Statistical validation of normal tissue complication probability models.

Area of Science:

• Radiation Oncology
• Biostatistics
• Medical Physics

Background:

• Normal tissue complication probability (NTCP) models are essential for predicting adverse events after radiation therapy.
• Validating these models is critical for ensuring patient safety and optimizing treatment plans.
• Established statistical approaches are needed for robust NTCP model assessment.

Purpose of the Study:

• To evaluate the effectiveness of double cross-validation and permutation tests for NTCP model validation.
• To determine the value of these statistical methods in assessing model reliability.
• To investigate their applicability in the context of head-and-neck cancer radiation therapy.

Main Methods:

• Utilized the LASSO (least absolute shrinkage and selection operator) penalized regression method to construct NTCP models.
• Focused on predicting xerostomia incidence following head-and-neck cancer radiation treatment.
• Employed likelihood function and area under the receiver operating characteristic curve (AUC) for model performance evaluation.

Main Results:

• Repeated double cross-validation highlighted the inherent uncertainty and instability of the developed NTCP models.
• Permutation testing demonstrated its capability to establish the statistical significance of model performance.
• These findings underscore the importance of rigorous validation techniques.

Conclusions:

• Repeated double cross-validation and permutation tests are strongly recommended for validating NTCP models.
• Implementing these methods prior to clinical deployment enhances model trustworthiness.
• This approach contributes to safer and more effective radiation therapy practices.