What's new?

This study makes an important step in predicting the efficacy of radioiodine therapy for patients with differentiated thyroid cancer by comparing multiple clinical models and identifying stimulated thyroglobulin as a reliable single predictor. A newly identified cutoff value of 7.22 ng/ml for stimulated thyroglobulin offers a practical tool for guiding treatment decisions and improving patient outcomes. Moreover, the study highlights the possible role of chronic lymphocytic thyroiditis in refining these predictive models, suggesting a path for further research. These insights could transform clinical approaches to differentiated thyroid cancer, optimizing long‑term outcomes and reducing unnecessary interventions. Future studies should focus on external validation of these models and the inclusion of chronic lymphocytic thyroiditis to further refine predictive accuracy and improve patient care.

Introduction

Differentiated thyroid cancer (DTC) stands as the predominant form of endocrine malignancy, accounting for approximately 90% of all thyroid neoplasms.¹ Its incidence has risen over the past 3 decades,² primarily due to the widespread use of ultrasound screening for thyroid abnormalities.^3,4 Encouragingly, DTC has an excellent mean 10‑year survival rate of 90%.⁵ Standard management protocols typically involve total or near‑total thyroidectomy, optionally followed by radioactive iodine (RAI) remnant ablation based on tumor stage and risk of recurrence.⁶

Although DTC generally has a favorable prognosis, it often demonstrates a tendency for persistence or recurrence. This underscores the need for long‑term surveillance, especially in higher‑risk patients.⁷ A key factor in effective management is RAI therapy,⁸ whose efficacy plays a critical role in determining patient outcomes.⁹ Indeed, an inadequate response to RAI therapy correlates with a markedly poorer prognosis.^10,11

Recent guidelines emphasize the importance of identifying patients who respond well to initial treatment.^3,12 This approach helps create personalized follow‑up plans, which can ease the emotional and financial strains of long‑term monitoring.^13,14 While the American Thyroid Association (ATA) guidelines predominantly rely on tumor, node, metastasis (TNM) staging and thyroglobulin (Tg) levels to guide decisions on RAI remnant ablation,³ the integration of predictive models has shown promise for refining prognostic accuracy.

Clinicopathological prediction models represent a notable advancement in improving the ability to forecast efficacy and long‑term prognosis of RAI therapy by integrating a range of patient‑specific characteristics.¹⁵ Incorporating these models into clinical practice facilitates a personalized medical approach, ensuring treatments are tailored to individual patient needs.¹⁶ However, to the best of our knowledge, no study has yet compared the efficacy of different DTC prediction models or established biomarker cutoff values to guide clinical decisions.

In this study, we aimed to evaluate the performance of various predictive models in determining the efficacy of RAI therapy, defined as no need for a second RAI therapy 12 months after the initial treatment and treatment‑free survival. The study population comprised a homogeneous cohort of patients diagnosed with DTC who underwent thyroidectomy and RAI remnant ablation. The second aim was to determine cutoff values for thyroid biomarkers to aid in this prediction.

Patients and methods

Results

Out of 744 patients with DTC, data on the need for a second dose of RAI at the 12‑month follow‑up visit were available for 656 individuals. Of these, 556 patients (84.8%) were diagnosed with papillary thyroid cancer. The study compared patients who required a second course of radioiodine therapy at 12‑month follow‑up after the initial radioiodine treatment (n = 285) with those who achieved an excellent response after the first course (n = 371). Out of the whole cohort, 88 patients (11.8%) were lost to follow‑up: 35 (39.8%) due to an absence at scheduled follow‑up visits, 6 (6.8%) due to incomplete Tg level measurements, and 47 (53.4%) due to insufficient information regarding subsequent RAI treatment. These patients did not differ from those who were followed in terms of age, sex, thyroid cancer histopathologic type, and disease staging. Of the patients ultimately included in our study, 9 underwent exogenous TSH stimulation, while 647 had endogenous stimulation. Of the 285 patients requiring repeat RAI, 212 achieved an excellent response after the second treatment, 37 required a third RAI, 6 required a fourth, and 30 required 5 or more RAI treatments. Among these, 5 patients required tyrosine kinase inhibitor treatment with sorafenib.

The analysis of several clinical and pathological features is summarized in Table 2. SMDs were calculated to determine associations with the need for a second therapy. There were no significant differences in age or histologic tumor type among the patients who achieved an excellent response after the first RAI course. Those requiring a second course were more frequently male (17.5% vs 7.8%; SMD, 0.3; P <⁠0.001). The proportion of patients with papillary thyroid cancer was similar in both groups (SMD, 0.061), with a slightly higher rate of the conventional variant of the tumor among those needing the second course (65% vs 59.1%), though this association was neither strong nor significant (SMD, 0.124).

In the group receiving the second RAI dose, more patients had tumors larger than 10 mm (50% vs 43%; SMD, 0.14), multifocal lesions (40.9% vs 29.6%; SMD, 0.24; P = 0.004), capsular infiltration (38.3% vs 28.8%; SMD, 0.203), and advanced disease stages (SMD, 0.362; P <⁠0.001), especially stage IV (9% vs 2.2%). Only the features with P values provided were significant. The mean RAI dose administered during the first RAI treatment was 85.43 mCi (3160.91 MBq), ranging from 30 to 150 mCi (1110–5550 MBq). There was no difference in mean RAI activity between the patients who required the second dose (82.83 mCi [3064.71 MBq]) and those who did not (85.11 mCi [3149.07 MBq]; P = 0.39).

A total of 240 patients had both DTC and CLT. Those who required the second RAI dose at 12 months were less likely to have CLT, as compared with those who did not (40.7% vs 47.4%; SMD, 0.14). The patients treated with the second RAI dose more frequently had distant metastases (20.7% vs 8.1%; SMD, 0.36; P = 0.003) and lymph node involvement (20.4% vs 6.2%; SMD, 0.43; P <⁠0.001).

Overall, the patients who required the second course of radioiodine therapy at 12 months were more often male and exhibited multifocal disease, advanced stage, distant metastases, lymph node involvement, and specific histologic features.

Cutoff values

Wherever possible, we determined cutoff values for various DTC biomarkers to predict RAI efficacy (Table 4 and Figure 1). sTg1 had the highest accuracy (0.628) with a mean cutoff value of 7.22 ng/ml. aTg1 showed a slightly lower accuracy (0.613) with a cutoff value of 138.22 IU/ml. The model incorporating the sTg1/sTSH1 ratio demonstrated an accuracy of 0.613 with a cutoff of 0.13 mg/IU. sTg1 × aTg1/sTSH1 or sTg1/sTSH1 had an accuracy of 0.617 with a cutoff of 3.88 mg/ml. The model evaluating changes in sTg over time (sTg2 − sTg1) had the lowest accuracy (0.587) with a cutoff of 0.53 ng/ml. Two additional models that integrated time‑based changes in sTg, sTSH, and aTg were not applicable for accuracy and cutoff determination.

Table 4. Cutoff values for various biomarkers for differentiated thyroid cancer with their accuracy

Biomarker	Accuracy	Cutoff
Abbreviations: NA, not applicable; others, see Table 1
sTg1	0.628	7.22
sTg1 × aTg1	0.613	138.22
	0.613	0.13
	0.617	3.88
sTg2‑sTg1	0.587	0.531
	NA	NA
	NA	NA
	NA	NA
sTg1+sTg1 × aTg1	0.617	148.48
	0.613	3.98

In summary, sTg1 alone or in combination with aTg1 and sTSH1 provided the most accurate cutoff points for predicting DTC outcomes, with sTg1 alone emerging as the most reliable single predictor.

Kaplan–Meier curves

Finally, the Kaplan–Meier curves were used to assess treatment‑free survival. In our dataset, a tumor smaller than 10 mm, female sex, stage I disease, absence of distant and lymph node metastases, and a Tg concentration below 7.22 ng/ml were associated with improved treatment‑free survival (Figure 2). For Tg concentrations above or below the cutoff of 7.22 ng/ml, the Kaplan–Meier curves diverged at the onset of follow‑up, and the difference between them remained significant at all subsequent time points when (beyond 8 months after initial treatment), as evidenced by nonoverlapping 95% CIs.

Discussion

In the present study, the assessed biomarkers did not significantly differ in their predictive value for the efficacy of RAI therapy in patients with DTC. Therefore, sTg alone serves as a reliable predictor of treatment outcomes. While including CLT status in the analysis appeared to improve specificity and sensitivity of the models predicting RAI efficacy, this trend did not reach significance. We demonstrated distinct cutoff values for various DTC biomarkers to identify patients at a higher risk of suboptimal response to RAI therapy.

The analyzed models of DTC biomarkers included sTg, aTg and / or sTSH in various combinations. All models incorporated Tg, a glycoprotein released by the thyroid follicular cells that generally reflects the amount of thyroid tissue.^43,45 Measurement of Tg concentration is the cornerstone of modern management of DTC, guiding both treatment decisions and follow‑up protocols.⁴⁵ Its clinical relevance was first reported in 1985.³³ sTg, whether initial or preablative, is a key marker for detecting residual disease following surgery. It can also signal the presence of occult distant metastases, even in the cases where morphological imaging yields no abnormal findings.²⁰

Depending on the studied population and the assay used, up to 25% of patients with DTC have detectable aTg at the time of initial diagnosis.⁴⁶ The presence of aTg remains a challenge in the follow‑up care of patients with DTC due to its potential interference.⁴⁷ An updated clinical and laboratory expert consensus strongly recommends routine testing for aTg alongside highly‑sensitive Tg measurements.⁴⁵ The sTg × aTg product has been shown to predict the efficacy and prognosis of RAI therapy in both aTg‑negative and aTg‑positive patients before their initial ¹³¹I treatment following thyroidectomy.³⁴ Therefore, this parameter can serve as a valuable clinical marker for monitoring patients with DTC.³⁴

Due to the potential stimulation of Tg production by TSH, sTg concentrations in patients undergoing pre‑RAI therapy may be influenced by their sTSH levels.²⁸ This has led to extensive research on the relationship between the sTg/sTSH ratio and the outcomes of initial RAI therapy.^26,29,44 A study by Ju et al²⁸ compared the predictive capabilities of sTg alone vs sTg/sTSH ratio for RAI ablation therapy outcomes. According to their findings, while both sTg concentrations and the sTg/sTSH ratio are valid predictors of the initial ¹³¹I therapy success, the model based solely on sTg demonstrated superior performance, as compared with the model incorporating the sTg/sTSH ratio.²⁸

Another predictive model incorporated the Tg reduction index, calculated as the difference between sTg concentrations after the first and after the second course of RAI therapy (sTg2 – sTg1). This index reflects the response to RAI therapy and the extent of tissue damage.²⁰ Response to RAI may be reflected by maximal sTg reduction, ideally to achieve undetectable values.²⁰ Additionally, Tg reduction index may provide insight into tumor differentiation and its ability to incorporate RAI,³ and could serve as a supplemental tool for predicting the efficacy of RAI therapy.²⁰

Although Tg has been widely recognized as a key parameter in predicting the efficacy of RAI therapy, its reliability as a standalone parameter predictor has been questioned due to the potential for analytical interference. However, our findings show that sTg alone can serve as a reliable predictor for the outcomes of the initial ¹³¹I ablative therapy. These results confirm and build upon the conclusions of Ju et al,²⁸ who compared predictive models incorporating sTg and the sTg/sTSH ratio. The study showed that both sTg and the sTg/sTSH ratio could be considered as predictors of RAI efficacy, although the model based solely on sTg performed even better than the one including the sTg/sTSH ratio.²⁸

The cutoff values for sTg concentration in predicting persistent disease have varied widely in previous studies, ranging from 0.27 to 30 ng/ml. A meta‑analysis by Webb et al,⁴⁸ which included 15 studies and 3900 patients, identified a cutoff of 10 ng/ml, demonstrating a strong negative predictive value. Another study proposed a cutoff of 13 ng/ml as predictive of disease recurrence.²⁰ However, many studies had relatively small sample sizes, ranging from 63 to 450 patients.²⁰

One of the largest studies, involving 1642 patients with DTC, reported an excellent response to the first RAI course in 855 patients. In this cohort, the cutoff for sTg concentration was 3.4 ng/ml (AUC, 0.789), while the cutoff for the sTg/sTSH ratio was 36.03 ng/mIU (AUC, 0.788).²⁸ Another study proposed a cutoff for the sTg/sTSH ratio of 1.5.³⁰

Our previous study³⁶ validated the predictive capability of sTg for detecting incomplete response to radioiodine therapy through ROC curve analysis, with the suggested optimal cutoff value of 8.17 ng/ml (sensitivity, 55%; specificity, 77%; positive predictive value, 42.1%; negative predictive value, 84.7%). An sTg level exceeding 8.17 ng/ml was associated with a 2.5‑fold increased risk of incomplete therapy response (P = 0.01).

In the current study, conducted in a larger and more homogenous cohort of 744 patients with DTC, the optimal sTg cutoff value was slightly lower at 7.22 ng/ml.

The relationship between CLT and DTC has been widely investigated.⁴⁹ In our comparison of predictive models using various DTC biomarkers, we included subgroups of patients with and without CLT. The patients without CLT showed a trend toward higher AUC values across all models, although this difference did not reach significance. Further research involving larger populations is needed to clarify the role of CLT in the models predicting the effectiveness of radioiodine therapy.

The use of the Kaplan–Meier curves for treatment‑free survival analysis provided a more detailed and nuanced understanding of the relationship between various parameters and the need for RAI treatment, as well as its efficacy. By considering treatment‑free survival as a time‑to‑event outcome rather than a binary one, we accounted for all treatments received during follow‑up, not just the need for a second RAI dose 12 months after the initial treatment. This approach yielded more granular data and improved the clarity and interpretability of our results.

We included both binary and linear analyses of the sTg concentration, which allowed for a more comprehensive evaluation of its impact on treatment‑free survival. These analyses offer valuable insights into the prognostic factors affecting treatment outcomes in patients with DTC, potentially informing more personalized treatment strategies and follow‑up protocols.

The Kaplan–Meier analysis confirmed that Tg concentration is a factor affecting treatment‑free survival, with sTg levels below the cutoff value of 7.22 ng/ml being associated with improved survival outcomes.

Our study has several limitations. First, we did not have an external dataset for the model validation. However, we implemented an internal cross‑validation procedure. The advantage of this approach is a more homogenous population that enhances reliability of the results, as all data were used for both training and validation. This method mitigates the risk of overfitting and provides a more robust evaluation of the model’s performance. Second, we lacked access to ultrasound data to assess thyroid remnant volume, a factor that can influence Tg concentrations. Third, the nonrandomized design and retrospective nature of patient inclusion present inherent biases. Nonetheless, the database was designed to include all patients consecutively admitted to the hospital, providing a representative model of the population. Finally, to account for missing values and to avoid bias associated with analyzing only the complete cases, we conducted multiple imputations.

By analyzing the performance of different clinicopathological models in predicting the efficacy of RAI therapy in patients with DTC, we aimed to highlight their potential for guiding clinical decision‑making and advancing the paradigm of personalized thyroid cancer management. Our findings indicate that all analyzed models have a comparable predictive value, supporting the use of sTg as a standalone parameter for predicting efficacy of RAI therapy. The identified cutoff value for sTg in this context is 7.22 ng/ml. We also established that future research should further investigate the influence of CLT status on the prediction of an excellent response to the first RAI dose in patients with DTC.