Predicting cancer mortality using machine learning methods: a global vs. Iran analysis.

Area of Science:

• Oncology
• Data Science
• Bioinformatics

Background:

• Cancer poses a significant global health challenge with varying outcomes worldwide.
• Regional disparities in cancer incidence, mortality, and survival necessitate localized analysis.
• Machine learning (ML) offers a powerful approach to analyze complex cancer data for improved predictions.

Purpose of the Study:

• To enhance the predictive accuracy of cancer mortality using ML models.
• To compare ML model performance on global versus Iran-specific cancer datasets.
• To investigate ML's utility in predicting Second Primary Cancer (SPC) risk.

Main Methods:

• Utilized datasets from Global Cancer Observatory (GLOBOCAN) and Iran National Cancer Registry (INCR).
• Evaluated XGBoost, Random Forest, and Support Vector Machines for cancer outcome prediction.
• Assessed ML model performance using metrics like [Formula: see text] and AUC-ROC.

Main Results:

• XGBoost demonstrated superior predictive performance globally ([Formula: see text] = 0.83, AUC-ROC = 0.93) compared to Iran-specific data ([Formula: see text] = 0.79, AUC-ROC = 0.89).
• Identified region-specific risk factors, such as Helicobacter pylori in Ardabil, influencing cancer outcomes.
• Key predictors for Second Primary Cancer (SPC) risk include radiation dose, age, and genetic mutations.

Conclusions:

• ML holds significant potential for personalized cancer treatment planning and improved patient care.
• Addressing data imbalances and regional disparities is crucial for effective ML implementation in oncology.
• Findings offer valuable insights for policymakers and healthcare providers to reduce the global cancer burden.