Artificial Intelligence Models to Predict Recurrence Risk Prediction in Early-Stage Non-Small Cell Lung Cancer: A...

Area of Science:

• Oncology
• Biostatistics
• Medical Informatics

Background:

• Early-stage non-small cell lung cancer (NSCLC) recurrence poses a significant clinical challenge.
• Accurate prediction of postoperative recurrence is crucial for personalized treatment strategies.

Purpose of the Study:

• To systematically evaluate predictive models for early-stage NSCLC recurrence risk.
• To assess the impact of integrating diverse data modalities on model performance.

Main Methods:

• A systematic literature search was conducted across PubMed, Embase, and Web of Science.
• Seventeen studies were included, with data extraction on study characteristics, data types, and performance metrics.
• Risk of bias was evaluated using the Prediction model Risk of Bias Assessment Tool (PROBAST)+AI.

Main Results:

• Random forest and random survival forest models showed robustness with single-modality data.
• Multimodality data integration significantly enhanced model performance (AUC 0.72-0.94).
• DeepRePath (XGBoost) achieved an AUC of 0.94 in pathological image analysis; graph neural networks performed well on CT data (AUC 0.785).
• Overfitting was a common issue, with several studies showing high bias risk in development and validation phases.

Conclusions:

• Predictive models demonstrate potential for accurate recurrence risk assessment in early-stage NSCLC.
• Multimodal data integration is key to improving the generalizability and predictive power of these models.