Genetic polymorphisms and early responses predict refractory hyperthyroidism during antithyroid drug therapy.

Antithyroid drugs (ATDs) are the first-line treatment for Graves' disease (GD), but variable outcomes require early risk stratification. We analyzed 578 adults, including 240 in a randomized trial of dose titration versus modified partial block-and-replace therapy and the remainder in a temporal observational cohort. Baseline characteristics, 3-month treatment indicators, and 13 GD-susceptibility single nucleotide polymorphisms (SNPs) were evaluated. An ATD-levothyroxine dose-correction method quantified thyroid function and autoantibody changes, informing clinical and SNP risk score-integrated models. The integrated model achieved areas under the curve (AUCs) of 0.841 in development and 0.805 in temporal validation, outperforming the clinical model (p < 0.001) with good calibration and net benefit. Intermediate-risk patients derived the greatest incremental benefit. Baseline free thyroxine (FT4) was the strongest predictor, and the SNP risk score ranked fourth. No inter-regimen efficacy or safety differences were observed. Restricted cubic spline analysis suggested that ATD therapy beyond 3 years may reduce recurrence risk, whereas modified partial block-and-replace showed no advantage over dose titration.