Lung Cancer Survival Estimation Using Data from Seven German Cancer Registries.

Area of Science:

• Oncology
• Medical Informatics
• Biostatistics

Background:

• Accurate prediction of cancer patient survival is vital for treatment planning and evaluation.
• Lung cancer prognosis remains a significant challenge in clinical oncology.
• Leveraging electronic health records and registry data can improve survival prediction models.

Purpose of the Study:

• To predict short-term (6, 12, 18, 24 months) survival probabilities for lung cancer patients.
• To identify key prognostic factors influencing lung cancer patient survival using machine learning.
• To enhance the interpretability of survival prediction models in oncology.

Main Methods:

• Utilized lung cancer data from seven German cancer registries.
• Applied data integration and preprocessing techniques.
• Employed a gradient boosting algorithm for survival prediction and permutation feature importance for model interpretability.

Main Results:

• The gradient boosting model successfully predicted patient survival at 6, 12, 18, and 24 months.
• Age at diagnosis and the presence of distant metastases were identified as the most impactful features for long-term survival.
• Permutation feature importance provided insights into the model's decision-making process.

Conclusions:

• Machine learning models, particularly gradient boosting, can effectively predict lung cancer survival.
• Age and metastatic status are critical determinants of long-term survival in lung cancer patients.
• Identified prognostic factors can inform future multivariate survival analyses and clinical decision-making.