Development and validation of a nomogram to predict failure of initial radioactive iodine therapy in differentiated thyroid cancer: a retrospective cohort study

Abstract

Given the scarcity of studies predicting iodine-131 treatment failure based on clinicopathological factors, this study aimed to determine whether clinicopathological features can predict iodine-131 treatment failure in differentiated thyroid carcinoma (DTC) patients. A total of 182 patients were analyzed, including 114 with favorable outcomes and 68 with resistance or poor outcomes. Patients were split into a training set (122 patients) and a validation set (60 patients). Logistic regression identified the number of lateral neck lymph node metastases (P = 0.001) and pre-radioactive iodine therapy serum thyroglobulin levels (P = 0.001) as independent predictors of treatment failure. The predictive model, visualized via a nomogram, achieved area under curves (AUCs) of 0.838 (95% CI, 0.759–0.917) and 0.766 (95% CI, 0.626–0.905) in the training and validation sets, respectively. Calibration curves showed good agreement between predicted and observed outcomes, and decision curve analysis confirmed the model’s clinical utility. Subgroup analyses yielded AUCs of 0.777 (95% CI, 0.648–0.906), 0.745 (95% CI, 0.646–0.843), 0.734 (95% CI, 0.649–0.820), and 0.911 (95% CI, 0.769–0.999) for male, female, age < 55, and age ≥ 55 groups, respectively. This model effectively predicts iodine-131 treatment failure risk in DTC patients, providing valuable information for clinical decision-making.

Introduction

Thyroid cancer (TC) is one of the most common endocrine malignancies, ranking seventh globally in incidence as of 2022, with cases rising rapidly¹. A study predicts that TC incidence in China will increase significantly over the next two decades, with the high-risk age group shifting from 45 to 65 years to 10–24 years². Another study found that TC rates are rising in many nations, particularly among those under 40, with the sharpest increases seen in South Korea, Poland, and China³. This highlights the urgent need for stronger prevention strategies targeting younger populations. Differentiated thyroid cancer (DTC), the most common form of TC, includes papillary thyroid cancer (PTC), follicular thyroid carcinoma (FTC), and Hürthle cell carcinoma. These cancers originate from thyroid follicular cells and make up over 90% of all TC cases⁴. While most cases of DTC progress slowly, certain histological subtypes—such as the tall cell, columnar cell, and hobnail variants of PTC, the widely invasive variant of FTC are more prone to extrathyroidal extension, vascular invasion, and distant metastasis. These subtypes are associated with higher recurrence rates and relatively poorer prognoses⁵.

Iodine-131 (I-131) therapy is a crucial postoperative measure for patients with intermediate- to high-risk DTC⁶. I-131 is a radioactive isotope that is effectively absorbed by thyroid tissue, particularly by DTC cells. It works by emitting radiation that damages the DNA of cancer cells, thereby reducing or eliminating tumors. Radioactive iodine therapy (RAIT) has been a cornerstone in the treatment of hyperthyroidism and DTC for nearly 80 years and remains a crucial component of DTC treatment⁷.

Common methods for evaluating the effectiveness of RAIT include measuring post-treatment serum thyroglobulin (Tg) and anti-thyroglobulin antibody (TgAb) levels, as well as performing neck ultrasound and iodine scintigraphy. However, some patients exhibit resistance or suboptimal responses to RAIT, prompting consideration of alternative therapies such as targeted treatments.

Patients who fail to respond to initial RAIT may require repeated radioiodine treatments or alternative therapies, potentially compromising their quality of life. However, reliable clinical indicators or molecular markers capable of accurately predicting or explaining individual variations in RAIT response among DTC patients are still lacking^8,9. Currently, predictive models for the efficacy of I-131 therapy remain limited. Some studies have investigated predictive factors and long-term outcomes in patients with elevated Tg levels but negative diagnostic whole-body iodine scans (Tg+/scan−)¹⁰. Other studies have relied on clinical and pathological indicators to identify DTC patients at risk of RAIT failure⁸. However, these studies lack external validation, and the performance of their models has not been adequately assessed, potentially limiting their generalizability.

This study aims to develop a predictive model based on clinical and pathological factors to identify risk factors for initial RAIT failure in post-operative DTC patients, using logistic regression analysis. To our knowledge, no previous studies have validated predictive models of I-131 resistance using external datasets. Independent predictors are visualized in a clinically applicable nomogram to facilitate early identification of patients at high risk of RAIT failure after their first treatment.

Results

Patient characteristics

After evaluating 230 patients, a total of 48 cases were excluded for not meeting the inclusion criteria. A total of 182 DTC patients underwent RAIT registered from January 2016 to June 2024 in Hunan Cancer Hospital were obtained. Patients in the training cohort underwent total thyroidectomy at Hunan Cancer Hospital (n = 122, age, 39.36 ± 12.31; range, 16–71 years). The mean follow-up time in train cohort was 17.27 ± 11.37 months. Patients in the validation cohort underwent total thyroidectomy at other hospitals (n = 60, age, 40.12 ± 11.16; range, 16–61 years). The mean follow-up time in validation cohort was 17.23 ± 13.44 months.

The baseline characteristics and pathological results of patients in the training and validation cohorts are shown in Table 1. There were no significant differences in clinical pathological characteristics between the two cohorts. The study included 182 DTC patients, with 122 in the training cohort (72 without RAIT failure and 50 with RAIT failure) and 60 in the validation cohort (42 without RAIT failure and 18 with RAIT failure). Of the 68 RAIT failure cases: 25 had locoregional recurrence, 5 developed distant metastases (3 with concurrent locoregional lesions), and 38 showed biochemical incomplete response (elevated serum Tg/TgAb without structural evidence). There was no significant difference in composition ratio of patients between the two cohorts.

Table 1.

Clinicopathological characteristics of patients in the training and validation cohorts.

Characteristic	Training(n = 122)	Validation(n = 60)	P value
Age, mean ± SD, years	39.36 ± 12.31	40.12 ± 11.16	0.689¶
Patients			0.192§
Non-RAIT failure	72 (59.02)	42 (70.00)
RAIT failure	50 (40.98)	18 (30.00)
Gender			0.173§
Male	41 (33.61)	14 (23.33)
Female	81 (66.39)	46 (76.67)
CI			0.428§
Positive	70 (57.38)	30 (50.00)
Negative	52 (42.62)	30 (50.00)
ETE			0.762§
Positive	30 (24.59)	16 (26.67)
Negative	92 (75.41)	44 (73.33)
Diameter, median ± interquartile range, cm	[1.50(1.00, 2.50)]	[1.50(1.00, 2.50)]	0.765‖
CLN metastasis			0.091§
N0	7 (5.74)	8 (13.33)
NX	115 (94.26)	52 (86.67)
LLN metastasis			0.151§
N0	45 (36.89)	29 (48.33)
NX	77 (63.11)	31 (51.67)
Number of LLN metastasis, median ± interquartile range	[3.00(0.00, 7.00)]	[1.50(0.00, 6.00)]	0.124‖
I-131 dosage, median ± interquartile range, mCi	[150(140.00, 150)]	[150(140, 150)]	0.099‖
Time from surgery to RAIT, median ± interquartile range, months	[4.00(3.00, 5.00)]	[4.00(2.25, 5.75)]	0.652‖
Previous thyroid disease			0.869§
Positive	41 (33.61)	21 (35.00)
Negative	81 (66.39)	39 (65.00)
Multifocality of lesions			0.873§
Positive	75 (61.48)	36 (60.00)
Negative	47 (38.52)	24 (40.00)
Follow-up time, median ± interquartile range, months	[14.00(8.75, 23.25)]	[11.50(9.00, 20.75)]	0.683‖
Maximum diameter of metastatic LLN, median ± interquartile range, mm	[10.50(0.00, 16.00)]	[7.00(0.00, 15.00)]	0.081‖
Tg before RAIT, median ± interquartile range, IU/ml	[2.80(0.13, 10.82)]	[2.67(0.21, 11.04)]	0.983‖
TGAb before RAIT, median ± interquartile range, IU/ml	[0.76(0.19, 5.50)]	[0.78(0.19, 3.88)]	0.942‖
Tg after RAIT, median ± interquartile range, IU/ml	[0.09(0.02, 0.91)]	[0.09(0.04, 0.53)]	0.905‖
TGAb after RAIT, median ± interquartile range, IU/ml	[0.90(0.29, 2.24)]	[0.90(0.29, 5.19)]	0.832‖

Data expressed as n (%), unless otherwise stated.

CLN, central lymph node; CI, capsular invasion; ETE, extrathyroidal extension; LLN, lateral neck lymph node; RAIT, radioactive iodine therapy; TG, thyroglobulin; TGAb, anti-thyroglobulin antibodies. ¶ By the Independent samples t-test. § By the Chi-square test. ‖ By the Mann–Whitney U test.

Construction of model

As shown in Table 2, in the univariate analysis, the status of lateral lymph node (LLN) metastasis (P = 0.001), the count number of LLN metastases (P = 0.001), the dosage of I-131 (P = 0.002), maximum diameter of metastatic LLN (P = 0.001), and Tg levels before RAIT (P = 0.001) were associated with RAIT failure in the training group. In binary logistic analysis (Forward: likelihood ratio [LR]), the status of LLN (P = 0.623), maximum diameter of metastatic LLN (P = 0.562), and the dosage of I-131 (P = 0.704) were excluded from consideration. The number of LLN metastases (P = 0.001, 95% confidence interval [CI] 1.091–1.357) and Tg levels prior to RAIT (P = 0.001, 95% confidence interval [CI] 1.045–1.176) were identified as independent predictors for distinguishing between patients without and with RAIT failure (Table 3). The calculation formulas are presented as follows: Logit (p)= −2.046 + 0.196 × the number of LLN metastases + 0.103 × Tg levels before RAIT. A nomogram was constructed based on the multivariable regression model to visualize the model (Fig. 1).

Table 2.

Clinicopathological characteristics in the training cohort: non-RAIT failure vs. RAIT failure.

Characteristic	Training(n = 122)		P value
	Non- RAIT failure (n = 72)	RAIT failure (n = 50)
Age, mean ± SD, years	39.74 ± 11.83	38.82 ± 13.07	0.688¶
Gender			0.245§
Male	21 (29.17)	20 (40.00)
Female	51 (70.83)	30 (60.00)
CI			0.711§
Positive	40 (55.56)	30 (60.00)
Negative	32 (44.44)	20 (40.00)
ETE			0.763§
Positive	17 (23.61)	13 (26.00)
Negative	55 (76.39)	37 (74.00)
Diameter, median ± interquartile range, cm	[1.45(0.93, 2.20)]	[1.65(1.28, 3.00)]	0.063‖
CLN metastasis			0.238§
N0	6 (8.33)	1 (2.00)
NX	66 (91.67)	49 (98.00)
LLN metastasis			0.001§
N0	37 (51.39)	8 (16.00)
NX	35 (48.61)	42 (84.00)
Number of LLN metastasis, median ± interquartile range	[0.00(0.00, 4.00)]	[6.50(3.00, 12.25)]	0.001‖
I-131 dosage, median ± interquartile range, mCi	[150(140, 150)]	[150(150, 150)]	0.002‖
Time from surgery to RAIT, median ± interquartile range, months	[4.00(3.00, 5.00)]	[3.00(2.75, 5.25)]	0.738‖
Previous thyroid disease			0.174§
Positive	28 (38.89)	13 (26.00)
Negative	44 (61.11)	37 (74.00)
Multifocality of lesions			0.707§
Positive	43 (59.72)	32 (64.00)
Negative	29 (40.28)	18 (36.00)
Tg before RAIT, median ± interquartile range, IU/ml	[1.10(0.08, 4.23)]	[12.92(0.44, 37.42)]	0.001‖
Maximum diameter of metastatic LLN, median ± interquartile range, mm	[0.00(0.00, 14.00)]	[10.00(10.00, 25.25)]	0.001‖
TGAb before RAIT, median ± interquartile range, IU/ml	[0.82(0.24, 4.87)]	[0.62(0.18, 8.37)]	0.963‖

Data expressed as n (%), unless otherwise stated. CI, capsular invasion; CLN, central lymph node; ETE, extrathyroidal extension; LLN, lateral neck lymph node; RAIT, radioactive iodine therapy; Tg, thyroglobulin; TGAb, anti-thyroglobulin antibodies. ¶ By the Independent samples t-test. § By the Chi-square test. ‖ By the Mann–Whitney U test.

Table 3.

Logistic regression analysis of risk factors for RAIT failure.

Variable	B	P	OR	95% CI for OR
Constant	−2.046	0.001	0.129
The number of LLN metastases	0.196	0.001	1.217	1.091	1.357
Tg level before RAIT	0.103	0.001	1.108	1.045	1.176

LLN, lateral neck lymph node; RAIT, radioactive iodine therapy; Tg, thyroglobulin.

Fig. 1.

Nomogram for identifying RAIT failure in DTC patients. DTC, differentiated thyroid cancer, LLN, lateral neck lymph node; RAIT, radioactive iodine therapy; Tg, thyroglobulin.

Model validation

The model passed the Hosmer-Lemeshow (H-L) test with P = 0.404. Receiver operating characteristic (ROC) analysis was performed on the model to identify RAIT failure. The model achieved an AUC of 0.838 (95% CI, 0.759–0.917) in the training cohort (Fig. 2a), and an AUC of 0.766 (95% CI, 0.626–0.905) in the validation cohort (Fig. 2b). Based on the Youden index, the threshold of the total points to predict RAIT failure was determined to be −0.585. As shown in Table 4, The Logit(p) model achieved accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of 78.69%, 62.00%, 90.28%, 81.58%, and 77.38% respectively, for the training group, and 78.33%, 72.22%, 80.95%, 61.90%, and 87.18% respectively, for the validation group. The calibration curves indicated a good agreement between the predicted probability of the nomogram and the observed probability in both the training and validation sets (Fig. 2c-d). Decision curve analysis (DCA) demonstrated that the nomogram model was effective in clinical practice for both the training and validation cohorts (Fig. 2e-f). The horizontal curve “none” represents the scenario where all samples are negative, and no individuals receive intervention, while the “ALL” line represents the scenario where all samples are positive, and everyone receives intervention. The lighter gray lines on both sides of the “ALL” line and the lighter red lines on both sides of the model represent the confidence intervals. The evaluation of model achieved an AUC of 0.777 (95%CI, 0.648–0.906), 0.745 (95%CI, 0.646–0.843), 0.734 (95% CI, 0.649–0.820), and 0.911 (95%CI, 0.769–0.999) in the male, female, age<55, and age ≥ 55 subgroups respectively (Fig. 3). The detailed comparison of statistical results is shown in Table 5.

Fig. 2.

Performance and Clinical Utility of the Nomogram in Training and Validation Cohorts. Receiver operating characteristic (ROC) curves of the model in training (A) and validation (B) cohorts. Calibration curves for the nomogram in training (C) and validation (D) cohorts. Decision curve analysis (DCA) of the nomogram’s clinical utility in training (E) and validation (F) cohorts. AUC, area under the ROC curve; CI, confidence interval; RAIT, radioactive iodine therapy.

Table 4.

Model performance for identifying RAIT failure in training and validation cohorts.

Index	Train cohort	Validation
TP	31	13
TN	65	34
FP	7	8
FN	19	5
Accuracy, %	78.69 (96/122)	78.33 (47/60)
Sensitivity, %	62.00 (31/50)	72.22 (13/18)
Specificity, %	90.28 (65/72)	80.95 (34/42)
PPV	81.58 (31/38)	61.90 (13/21)
NPV	77.38 (65/84)	87.18 (34/39)

FP, False Positive; FN, False Negative; NPV, negative predictive value; PPV, positive predictive value; TP, True Positive; TN, True Negative.

Fig. 3.

Receiver operating characteristic (ROC) curves of the model in subgroups by gender and age: (A) Male, (B) Female, (C) Age < 55, (D) Age ≥ 55. AUC, area under the ROC curve; CI, confidence interval.

Table 5.

Performance of the model for identifying RAIT failure in subgroups.

Index	Male	Femal	Age<55	Age ≥ 55
TP	18	26	36	8
TN	25	74	85	14
FP	4	11	14	1
FN	8	16	23	1
Accuracy, %	78.18 (43/55)	78.74 (100/127)	76.58 (121/158)	91.67 (22/24)
Sensitivity, %	69.23 (18/26)	61.90 (26/42)	61.02 (36/59)	88.89 (8/9)
Specificity, %	86.21 (25/29)	87.06 (74/85)	85.86 (85/99)	93.33 (14/15)
PPV	81.82 (18/22)	70.27 (26/37)	72.00 (36/50)	88.89 (8/9)
NPV	75.76 (25/33)	82.22 (74/90)	78.70 (85/108)	93.33 (14/15)

FP, False Positive; FN, False Negative; NPV, negative predictive value; PPV, positive predictive value; TP, True Positive; TN, True Negative.

Discussion

According to the joint statement from relevant associations in 2019 ⁹, RAIT should be goal-oriented and classified into three categories: remnant ablation, adjuvant treatment, and treatment of known disease. Since its first use for hyperthyroidism and TC, RAIT has remained central in treating DTC. As the importance of predicting clinical outcomes becomes increasingly evident, risk stratification has emerged as a key component in clinical management and RAIT guidelines for patients with DTC^11,12. Notably, the 2019 ATA guidelines emphasize factors such as lymph node metastases and serum Tg levels as important indicators for tailoring RAIT, which aligns well with our findings¹³. While previous studies have identified factors such as tumor size > 1 cm, multifocality, the presence of LLN metastases, postoperative stimulated Tg > 10 ng/mL, and male gender as predictors of recurrence^14,15. The present study contributes additional insights in two respects. First, we evaluated the number of LLN metastases and pre-RAIT serum Tg levels as continuous variables, providing a more detailed representation of the biological gradient of recurrence risk. Second, we validated our model using an independent external cohort, thereby enhancing its generalizability and clinical relevance.

Changes in Tg levels can effectively assess treatment efficacy; a decline typically indicates successful removal or suppression of tumor cells. Elevated Tg levels point to a potential risk of recurrence, guiding doctors in determining the necessity and intensity of RAIT. Regular follow-up monitoring of Tg levels is crucial for evaluating long-term prognosis, and any significant increases should prompt further investigation^16,17. Previous studies have also concluded that pre-RAIT Tg levels play a crucial role in predicting treatment outcomes. Klain and colleagues evaluated the outcomes of low- and intermediate-risk pediatric DTC patients and found that those with postoperative Tg levels greater than 10 ng/ml before RAIT were more likely to develop structural persistent disease. Multivariate analysis identified postoperative stimulated Tg > 10 ng/ml as an independent predictor of persistent disease¹⁴. Another study indicates that low postoperative Tg levels can serve as a basis for reclassifying intermediate-risk patients as low risk, aiding in the formulation of more precise treatment plans¹⁸. Some researchers found that tumor multifocality, lymph node invasion at the time of surgery, and elevated postoperative Tg and TgAb levels are significantly associated with treatment resistance⁸. In conjunction with the above studies and our research findings, it can be concluded that post-surgery pre-RAIT Tg levels serve as an important indicator for evaluating disease prognosis and treatment outcomes.

LLN metastasis holds significant clinical implications in TC. Positive LLN often serves as an indicator of the malignancy, suggesting that patients may face a poorer prognosis. The presence of LLN metastasis influences treatment strategies, as surgeons may need to perform more extensive lymphadenectomy to mitigate the risk of recurrence¹⁹. Numerous studies support this view. A study examined the recurrence/persistence risk linked to large-volume LLN metastases in PTC patients. Among 448 patients with confirmed LLN metastases, the recurrence/persistence risk was 25.1%, with significantly higher rates in those with large-volume LLN metastases compared to low-volume LLN metastases (37.3% vs. 17.1%)²⁰. Studies have shown that in patients with TC, the rate of lymph node positivity is high and increases with both tumor size and T stage^21,22. The above studies also confirm that lymph node status and lymph node burden are closely related to disease prognosis.

In our study, univariate analysis indicated that LLN status, the maximum diameter of metastatic LLN, and RAIT dosage were associated with treatment failure in the training cohort; however, these associations were not statistically significant in the multivariate analysis. The number of LLN metastases may have had a confounding effect on both LLN status and the maximum diameter of metastatic LLN, potentially accounting for the lack of significance of these two variables in the multivariate model. Similar to our findings, a previous study⁸ also reported an association between RAIT dosage and treatment failure; however, in their analysis, RAIT dosage was ultimately retained as an independent predictor. This discrepancy may be partly explained by the fact that RAIT dosage is closely related to individual patient characteristics and disease severity, with higher doses potentially reflecting a more aggressive disease profile. Gender is commonly used as a stratification factor, and age 55 also serves as a key threshold in TNM (Tumor, Node, Metastasis) clinical staging²³. In patients with PTC, our model demonstrated good performance across male, female, age < 55 years, and age ≥ 55 years subgroups.

Our study has several limitations. First, as a retrospective analysis from a single institution with a relatively short median follow-up, the generalizability of findings may be constrained by both selection bias and the limited observation period for treatment failure assessment. Although we included all eligible patients with complete follow-up data (including serum Tg, anti-TgAb levels, and imaging), this design may affect the model’s long-term predictive performance and external validity. Multi-center prospective studies with extended follow-up periods are warranted to validate these findings across diverse populations. Second, Although BRAF V600E mutation testing was performed in a subset of patients (n = 21), the number of cases (11 treatment-responsive and 10 RAIT failure) was insufficient to support robust statistical analysis. Therefore, this variable was not incorporated into the current model. Experimental and clinical studies have established that the BRAF V600E mutation contributes to radioiodine resistance through molecular mechanisms involving suppression of sodium-iodide symporter (NIS) expression, disruption of iodine metabolism pathways, and functional cooperation with other oncogenic alterations²⁴. Future studies should prioritize integrating these molecular features with conventional clinicopathological factors to optimize predictive modeling. We currently maintain Tg levels and the number of LLN metastases as continuous variables, primarily because existing evidence has not yet established optimal cutoff values for predicting RAIT failure. Using continuous variables allows for maximal preservation of prognostic information. Considering clinical application requirements, future multicenter studies with expanded sample sizes will systematically evaluate the predictive performance of different cutoff values using ROC curve analysis and other methods, with particular focus on the clinical decision-making value of categorical variables and comparative assessment of their practicality and predictive accuracy versus continuous variable models.

In summary, we developed a predictive model based on clinicopathological data to estimate the risk of first-time RAIT failure in DTC patients following total thyroidectomy. To our knowledge, no previous studies have externally validated predictive models for RAIT efficacy in DTC. Our study is the first to address this important step. Our model demonstrated good discriminative performance across sex- and age-based subgroups, suggesting robustness and potential applicability to a broad patient population. Clinically, early identification of patients at high risk for RAIT failure may assist physicians in selecting more appropriate treatment strategies—such as closer post-treatment monitoring, considering alternative or adjunctive therapies, or deferring unnecessary RAIT in selected cases. This approach may help optimize healthcare resource allocation and ultimately improve patient quality of life. However, while our findings demonstrate the potential clinical utility of this model, further prospective validation is required before it can be routinely implemented as a decision-making tool in DTC management.

Materials and methods

Patients

We retrospectively reviewed the medical records of 230 cases of DTC who underwent RAIT at the thyroid department of Hunan Cancer Hospital from January 2016 to June 2024. The administered RAI activity was determined empirically based on clinical stratification, followed by post-therapy whole-body scanning to assess iodine avidity. All patients underwent total thyroidectomy and this retrospective study was approved by the Ethics Committee of Hunan Cancer Hospital, with a waiver of the informed consent requirement. The inclusion criteria were as follows: (i) patients with complete clinical and pathological information; (ii) patients diagnosed with DTC after undergoing total thyroidectomy. The exclusion criteria were as follows: (i) missing clinicopathological information; (ii) patients with less than 6 months of follow-up or no follow-up after the first RAIT treatment; (iii) patients with radiologically suspected residual disease on postoperative imaging. Patients with missing data were excluded from the analysis (complete-case analysis approach). The flowchart of the study population is shown in Fig. 4.

Fig. 4.

Flow chart of the study population. DTC, differentiated thyroid cancer; I-131, iodine-131.

The baseline clinicopathological information was retrieved from patient medical records, encompassing tumor diameter, capsular invasion, extrathyroidal extension, status of central lymph nodes and LLN, count of LLN metastases, maximum diameter of metastatic LLN, age, gender, previous thyroid disease, multifocality of lesions, I-131 dosage, time from surgery to RAIT, and postsurgery serum Tg and TgAb levels both before and after RAIT. Radioiodine-refractory DTC has been clearly defined in multiple clinical trials²⁵. The diagnosis requires exclusion of thyroid remnant tissue, inadequate thyroid-stimulating hormone (TSH) stimulation, and excessive stable iodine levels, and is established when either of the following criteria is met: (1) all known lesions demonstrate no radioiodine avidity; or (2) despite partial or complete radioiodine avidity, disease progression occurs after RAIT, as evidenced by at least one of the following: persistently elevated Tg or thyroglobulin antibody TgAb levels, enlargement of existing lesions or emergence of new lesions, worsening of pre-existing disease-related symptoms, development of new disease-related symptoms, or disease-related mortality. Furthermore, based on relevant authoritative guidelines and literature^26–28we have specified the Tg threshold criteria, defining treatment resistance or failed treatment as the presence of any of the following conditions: after undergoing RAIT, patients show suppressed Tg levels ≥ 1 ng/L, stimulated Tg levels ≥ 10 ng/L, a rising trend in TgAb levels, or imaging evidence of structural or functional disease.

Statistical analysis

Both univariate and multivariate analyses were performed using IBM SPSS Statistics version 21 (SPSS Inc., Chicago, IL). All levels of statistical significance are bilateral, with a P value less than 0.05. In univariate analysis, the Mann–Whitney U test was employed to compare characteristics among different groups for continuous variables (not normally distributed), while the independent samples t-test was utilized for continuous variables that followed a normal distribution. The chi-square test was applied to assess differences in categorical variables.

The characteristics that showed statistically significant differences between groups in the univariate analysis were included in the subsequent binary logistic regression analysis. A predictive model for identifying RAIT failure was developed using binary logistic analysis (Forward: LR). Independent predictive factors derived from multivariate analysis were utilized to construct the predictive equation, and the odds ratio (OR) was used to measure the effect of the independent variable on the probability of the dependent variable occurring.

The nomogram was plotted using R software version 4.4.1 (https://www.r-project.org/) to visualize the model. The nomogram assigns each variable a corresponding score, and the sum of all variable scores is defined as the total score. By drawing a vertical line from the total score axis, the estimated probability of RAIT failure can be obtained. The model’s fit was evaluated using the H-L goodness-of-fit test. Both the internal training set and the external validation set were utilized to assess the model’s performance. The evaluation of the predictive model’s performance was conducted through measures of AUCs and calibration curves. Additionally, DCA was employed to evaluate the clinical validity and net benefit of the nomogram²⁹. The diagnostic sensitivity, specificity, PPV, NPV, and accuracy were calculated.

Author contributions

Yulong Tang, Investigation, Data Curation, Visualization, Resources, Statistical analysis, Writing - Original Draft. Peng Wu, Investigation, Data Curation, Validation, Funding acquisition. Shiwei Zhou, Resources, Funding acquisition. Hui Li, Investigation, Data Curation. Xiaohua Song, Investigation, Resources. Wu Li, Data Curation, Visualization, Writing-Reviewing and Editing. Xiaowei Peng, Writing- Reviewing and Editing, Supervision, Project administration, Funding acquisition. All authors reviewed the manuscript.

Funding

This research was funded by the Hunan Provincial Natural Science Foundation of China (Grant No. 2023JJ40419, 2023; 2023JJ60331, 2023), Hunan Cancer Hospital Climb Plan (Grant No. ZX2021004, 2021), Health Research Project of Hunan Provincial Health Commission (Grant No. W20243236 and R2023115, 2023).

Data availability

The data that support the findings of this study are available from the corresponding author, Xiaowei Peng, upon reasonable request.

Declarations

Competing interests

The authors declare no competing interests.

Conflict of interest disclosure

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Ethics approval statement

This study was conducted in accordance with the ethical standards formulated in the Helsinki Declaration and approved by the Ethics Committee of the Hunan Cancer Hospital.