This study utilized SEERStat software (version 8.3.6) to retrieve medical records of 114,911 patients diagnosed with DTC from 2010 to 2019 within the SEER database. The ICD codes for DTC included 8050/3, 8260/3, 8330/3, 8331/3, 8332/3, 8335/2, 8339/3, 8340/2, 8340/3, 8341/2, 8341/3, 8342/3, 8343/2, 8343/3, and 8344/3. We next excluded patients who received radiotherapy (n=48,995); patients with unknown staging or stage 3 or 4 (n=34,748); patients with T0 and TX stages (n=599); patients who did not undergo surgical treatment (n=737); patients with surgery codes 13 (Local tumor destruction), 80 (Thyroidectomy, NOS), 90 (Surgery, NOS), and 99 (Unknown if surgery performed) (n=209); patients with unknown bone metastasis status (n=82); patient with brain metastasis (n=1); patient with unknown liver metastasis status (n=1); patients with unknown lung metastasis status (n=8); patients diagnosed by non-histological methods (only Positive histology included) (n=368); patients with unknown tumor size (n=336); patients with NX stage (n=109); patients who received chemotherapy (n=32); patients with low differentiation or unknown differentiation status (n=21,437); patients aged 90+(n=3); patients with unknown cause of death (n=12); patients with M1 stage (n=4). Ultimately, 7,230 patients were included in the study. All included patients were those who underwent surgical treatment alone, were classified as stage 1 or 2, and had no distant metastasis.

We gathered demographic characteristics and clinical indicators for the DTC patients included in the study. These variables encompass a comprehensive range of patient information. Specifically, the variables include: age, gender, race, marital status, median household income, county type, grade, histologic, clinical stage, T stage, N stage, multifocality, neck dissection, number of lymph nodes removed, surgical treatment methods, treatment delay days, tumor size (cm), positive lymph node numbers (PLNNs), examined lymph node number (ELNN), and long-term outcomes (survival time, vital status, causes of death). Notably, the surgical treatment methods are classified into three types according to the study methodology of Cao et al. (11). Surgery 1: lobectomy, isthmectomy, or removal of less than a lobe. Surgery 2: subtotal or near total thyroidectomy, or removal of a lobe and partial removal of the contralateral lobe (subtotal thyroidectomy: affected thyroid gland lobe (including capsule) + thyroid isthmus + thyroid vertebral lobe + contralateral partial gland lobe; near total thyroidectomy: almost all thyroid tissues were removed, and about 1 gram of normal tissue was retained on each side to protect parathyroid gland and recurrent laryngeal nerve). Surgery 3: total thyroidectomy (complete removal of all thyroid tissue).

The causes of death were categorized into two main types: cancer-specific death and non-specific death. Within the non-specific causes, we focused on several significant categories: cardiovascular diseases (as per the SEER database), infectious diseases, respiratory diseases, and gastrointestinal diseases. Following the methodology outlined in the study by Wang et al. (12), the causes of death recorded in the SEER database were classified as follows: Cardiovascular diseases: causes labeled as aortic aneurysm and dissection, cerebrovascular diseases, diseases of the heart, hypertension without heart disease, and other diseases of arteries or arterioles capillaries. Infectious diseases: causes labeled as pneumonia and influenza, septicemia, and other infectious and parasitic diseases including human immunodeficiency virus (HIV). Respiratory diseases: causes labeled as lung and bronchus, chronic obstructive pulmonary disease, and Allied Cond. Gastrointestinal diseases: causes labeled as colon excluding rectum, stomach, and chronic liver disease and cirrhosis.

Data processing and analysis were performed using SPSS statistical analysis software. Categorical data were expressed as numbers and percentages (%), and comparisons between different surgical groups were conducted using the Chi-square test. Continuous variables were analyzed by one-way analysis of variance (ANOVA) and were expressed as median [25%, 75%]. In the competing risks model, DTC-related death was considered the event of interest, while death from other causes was treated as competing events. Survival was treated as a censored event. The Fine-Gray method was used for statistical testing in the competing risks model. This analysis was implemented using the “cmprsk” package in R software. Different variables were adjusted in the various risk models (specific model adjustments can be copied from the original text). Three methods including Logistic, random forest, and XGBoost were then used to rank the importance of variables to explore the contribution of surgical approaches to long-term survival in patients. P < 0.05 indicated that the difference was statistically significant.

In this study, a total of 7,230 patients were enrolled. Those patients only received the surgical treatment (without radiotherapy or chemotherapy), were in the early stage of DTC, and had no distant metastasis. Among them, 1,512 patients received Surgery 1 (lobectomy, isthmectomy, or removal of less than a lobe); 249 patients received Surgery 2 (subtotal or near total thyroidectomy, or removal of a lobe and partial removal of the contralateral lobe); and 5,469 patients received Surgery 3 (total thyroidectomy). Except for the race, marital status, clinical stage, and multifocality, the remaining variables all showed significant differences among three surgery subgroups ( Table 1 , all P<0.05). The long-term outcomes among three groups were different (P<0.001).

We then explored the association between the types of surgery and long-term outcomes in patients with DTC by competitive risk model. The univariable analysis showed that types of surgery were significantly related to DTC-related death (stat=10.706, P=0.005). It was also associated with the other causes death of patients (stat=21.310, P<0.001). These results showed the importance of types of surgery on the long-term outcomes of patients with DTC.

Next, we performed the multivariable competitive risk model analysis to explore the independent association between the types of surgery and long-term outcomes ( Table 2 ). The results indicated that Surgery 2 was related to a higher risk of DTC-related death in adjusted model 1 (HR=7.966, P=0.003), model 2 (HR=6.242, P=0.012), and model 3 (HR=6.198, P=0.010). However, Surgery 3 was related to a lower risk of other causes death in adjusted model 1 (HR=0.786, P=0.037), model 2 (HR=0.729, P=0.008), and model 3 (HR=0.721, P=0.006). In addition, age, gender, and N stage were the common independent factors of DTC and other causes death (all P<0.05, data not shown).

We subsequently explored the influence of types of surgery on the cumulative incidence rate of DTC and other causes death. Among whole populations ( Figure 1A ), Surgery 2 caused the highest cumulative incidence rate of DTC-related death while Surgery 3 caused the lowest other causes death, which were consistent with the results obtained from the above multivariable competitive risk analysis. Age, gender, and N stage were the common independent factors of DTC and other causes death. Therefore, we also explored the influence of types of surgery on the long-term outcomes among these subgroups. Its influence in patients with N0 stage and these patients with age>55 years old ( Figures 1B, D ) was similar with that of whole populations. However, among female patients ( Figure 1C , P<0.001) and those with age ≤ 55 years old ( Figure 1D , P=0.043), types of surgery were only related to the other causes death and Surgery 2 caused the highest cumulative incidence rate of other causes death.

Further, we refined the death of other causes to four categories, including cardiovascular diseases related death, infectious diseases related death, respiratory diseases related death, and gastrointestinal diseases related death. We explored the influence of types of surgery on the cumulative incidence of DTC-related death and four types of causes death by competitive risk analysis. The results showed that types of surgery can significantly influence the cumulative incidence of DTC-related death no matter what kind of disease was set as its competitive event ( Figures 2A-D , all P<0.05). However, our result also showed potential influence of types of surgery on the cardiovascular diseases related death (P=0.046), which suggested the importance of types of surgery on the cardiovascular diseases related death.

We next performed a multivariable competitive risk model analysis to explore the independent association between the types of surgery and DTC or cardiovascular diseases related death ( Table 3 ). When cardiovascular diseases related death was detailly set as the competitive event of DTC-related death, the types of surgery were still related to the DTC-related death both in model 1 (HR=7.765, P=0.004) and model 2 (HR=6.186, P=0.012). However, the independent association between types of surgery and cardiovascular diseases related death was not observed after adjusting for several variables (all P>0.05). These results further highlighted the importance of types of surgery on the DTC-related death and their stable association.

Above results have demonstrated significant association between types of surgery and long-term outcomes of patients with DTC. We finally evaluated the importance degree of types of surgery among all variables by 3 machine learning algorithms based on Logistic, Random forest, and XGBoost methods ( Table 4 ). Among top 5 variables from 3 algorithms, gender was the common factor, implying the importance of gender on the long-term outcomes of patients. However, the ranking of the types of surgery was not observed among the top 10 variables. The results highlight the potential for surgical modality to be neglected in longitudinal assessments of patient prognosis.