Nomogram for survival prediction in metastatic differentiated thyroid cancer and survival comparison between systemic chemotherapy and local radiotherapy in high-risk patients Authors: Jian Xu Published: 2025 ABSTRACT AbstractBackgroundDistant metastases are the leading cause of death for patients with differentiated thyroid cancer (DTC), and surgery and radioactive iodine (RAI) therapy are effective treatments for them. However, there remains a large proportion of patients with poor prognoses, and there exists no consensus on whether systemic chemotherapy or beam radiation should be administrated to improve their prognoses.MethodThe clinicopathology information of patients with metastatic DTC (mDTC) was retrospectively collected and analyzed. A nomogram was created to predict their prognosis and divided patients into high- and low-risk groups, and the external validation was performed.ResultsThe Cox regression showed that sex, age, metastatic sites, surgery, and radiotherapy were independent influencing factors affecting OS. Fine-gray competing risk analysis showed that age, T-stage, N-stage, metastatic sites, and radiotherapy were independent risk factors for TCSD. The nomogram we constructed proved to be effective and robust with high efficiency, with C-indexes of 0.785 (95% CI = 0.752–0.813) and 0.794 (95%CI = 0.766–0.831) on the train and test sets, respectively. mDTC patients were further classified into high- and low-risk groups based on the prediction of nomogram, those in the high-risk group had significantly worse OS and higher TCSD than those in the low-risk group (P < 0.001), and chemotherapy significantly improved OS and reduced TCSD for patients in the high-risk group, whereas local beam radiation did not provide any survival benefit. In the external validation dataset, the 1-, 3-, and 5-year AUC values of the model ranged from 0.833 to 0.873, which further proved the efficiency of our model.ConclusionThis study developed an accurate prognostic model for patients with mDTC, which may be applicable in future clinical practice for prognosis and death prediction. For high-risk patients, as predicted by nomogram, systemic chemotherapy was the recommended therapy option for them.Keywords: Differentiated thyroid cancer, Distant metastases, Chemotherapy, Radioactive iodine, Nomogram, SEER Introduction The past thirty years have witnessed a steady, worldwide upward trend in the incidence of thyroid cancer (TC), though recent data indicate stabilization in some regions due to improved screening practices [1, 2]. DTC, as the most common type of TC, accounts for over 95% of cases [3]. It originates from thyroid follicular epithelial cells and can be classified into different subtypes, mainly including papillary thyroid carcinoma (PTC, 80%), and follicular thyroid carcinoma (FTC, 15%). The standard treatment procedure of DTC involves risk-oriented surgery, postoperative I131, and thyroid hormone therapy, and the patients always have a favorable prognosis in the vast majority of cases [4]. However, this favorable long-term prognosis is less seen for those with distant metastases (DM) [5–11]. About half of DM are found at the initial diagnosis of TC, while the others are discovered later, sometimes even decades following initial treatment [12]. The most common metastasis sites are lungs (49% of all cases) and bones (25% of all cases), or both of the two sites (15% of all cases) [13]. The routine management of patients with metastatic DTC (mDTC) consists of thyroidectomy with or without lymph node dissection, followed by radioactive iodine (RAI) [14], and long-term survival for this patient population ranges from 31% to 43% [15]. RAI destroys occult microscopic foci of neoplastic cells within the thyroid remnant or elsewhere in the body so as to improve the long-term prognosis. Although RAI remains the mainstay of treatment for mDTC, the clinical prognosis is variable and depends on the response to RAI [16]. Shaha et al. [17] proved that DTC patients with noniodine-avid metastasis have a poorer prognosis than patients with radioactive iodine-avid (RAI-A) metastasis. When it comes to RAI-refractory (RAI-R) diseases, the prognosis is always extremely poor, and individualized treatments are demanded, current optional therapy includes locoregional therapy (re-operation, beam radiation therapy) and system therapy (chemotherapy) [18]. There remains no standard treatment for mDTC patients who have poor outcomes or respond poorly to RAI, and extremely few studies have looked into this particular cohort of patients due to its minority and intractable nature. Although it cannot represent the majority of patients, RAI-R has an estimated incidence of 4–5 cases per million population [18]. To assist clinicians in selecting optimal treatments based on prognostic predictions, this study established an accurate model for mDTC patients. Afterwards, the patients were divided into high- and low-risk groups based on the prediction of our model, and the more effective treatment procedure (systemic chemotherapy or local radiotherapy) for the patients in the high-risk group was determined. Our nomogram, while based on classical regression, provides a foundation that could be integrated with ML models for even greater accuracy, distinguishing it from prior tools by its focus on high-risk mDTC subgroups. Materials and methods Data sources and study design All data about patients with mDTC analyzed in this study were obtained from the publicly available Surveillance, Epidemiology, and End Results (SEER) database [SEER 17 Regs Study Data (Changes 2010–2021); version 8.4.3], a population-based registry covering approximately 28% of the U.S. population, with high-quality data on cancer incidence, treatment, and outcomes [19]. Inclusion criteria: (1) Patients with morphological and histopathological evidence of diagnosis according to the International Classification of Diseases of Cancer, Third Revision (ICD-O-3) for DTC (e.g., site code C73.9 for thyroid; histology codes 8050/3, 8260/3, 8340/3, 8341/3, 8342/3, 8343/3, 8344/3 for papillary; 8330/3, 8331/3, 8332/3, 8335/3 for follicular); (2) Patients with primary DTC with M1 staged according to the American Joint Committee on Cancer (AJCC) 7th edition staging group (2010–2015), SEER combined staging group (2016–2017), and 2018 Extent of Disease (EOD) staging group (2018+) classification. Exclusion criteria were (1) Patients with non-primary thyroid cancer according to international rules; (2) Patients with a previous history of cancer; (3) Patients with unknown status of surgery or radiotherapy; (4) Patients with unknown survival time; and (5) Patients with T0 stage. In the survival analysis, the primary outcomes included OS and thyroid cancer-specific death (TCSD). OS was determined by all causes of death, and TCSD was death caused by thyroid cancer. Statistical analysis Categorical variables were expressed as frequencies and percentages. Comparisons between groups of categorical data were achieved using the χ2 test or Fisher’s exact test. For numerical variables, comparisons were performed using independent t-tests or Mann-Whitney U tests depending on normality, assessed via Shapiro-Wilk tests. Survival rates were compared using Kaplan-Meyer (K-M) curves, and statistical significance was tested using the log-rank test. Univariate and multivariate Cox analyses were performed to assess the risk of death and to identify independent prognostic factors. While assessing the risk of TCSD, the Fine-gray competing risk model was employed due to the presence of competing events. Tables 2 and 3 were performed on the whole dataset, and Tables 4 and 5 were performed on the predicted high-risk patient within the test set. Patients were randomized into train and test sets in a ratio of 7:3. Data preparation included age categorization (stratified into 10-year intervals), imputation of missing values (< 5% of cases) via mean substitution for continuous variables and mode for categorical, random shuffling before 70:30 train/test split to avoid bias. After determination of the independent risk factor for OS and TCSD among patients with mDTC, a nomogram was constructed to predict their survival. The nomogram incorporated the clinicopathological variables that were statistically significant in the multivariate Cox analysis and competing risk analysis, including age at diagnosis, T and N stage, sites of metastasis, sex, radiotherapy, and primary tumor surgery, so as to predict the OS of patients. The accuracy and performance of our nomogram were evaluated by consistency index (C-index), receiver operating characteristic (ROCs) curve (pROC package), decision curve (DecisionCurve package), and calibration curve analysis (rms package). Further, the mDTC patients were classified into high-risk and low-risk groups according to the results of our nomogram, and the OS and TCSD of patients in the high-risk group receiving localized beam radiation and systemic chemotherapy were then assessed. All statistical calculations were achieved by the R software (version 4.0.2), and statistical significance was defined as a two-sided tail value of less than 0.05. Table 2Univariate and multivariate Cox analysis of characteristics extracted from SEER databaseUnivariate Cox analysisMultivariate Cox analysisHR95%CIP ValueHR95%CIP ValueSex Femalereferencereference Male1.2651.066–1.501.066.501**1.3851.119–1.714.119.714**Marital status Marriedreference Others1.1800.991–1.406.991.4060.063///Age at diagnosis <40referencereference 40–495.7002.895–11.220.895.220***4.0551.952–8.423.952.423** 50–598.4074.457–15.860.457.860***4.8682.464–9.618.464.618* 60–6911.6706.317–21.560.317.560***5.3042.735–10.285.735.285*** 70–7914.5137.872–26.760.872.760***6.9423.589–13.425.589.425*** >7924.83313.301–46.360.301.360***9.8844.933–19.803.933.803***Race Whitereference Black1.0450.760–1.438.760.4380.785/// Other1.1210.899–1.397.899.3970.309///T stage T1referencereference T20.9710.599–1.574.599.5740.9050.6980.420–1.161.420.1610.166 T31.3920.944–2.051.944.0510.0951.2510.820–1.908.820.9080.299 T42.6081.797–3.784.797.784***1.4160.934–2.146.934.1460.101N stage N0reference N10.9450.779–1.146.779.1460.567///Histological type Follicular adenocarcinomareferencereference Papillary adenocarcinoma0.7800.645–0.944.645.944*1.1010.852–1.422.852.4220.461Site of metastasis Onereferencereference Two2.2531.817–2.792.817.792***1.3151.007–1.717.007.717* Three or more3.7382.752–5.076.752.076***2.2221.534–3.220.534.220***Surgery on regional lymph nodes Noreferencereference 1–3 regional lymph nodes removed0.4120.301–0.563.301.563***0.8820.588–1.321.588.3210.541 4 or more regional lymph nodes removed0.3740.306–0.456.306.456***1.0470.760–1.444.760.4440.777Surgery on primary site Noreferencereference Yes0.2640.222–0.314.222.314***0.5350.383–0.746.383.746***Chemotherapy Noreferencereference Yes1.9331.539–2.427.539.427***0.9910.735–1.335.735.3350.949Radiotherapy Noreferencereference Beam radiation1.3771.127–1.682.127.682**1.1770.891–1.554.891.5540.251 Radioactive iodine0.2750.221–0.342.221.342***0.3930.293–0.529.293.529****P< 0.05, ** P< 0.01, ***P< 0.001 Univariate and multivariate Cox analysis of characteristics extracted from SEER database *P< 0.05, ** P< 0.01, ***P< 0.001 Table 3Univariate and multivariate Fine-Gray competing risk analysis of characteristics extracted from SEER databaseUnivariate analysisMultivariate analysisHR95%CIP ValueHR95%CIP ValueSex Femalereferencereference Male1.2901.050–1.570.050.570*0.9370.714–1.230.714.2300.640Marital status Marriedreference Others1.1000.894–1.350.894.3500.380///Age at diagnosis <40referencereference 40–497.3103.230–16.500.230.500***5.6452.315–13.765.315.765*** 50–598.1103.700–17.800.700.800***5.2802.243–12.429.243.429*** 60–6912.4605.830–26.600.830.600***5.1102.198–11.880.198.880*** 70–7912.5405.880–26.800.880.800***6.8152.939–15.804.939.804*** >7915.7307.230–34.200.230.200***6.8512.724–17.228.724.228***Race Whitereference Black0.8000.539–1.190.539.1900.270/// Other0.9800.759–1.270.759.2700.880///T stage T1referencereference T21.0300.536–1.970.536.9700.9300.7450.353–1.573.353.5730.440 T31.8601.107–3.120.107.120*1.6390.928–2.895.928.8950.089 T43.7402.265–6.170.265.170***1.8861.053–3.379.053.379*N stage N0referencereference N11.3001.030–1.640.030.640*1.5871.122–2.244.122.244**Histological type Follicular adenocarcinomareference Papillary adenocarcinoma1.0400.822–1.310.822.3100.760///Site of metastasis Onereferencereference Two2.3601.850–3.020.850.020***1.4871.055–2.097.055.097* Three or more3.6702.630–5.130.630.130***1.7381.035–2.919.035.919*Surgery on regional lymph nodes Noreferencereference 1–3 regional lymph nodes removed0.5080.359–0.719.359.719***0.8540.534–1.364.534.3640.510 4 or more regional lymph nodes removed0.4580.364–0.577.364.577***0.7670.491–1.199.491.1990.250Surgery on primary site Noreferencereference Yes0.3340.274–0.408.274.408***0.8050.505–1.281.505.2810.360Chemotherapy Noreferencereference Yes2.4001.880–3.050.880.050***1.3700.952–1.971.952.9710.090Radiotherapy Noreferencereference Beam radiation1.6221.283–2.052.283.052**1.3070.931–1.836.931.8360.120 Radioactive iodine0.3310.257–0.428.257.428***0.4860.340–0.693.340.693****P< 0.05, **P< 0.01, ***P< 0.001 Univariate and multivariate Fine-Gray competing risk analysis of characteristics extracted from SEER database *P< 0.05, **P< 0.01, ***P< 0.001 Table 4Univariate and multivariate Cox analysis of characteristics in the high-risk groupUnivariate Cox analysisMultivariate Cox analysisHR95%CIP ValueHR95%CIP ValueSex Femalereference Male1.0230.748–1.398.748.3980.888///Age at diagnosis 40–49reference 50–590.8730.371–2.055.371.0550.756/// 60–690.7820.353–1.730.353.7300.544/// 70–790.7770.351–1.723.351.7230.535/// >790.8870.393–2.002.393.0020.773///T stage T1reference T20.4720.218–1.122.218.1220.905/// T30.4560.244–1.089.244.0890.098/// T40.6790.387–1.190.387.1900.176///N stage N0reference N10.9490.686–1.311.686.3110.749///Site of metastasis Onereferencereference Two1.2050.836–1.738.836.7380.3171.2860.890–1.859.890.8590.181 Three or more1.8191.166–2.838.166.838**2.0751.320–3.263.320.263**Surgery on primary site Noreferencereference Yes0.6760.494–0.924.494.924*0.6140.446–0.846.446.846**Chemotherapy Noreferencereference Yes0.7090.487–0.952.487.952*0.6120.416–0.899.416.899*Radiotherapy Noreference Beam radiation0.8940.651–1.227.651.2270.487/// Radioisotopes0.8650.376–1.987.376.9870.732///*P< 0.05, **P<0.01, ***P< 0.001 Univariate and multivariate Cox analysis of characteristics in the high-risk group *P< 0.05, **P<0.01, ***P< 0.001 Table 5Univariate and multivariate Cox analysis of characteristics in the low-risk groupUnivariate Cox analysisMultivariate Cox analysisHR95%CIP ValueHR95%CIP ValueSex Femalereference Male1.2760.993–1.640.993.6400.566///Age at diagnosis <40referencereference 40–492.9581.201–7.288.201.288*2.3090.931–5.728.931.7280.071 50–594.3222.085–8.962.085.962***3.5311.693–7.363.693.363*** 60–696.6743.341–13.332.341.332***4.5712.259–9.248.259.248*** 70–798.6114.305–17.224.305.224***6.4793.197–13.130.197.130*** >7911.9765.777–24.827.777.827***7.7913.667–16.550.667.550***T stage T1referencereference T21.6620.880–3.138.880.1380.1171.1710.615–2.227.615.2270.631 T31.6110.929–2.791.929.7910.0891.6870.968–2.937.968.9370.065 T43.0071.757–5.146.757.146***1.8881.086–3.284.086.284*N stage N0reference N10.9380.720–1.223.720.2230.638///Site of metastasis Onereferencereference Two1.8631.350–2.572.350.572***1.0860.852–1.128.852.1280.181 Three or more3.2501.963–5.380.963.380***1.6970.975–2.096.975.0960.112Surgery on primary site Noreferencereference Yes0.2340.178–0.309.178.309***0.5670.411–0.782.411.782***Chemotherapy Noreferencereference Yes1.7921.305–2.459.305.459***1.5611.178–2.765.178.765*Radiotherapy Noreferencereference Beam radiation1.2150.898–1.643.898.6430.2061.2480.896–1.739.896.7390.189 Radioisotopes0.2830.208–0.385.208.385***0.4380.310–0.619.310.619****P< 0.05, **P< 0.01, ***P< 0.001 Univariate and multivariate Cox analysis of characteristics in the low-risk group *P< 0.05, **P< 0.01, ***P< 0.001 Results Clinical characteristics of mDTC patients A total of 1,236 eligible mDTC patients from the SEER database (2010–2021) were included in this study. The clinicopathological characteristics of these patients are detailed in Table 1. Among them, 721 patients (58.3%) were female and 515 (41.7%) of them were male. As for histological type, 345 patients (27.9%) had FTC and 891 cases (72.1%) had PTC. For various treatment procedures, 773 cases (62.5%) received primary tumor surgery, 161 cases (13.0%) underwent system chemotherapy, 481 cases (38.9%) had RAI therapy, and 247 cases (20.0%) received beam radiation. Note that percentages exceed 100% due to combinations of therapies; for example, 37.5% of patients received both surgery and RAI. Table 1Characteristics of patients with mDTCN=1236%Sex Female72158.3% Male51541.7%Marital status Married55244.7% Others56045.3% Unknown12410.0%Age at diagnosis <4016613.4% 40–491169.4% 50–5918715.1% 60–6928322.9% 70–7926721.6% ≥8014311.6%Race White85369.0% Black887.1% Others21217.2% Unknown90.7%T stage T11189.5% T212510.1% T330925.0% T438230.9% Unknown22818.4%N stage N034427.8% N167954.9% Unknown13911.3%Histological type Follicular adenocarcinoma34527.9% Papillary adenocarcinoma89172.1%Site of metastasis One77963.0% Two17414.1% Three or more625.0% Unknown14711.9%Surgery on regional lymph nodes No50040.5% 1–3 regional lymph nodes removed12410.0% 4 or more regional lymph nodes removed46137.3% Unknown776.2%Surgery on primary site No46337.5% Yes77362.5%Chemotherapy No107587.0% Yes16113.0%Radiotherapy No50841.1% Beam radiation24720.0% Radioactive iodine48138.9% Characteristics of patients with mDTC Univariable and multivariable Cox regression analysis To identify the clinicopathological variables that significantly affect the OS of mDTC patients, univariable Cox regression analysis was conducted (Table 2), and the following variables were identified: sex, age at diagnosis, T stage, histological type, site of metastasis, and treatment information (surgery on regional lymph nodes, surgery on primary site, chemotherapy, radiotherapy). Further, to eliminate confounding factors and determine the independent OS influencing factors, the multivariable Cox regression analysis was performed incorporating the variables identified by the univariable Cox analysis (Table 2). Analyses used the full dataset for exploratory univariable and multivariable. The results showed that male (HR = 1.385, 95% CI = 1.119–1.714, P < 0.01), older age (compared with age < 40 years; aged > 79 years: HR = 9.884, 95% CI = 4.933–19.803, P < 0.001), and more sites of metastasis (compared with one metastatic site; three or more metastatic sites: HR = 2.222, 95% CI = 1.534–3.220, P < 0.001) led to worse OS, while performance of primary tumor surgery (HR = 0.535, 95% CI = 0.383–0.746, P < 0.001) and RAI therapy (HR = 0.393, 95% CI = 0.293–0.529, P < 0.001) can improve the survival of mDTC patients. Univariate and multivariate Fine-Gray competing risk analysis To further investigate those with a higher risk of TCSD, the Fine-gray competing risk model was introduced for the patients with mDTC. Analyses used the full dataset for exploratory univariable and multivariable. The univariate competing risk analysis showed that sex, age at diagnosis, T and N stage, sites of metastasis, surgery on regional lymph nodes, surgery on primary site, chemotherapy, and radiotherapy were variables that were significantly related to the TCSD of patients with mDTC (Table 3). The above-mentioned variables were incorporated into the multivariate competing risk analysis, and the results showed that older age at diagnosis (compared with age < 40 years; aged > 79 years: HR = 6.851, 95% CI = 2.724–17.228, P < 0.001), advanced T stage and N stage, site of metastasis (compared with one metastatic site; three or more metastatic sites: HR = 1.738, 95% CI = 1.035–2.919, P < 0.05) were related to higher risk of TCSD for patients with mDTC. While only RAI therapy can decrease the TCSD risk for patients with mDTC. Construction and evaluation of the nomogram to predict the survival of patients with mDTC The graph proved an outstanding performance of our nomogram in terms of predicting survival of patients with mDTC, with C-indexes of 0.785 (95% CI = 0.752–0.813) and 0.794 (95% CI = 0.766–0.831) on the train and test sets, respectively (Fig. 1). Moreover, the AUC value at 3-, 5-, and 7-year were 0.825, 0.835, 0.832 in the train set (Fig. 2A), and 0.872, 0.868, 0.901 in the test set (Fig. 2B). Further validation of our nomogram was achieved by the calibration curve. Our calibration curve greatly matches the standard line for both the train set and test set at 3-,5-, and 7-year (Fig. 3), denoting that our model can greatly predict the actual distribution of the dataset and therefore the prediction probability of our nomogram agrees well with the actual OS. Finally, decision curve analysis was conducted to assess the clinical utility of the nomogram, which shows a more net benefit than full or no treatment across a threshold probability range in the train and test set (Fig. 4). Nomogram to calculate the survival probability for patients with metastatic differentiated thyroid cancers at 3-year, 5-year, and 7-year. To calculate the total score, a vertical line was drawn from each variable to the point axis at the top, and the individual points for each variable was added. Further, the patients were categorized into high- and low-risk groups according to the total score axis ROC curves for the nomogram on the train and test set. (A) ROC curve on the train data; (B) ROC curve on the test data; ROC: receiver operating characteristic curve; AUC: area under the curve Calibration curves of the nomogram’s predicted results in the train and test set Calibration curves on the train data at (A) 3-year, (C) 5-year, and (E) 7-year; Calibration curves on the test data at (B) 3-year, (D) 5-year, and (F) 7-year Decision curves of the nomogram’s predicted results in the train and test set Decision curves on the train data at (A) 3-year, (C) 5-year, and (E) 7-year; Decision curves on the test data at (B) 3-year, (D) 5-year, and (F) 7-year Patients were divided into high- and low-risk groups using the median total nomogram score as the cutoff, ensuring balanced groups for comparison. K-M survival analysis was performed for patients in the high- and low-risk groups (Fig. 5A), and the results showed that mDTC patients in the low-risk group have significantly better OS compared with those in the high-risk group (P < 0.001). The 10-year survival probability of mDTC patients in the low-risk group is over 50%, while this data for the high-risk group is less than 10%. In the high-risk group, 5-year TCSD was 65%, compared to 25% in low-risk; nomogram-predicted death rates aligned with observed (calibration curves). Figure 5B indicates that mDTC patients in the high-risk group have a significantly higher risk of TCSD and other cause-related death compared with those in the low-risk group (P < 0.001), and the TCSD of patients in the high-risk group significantly outdistance the low-risk group as the survival time prolongs. K-M survival analysis and Fine-gray competition risk analysis of patients in the high- and low-risk groups. (A) K-M survival analysis of patients in the high- and low-risk groups; (B) Fine-gray competition risk analysis of patients in the high- and low-risk groups Independent prognosis factors affecting the survival of mDTC patients in the High-risk group The above results have proved that mDTC patients in the high-risk group have an extremely poor prognosis, but there exists no consensus on the further treatment of these patients. Therefore, Cox analysis and competing risk analysis were both performed on those in the high-risk group. The univariate Cox regression analysis showed that the site of metastasis, primary tumor surgery, and chemotherapy can significantly affect the OS of mDTC patients (Table 4). Further multivariate Cox analysis proved that two or more sites of metastasis significantly correlated with poor OS (compared with one metastatic site; HR = 2.075, 95% CI = 1.320–3.263, P < 0.001). For various therapies, primary tumor surgery (HR = 0.614, 95% CI = 0.446–0.846, P < 0.01) and chemotherapy (HR = 0.612, 95% CI = 0.416–0.899, P < 0.05) can significantly improve the OS of patients, while the patients can not gain any survival benefit from radiotherapy, including beam radiation and RAI (P > 0.05). In contrast to high-risk, low-risk patients showed benefits from RAI (HR = 0.438, P < 0.001) but not chemotherapy (HR = 1.561, P < 0.05) (Table 5). To examine the TCSD of mDTC patients in the high-risk group, the univariate and multivariate Fine-Gray competing risk analysis was conducted on this particular subgroup of patients (Table 6). The univariate competing risk analysis showed that sites of metastasis and chemotherapy can affect the TCSD of patients in the high-risk group. Further multivariate analysis indicated that more sites of metastasis significantly correlated with poor TCSD (compared with one metastatic site; two metastatic site sites: HR = 1.758, 95% CI = 1.141–2.710, P < 0.05; three or more sites: HR = 2.819, 95% CI = 1.688–4.710, P < 0.001). Only chemotherapy (HR = 0.695, 95% CI = (HR = 0.614, 95% CI = 0.446–0.846, P < 0.05) can improve the TCSD of this population. Table 6Univariate and multivariate Fine-Gray competing risk analysis of characteristics in the high-risk groupUnivariate analysisMultivariate analysisHR95%CIP ValueHR95%CIP ValueSex Femalereference Male0.9930.698–1.410.698.4100.970///Age at diagnosis 40–49reference 50–590.7790.349–1.740.349.7400.540/// 60–690.8670.432–1.740.432.7400.690/// 70–790.8130.405–1.630.405.6300.560/// >790.5990.278–1.290.278.2900.190///T stage T1reference T20.4090.114–1.470.114.4700.170/// T31.2480.523–2.980.523.9800.620/// T41.6340.712–3.750.712.7500.250///N stage N0reference N11.1600.793–1.690.793.6900.450///Site of metastasis Onereferencereference Two1.7801.180–2.700.180.700**1.7581.141–2.710.141.710* Three or more2.4901.620–3.830.620.830***2.8191.688–4.710.688.710***Surgery on primary site Noreferencereference Yes0.9360.624–1.400.624.4000.7500.7560.491–1.160.491.1600.200Chemotherapy Noreferencereference Yes0.8770.651–0.983.651.983*0.6950.470–0.995.470.995*Radiotherapy Noreference Beam radiation1.2250.851–1.760.851.7600.270/// Radioisotopes0.8060.392–1.660.392.6600.560///*P< 0.05, **P< 0.01, ***P< 0.001 Univariate and multivariate Fine-Gray competing risk analysis of characteristics in the high-risk group *P< 0.05, **P< 0.01, ***P< 0.001 External validation The external validation cohort of 117 mDTC patients was retrospectively collected from our institution (Shaoxing People’s Hospital) between 2015 and 2023, with data on clinicopathology, treatments, and outcomes gathered via electronic medical records review. Our model showed high efficiency in the external independent dataset, and the AUC value at 3-, 5-, and 7-year were 0.873, 0.850, and 0.833, respectively (Fig. 6A). Further, these patients were divided into high-and low-risk group according to the nomogram, and those in the high-risk group showed significantly worse OS than the low-risk group (P < 0.001) (Fig. 6B). In the whole external cohort, chemotherapy can not improve the OS of patients (P = 0.680), while it can significantly improve the OS of patients in the high-risk group (P = 0.028) (Fig. 6CD). The cohort reflects the heterogeneity and treatment patterns of mDTC patients in the real-world, confirming the generalizability of our model, and most importantly, proving that our model can effectively identify those patients who can benefit from the chemotherapy. External validation of nomogram. (A) ROC curve for 3-, 5-, and 7-year survival predictions in the external cohort (n = 117). AUC values were 0.873 (3-year), 0.850 (5-year), and 0.833 (7-year); (B) K-M survival analysis of mDTC patients in the high- and low-risk groups; (C) K-M survival analysis of mDTC patients with and without chemotherapy; (D) K-M survival analysis of mDTC patients with and without chemotherapy separately in the high-risk and low-risk group Discussion DTC was characterized by a slow progressive course and proved to be one of the most curable cancers, with a 10-year survival rate up to 80%–90% [5, 20]. DM only occurs in less than 10% of patients with DTC, but it represents the most frequent cause of thyroid cancer-related death [21]. The overall mortality rate of mDTC at 5 and 10 years after diagnosis were 65% and 75%, respectively [22]. Although patients with mDTC always have a favorable OS and disease-free survival (DFS) [23, 24] compared with other metastatic tumors, there remains a group of patients who have a very dismal prognosis or respond very poorly to various treatments. So far, there have been very few studies looking primarily at the clinical management and outcomes of patients with mDTC, not to mention developing personalized treatment for different subgroups. Based on a large population cohort, this study focused on mDTC patients and tried to cast light on the personalized therapy that may improve the prognosis of those patients. The Cox regression analysis showed that primary tumor surgery and RAI therapy could improve the OS of mDTC patients, while males, old age, and more sites of metastasis were related to poor survival. The independent risk survival factors we discovered were basically proved by former studies [25]. Our Fine-gray competing risk model showed that older age at diagnosis, advanced T (T4) and N (N1) stage, and more sites of metastasis were related to a higher risk of TCSD for patients with mDTC. At the same time, only RAI therapy can decrease the TCSD risk for patients with mDTC. Age was found to be a strong prognostic factor, and older age always companies poorer OS and TCSD than younger ones. Even in the presence of metastatic disease, age can still maintain its strong prognostic impact as proved by both the present study and also the previous ones [15, 26], and DM in old patients was proved to be correlated with a 5-year mortality rate close to 50% [14]. There might be several underlying reasons. First, it is believed that DTC patients with younger age had higher rates of RAI-A in metastatic lesions compared to older patients [27, 28], and those younger patients tend to respond better to RAI [13, 28, 29]. Ito et al. [28, 30] proved that RAI-A metastasis was more prevalent among young patients (< 55 years old) and none of these patients with RAI-A metastasis died of thyroid cancer. Additionally, due to the fact that the uptake of I131 is age-dependent, the effectiveness of RAI therapy decreases in the elder group [31]. Third, the advancing age itself carries risks of increased medical comorbidities and death. This may be another reason why mDTC patients have deteriorated prognoses with aging. Besides age, OS and TCSD benefits gained from primary tumor surgery and RAI we discovered were in accord with the present guidelines [4, 18]. For patients with mDTC, although RAI is advocated as an adjuvant treatment, this procedure has also been questioned in the past decade [14]. On the one hand, RAI can lead to short-term morbidity and an increased risk of second cancer development [32, 33]. Guidelines now recommend selective use of RAI, based on individual risk [12]. On the other hand, for patients with mDTC, the clinical prognosis is variable and depends on the response to RAI [16]. When it comes to RAI-R disease, individualized treatments are demanded. RAI-R was diagnosed while DM failed to concentrate RAI or structural progression developed within 6–12 months after RAI administration [13, 34], and the diagnosis always denotes a bleak prognosis. For DTC patients with RAI-A metastatic disease, I131 therapy is the dominant treatment modality [12, 35]. The rate of RAI-A disease was demonstrated to be approximately 67% [26]. Durante et al. proved an estimated 10-year survival rate of 92% for mDTC patients who achieved remission with RAI compared to an estimated 10-year survival rate of 10% for mDTC patients without iodine uptake [13], and they also discovered the patients lacked iodine uptake account 23% among those with PTC, 8% among those with well-differentiated FTC, and 47% of those with poorly differentiated FTC. Since those with RAI-R can not benefit from RAI but may gain high toxicity, we cannot blindly administrate RAI to every single patient with mDTC, and therefore grouping patients at different risk level is imperative to conduct individualized treatment. To better develop individualized treatment, an effective and robust nomogram was constructed, our nomogram effectively identifies key factors like age (>79 years: highest points) and multiple metastases, enabling clinicians to prioritize interventions. Compared to ML-based thyroid studies [36–38], our nomogram offers interpretable scores, improving on black-box models while achieving comparable AUC value (0.83–0.87). Patients were divided into high-risk and low-risk groups, accordingly. The survival analysis of these two groups showed significantly poor survival in the high-risk group, and further Cox regression analysis on the mDTC patients in the high-risk group showed that only primary tumor surgery and chemotherapy can significantly improve the OS, while neither beam radiation nor RAI can provide survival benefits. Further, competing risk analysis showed that only chemotherapy can decrease the TCSD of mDTC patients in the high-risk group. These results indicate that those who have poor survival always respond poorly to the RAI, and therefore we assumed that the patients in the high-risk group might overlap those who are RAI-R. Previous studies have found that up to 50% of the patients in the RAI-A subgroup achieved complete remission or stabilization over a long-term period following RAI, leading to a favorable OS and disease-free survival (DFS) [22, 39, 40]. However, those proved to be RAI-R are no longer treated with RAI [4, 41], and there is no consensus for the management of these patients while both beam radiation [42] and system chemotherapy [20, 43, 44] are recommended by previous reports. Our results proved that mDTC patients in the high-risk group (assumed those who respond poorly to RAI) also respond poorly to beam radiation, and chemotherapy is the sole solution to improve OS and decrease TCSD at the same time. The powerlessness of beam radiation was also proved in previous articles believing that external beam radiation therapy was associated with poorer survival outcomes in patients with Stage IV DTC [45]. Historically, cytotoxic chemotherapy has a limited role for DTC, due to both limited need and limited efficacy when used. However, cytotoxic agents including pemetrexed and doxorubicin may have a role in the treatment of patients with progressing and symptomatic metastatic disease with RAI-R as well as the newer, emerging targeted therapies [46]. While our findings highlight the benefits of systemic chemotherapy in high-risk patients, emerging targeted therapies like TKIs (e.g., sorafenib and lenvatinib) have shown superior response rates and progression-free survival in RAI-refractory mDTC compared to traditional cytotoxics. Future studies should compare TKIs directly with chemotherapy in high-risk cohorts identified by our nomogram to refine treatment algorithms. In summary, we identified a group of high-risk mDTC patients who respond poorly to RAI and beam radiation, and in this circumstance, only chemotherapy can generate survival benefits. Effectively screening for these patients can assist in personalized therapy development, avoiding RAI toxicity, and highlighting the effectiveness of chemotherapy for them. Limitation The limitations of this study are its retrospective nature, and the deficiencies inherent in the SEER database, such as the dosage of RAI was not available as SEER is a fairly static database that lacks such information. Additionally, the SEER database lacks granular data on chemotherapy regimens, including types (e.g., doxorubicin or pemetrexed), doses, uniformity across patients, and associated adverse effects, which limits our ability to provide detailed recommendations on specific agents. Gene stuff (e.g., BRAF mutations) is also important, but the SEER database does not include either. Conclusion Primary tumor surgery and RAI remain the most universal therapies for patients with mDTC, while individualized procedures should be designed for those who respond poorly. This study developed an effective prognostic model for patients with mDTC cancer, which can accurately discriminate the high- and low-risk populations. It was proved that patients in the high-risk group could not benefit from beam radiation and RAI, and only systemic chemotherapy can improve their OS and TCSD. Hopefully, our model can be applicable in future clinical practice to predict the risk of patients’ death and help in personalized therapeutic method design. Acknowledgements We thank all staff of the SEER database for their contribution to data collection, maintenance, distribution, and so on. Also, we would like to thank all the developers of the R programming package for selflessly sharing their code. Author’s contributions Conceptualization, methodology; formal analysis, data curation, writing—original draft preparation, writing—review and editing, supervision: Jian Xu. The author has read and agreed to the published version of the manuscript. Funding This work was not funded. Data availability Publicly available datasets were analyzed in this study. This data can be found here: https://seer.cancer.gov. Declarations Ethics approval and consent to participate Ethical review and approval were waived for this study due to the fact that the data are fully de-identified and no intervention on patients was performed. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement Publicly available datasets were analyzed in this study. This data can be found here: https://seer.cancer.gov.