Modeling the Pretest Probability of Identifying Druggable Mutations in Lung Cancer Using Nationwide Comprehensive...

Area of Science:

• Oncology
• Genomics
• Artificial Intelligence

Background:

• Comprehensive genomic profiling (CGP) is crucial for precision medicine in lung cancer.
• Clinical implementation of CGP is limited by uncertainty in identifying actionable mutations.
• Predicting the probability of actionable mutations before CGP is needed.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI)-based tool to predict the pretest probability of identifying druggable mutations before CGP.
• To assess the clinical utility of AI in guiding CGP decisions for lung cancer patients.

Main Methods:

• Developed an eXtreme Gradient Boosting (XGBoost) prediction model using pre-CGP clinical variables from 3,470 lung cancer patients.
• Employed explainable artificial intelligence (XAI) to identify key predictors.
• Validated the model in an independent cohort of 1,307 patients using Brier score.

Main Results:

• The AI model achieved an AUROC of 0.85 in the overall validation cohort.
• Key predictors included sex, smoking history, histology, and metastatic sites.
• The model demonstrated good performance in predicting druggable mutations, with Brier scores of 0.19 and 0.16 in independent test cohorts.

Conclusions:

• An AI-based tool utilizing pre-CGP clinical data can effectively predict the probability of identifying druggable mutations.
• This AI tool has the potential to broaden CGP implementation in lung cancer.
• Improved CGP implementation can enhance patient access to targeted therapies.