Multimodal survival prediction in advanced pancreatic cancer using machine learning

Area of Science:

• Oncology
• Machine Learning
• Prognostics

Background:

• Current risk scores are inadequate for advanced pancreatic ductal adenocarcinoma (PDAC) survival assessment.
• Existing methods fail to utilize comprehensive clinical data for individual patient risk stratification.

Purpose of the Study:

• To develop and validate a machine learning model for predicting survival risk in advanced PDAC.
• To assess the utility of clinical, imaging, and molecular parameters in PDAC prognostication.

Main Methods:

• A random survival forest model was constructed using clinical data from 203 advanced PDAC patients.
• Key parameters included age, CA19-9, C-reactive protein, metastatic status, neutrophil-to-lymphocyte ratio, and total serum protein.
• Subgroup analyses incorporated imaging and molecular data, including KRAS mutational status.

Main Results:

• The clinical parameter model achieved a c-index of 0.71, outperforming AJCC staging and mGPS.
• External validation in a liver metastases cohort yielded a c-index of 0.67.
• KRAS p.G12D mutation improved prediction; primary tumor radiomics showed minimal benefit.

Conclusions:

• Multimodal data combined with machine learning show promise for personalized prognostication in advanced PDAC.
• This approach enables more accurate individual risk assessment at the time of diagnosis.