Impact of statistical learning methods on the predictive power of multivariate normal tissue complication probability...

Area of Science:

• Oncology
• Radiotherapy
• Biostatistics

Background:

• Accurate prediction of normal tissue complication probability (NTCP) is crucial for optimizing radiotherapy.
• Multivariate NTCP models aid in predicting treatment toxicity.
• The choice of statistical learning method can influence model performance.

Purpose of the Study:

• To evaluate the predictive performance of different statistical learning methods for multivariate NTCP models.
• To compare stepwise selection, LASSO, and BMA in predicting xerostomia after head and neck cancer radiotherapy.

Main Methods:

• Development of NTCP models for xerostomia using stepwise selection, LASSO, and Bayesian model averaging (BMA).
• Radiotherapy treatment for head and neck cancer patients was the clinical context.
• Repeated cross-validation was employed for robust performance evaluation and comparison.

Main Results:

• LASSO and BMA methods demonstrated significantly better predictive power compared to stepwise selection.
• The LASSO method produced models that were interpretable, similar to stepwise selection.
• BMA models, while predictive, were less intuitive to interpret than LASSO or stepwise models.

Conclusions:

• Standard stepwise selection may be inadequate for developing effective NTCP models.
• The LASSO method offers a recommended approach for NTCP modeling due to its balance of predictive accuracy and interpretability.
• Further investigation into advanced statistical learning techniques is warranted for improved radiotherapy outcome prediction.