Predictive machine learning algorithms for depression and anxiety disorders in six cancer types: a comprehensive...

Area of Science:

• Oncology
• Psychiatry
• Data Science

Background:

• Machine learning (ML) applications in medical diagnosis are extensive, but predicting post-cancer mental health issues like depression and anxiety remains under-explored.
• Cancer patients face significant risks for developing depression and anxiety, impacting treatment adherence and overall quality of life.

Purpose of the Study:

• To develop and validate ML models for predicting depression and anxiety disorders in cancer patients within one year of diagnosis.
• To identify key demographic, clinical, and quality of care factors contributing to post-cancer depression and anxiety.

Main Methods:

• A longitudinal, cross-institutional study involving 24,580 cancer patients across three medical centers in Taiwan (2017-2022).
• Development and comparison of multiple ML algorithms including Logistic Regression, Random Forest, K-Nearest Neighbor, Adaptive Boosting, and Extreme Gradient Boosting (XGBoost).
• Model performance was assessed using accuracy, precision, recall, F1 score, and AUROC, with statistical analysis via SPSS and Python.

Main Results:

• The XGBoost model demonstrated superior performance, achieving 98.30% accuracy, 98.17% precision, 98.30% F1 score, and 99.72% AUROC.
• Feature importance analysis highlighted tumor size, patient age, and body mass index as the most significant predictors of depression and anxiety.

Conclusions:

• ML models effectively predict depression and anxiety in cancer patients using longitudinal data, offering valuable insights into risk factors.
• These predictive models can enhance mental health care, inform evidence-based guidelines, and improve patient outcomes throughout cancer treatment.