Thyroid dysfunction after PD-1 inhibitor therapy: selection into post-treatment monitoring and thyroid-axis trajectories in a single-centre real-world cohort study

Background Laboratory surveillance affects whether thyroid dysfunction after immune checkpoint inhibitor (ICI) therapy is detected. Incomplete or symptom-driven retesting can leave events unrecorded and complicate comparisons between real-world studies. We examined observed thyroid events, monitoring selection and exploratory baseline associations. Methods This single-centre retrospective study identified 2,634 ICI-exposed patients; 812 had normal TSH and FT4 in the most recent qualifying pre-infusion pair during Days − 90 to 0 and met the remaining baseline criteria, and 627 had 365 days of administrative follow-up opportunity. The primary cohort required a TSH retest within Days 1–365 ( n = 240). Clinically adjudicated events were analysed using exploratory logistic regression, Benjamini–Hochberg (BH) adjustment and a limited-adjustment Firth model. Inverse probability weighting (IPW) standardized recorded-event proportions for measured baseline differences. Sensitivity analyses required paired retesting or baseline reports strictly before the ICI calendar date. Results Of 627 baseline-eligible patients with 365-day administrative follow-up opportunity, 240 (38.3%) had a qualifying TSH retest within Days 1–365. Thirty recorded events occurred among these 240 patients (12.5%; Wilson 95% CI 8.9%–17.3%), an observed detection proportion under incomplete monitoring. Among 210 patients with no recorded event, 196 (93.3%) had their last available thyroid test before Day 365. Baseline-window antibody positivity was associated with recorded events (OR 8.73, 95% CI 3.44–22.14; BH q < 0.001), as was higher baseline TSH (OR per 1 mIU/L 1.69, 95% CI 1.25–2.27; q = 0.0027). Both associations persisted in the limited-adjustment Firth analysis. Capped IPW yielded 12.6% (bootstrap 95% CI 8.5%–17.1%). The paired-retest analysis recorded 30/239 events; the strictly prior-calendar-day baseline analysis recorded 22/171 events. Conclusions Approximately one in eight patients monitored within one year had a recorded study-defined event. This is not a fully ascertained one-year incidence estimate. Baseline antibodies and TSH showed exploratory associations, which may partly reflect differential detection. Real-world thyroid toxicity studies should report monitoring coverage, laboratory timing and explicit outcome definitions. Clinical trial number Not applicable.