Thyroid cancer has demonstrated the most rapid rise in incidence among endocrine malignancies globally in recent decades. By 2020, the global incidence surpassed 500,000 new cases annually, with over 200,000 in China alone (1, 2). Despite differentiated thyroid cancer (DTC) comprising the majority of cases, often associated with favorable outcomes and low mortality, a subset of patients will experience relapse or metastasis, developing resistance to radioactive iodine (RAI), thus creating a substantial therapeutic obstacle (3). Postoperative management of DTC now primarily relies on risk-based assessments, prioritizing individualized approaches and ongoing adjustments. Prognostic evaluation is currently guided by the 8th edition of the American Joint Committee on Cancer Tumor-Node-Metastasis (AJCC TNM) staging system and the 2015 American Thyroid Association risk stratification system (ATA RSS), both of which are widely adopted methodologies (4). The 8th edition of the AJCC TNM staging system provides a more precise reflection of mortality risk. However, for evaluating postoperative recurrence risk in DTC patients with favorable prognoses, greater reliance on supplementary risk assessment tools is warranted (5). The 2015 ATA RSS serves as a key prognostic framework following DTC surgery. In a study with a median follow-up of 4 years, 31% of patients in the ATA high-risk group achieved no evidence of disease (NED), compared to 52% in the intermediate-risk group and 78% in the low-risk group (6). A separate short-term study reported recurrence rates of 22.8%, 6.7%, and 2.9% for the high-, intermediate-, and low-risk groups, respectively (7). These findings indicate that the ATA RSS stratifies recurrence risk with greater precision. In various regions, including China, the ATA RSS remains a foundational tool for postoperative risk assessment. However, the ATA RSS framework omits several critical individual variables, such as pathological subtypes, multifocality, extranodal invasion, and molecular profiles, resulting in substantial prognostic variability even among cases classified within the same risk group (4). Continued research and accumulating evidence are required to refine and enhance the accuracy of this stratification system. In the three risk classes, persistent and recurrent lesions frequently occur in high-risk DTC cases, posing significant obstacles to effective treatment (8, 9). Further investigation into prognostic factors for patients within this risk category remains essential.

Research into the molecular mechanisms of thyroid cancer has evolved over several decades, with early studies identifying key mutations in BRAF and RAS, along with fusion events in the RET and NTRK genes (10, 11). These genetic alterations critically affect the clinical and pathological characteristics of thyroid cancer by activating the mitogen activated protein kinase (MAPK) and/or PI3K/AKT/mTOR signaling pathways. Such oncogene mutations are present throughout all stages of thyroid cancer and are now recognized as early drivers of tumorigenesis (12). Molecular diagnostic tools that detect common early driver genes, such as BRAF V600E, RAS, RET, and PAX8/PPARG, when combined with fine needle aspiration biopsy (FNAB), have been shown to substantially enhance diagnostic accuracy, a finding supported by numerous studies and widely accepted in the field (13–16). Some of these driver alterations, such as BRAF V600E, RET, NTRK fusion, and ALK fusion, also serve as therapeutic targets for selective kinase inhibitors. Compared to multi-targeted kinase inhibitors, these agents are often preferred due to their superior efficacy and safety profiles, underscoring the heightened clinical value of molecular testing in managing aggressive and RAI-refractory thyroid cancers (17). Mutations in the PI3K/AKT/mTOR pathway, including PIK3CA, AKT1 and PTEN are infrequent in thyroid cancer, though they can act as early driver mutations in certain RAS-like subtypes. More commonly, however, they co-occur with other driver mutations in advanced, poorly differentiated cases (12, 18). Additionally, mutations in the TERT promoter (TERTp) and tumor suppressor genes like TP53 and CNKN2A frequently co-mutate with driver genes. These mutations are closely linked to adverse prognostic indicators, such as recurrence, metastasis, and iodine resistance. Importantly, they are regarded as late-stage mutations that drive thyroid cancer dedifferentiation and progression to high-grade or undifferentiated forms (12, 18). These previous findings have enabled prognostic stratification of certain thyroid cancers through the integration of specific genetic markers in molecular testing. In recent years, several studies have sought to elucidate the association between the molecular profiles of thyroid cancer and clinical risk and prognosis. Yip et al. conducted a case-control study analyzing the genetic variation in DTC patients with distant metastases, proposing molecular risk stratification based on these variations, with distinct groups showing differing predicted rates of distant metastasis (19). Liu et al. and Schumm et al. further validated the relationship between this molecular risk stratification and tumor recurrence through retrospective cohort studies (7, 20). Collectively, these studies have highlighted the feasibility of molecular testing in enhancing current risk assessment models and introduced novel perspectives for its application in thyroid cancer management.

The molecular profiles of ATA high-risk DTC patients and its association with prognosis remain unexplored. This retrospective cohort study analyzed the clinical characteristics and follow-up data of ATA high-risk individuals who underwent 18-gene next-generation sequencing (NGS) test, aiming to evaluate the potential of incorporating molecular profiles to refine postoperative recurrence risk stratification in this high-risk population.

ATA high-risk cases in DTC pose significant treatment challenges and are associated with poorer outcomes. Research on these high-risk cases remains limited, with considerable variation in reported characteristics across studies. A Dutch single-center study by Van Velsen et al., involving 236 ATA high-risk DTC patients, identified FTC in 32% of cases, ETE in 32%, cervical lymph node dissection in 45%, and distant metastasis in 33% (19). In contrast, a retrospective single-center study by Shah et al. in the United States, analyzing 320 ATA high-risk DTC cases, reported FTC in 4.1%, ETE in 94%, cervical lymph node metastasis (N1a+N1b) in 62.5%, and distant metastasis in 13.4%. The cohort characteristics observed in this study align with those reported by Shah et al., with a minimal incidence of FTC, most complicated by ETE, and 14.8% presenting with distant metastasis. These variations likely stem from differences in the pathological composition. In contrast, the higher proportion of FTC in van Velsen et al.’s study may explain the lower rates of ETE and lymph node metastasis, alongside a higher incidence of distant metastasis. Notably, the rate of cervical lymph node metastasis in this study exceeded that of the two aforementioned studies. Potential explanations include: (1) China’s diagnostic and treatment protocols still recommend routine prophylactic central cervical lymph node dissection, leading to the detection of subclinical metastases in cases, including those initially staged as cN0; (2) this study employed a more positive pre-treatment evaluation strategy for lateral cervical lymph nodes. For suspicious lesions detected on ultrasound or CT, ultrasound-guided FNAB combined with Tg measurement in puncture eluent enhanced the detection rate of lateral cervical lymph node metastasis (25). Additionally, 64.8% of the cases in this study involved multifocal lesions, with more than half (54.6%) exhibiting lateral cervical lymph node metastasis, and 44.4% showing ENE. These features align with previous reports on the characteristics of advanced DTC (3).

The NGS molecular assay used in this study incorporated an 18-gene panel, shown in prior research to improve diagnostic precision for FNAB (15). In contrast to widely adopted large NGS panel like ThyroSeq V3 and Afirma GSC, this assay targets the 18 most prevalent genes in Chinese thyroid cancer cases, optimizing cost-effectiveness. Previous studies employing large NGS panel reported variant detection rates ranging from 86.5% to 95.2% (20, 26, 27). In this study, clinically significant variants were identified in 87.8% of ATA high-risk DTC cases, with driver variants detected in 92.6%, reflecting comparable detection efficiency. Additionally, this molecular test covers several key prognosis-related genes, including TERTp, TP53, PIK3CA, AKT1, and PTEN, enhancing the ability to identify high-risk molecular profiles. Nevertheless, given the limited scope of molecular testing in this study, we believe that the molecular backgrounds of the “no alteration detected” cases remains unknown and may vary on an individual basis, which is why they were excluded from the study. Among the ATA high-risk cohort, the most prevalent driver variants were BRAF V600E, NRAS, KRAS mutations, and RET and NTRK fusions, aligning with findings from a recent Chinese study focused primarily on DTC (26). This distribution is also consistent with another report from China involving PTC cases with high recurrence risk (28). In this cohort, TERTp and PIK3CA mutations were identified in 24.1% and 5.6% of cases, respectively, notably exceeding the frequencies reported in other cohorts with varying recurrence risks. For instance, Du et al. reported TERTp and PIK3CA mutations in 6.3% and 1.5% of cases, respectively (26). Similarly, in a meta-analysis focusing solely on PTC, the average incidence of TERTp mutations was 10.1% (21). The elevated frequency of high-risk molecular alterations in the ATA high-risk DTC cohort suggests that prognostic gene detection in this group may hold enhanced clinical relevance. This study found that high-risk molecular alterations were consistently associated with BRAF V600E or NRAS mutations, while no such mutations were detected in the 15 cases exhibiting RET or NTRK fusions, aligning with findings from other large-scale studies. For instance, TCGA reports indicate that TERTp mutations are linked to driver mutations or arm-level somatic copy number alterations, rather than BRAF mutations or gene fusions (29). In the study by Du et al., only one case of RET fusion and one of NTRK1 fusion were accompanied by TERTp mutations (26). However, other studies have noted the co-existence of RET fusions and TERTp mutations, which may impact prognosis and the response to RET and MEK targeted therapies (30). Further investigation into the association between gene fusion variants and late-stage alterations, such as TERTp mutations, is warranted as additional data becomes available.

A 2021 study by Yip et al. first proposed a novel molecular risk classification divided into three categories based on ThyroSeq, version 3 molecular testing and ThyroSeq Cancer Risk Classifier (19). This classification has since been adopted in several subsequent studies, with its clinical relevance consistently validated (7, 20, 22). The present study adapted this classification, modifying the risk grouping to align with the specific molecular tests used and the unique characteristics of the study cohort. Due to the limitations of molecular testing in this study in detecting certain early driver genes (e.g., ALK, DICER1, EIF1AX, etc.) and specific variant forms (e.g., gene expression alterations, copy number alterations, etc.), “concurrent early variants” were not considered a requirement for inclusion in the HMR group. Additionally, possibly influenced by the cohort characteristics, only two cases exhibited “RAS-like” molecular profiles, which were defined as “low molecular risk”. Consequently, the sample size for analyzing “low-risk” cases was insufficient, the non-HMR cases were not further stratified into “low-risk” or “intermediate-risk”. In the study by Liu et al., increasing molecular risk was associated with higher mean patient age, a lower proportion of female patients, larger median tumor size, and more frequent occurrences of aggressive PTC subtypes, ETE, cervical lymph node metastasis, and synchronous distant metastasis (22). Schumm et al. also identified a correlation between higher molecular risk groups and more aggressive clinical features, including larger primary tumors, ETE, and positive surgical margins (20). Similarly, in the HMR group of this study, clinical parameters such as age, tumor size, aggressive pathological subtypes, and distant metastasis demonstrated comparable patterns. Although TERTp mutations are well-documented as being significantly associated with RAI resistance (21, 31), the relationship between molecular risk stratification and RAI refractory status has not been explored. In this cohort, 25 patients underwent therapeutic RAI for unresectable lesions. While the HMR group exhibited a higher prevalence of RAIR cases, the difference was not statistically significant, likely due to the limited sample size. Further investigation is required to determine whether MCGs can reliably predict RAI treatment efficacy, with larger sample sizes necessary.

Previous studies have demonstrated a significant correlation between high-risk molecular variants and both RFS and SR following DTC surgery, with a strong link to distant metastasis. For example, Liu et al. observed a substantially lower 36-month RFS in the high molecular risk group compared to the intermediate and low-risk groups, with a recurrence risk exceeding threefold that of the latter two groups (22). Similarly, Schumm et al. found that patients in the high molecular risk group exhibited an elevated risk of SR (HR: 9.31) and distant metastasis (HR: 42.7) compared to those in the intermediate-risk group (20). In our study, a median follow-up of 32.5 months revealed significant differences in key prognostic indicators, including local recurrence, metachronous distant metastasis, and RFS across various MCGs following initial surgery. Notably, 80% of metachronous distant metastasis cases occurred in the HMR group. In contrast, the non-HMR group exhibited markedly lower rates of SR (10.1%) and metachronous distant metastasis (2.9%) compared to the overall study cohort, and significantly below historical data for ATA high-risk DTC (8, 9). All three recorded deaths occurred within the HMR group. Although between-group differences in mortality were significant, the limited follow-up duration and small number of death cases precluded survival analysis. These results indicate that for ATA high-risk DTC, the presence of high-risk variants serves as a strong predictor of adverse prognostic events during the initial treatment phase.

Some prognostic factors, such as age, tumor size, and pathological subtype, exhibit variability across different MCGs as previously noted, are currently employed to evaluate prognostic risk during the initial treatment phase of DTC (23). We hypothesized that high-risk genetic alterations contribute to the correlation between risk-related clinical features and poor prognosis, a relationship supported by emerging evidence. For instance, age is widely recognized as a key prognostic factor in DTC, as demonstrated in numerous studies (5, 32), and is a critical factor in the 8th AJCC TNM staging system. Nevertheless, research by Heo et al. found that TERTp mutations mediated the effect of age at diagnosis on the mortality rate by 36% in DTC (33). In this study, while age, tumor size, pathological subtype, and N stage appeared to be associated with RFS during initial univariate analysis, only MCGs (HR: 5.77) and tumor size (HR: 1.32) remained independently linked to RFS after incorporating the MCGs variable into a multivariable model. This finding aligns with reports by Liu et al. and Schumm et al., whose multivariate models similarly indicated that recurrence risk was solely related to tumor size and molecular risk (7, 20). Despite differences in cohort composition, the consistent results across these three DTC studies suggest that molecular profiles may be independently and more directly predictive of prognosis.

Several studies on “molecular risk stratification” have highlighted the importance of integrating routine molecular testing via preoperative FNAB, as it offers valuable insights into the molecular risk stratification prior to surgery, thereby shaping the initial treatment approach (7, 20, 22). However, despite its efficacy in informing clinical decisions, the broad adoption of NGS with large panels is hindered by substantial costs and limited access, restricting its routine application in many regions for thyroid cancer management. This study indicated that when NGS testing was employed for prognostic risk stratification in thyroid cancer, the use of small panels targeting key prognostic loci (such as TERTp, PIK3CA, TP53, AKT1, etc.) offered comparable efficiency in risk grouping. In regions where routine preoperative molecular testing is not feasible, a more targeted and cost-effective approach would be better to perform supplementary postoperative molecular testing specifically for ATA high-risk cases, allowing for accurate risk stratification. Given the elevated risk of recurrence and distant metastasis in patients with both high-risk ATA RSS and adverse molecular profiles, closer follow-up is warranted. Further research is needed to determine whether more positive RAI adjuvant therapies or alternative treatments may offer enhanced benefits for this subgroup.

Despite strengths, several limitations affect this study. First, as a retrospective cohort study, only cases with completed molecular testing were included, introducing potential selection bias due to the two-stage process of physician explanation and patient consent required for testing. Second, compliance issues led to incomplete RAI treatment in some cases, potentially skewing overall prognostic outcomes compared to patients who received the full standard treatment. Additionally, this non-compliance complicates a standardized analysis of postoperative serum Tg levels and biochemical incomplete responses within the study cohort. Given the lack of significant differences in case distribution across various MCG groups, this factor is unlikely to decisively affect the prognostic variation between the groups. Additionally, the small sample size limits the feasibility of performing propensity score matching based on different characteristics, which could have enhanced the precision of the statistical analysis. Furthermore, the short follow-up period constrains the study to short-term prognostic outcomes, and the long-term prognostic implications of different MCGs will require extended follow-up for more comprehensive assessment.

In conclusion, our study indicated a higher incidence of structural recurrence and poorer RFS in ATA RSS high risk DTC presented with high molecular risk. Molecular testing and MCGs offer significant potential for refining prognosis in high-risk DTC cases and merit broader application. The potential need for specialized postoperative management strategies for ATA high-risk cases with high molecular risk warrants further investigation.