Early thyroid-stimulating hormone dynamics as an independent predictor of immune-related hypothyroidism during nivolumab therapy

Immune checkpoint inhibitors (ICIs) have transformed the treatment landscape of advanced solid tumors, yet immune-related adverse events remain a major challenge. Thyroid dysfunction is among the most frequent endocrine toxicities associated with programmed cell death-1 (PD-1) inhibitors, often evolving dynamically and remaining early immune-mediated thyroid injury. Identifying an early and reliable biomarker for immune-related hypothyroidism could improve patient monitoring and treatment safety. This retrospective, single-center cohort study included 214 adult patients with solid tumors treated with nivolumab between January 2020 and January 2025. Thyroid function tests obtained at baseline and at the first month of treatment were analyzed. Early thyroid-stimulating hormone change (ΔTSH) was defined as change from baseline, calculated as first-month TSH minus baseline TSH. The primary endpoint was the development of de-novo or deteriorated hypothyroidism. Time-to-event analyses were performed using Cox proportional hazards regression. The discriminative performance of ΔTSH was assessed using receiver operating characteristic (ROC) curve analysis, with internal validation using 2,000 bootstrap resamples. During follow-up, nivolumab-related thyroid adverse events occurred in 37 patients (17.3%), with de novo or deteriorated hypothyroidism observed in 27 patients. Median time to hypothyroidism was 12 weeks (IQR: 8–24). In multivariable Cox regression, ΔTSH remained independently associated with subsequent hypothyroidism (adjusted HR per 1 mU/L increase: 1.060; 95% CI: 1.036–1.085; p < 0.001), together with age. Exploratory ROC analysis demonstrated high apparent discrimination of ΔTSH for subsequent hypothyroidism (AUC 0.952, 95% CI 0.925–0.979). Internal validation using 2,000 bootstrap resamples showed a mean AUC of 0.952 (95% bootstrap CI: 0.922–0.976). Early ΔTSH was associated with subsequent immune-related hypothyroidism in this retrospective cohort. However, the modest sample size of 214 patients and the limited number of hypothyroidism events ( n = 27) mean that the observed association and apparent discriminative performance should be regarded as preliminary, exploratory evidence. Bootstrap resampling does not overcome these data limitations. Larger cohorts with more outcome events are needed to confirm predictive performance before clinical application. Not applicable.