Background
Immune checkpoint inhibitors have become one of the fundamental therapeutic approaches in contemporary oncology by significantly prolonging survival in patients with advanced solid tumors. The programmed cell death-1 (PD-1) inhibitor nivolumab has increased durable response rates across a variety of malignancies; however, through the deliberate suppression of immune tolerance, it also introduces a distinct spectrum of toxicities. These immune-related adverse events (irAEs) arise via pathophysiological mechanisms that differ from those of conventional cytotoxic therapies and, when not recognized at an early stage, may negatively impact treatment continuity and patient safety [1, 2].
Among irAEs, the endocrine system is one of the most frequently affected organ systems, with thyroid gland involvement being of particular clinical importance due to both its high incidence and heterogeneous clinical course [3]. Nivolumab-related thyroid dysfunction most commonly arises on the basis of T cell–mediated destructive thyroiditis, encompassing a broad clinical spectrum that may progress from a transient hyperthyroid phase to permanent hypothyroidism [4]. Notably, in many patients this process initially evolves in the presence of normal thyroid function tests and passes through an early immune-mediated thyroid injury before clinically overt dysfunction becomes apparent [5].
This characteristic trajectory renders the early prediction of thyroid-related irAEs clinically critical.
In contrast, the existing literature on the prediction of immunotherapy-related thyroid dysfunction has predominantly focused on the absolute values of thyroid function tests [6]. Osorio et al. demonstrated that thyroid dysfunction in patients treated with immune checkpoint inhibitors is primarily defined on the basis of biochemical abnormalities identified in the absolute levels of TSH and free thyroxine (fT4) [7]. Similarly, Iyer et al. reported that the clinical diagnosis of thyroid dysfunction in patients receiving immune checkpoint inhibitors largely relies on absolute TSH and fT4 values and that a substantial proportion of these cases ultimately result in permanent hypothyroidism [8].
However, considering that immune-mediated tissue injury represents a biologically dynamic process, it is evident that single time-point measurements are insufficient to adequately capture this evolution.
TSH is the most sensitive feedback component of the hypothalamic–pituitary–thyroid axis and is capable of exhibiting early changes even under conditions of minimal thyroidal stress [9]. This physiological characteristic suggests that early alterations in TSH may provide a measurable biological signal preceding the development of clinically overt hypothyroidism.
In this context, changes in thyroid-stimulating hormone occurring after treatment initiation (ΔTSH) have been proposed as a potential biomarker for predicting the development of immune-related hypothyroidism (ir-hypothyroidism). However, studies systematically evaluating the predictive performance of ΔTSH remain limited, and the available evidence lacks generalizability to clinical practice due to substantial methodological heterogeneity. This highlights the ongoing need for an accessible, reliable, and easily applicable biomarker that enables early and accurate risk stratification in patients receiving immunotherapy.
In this study, we aimed to determine the predictive value of early changes in thyroid-stimulating hormone (ΔTSH) for the development of immune-related hypothyroidism (ir-hypothyroidism) in patients receiving nivolumab therapy. In addition, we evaluated whether ΔTSH represents an independent risk determinant in multivariable models after adjustment for clinical and laboratory parameters. Overall, this study seeks to provide a dynamic biomarker-based approach that may facilitate the early identification of immunotherapy-related thyroid toxicities and can be readily integrated into routine clinical practice.
Methods
Study design and population
This study was designed as a retrospective, single-center cohort study conducted to evaluate the predictive value of early changes in thyroid-stimulating hormone (ΔTSH) for the development of immune-related thyroid dysfunction in patients with solid tumors treated with nivolumab. Adult patients who were consecutively initiated on nivolumab therapy at the Medical Oncology Clinic of our institution between January 2020 and January 2025 were screened. Nivolumab was administered every 2 weeks in all patients included in the study.
A total of 214 patients who met the predefined inclusion and exclusion criteria constituted the study population. The patient selection process and the formation of the study cohort are illustrated in Fig. 1.
Inclusion criteria were defined as age ≥ 18 years, initiation of nivolumab therapy for systemic antineoplastic treatment, availability of baseline thyroid-stimulating hormone (TSH) and free thyroxine (fT4) measurements obtained prior to treatment initiation, and the presence of repeated TSH and fT4 measurements at the first month of therapy. In addition, complete access to demographic characteristics, comorbidities, and laboratory data in the electronic medical records was required for study inclusion.
Patients without available baseline or first-month thyroid function tests, those who received only a single dose of nivolumab, and patients with incomplete follow-up data due to a change in treatment center during the course of therapy were excluded from the study. In addition, patients who experienced acute conditions or underwent interventions within the first month that could affect thyroid function—such as parenteral iodine administration or high-dose systemic corticosteroid therapy—were not included in the analyses. Malignancies with distinctly different biological response profiles to immunotherapy (e.g., non-Hodgkin lymphoma) were also excluded from the study.
Data collection and definitions
Demographic characteristics, comorbidities, tumor histologies, previously administered systemic treatment lines, laboratory parameters, and information regarding immune-related thyroid adverse events were retrospectively obtained from electronic medical records. This retrospective observational study was conducted in accordance with the principles of the Declaration of Helsinki, and its reporting followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.
As this was a retrospective cohort study, no a priori sample-size calculation was performed. The study sample comprised all consecutive patients who met the predefined eligibility criteria during the study period.
Laboratory assessments were based on measurements obtained prior to nivolumab treatment (baseline) and at the first month of therapy. Changes in thyroid-stimulating hormone (ΔTSH) were calculated as the first-month TSH concentration minus the baseline TSH concentration. Accordingly, positive ΔTSH values indicated an increase in TSH, whereas negative values indicated a decrease.
TSH and fT4 were measured in the clinical biochemistry laboratory of İzmir Atatürk Training and Research Hospital using direct chemiluminescence immunoassays on the ADVIA Centaur XP platform (Siemens Diagnostics, Tarrytown, USA). The laboratory reference intervals were 0.35–5.30 mU/L for TSH and 0.58–1.38 ng/dL for fT4. Blood samples were obtained during routine clinical care.
Uniform fasting status and blood collection times could not be verified retrospectively. Cohort-specific analytical imprecision was not quantified. Immune checkpoint inhibitor–related hypothyroidism was classified into three clinical categories in accordance with current guidelines: de novo primary hypothyroidism, deterioration of pre-existing hypothyroidism, and central hypothyroidism secondary to hypophysitis [10, 11].
For analytical purposes, the primary outcome event was defined as the development of hypothyroidism (de novo or deteriorated).
Hypothyroidism was defined biochemically as a serum TSH concentration above the upper limit of the laboratory reference range accompanied by a free T4 (fT4) concentration below the lower limit of the laboratory reference range. Isolated TSH elevation with a normal fT4 concentration was not considered an outcome event. Patients with pre-existing hypothyroidism who developed elevated TSH levels together with low fT4 concentrations during nivolumab treatment despite continued thyroid hormone replacement therapy at unchanged doses were classified as having deteriorated hypothyroidism.
Central hypothyroidism was considered separately from primary hypothyroidism. One patient developed central hypothyroidism in the setting of hypophysitis and panhypopituitarism, characterized by low fT4 with an inappropriately low TSH concentration. This patient was not included in the regression analyses evaluating predictors of primary hypothyroidism.
Accordingly, the regression analyses included patients who developed de novo or deteriorated primary hypothyroidism.
Thyroid function parameters were evaluated regardless of baseline thyroid status, and the response of the hormonal axis to immunotherapy was analyzed in a comprehensive manner. To capture early hormonal dynamics, first-month measurements were selected as the primary time point of interest. Time to hypothyroidism was calculated as the interval between the initiation date of nivolumab therapy and the date on which hypothyroidism was diagnosed.
Study endpoints
The primary endpoint of the study was the association between early ΔTSH and the subsequent development of nivolumab-related hypothyroidism. Secondary analyses included assessment of the discriminative ability of ΔTSH for subsequent hypothyroidism, time to hypothyroidism, and exploratory assessment of overall mortality.
Importantly, no de novo hypothyroidism events occurred before completion of the first-month thyroid function assessment; the earliest de novo hypothyroidism event was observed at week 5.
Statistical analysis
All statistical analyses were performed using IBM SPSS Statistics for Windows, Version 24.0 (IBM Corp., Armonk, NY, USA). The distribution of continuous variables was assessed using the Kolmogorov–Smirnov test; normally distributed data were reported as mean ± standard deviation, whereas non-normally distributed data were expressed as median (interquartile range, IQR). Baseline and first-month laboratory values were compared using paired t-tests or Wilcoxon signed-rank tests, as appropriate.
The development of hypothyroidism was treated as a time-dependent event, and potential risk factors were evaluated using Cox proportional hazards regression models. Age, sex, Charlson Comorbidity Index, and ΔTSH were included a priori in the multivariable model; model assumptions were assessed prior to analysis. Results were reported as hazard ratios (HRs) with 95% confidence intervals (CIs).
To assess the internal stability of the multivariable Cox model, bootstrap resampling with 2,000 samples was performed, and bias-corrected and accelerated (BCa) 95% confidence intervals were calculated.
The discriminative ability of ΔTSH for subsequent hypothyroidism was explored using ROC curve analysis. The AUC with 95% CI was calculated, and an exploratory threshold was identified using the Youden index. The internal stability of the discriminative performance of ΔTSH was assessed using 2,000 bootstrap resamples.
Bootstrap estimates of the AUC and the distribution of ROC-derived thresholds were examined to assess the robustness of discrimination and threshold stability. Overall survival was analyzed using the Kaplan–Meier method, and survival curves were compared using the log-rank test. A two-sided p value < 0.05 was considered statistically significant for all analyses.
Results
Patient characteristics
A total of 214 patients receiving nivolumab therapy were included in the study. The mean age of the study population was 64.9 ± 9.4 years, and 78.5% of the patients were male (n = 168). The median body mass index (BMI) was 24.2 (range: 13.8–48.4).
The median follow-up duration was 96 weeks (interquartile range [IQR]: 60–188).
The majority of patients had a good performance status: 167 patients (78.0%) had an Eastern Cooperative Oncology Group (ECOG) performance status of 0, 45 patients (21.1%) had an ECOG score of 1, and only 2 patients (0.9%) were classified as ECOG 2. No patients with an ECOG performance status greater than 2 were identified. The mean Charlson Comorbidity Index (CCI) score was 9.3 ± 1.7.
The distribution of histopathological diagnoses was heterogeneous, with lung squamous cell carcinoma (n = 68, 31.8%) and lung adenocarcinoma (n = 55, 25.7%) being the most common malignancies. These were followed by malignant melanoma (n = 43, 20.0%) and renal cell carcinoma (n = 40, 18.7%). Less frequent diagnoses included multiple primary malignancies (n = 6, 2.8%), small cell lung carcinoma (n = 1, 0.5%), and head and neck squamous cell carcinoma (n = 1, 0.5%).
Comorbid conditions were common in the study population. The most frequently observed comorbidities were hypertension (n = 95, 44.4%), chronic obstructive pulmonary disease (n = 57, 26.6%), diabetes mellitus (n = 56, 26.2%), coronary artery disease (n = 46, 21.5%), and dyslipidemia (n = 37, 17.3%). Pre-existing thyroid disease was identified in 23 patients (10.7%), including hypothyroidism in 20 patients (9.3%) and hyperthyroidism in 3 patients (1.4%).
Prior to nivolumab initiation, the majority of patients had received one line of systemic therapy (n = 148, 69.2%); 21.0% (n = 45) had received two lines, 7.9% (n = 17) three lines, and 1.4% (n = 3) more than three lines of systemic treatment. One patient (0.5%) received nivolumab without prior systemic therapy.
The median duration of nivolumab treatment was 17 weeks (IQR: 11–42). During follow-up, nivolumab-related thyroid toxicity was observed in 37 patients (17.3%), disease progression occurred in 103 patients (48.1%), and a total of 147 patients (68.7%) died over the follow-up period. The baseline demographic, clinical, and treatment-related characteristics of the study population are summarized in Table 1.
Table 1Baseline demographic, clinical, and treatment-related characteristics of the study populationVariablesTotal (n = 214)Age, mean ± SD64.9 ± 9.4Sex, n (%)MaleFemale168 (78.5)46 (21.5)BMI, median (min-max)24.2 (13.8–48.4)Observation duration, weeks, median (IQR)96 (60–188)ECOG scale, n (%)012> 2167 (78.0)45 (21.1)2 (0.9)NoneCCI, mean ± SD9.3 ± 1.7Histopathological types, n (%)Lung adenocarcinomaLung squamous cell carcinomaLung small cell carcinomaRenal cell carcinomaMalign melanomaMultiple primary malignancyHead-neck squamous cell carcinoma55 (25.7)68 (31.8)1 (0.5)40 (18.7)43 (20.0)6 (2.8)1 (0.5)Comorbidities, n (%)HypertensionDiabetes mellitusDyslipidemiaCoronary artery diseaseAtrial fibrillationCongestive heart failurePeripheral artery diseaseChronic kidney diseaseCOPDCerebrovascular diseaseLiver cirrhosisDementiaConnective tissue disorderHypothyroidismHyperthyroidism95 (44.4)56 (26.2)37 (17.3)46 (21.5)6 (2.8)13 (6.1)10 (4.7)11 (5.1)57 (26.6)14 (6.5)4 (1.9)5 (2.3)7 (3.3)20 (9.3)3 (1.4)Number of treatment regimens before nivolumab, n(%)0123> 31 (0.5)148 (69.2)45 (21.0)17 (7.9)3 (1.4)Nivolumab treatment duration, weeks, median (IQR)17 (11–42)Adverse effect on thyroid gland by nivolumab, n (%)37 (17.3)Progression, n (%)103 (48.1)Mortality, n (%)147 (68.7)Abbreviations:BMI, body mass index; CCI, Charlson Comorbidity Index; COPD, chronic obstructive pulmonary disease; ECOG, Eastern Cooperative Oncology Group; IQR, interquartile range; SD, standard deviation
Changes in laboratory parameters following nivolumab treatment
A comparison of laboratory parameters measured before nivolumab initiation and at the first month of therapy is presented in Table 2. At the first month of treatment, a significant increase in serum TSH levels was observed compared with baseline values [median: 2.20 (IQR: 1.12–6.77) mU/L vs. 1.46 (IQR: 0.83–2.47) mU/L; p < 0.001].
In parallel, fT4 levels demonstrated a modest but statistically significant decrease [median: 0.93 (IQR: 0.74–1.15) ng/dL vs. 0.98 (IQR: 0.83–1.18) ng/dL; p = 0.039].
Table 2Comparison of baseline and first-month laboratory parameters following nivolumab treatment (n = 214)VariablesBaseline1st monthpTSH, mU/L, median (IQR)1.46 (0.83–2.47)2.20 (1.12–6.77)< 0.001fT4, ng/dL, median (IQR)0.98 (0.83–1.18)0.93 (0.74–1.15)0.039Creatinine, mg/dL, median (IQR)0.82 (0.80–1.11)0.98 (0.81–1.19)0.005FBG, mg/dL, mean ± SD117 ± 41115 ± 370.595ALT, U/L, median (IQR)14 (10–21)14 (11–21)0.215AST, U/L, median (IQR)18 (13–21)18 (14–26)0.114GGT, U/L, median (IQR)28 (19–45)27 (18–57)0.115ALP, U/L, median (IQR)83 (66–107)88 (68–124)0.003LDH, U/L, median (IQR)184 (153–228)189 (151–260)0.031Total bilirubin, mg/dL, median (IQR)0.52 (0.40–0.68)0.53 (0.41–0.74)0.073Albumin, g/dL, mean ± SD3.9 ± 0.43.8 ± 0.6< 0.001WBC, 109/L, mean ± SD7.7 ± 3.28.4 ± 3.80.008NEU, 109/L, median (IQR)4.7 (3.3–6.3)5.4 (3.8–6.7)< 0.001LYM, 109/L, median (IQR)1.53 (1.0-2.1)1.39 (0.9–1.9)0.739Hemoglobin, g/dL, mean ± SD12.2 ± 1.912.1 ± 2.00.744Platelets, 109/L, mean ± SD283 ± 112287 ± 1230.673Abbreviations: ALP, alkaline phosphatase; ALT, alanine transaminase; AST, aspartate transaminase; FBG, fasting blood glucose; fT4, free thyroxine; GGT, gamma-glutamyl transferase; LDH, lactate dehydrogenase; LYM, lymphocytes; NEU, neutrophils; IQR, interquartile range; SD, standard deviation; TSH, thyroid-stimulating hormone; WBC, white blood cells
When biochemical parameters were evaluated, creatinine levels showed a significant increase at the first month compared with baseline values [median: 0.98 (IQR: 0.81–1.19) mg/dL vs. 0.82 (IQR: 0.80–1.11) mg/dL; p = 0.005], whereas no significant change was observed in fasting blood glucose levels (p = 0.595). Among liver function tests, ALP (p = 0.003) and LDH (p = 0.031) levels increased significantly, while no significant changes were detected in ALT, AST, GGT, or total bilirubin levels.
When hematological parameters were examined, a significant increase in total leukocyte count was observed at the first month compared with baseline values [8.4 ± 3.8 × 10⁹/L vs. 7.7 ± 3.2 × 10⁹/L; p = 0.008], and this increase was primarily driven by a rise in neutrophil count [median: 5.4 (IQR: 3.8–6.7) ×10⁹/L vs. 4.7 (IQR: 3.3–6.3) ×10⁹/L; p < 0.001].
No significant differences were detected between baseline and first-month values in lymphocyte count, hemoglobin level, or platelet count (all p > 0.05). Serum albumin levels demonstrated a significant decrease at the first month (p < 0.001).
Characteristics of patients who developed nivolumab-related thyroid adverse events
A total of 37 patients (17.3%) who developed thyroid-related adverse events during nivolumab therapy were evaluated. The mean age of this subgroup was 68.4 ± 6.3 years; 30 patients (81.1%) were male and 7 (18.9%) were female. The median follow-up duration was 152 weeks (IQR: 80–248).
When thyroid dysfunction patterns were examined, de novo hypothyroidism was observed in the majority of patients (n = 24, 64.9%). This was followed by de novo hyperthyroidism related to thyroiditis (n = 9, 24.3%). Deterioration of pre-existing hypothyroidism was identified in 3 patients (8.1%), whereas central hypothyroidism secondary to hypophysitis was observed in only 1 patient (2.7%).
No cases of deterioration of pre-existing hyperthyroidism were detected.
The median time from nivolumab initiation to the onset of thyroid dysfunction was 12 weeks (IQR: 8–24). Systemic glucocorticosteroid therapy for thyroid-related adverse events was administered in only 2 patients (5.4%). During the follow-up period, the mortality rate in this subgroup was 73.0% (n = 27).
The clinical characteristics of patients who developed nivolumab-related thyroid adverse events are summarized in Table 3.
Table 3Clinical characteristics of patients who developed nivolumab-related thyroid adverse eventsVariablesTotal (n = 37)Age, mean ± SD68.4 ± 6.3Sex, n (%)MaleFemale30 (81.1)7 (18.9)Observation duration, weeks, median (IQR)152 (80–248)Central hypothyroidism due to hypophysitis, n (%)1 (2.7)Deteriorated hypothyroidism, n (%)3 (8.1)Deteriorated hyperthyroidism, n (%)0 (0)De-novo hypothyroidism, n (%)24 (64.9)De-novo hyperthyroidism due to thyroiditis, n (%)9 (24.3)Nivolumab treatment duration till adverse effect was developed, weeks, median (IQR)12 (8–24)Glucocorticosteroid treatment for adverse effect of nivolumab, n (%)2 (5.4)Mortality, n (%)27 (73.0)Abbreviations:IQR, interquartile range; SD, standard deviation
Predictors of hypothyroidism development following nivolumab therapy
Factors associated with the development of de novo or deteriorated hypothyroidism following nivolumab treatment were evaluated using Cox proportional hazards regression analysis (Table 4). In univariate analysis, increasing age [HR: 1.084; 95% CI: 1.032–1.138; p = 0.001], Charlson Comorbidity Index (CCI) [HR: 1.298; 95% CI: 1.036–1.626; p = 0.023], and early changes in thyroid-stimulating hormone observed after treatment initiation (ΔTSH, per 1 mU/L increase) [HR: 1.014; 95% CI: 1.008–1.021; p < 0.001] were found to be associated with the development of hypothyroidism.
Table 4Cox proportional hazards analysis of predictors for hypothyroidism (de-novo and deteriorated) following nivolumab treatmentVariablesUnivariateMultivariateHR95% CIpHR95% CIpAge, (per a year increase)1.0841.032–1.1380.0011.0921.036–1.1520.001Female sex (vs. male)1.2540.532–2.9550.6051.9670.696–5.5550.202Age-independent CCI, (per one point increase)1.2981.036–1.6260.0230.9190.643–1.3130.643Baseline TSH, (per 1 mU/L increase)0.9780.896–1.0690.629Baseline fT4, (per 1 ng/dL increase)0.9780.435–2.2020.957Baseline WBC, (per 1 × 109/L increase)1.0080.903–1.1260.881Baseline NEU, (per 1 × 109/L increase)0.9850.852–1.1390.838Baseline LYM, (per 1 × 109/L increase)1.2880.844–1.9650.240Baseline platelets, (per 1 × 109/L increase)0.9980.994–1.0020.241ΔTSH, (per 1 mU/L increase)1.0141.008–1.021< 0.0011.0601.036–1.080< 0.001ΔfT4, (per 1 ng/dL decrease)N/AΔWBC, (per 1 × 109/L increase)1.0180.916–1.1330.735ΔNEU, (per 1 × 109/L increase)1.0220.912–1.1450.710ΔLYM, (per 1 × 109/L increase)1.0330.607–1.7590.904ΔPlatelets, (per 1 × 109/L increase)1.0010.998–1.0040.514Abbreviations:CCI, Charlson Comorbidity Index; CI, confidence interval; fT4, free thyroxine; HR, hazard ratio; LYM, lymphocytes; NEU, neutrophils; N/A, not applicable due to overfitting; TSH, thyroid-stimulating hormone; WBC, white blood cells
Baseline TSH and free thyroxine (fT4) levels, as well as baseline leukocyte, neutrophil, lymphocyte, and platelet counts, were not found to be associated with the development of hypothyroidism. Similarly, early changes in hematological parameters (ΔWBC, ΔNEU, ΔLYM, and Δplatelets) were not significant predictors of hypothyroidism development. The ΔfT4 variable was not included in multivariable analyses due to overfitting of the model.
In multivariable Cox regression analysis, age (adjusted HR: 1.092; 95% CI: 1.036–1.152; p = 0.001) and ΔTSH (adjusted HR per 1 mU/L increase: 1.060; 95% CI: 1.036–1.085; p < 0.001) remained independently associated with the development of hypothyroidism. In contrast, sex (adjusted HR: 1.967; 95% CI: 0.696–5.555; p = 0.202) and age-independent CCI (adjusted HR: 0.919; 95% CI: 0.643–1.313; p = 0.643) were not independently associated with hypothyroidism.
In a sensitivity analysis excluding the three patients with deteriorated pre-existing hypothyroidism, ΔTSH remained independently associated with de novo hypothyroidism (adjusted HR 1.065, 95% CI 1.040–1.091; p < 0.001).
The stability of the multivariable Cox model was further assessed using 2,000 bootstrap resamples. The association between ΔTSH and subsequent hypothyroidism remained statistically significant after bootstrap resampling (BCa 95% CI for β: 0.032–0.413; p = 0.003).
Survival according to hypothyroidism development and exploratory discriminative analysis of ΔTSH
Overall mortality following nivolumab therapy was analyzed according to the development of de novo or deteriorated hypothyroidism using the Kaplan–Meier method (Fig. 2). No statistically significant difference in overall mortality was observed between patients who developed hypothyroidism and those who did not (log-rank p = 0.176).
In exploratory ROC analysis, ΔTSH showed high apparent discrimination for subsequent nivolumab-related hypothyroidism, with an AUC of 0.952 (95% CI: 0.925–0.979, p < 0.001) (Fig. 3). Internal validation using 2,000 bootstrap resamples demonstrated a mean AUC of 0.952 (95% bootstrap CI: 0.922–0.976), indicating that the overall discriminative performance was relatively stable across bootstrap samples.
performance of ΔTSH for predicting nivolumab-related hypothyroidism (AUC 0.952; 95%.
CI: 0.925–0.979; p < 0.001).
ΔTSH threshold of ≥ 1.19 mU/L, sensitivity was 96.3%, specificity was 85.6%, positive predictive value was 49.1%, and negative predictive value was 99.4%. Increasing the threshold to ≥ 2.05 and ≥ 2.96 mU/L resulted in progressively higher specificity at the expense of lower sensitivity. Importantly, bootstrap resampling demonstrated variability in ROC-derived optimal thresholds, indicating that no single ΔTSH threshold could be considered internally robust.
Therefore, the evaluated thresholds should be regarded as exploratory and hypothesis-generating rather than validated clinical cut-offs.
Internal validation using 2,000 bootstrap resamples demonstrated stable overall discrimination, with a mean bootstrap AUC of 0.952 (95% bootstrap CI: 0.922–0.976). In contrast, greater variability was observed in the ROC-derived optimal thresholds, supporting cautious interpretation of any single cut-off value.
Table 5Exploratory discriminative performance of different ΔTSH thresholds for subsequent nivolumab-related hypothyroidismΔTSH thresholdSensitivity (%)Specificity (%)PPV (%)NPV (%)≥ 1.1996.385.649.199.4≥ 2.0585.291.459.097.7≥ 2.9659.394.761.594.1
Discussion
In this study, early ΔTSH was associated with the subsequent development of immune-related hypothyroidism after adjustment for age, sex, and comorbidity burden. This observation is compatible with the dynamic changes in thyroid function that may occur following immune checkpoint inhibition. However, given the modest sample size and the limited number of hypothyroidism events, the findings should be interpreted as preliminary and exploratory rather than conclusive evidence of predictive utility.
Immune-related thyroid dysfunction has been described as one of the earliest and most frequently observed endocrine toxicities following PD-1 blockade [12]. Experimental models and studies conducted in human thyroid tissues have demonstrated that PD-1 inhibitors enhance CD8⁺ T cell–mediated cytotoxicity within the thyroid gland, disrupt follicular epithelial integrity, and promote the formation of an interferon-dominant inflammatory microenvironment [13, 14]. In addition, experimental models using PD-1 inhibitors have shown infiltration of cytotoxic CD4⁺ T cells into thyroid follicles, increased granzyme B expression, and a predominantly interferon-γ–driven inflammatory response; this process has been reported to follow a rapidly progressive and destructive inflammatory pattern distinct from that of classical autoimmune thyroiditis [15].
The observation that thyroid dysfunction in clinical cohorts frequently emerges within the first weeks to months of treatment further supports the early activation of this immune-mediated injury [16].
Within this biological framework, early changes in TSH occupy a central position in the pathophysiological cascade. The hypothalamic–pituitary–thyroid axis represents one of the most sensitive feedback systems in the human body, generating a disproportionate (logarithmic) TSH response to even minimal alterations in thyroid hormone production [9]. Physiological studies have demonstrated that small reductions in thyroid hormone synthesis can lead to marked increases in TSH secretion, even while peripheral fT4 levels remain within the normal range [17, 18].
Accordingly, TSH may be regarded as an early and sensitive biochemical reflection of early immune-mediated thyroid injury, preceding the development of clinically or biochemically overt hypothyroidism.
The lack of an association between baseline TSH and fT4 levels and the development of hypothyroidism in this study indicates that single time-point measurements are insufficient to capture the underlying immune-mediated dynamic process. In contrast, the association between early ΔTSH and subsequent hypothyroidism observed during the first month of therapy supports the concept that immune-related thyroid injury represents a time-dependent biological process rather than a static predisposition. In this context, ΔTSH may be positioned as a measurable intermediate phenotype reflecting the early phase of an inflammatory process initiated within the thyroid gland, before clinical or biochemical thresholds of overt dysfunction are exceeded.
Previous studies aimed at predicting immune-related thyroid dysfunction have predominantly focused on thyroid autoantibodies [19, 20]. Although the presence of anti–thyroid peroxidase and anti–thyroglobulin antibodies has been associated with an increased risk in some studies, the consistency of this association remains limited, and a substantial proportion of antibody-negative patients have also been shown to develop thyroid dysfunction [21]. Moreover, autoantibody testing is not a component of standard monitoring algorithms in routine clinical practice [22].
In contrast, TSH measurement is included in routine surveillance as recommended by current guidelines and enables dynamic risk assessment without additional cost or infrastructural requirements [23]. In this regard, ΔTSH represents not only a biologically meaningful but also a clinically pragmatic biomarker that can be applied across all levels of the healthcare system. Importantly, ΔTSH appears to reflect a treatment-specific dynamic immune response rather than pre-existing autoimmune susceptibility in predicting immunotherapy-related thyroid dysfunction.
The persistence of the association between ΔTSH and the development of hypothyroidism in multivariable time-dependent analyses, independent of age, sex, and comorbidity burden, suggests that this parameter does not merely reflect overall disease severity or nonspecific systemic inflammation. These findings are consistent with the possibility that ΔTSH may reflect early changes within the thyroid axis during nivolumab therapy. Accordingly, ΔTSH should be regarded not simply as a concomitant laboratory fluctuation but as a distinct biological signal that reflects the temporal dynamics of the underlying pathophysiological process.
Recent studies have demonstrated that thyroid dysfunction developing after immune checkpoint inhibitor therapy may be predicted by dynamic changes in TSH observed during the early phase of treatment; however, the quantitative characterization of these biochemical changes and the establishment of clinically applicable cut-off values remain limited [24–27]. Further studies are therefore needed to determine whether reproducible and clinically applicable ΔTSH thresholds can be established.
The exploratory ROC analysis showed high apparent discrimination of ΔTSH for subsequent hypothyroidism in this cohort. Although AUC estimates were similar across 2,000 bootstrap resamples, this finding describes internal stability within the available data and does not establish the reproducibility of predictive performance in a larger population. Bootstrap resampling cannot compensate for the modest original sample size or the limited number of outcome events, and the apparent performance may be optimistic.
Variability in the ROC-derived thresholds further limits interpretation of any single cut-off. Accordingly, both the observed discrimination and the evaluated thresholds, including ≥ 1.19 mU/L, should be regarded as preliminary and exploratory. Larger independent cohorts with more hypothyroidism events are needed to confirm predictive performance and assess threshold reproducibility before clinical application.
The absence of a significant association between the development of hypothyroidism and overall survival is consistent with the existing literature indicating that immune-related endocrine toxicities are generally manageable adverse events [28, 29]. This finding suggests that the value of ΔTSH should be interpreted not as a prognostic marker but rather within the context of early detection and optimization of safe clinical monitoring. In other words, the primary clinical contribution of ΔTSH lies not in predicting survival outcomes, but in facilitating the uninterrupted and safe continuation of immunotherapy.
The findings of this study indicate that immune-related hypothyroidism following nivolumab therapy represents an early-onset, dynamic biological process that becomes clinically measurable through ΔTSH. This hormone dynamics–based approach supports a shift toward temporal biomarker–driven monitoring in the assessment of immunotherapy-related endocrine toxicities. The use of an easily accessible parameter such as ΔTSH for early risk stratification may facilitate closer clinical surveillance and timely endocrine evaluation.
Although female sex has been associated with an increased risk of immune-related thyroid dysfunction in previous studies, including Zhao et al., sex was not independently associated with hypothyroidism in our cohort [16]. This discrepancy may partly reflect the marked male predominance in our study population, which may have limited our ability to detect sex-related differences.
The principal limitation of this study is the modest sample size of 214 patients and, more importantly, the limited number of primary outcome events: 27 cases of de novo or deteriorated hypothyroidism. These data limitations constrain the precision and stability of the multivariable estimates and increase the risk of overfitting and optimistic estimates of discriminative performance. Although bootstrap resampling was used to assess internal stability, it repeatedly samples the same original observations and cannot compensate for the limited amount of independent information.
Consequently, the observed associations, AUC estimates, and ROC-derived thresholds should be regarded as preliminary, exploratory findings. Larger cohorts with more outcome events are needed to confirm predictive performance; the need for additional data is not addressed by bootstrap resampling alone. The retrospective, single-center design further limits causal inference and generalizability.
In addition, requiring first-month thyroid function measurements to calculate ΔTSH may have introduced selection bias by excluding patients with early follow-up discontinuation. Accordingly, these findings require validation in larger independent cohorts. Another limitation relates to the assessment of patients with pre-existing hypothyroidism.
Deterioration of hypothyroidism was defined biochemically by elevated TSH and reduced fT4 concentrations despite ongoing thyroid hormone replacement therapy at unchanged doses. However, because of the retrospective design, potential confounding factors that may influence thyroid hormone status, such as medication adherence, gastrointestinal absorption disorders, concomitant medications (e.g., proton pump inhibitors, iron or calcium supplements), weight changes, and other clinical conditions, could not be systematically evaluated. Therefore, although these patients were included as deteriorated hypothyroidism cases, alternative contributors to worsening thyroid function cannot be completely excluded.
An additional limitation is the inherent biochemical relationship between ΔTSH and the definition of hypothyroidism, as both involve TSH measurements. Although ΔTSH was assessed before the occurrence of de novo hypothyroidism, this shared biochemical component may have contributed to the high apparent discrimination observed in the ROC analysis. Therefore, the ROC-derived threshold should be considered exploratory and requires prospective external validation.
Furthermore, the lack of comprehensive immunological profiling constrains mechanistic insights. These findings should therefore be interpreted cautiously and warrant prospective validation.
Conclusion
In this retrospective cohort, early ΔTSH was associated with subsequent immune-related hypothyroidism during nivolumab therapy. However, the modest sample size and the limited number of hypothyroidism events restrict the conclusions that can be drawn regarding its predictive performance. The findings should therefore be regarded as preliminary and exploratory, despite the use of bootstrap resampling to assess internal stability.
Larger studies with more outcome events are needed to confirm the observed performance and determine whether ΔTSH can meaningfully inform clinical monitoring.
Fig. 1Patient selection flowchart of the study cohort
Fig. 2Kaplan–Meier curves demonstrating overall mortality according to the development of hypothyroidism (de-novo and deteriorated) following nivolumab treatment. No statistically significant difference in overall mortality was observed between patients with and without hypothyroidism (log-rank p = 0.176)
Fig. 3Receiver operating characteristic (ROC) curve showing the discriminative performance of ΔTSH for predicting nivolumab-related hypothyroidism (AUC 0.952, 95% CI 0.925–0.979; p < 0.001)
Supplementary Information
Below is the link to the electronic supplementary material.
Supplementary Material 1