Predicting severe toxicity after head-and-neck cancer RT: Validation of the synergist role of a biological marker and dosimetry

To assess the value of biologically enriched Normal Tissue Complication Probability (NTCP) models integrating clinical, dosimetric, and biological markers, specifically the RadioDtect© assay based on phosphorylated ATM (Ataxia-Telangiectasia Mutated, pATM) quantification.

We considered 67 patients with head-and-neck cancer receiving curative-intent radiotherapy. We developed logistic NTCP models for severe acute/late toxicities (grade ⩾3) using (i) the pATM-based RadioDtect test, (ii) clinical-dosimetric variables, and (iii) a combined model. Model performance was evaluated using AUC (Area Under the receiver operating characteristic Curve), calibration, and Net Reclassification Improvement (NRI), with internal validation via bootstrapping/permutation.

Acute and late toxicity occurred in 70% and 15% of patients. The RadioDtect test alone yielded moderate performance for late toxicity (AUC=0.65) and low discriminatory power for acute toxicity (AUC=0.57). Clinical-dosimetric models for acute toxicity demonstrated substantial predictive value (AUC=0.77), primarily driven by doses to the pharyngeal constrictor muscles and parotid glands. Integrating RadioDtect modestly improved discrimination for acute toxicity (AUC=0.82) but added no clinical utility (NRI=0%). Conversely, for late toxicity, doses showed poor association and discrimination for the occurrence of severe side effects. The biological refined NTCP model showed improved discrimination (AUC=0.76) and meaningful clinical utility (NRI=46.3%).

The RadioDtect assay adds limited value for predicting acute toxicity in clinical settings entailing large volumes of organs at risk irradiated at high doses, but enhances NTCP models for late toxicity prediction.