External validation of a normal tissue complication probability model for radiation-induced hypothyroidism in an...

A normal tissue complication probability (NTCP) model for radiation-induced hypothyroidism (RIHT) was previously derived in patients with squamous cell carcinoma of the head and neck (HNSCC) discerning thyroid volume (Vthyroid), mean thyroid dose (Dmean), and latency as predictive factors. The purpose of this study was to test the performance of this model in an independent cohort of patients receiving primary radiotherapy (RT) for HNSCC.

A validation cohort of 198 patients with HNSCC was included after plasma thyrotropin (TSH) assessment. RIHT was defined as TSH > 4.0 mU/l from blood samples obtained during follow-up. A new mixture NTCP model was developed from the validation cohort after multivariable analysis. Due to only one follow-up TSH assessment in the validation cohort, the time factor derived from the original cohort was fixed in a mixture model and applied for the NTCP validation. Association between model predictions of the initial model and observed clinical outcome in the validation cohort was investigated by applying the previous model (Vthyroid, Dmean and time) on the new cohort and comparing it to the clinical outcome.

Both Dmean and Vthyroid were confirmed as significant risk factors for RIHT in the validation cohort, odds ratio (OR) 1.19 (1.1-1.37) and OR 0.75 (0.57-0.9), respectively. A small difference in overall probability of RIHT was observed between the cohorts, further analysis indicated this to be related to less frequent blood tests in the validation cohort relative to the original cohort. However, Pearson's correlation coefficients between model and clinical outcome were high: r = 0.97 estimated by the original model versus the original cohort, and r = 0.97 estimated by the original model versus the new cohort.

Dmean and Vthyroid were significant predictors of RIHT in both cohorts. The original NTCP model demonstrated external validity owing to high Pearson's correlation coefficients between estimated and observed incidence rates of RIHT in the original as well as in the validation cohort. This model may facilitate clinically relevant estimations of RIHT after RT to the neck.