Immune-Related Thyroid Dysfunction in PD-L1 High Non-Oncogene-Addicted NSCLC Treated with First-Line Pembrolizumab: Incidence, Timing, and Predictive Impact

In metastatic NSCLC with high PD-L1 expression (TPS ≥ 50%), pembrolizumab monotherapy yields durable benefit in a subset of patients. Immune-related thyroid dysfunction (irTD) is common during PD-1/PD-L1 blockade, but its predictive value remains uncertain. We conducted a retrospective, single-center study including 363 patients with metastatic NSCLC, PD-L1 TPS ≥ 50%, and no actionable oncogenic drivers treated with first-line pembrolizumab. Thyroid function tests were performed at baseline and every six weeks. irTD was defined based on laboratory abnormalities with or without clinical symptoms and classified as early onset (≤90 days) or late onset (>90 days). Progression-free survival (PFS) was estimated using Kaplan–Meier methods and compared using log-rank tests. Cox proportional hazards models included irTD as a time-varying covariate. Landmark analyses at 3 and 6 months reduced immortal-time bias. Events were graded according to CTCAE v5.0. Among 363 eligible patients, irTD occurred in 110 (30.3%); median onset was 114 days (range 21–550). Median cohort PFS was 9.8 months (95% CI 7.26–12.34). Patients with irTD had significantly longer PFS than those without irTD: 26.33 (95% CI 19.09–33.57) vs. 6.16 months (95% CI 4.70–7.63), with an HR of 0.378 (95% CI 0.280–0.511; p < 0.001). Landmark analyses confirmed benefit at 3 months (28.4 vs. 13.7 months; HR 0.490, p < 0.001) and 6 months (29.0 vs. 20.5 months; HR 0.587, p < 0.001). PFS did not differ by irTD timing (early vs. late; HR 0.926, p = 0.682). Poor ECOG PS (≥2) was associated with worse outcomes and a lower incidence of irTD. We found that irTD is common, clinically manageable, and strongly associated with improved PFS in PD-L1-high metastatic NSCLC treated with pembrolizumab. Thyroid dysfunction may serve as a feasible on-treatment biomarker of effective immune activation, warranting further prospective validation.