Machine learning application in personalised lung cancer recurrence and survivability prediction

Area of Science:

• Oncology
• Bioinformatics
• Artificial Intelligence

Background:

• Machine learning (ML) is increasingly used for cancer prognosis prediction, aligning with personalized medicine.
• Reliable ML models for predicting cancer outcomes in clinical settings remain a challenge.

Purpose of the Study:

• To develop and compare ML models for predicting recurrence and survivability in lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC).
• To identify key genomic, clinical, and demographic predictors of cancer outcomes.

Main Methods:

• Integrated genomic, clinical, and demographic data from The Cancer Genome Atlas (TCGA) for LUAD and LUSC patients.
• Incorporated copy number variation (CNV) and mutation data for 15 selected genes.
• Compared three ML algorithms: decision tree methods, neural networks, and support vector machines.

Main Results:

• Decision tree models, while not superior in accuracy, effectively identified important predictors of recurrence and survivability.
• Key predictors included genomic information (e.g., KRAS, EGFR, TP53), clinical status (TNM stage, radiotherapy), and demographics (age, gender).
• These findings apply to both early-stage LUAD and LUSC.

Conclusions:

• ML models show potential in assisting clinicians with personalized patient management, including follow-up schedules and social care planning.
• Identifying crucial predictors enhances the understanding of factors influencing lung cancer prognosis.