy (ACC) of 0.75 (95% CI 0.62-0.88) with a sensitivity (SEN) of 0.82 (95% CI 0.77-0.87) and a specificity (SPE) of 0.73 (95% CI 0.66-0.80) for prediction of EGFR-TKI response. The meta-analysis of diagnostic odds ratios (DOR) exhibited a pooled DOR of 12.41 (95% CI 7.32-21.04).
ML-based models show promising ability to predict EGFR-TKI response in LCBM, supporting their potential to guide treatment selection. However, their use in clinical practice remains limited by small retrospective datasets and lack of external validation.
Authors
Hajikarimloo Hajikarimloo, Mohammadzadeh Mohammadzadeh, Hashemi Hashemi, Tos Tos, Bahrami Bahrami, Najari Najari, Ebrahimi Ebrahimi, Hasanzade Hasanzade, Habibi Habibi
View on Pubmed