The results of our study show that analyzed biomarkers did not significantly differ in their value in prediction of RAI therapy efficacy in patients with differentiated thyroid cancer. Therefore, sTg alone can be used as a reliable predictor of treatment effect. Inclusion of CLT status in patients with DTC tends to increase specificity and sensitivity of models in predicting RAI efficacy although without significance. We demonstrate different cut-off values for various DTC biomarkers to prognose the poorer performance of RAI.
The analyzed models of DTC biomarkers included Tg, aTg and/or TSH in different variations. All of the models incorporated thyroglobulin, which is a glycoprotein released by the thyroid follicular cells generally linked to the quantity of thyroid tissue [46, 47]. The measurement of thyroglobulin itself is fundamental in the contemporary management of differentiated thyroid cancer, guiding treatment decisions and follow-up protocols based on its results [46]. Its role has been reported already in 1985 [34]. Stimulated thyroglobulin, initial stimulated or preablative stimulated thyroglobulin just before RAI serve as a marker for residual disease following surgery and can indicate the presence of hidden distant metastases, even in cases where morphological imaging yields no abnormal findings [21].
Depending on the studied population and the assay utilized, as many as 25% of patients diagnosed with DTC have aTg present at the time of initial diagnosis [48]. The risk of aTg interference poses a substantial challenge in the follow-up care of DTC patients [49]. An updated clinical and laboratory expert consensus concerning thyroglobulin and thyroglobulin antibody highly recommends testing for the presence of aTg routinely with highly-sensitive thyroglobulin measurement [46]. The sTg×aTg product predicted the effectiveness and prognosis of 131I therapy in both aTg-negative and aTg-positive DTC patients prior to their initial 131I therapy post-thyroidectomy [35]. Therefore, it can serve as a valuable clinical marker for monitoring DTC patients [35].
Due to the potential stimulation of Tg production by TSH, Tg concentrations in pre-RAI therapy patients may be influenced by their TSH levels [29]. Consequently, extensive research has been conducted on the correlation between the Tg/TSH ratio and the outcomes of initial RAI therapy [27, 30, 45]. The study of Ju et al. compared the predictive model capability to the RAI ablation therapy of Tg and Tg/TSH [29]. According to their results, while both Tg concentrations and Tg/TSH ratios could serve as predictors for the outcomes of the initial 131I ablative therapy, the model incorporating Tg alone exhibited superior performance compared to the model incorporating Tg/TSH ratios [29].
Another model incorporated the Thyroglobulin Reduction Index (TRI) – sTg2-sTg1 (the difference between concentration of stimulated thyroglobulin after the second course of RAI therapy) and the concentration of stimulated thyroglobulin after the first course of RAI), which indicates the response to RAI therapy and the extent of tissue damage [21]. Response to RAI may be reflected by maximal Tg reduction to obtain undetectable values [21]. TRI may also show tumor differentiation and its capacity to incorporate RAI[3] and could be a supplemental tool in predicting RAI efficacy [21].
Although Tg has been the most recognized parameter to analyze in prediction of RAI therapy efficacy, its role as a single parameter of this prediction was questionable due to the risk of analytical interference. The results of our study show that Tg alone may serve as a predictor for the outcomes of the initial 131I ablative therapy. It confirms and expands the conclusions of Ju et al. who compared models including Tg and Tg/TSH showing that both Tg and Tg/TSH rations could be considered predictors of the effects of RAI. However, in their study, the prediction model including Tg performed even better than model including Tg/TSH [29]
The cut-off for Tg concentration to predict a persistent disease suggested in previous studies varied from 0.27 to 30 mg/l. In a meta-analysis by Webb et al. [50], which included 15 studies involving 3900 patients, a cut-off of 10 mg/L was determined, demonstrating a good negative predictive value (NPV). Another study proposed a cut-off of 13 mg/L as predictive of disease recurrence [21]. However, the number of patients included in many studies was low, ranging from 63 to 450 patients [21]. In one of the biggest studies, enrolling 1642 DTC patients, the first RAI had an excellent response in 855 patients. The cut-offs for Tg level and Tg/TSH ratio were 3.40 ng/ mL (AUC: 0.789) and 36.03 ng/mIU (AUC: 0.788), respectively[29]. Another study showed a cut-off for Tg/TSH as 1.50 [31].
Our previous study [37] showed that predictive capability of sTg for detecting incomplete response to radioiodine has been validated through ROC curve analysis, with the suggested optimal cut-off value being 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 raises the risk of incomplete therapy response by 2.5 times (P = 0.002).
Our current study indicates the cut-off value of Tg as 7.22 ng/mL in a big homogenous population of 744 patients with DTC.
The relationship of CLT and DTC has been a subject of many studies [51]. According to that, in our comparison of models with different DTC biomarkers, we included the subgroup of patients with and without CLT, patients lacking CLT exhibited a trend toward higher AUC values across all models, although this contrast did not reach statistical significance. However, the inclusion of CLT in the models predicting radioiodine effectiveness should be further studied in bigger populations.
Limitations
Our study has its limitations. First, we did not have an external dataset for model validation. However, we implied an internal cross validation procedure. The advantage of this attitude is a more homogenous population which enables reliability of the results, as all data were used for both training and validation. The approach mitigates overfitting risks while delivering a more resilient appraisal of the model's performance. Other limitations were the nonrandomization and the retrospective nature of the inclusion. However, the nature of the database was to include all patients consecutively admitted to the hospital making a representative model of patients’ population. Third, to account for missing values and to avoid bias associated with analyzing just the complete cases, we conducted multiple imputations.