A comprehensive review of 592 prediction models designed to estimate radiation-induced side effects in patients with head and neck cancer found that only a small proportion demonstrated acceptable methodological quality. Of these, just nine models were externally validated at least twice, with moderate discriminative performance for predicting xerostomia, dysphagia, hypothyroidism, and temporal lobe injury. However, calibration data were inconsistently reported, and many validation studies showed high risk of bias, mainly due to small sample sizes and analytical limitations.
Across more than 140,000 patients in developmental cohorts and over 34,000 in external validations, the analysis showed that most published NTCP models lack robust evidence needed for clinical implementation. While a few models—such as those predicting xerostomia and temporal lobe injury—performed acceptably, overall data highlight a need for more high-quality external validations and improved reporting standards before these tools can be reliably used in treatment planning.