Construction and external validation of a nomogram for predicting 10-year recurrence risk after radical resection in Chinese patients with stage III colorectal cancer

Introduction

Colorectal cancer (CRC) ranks among the leading causes of cancer incidence and mortality worldwide (1). In 2020, approximately 1.9 million new CRC cases and 935,000 deaths were reported globally, making it the third most common malignancy and the second leading cause of cancer-related death (2). Stage III CRC, characterized by lymph node metastasis [American Joint Committee on Cancer (AJCC) stage III] and deeper tumor invasion, is associated with markedly worse prognosis. Despite substantial progress in comprehensive treatment strategies for CRC over recent decades, accurate prediction of postoperative recurrence risk in stage III patients remains challenging (3). Recurrence is one of the major determinants of long-term survival in CRC; it has been reported that approximately 20–35% of patients with stage II–III CRC who undergo curative resection experience recurrence within 5 years postoperatively (4). While most recurrences occur within the first 3–5 years, late recurrences beyond 5 years are not uncommon, underscoring the need for long-term surveillance to monitor disease progression (5).

In a large population-based study, the 5-year cumulative recurrence rate of stage III colon cancer decreased from 35.3% to 24.6% with advances in diagnosis and treatment (6); however, recurrence risk remains substantial for high-risk patients. Analysis of the International Duration Evaluation of Adjuvant chemotherapy (IDEA) trial data revealed marked heterogeneity in prognosis across subgroups of stage III colon cancer: the lowest-risk group (T1N1a) had a 5-year disease-free survival rate of as high as approximately 89%, whereas the highest-risk group (T4N2b) had only about 31% (7). These findings indicate that reliance solely on tumor-node-metastasis (TNM) staging is insufficient to identify high-risk individuals within stage III patients, highlighting the need for more refined risk stratification tools. In recent years, multivariable prognostic models such as nomograms have demonstrated significant value in CRC prognosis assessment, integrating multiple clinicopathological factors to provide individualized risk estimation, often with superior predictive performance compared to traditional TNM staging (4). However, studies focusing on long-term recurrence risk prediction after curative resection for stage III CRC remain limited, particularly those based on large-scale populations with independent external validation.

This study was based on large-scale data from the U.S. Surveillance, Epidemiology, and End Results (SEER) cancer registry, extracting follow-up records of Chinese patients with stage III CRC who underwent curative resection. A nomogram model was developed to predict 5- and 10-year recurrence probabilities, and external validation was performed in a retrospective institutional cohort. Our objective was to establish a reliable recurrence risk prediction model to assist clinicians in achieving more refined risk assessment for postoperative surveillance planning and adjuvant treatment decision-making. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2402/rc).

Methods

Study design

This study was a retrospective multi-cohort analysis aimed at developing a nomogram to predict 5- and 10-year recurrence probabilities after curative resection of stage III CRC using large-scale SEER data, with external validation in an independent institutional cohort. The overall workflow included: screening and selection of eligible cases from the SEER cohort to construct the training set; conducting baseline comparisons, univariate and multivariate Cox regression analyses; generating the nomogram; and evaluating its discrimination, calibration, and clinical utility through decision curve analysis (DCA). The model was subsequently applied to a retrospective follow-up cohort from The Ninth Hospital of Hangzhou for external validation (assessing discrimination, calibration, and net clinical benefit). Additionally, 10-year cumulative recurrence risk curves were plotted for both cohorts, and subgroup analysis curves were generated from the SEER dataset. Importantly, because SEER does not routinely capture clinically adjudicated recurrence, recurrence-related outcomes in population-based registries are commonly operationalized using available proxy indicators (e.g., subsequent cancer diagnoses and cancer-specific mortality), as detailed in Figure 1 (8).

Figure 1 The flow diagram of participant inclusion and exclusion. A1, dataset of patients who experienced a second diagnosis of colorectal cancer within 10 years; A2, dataset of patients who died from colorectal cancer within 10 years; A3, merged proxy recurrence dataset; B1, dataset of patients who did not experience a second diagnosis of colorectal cancer within 10 years; B2, dataset of patients who did not die from colorectal cancer within 10 years; B3, merged non-recurrence dataset. ICD-O-3, International Classification of Diseases for Oncology, Third Edition; SEER, Surveillance, Epidemiology, and End Results; WHO, World Health Organization.

Study population

Data were obtained from the SEER Research Plus, 17 Registries, Nov 2024 Sub [2000–2022], restricted to CRC cases [site recode International Classification of Diseases for Oncology, Third Edition/World Health Organization (ICD-O-3/WHO) 2008: colorectum] and limited to the Chinese population. To distinguish between “individuals with potential multiple diagnoses” and “single primary cases”, two types of records were extracted: (I) 1st of 2 or more primaries, 2nd of 2 or more primaries (N=3,672); and (II) one primary only (N=6,244).

Inclusion criteria were: initially diagnosed as AJCC stage III, receipt of curative surgical resection, age ≥18 years, and first diagnosis between 2000 and 2012 (to ensure a 10-year observation window).

Exclusion criteria included: duplicate cases; initial diagnosis not stage III; no curative resection performed; diagnosis of a second primary tumor/recurrence or death occurring within 1 month after the initial diagnosis (to avoid misclassification of synchronous lesions or perioperative outcomes); and records where the “second admission” was used as the index date. The complete SEER inclusion and exclusion procedure is illustrated in Figure 1. After sequential screening and merging, a final SEER analysis cohort of N=1,114 was obtained.

External validation cohort (institutional cohort): consecutive patients with pathologically confirmed AJCC stage III CRC who underwent curative resection at The Ninth Hospital of Hangzhou between 2012 and 2014 were retrospectively identified through the hospital medical records system. Clinicopathological and treatment data were extracted, and long-term follow-up information was supplemented via outpatient visits, telephone contacts, or inpatient reassessments to determine 10-year recurrence outcomes and timing. Consistent with the SEER cohort, major exclusion criteria included missing key outcome or exposure data, perioperative death, and clear evidence of synchronous second primaries. A total of 92 cases were included as the external validation cohort.

Data extraction and collection

In the “multiple primaries” branch (N=3,672), after de-duplication, 2,692 cases were retained. We further excluded patients with initial diagnosis other than stage III (n=778), those who did not undergo curative surgery (n=2), cases with a second diagnosis/recurrence recorded within 1 month (n=100), and individuals whose “second diagnosis” served as the index event (n=50), resulting in N=50. Among these, 49 experienced a second diagnosis within 10 years (defined as recurrence, A1 =49), while 1 did not (B1 =1).

In the “single primary” branch (N=6,244), exclusions were applied for initial diagnosis other than stage III (n=5,109), absence of curative surgery (n=33), and death within 1 month (n=19), yielding N=1,083. Of these, 333 patients died of CRC within 10 years (A2 =333), and 750 did not (B2 =750). A1 and A2 were then merged into A3 =382, while B1 and B2 were merged into B3 =751. After additionally excluding 19 cases with first diagnosis outside 2000–2012, the final analytic cohort comprised N=1,114 (see Figure 1).

Institutional patient data collection: demographic, oncological, and treatment information was extracted from the electronic medical record system according to a standardized variable dictionary. Follow-up data were obtained through a combination of outpatient visits and telephone follow-up. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of The Ninth Hospital of Hangzhou (No. 2025-024). Because of the retrospective nature of the study and the use of de-identified data, the requirement for informed consent was waived by the ethics committee.

Statistical analysis

The index date was defined as the date of first curative surgery, and the event time as the date of first confirmed recurrence or second diagnosis. Patients without recurrence were censored at the last follow-up or at 10 years. Baseline characteristics were summarized as frequencies (percentages), and group comparisons were performed using the χ² test or Fisher’s exact test (when expected counts <5).

Cox proportional hazards models were constructed to identify risk factors for recurrence. Variables with statistical significance in univariate analysis (or clinical relevance) were entered into multivariable modeling, with hazard ratios (HRs) and 95% confidence intervals (CIs) reported. Two-sided P<0.05 was considered statistically significant. Based on the regression coefficients from the multivariable Cox model, a nomogram was developed to estimate individualized 5- and 10-year recurrence probabilities.

Model performance was assessed as follows: discrimination was evaluated using time-dependent receiver operating characteristic (ROC) analysis to calculate area under the curve (AUC) at 5 and 10 years; calibration was assessed by comparing predicted probabilities with observed outcomes using calibration plots; and clinical utility was evaluated by DCA across a range of threshold probabilities. The nomogram was then applied to the institutional validation cohort, where discrimination, calibration, and DCA were reassessed.

Additionally, cumulative 10-year recurrence risk curves were generated for the SEER cohort, and stratified Kaplan-Meier curves [by age, carcinoembryonic antigen (CEA) status, histology, T/N stage, and tumor size] were constructed with log-rank tests performed for group comparisons. All statistical analyses and visualizations followed standard survival analysis procedures. Given the retrospective design, all P values were interpreted as exploratory.

Results

Baseline characteristics of the SEER cohort

A total of 1,114 patients with stage III CRC who underwent curative resection were included in the SEER training cohort, among whom 363 (32.6%) experienced recurrence within 10 years. Baseline comparisons revealed significant differences between the recurrence and non-recurrence groups with respect to age, histological type, T stage, N stage, CEA status, tumor size, and marital status, whereas no significant differences were observed for sex, receipt of radiotherapy or chemotherapy, year of diagnosis, or primary tumor site.

Patients aged ≥80 years had a 10-year recurrence rate of 48.1%, significantly higher than that of patients aged 25–39 years (24.4%) and 40–59 years (28.3%) (P<0.001). Mucinous adenocarcinoma (MAC) and other rare histological subtypes showed recurrence rates of 43.2% and 57.1%, respectively, compared with 31.2% for conventional adenocarcinoma (P=0.005). Recurrence rates increased with deeper tumor invasion and greater nodal burden: 10.0% for T1, 18.7% for T2, 32.1% for T3, and 53.8% for T4 (P<0.001); and 26.6% for N1 versus 45.3% for N2 (P<0.001). Patients with positive CEA had a recurrence rate of 43.1%, compared with 27.5% in CEA-negative cases (P=0.03). Tumors >3 cm were associated with a higher recurrence rate (35.6%) compared with tumors ≤3 cm (24.9%) (P<0.001). Regarding marital status, widowed patients had a recurrence rate of 43.0%, higher than that of married patients (30.5%) (P=0.041). Radiotherapy and chemotherapy showed no significant differences between groups in the overall cohort (P=0.57 and 0.08, respectively) (Table 1).

Univariate and multivariate Cox regression analyses

Univariate analysis demonstrated that advanced age, MAC/other histological subtypes, T3–T4 stage, N2 stage, positive CEA status, and tumor diameter >3 cm were all associated with an increased risk of 10-year recurrence (Table 2). In the multivariate model, age ≥80 years remained an independent risk factor (HR =3.89, 95% CI: 1.97–7.70, P<0.001). Patients with “other” histological subtypes had a significantly higher risk compared with conventional adenocarcinoma (HR =1.92, 95% CI: 1.06–3.49, P=0.03), while MAC showed a borderline association (HR =1.41, P=0.058). Compared with T1 stage, both T3 and T4 stages were associated with significantly increased recurrence risk (HR =2.67, 95% CI: 1.25–5.73, P=0.01; HR =5.03, 95% CI: 2.30–11.01, P<0.001, respectively). Similarly, N2 stage conferred a significantly higher risk than N1 (HR =1.82, 95% CI: 1.47–2.25, P<0.001). After multivariable adjustment, positive CEA status (HR =1.45, P=0.10) and tumor diameter >3 cm (HR =1.26, P=0.08) did not reach statistical significance. Marital status also lost significance in the multivariate model. These findings suggest that the primary independent drivers of postoperative recurrence in stage III CRC are tumor burden-related factors (T and N stage), followed by very advanced age and certain unfavorable histological subtypes.

Baseline results of the external validation cohort

A total of 92 patients with stage III CRC who underwent curative resection at our institution between 2012 and 2014 were included, among whom 30 experienced recurrence within 10 years, yielding an overall recurrence rate of 32.6% (Table 3). Due to the limited sample size, most variables showed no statistically significant differences between the recurrence and non-recurrence groups. Notably, the recurrence rate in N2 patients was 55.0%, compared with 30.4% in N1 patients, showing a near-significant trend (χ²=3.57, P=0.059). Similarly, both T stage and MAC demonstrated trends toward increased recurrence risk consistent with the SEER findings. Patients who received chemotherapy or radiotherapy exhibited lower overall recurrence rates, although these differences did not reach statistical significance (chemotherapy P=0.23; radiotherapy P=0.15).

Construction and performance evaluation of the prediction model

A nomogram was developed based on the significant variables identified in the multivariate Cox regression analysis from the SEER training cohort, incorporating six predictors: age, histological type, T stage, N stage, CEA status, and tumor size (marital status was excluded as it was not significant after adjustment; see Figures 2,3). The nomogram provided individualized estimates of recurrence probability at 5 and 10 years. In terms of discrimination, the AUC was 0.74 (95% CI: 0.70–0.78) at 5 years and 0.73 (95% CI: 0.70–0.77) at 10 years in the training cohort. In the external validation cohort, the AUCs for both 5- and 10-year predictions were 0.77 (95% CI: 0.63–0.90 and 0.64–0.91, respectively) (Figure 3A-3D). For calibration, the 5- and 10-year calibration curves in the training cohort showed excellent agreement with the ideal line, while the external validation cohort demonstrated a similar trend though with wider CIs (Figure 3E-3H). DCA indicated that in the training cohort, the model provided greater net benefit compared with “treat-all” or “treat-none” strategies within a threshold probability range of approximately 0.10–0.60. In the external validation cohort, a comparable net benefit was observed within the threshold probability range of approximately 0.10–0.40 (Figure 3I-3L).

Figure 2 Nomogram based on SEER cohort for predicting the 5- and 10-year probability of recurrence in patients with stage III colorectal cancer. AJCC, American Joint Committee on Cancer; CEA, carcinoembryonic antigen; N, node; NOS, not otherwise specified; SEER, Surveillance, Epidemiology, and End Results; T, tumor.

Figure 3 Discrimination, calibration, and clinical utility of the nomogram for predicting 5- and 10-year recurrence in stage III colorectal cancer patients. (A,B) ROC curves of the nomogram for predicting 5-year (A) and 10-year (B) recurrence in the SEER training cohort, with AUCs of 0.74 (95% CI: 0.70–0.78) and 0.73 (95% CI: 0.70–0.77), respectively. (C,D) ROC curves of the nomogram for predicting 5-year (C) and 10-year (D) recurrence in the external validation cohort (institutional data), with AUCs of 0.77 (95% CI: 0.63–0.90) and 0.77 (95% CI: 0.64–0.91), respectively. (E,F) Calibration plots for 5-year (E) and 10-year (F) recurrence in the SEER training cohort, showing good agreement between predicted probabilities and observed outcomes. (G,H) Calibration plots for 5-year (G) and 10-year (H) recurrence in the external validation cohort, demonstrating overall consistency between predicted and observed probabilities, although with wider confidence intervals due to limited sample size. (I,J) DCA for 5-year (I) and 10-year (J) recurrence in the SEER training cohort, indicating that the nomogram provides higher net clinical benefit across a wide range of threshold probabilities compared with the “treat all” or “treat none” strategies. (K,L) DCA for 5-year (K) and 10-year (L) recurrence in the external validation cohort, further confirming the clinical usefulness of the nomogram in individualized risk stratification. AUC, area under the curve; CI, confidence interval; DCA, decision curve analysis; ROC, receiver operating characteristic; SEER, Surveillance, Epidemiology, and End Results.

Ten-year recurrence risk curves in the SEER and external validation cohorts

Kaplan-Meier cumulative recurrence curves for both cohorts (Figure 4) demonstrated a continuous accumulation of postoperative recurrence risk over time, with the first 5 years representing the period of highest incidence, followed by a flattening of the slope thereafter. In the SEER training cohort, the cumulative recurrence probability at 10 years was approximately 45–50%. In the external validation cohort, the 10-year cumulative recurrence probability was also close to 50%, showing an overall trend consistent with the SEER cohort, although with wider CIs due to the smaller sample size.

Figure 4 Kaplan-Meier curves of recurrence probability within 10 years after curative resection for stage III colorectal cancer. (A) Cumulative probability of recurrence in the SEER training cohort (n=1,114). The recurrence risk steadily increased over time, reaching approximately 45–50% at 10 years, with most events occurring within the first 5 years and then plateauing thereafter. (B) Cumulative probability of recurrence in the external validation cohort (institutional data, n=92). A similar upward trend was observed, with recurrence risk approaching 50% at 10 years. Wider confidence intervals reflect the smaller sample size in the validation set. SEER, Surveillance, Epidemiology, and End Results.

Subgroup analysis in the SEER cohort

Stratified Kaplan-Meier curves and log-rank tests further confirmed the associations between key clinicopathological features and recurrence risk (Figure 5). Compared with patients aged 25–59 years, those aged ≥60 years had a higher recurrence risk (HR =1.574, 95% CI: 1.262–1.963, P<0.001). Patients with positive CEA status were at increased risk compared with CEA-negative patients (HR =1.686, 95% CI: 1.095–2.596, P=0.02). MAC was associated with higher recurrence risk relative to conventional adenocarcinoma (HR =1.557, 95% CI: 1.098–2.207, P=0.01). The most pronounced differences were observed in tumor burden–related stratifications: T3–T4 versus T1–T2 showed significantly elevated recurrence risk (HR =2.877, 95% CI: 1.980–4.181, P<0.001), and N2 stage conferred a markedly higher risk than N1 (HR =1.987, 95% CI: 1.616–2.445, P<0.001). In addition, tumor diameter ≥3 cm was associated with greater recurrence risk compared with <3 cm (HR =1.576, 95% CI: 1.228–2.022, P<0.001).

Figure 5 Subgroup analyses of recurrence probability within 10 years after curative resection for stage III colorectal cancer in the SEER training cohort. (A) Age: patients aged ≥60 years had significantly higher recurrence risk compared with those aged 25–59 years (HR =1.574, 95% CI: 1.262–1.963, log-rank P<0.001). (B) CEA status: CEA-positive patients showed a higher probability of recurrence than CEA-negative patients (HR =1.686, 95% CI: 1.095–2.596, log-rank P=0.02). (C) Histology: mucinous adenocarcinoma was associated with a higher recurrence risk than adenocarcinoma, NOS (HR =1.557, 95% CI: 1.098–2.207, log-rank P=0.01). (D) T stage: patients with T3–T4 tumors exhibited significantly increased recurrence risk compared with those with T1–T2 disease (HR =2.877, 95% CI: 1.980–4.181, log-rank P<0.001). (E) N stage: N2 disease was associated with markedly higher recurrence probability compared with N1 (HR =1.987, 95% CI: 1.616–2.445, log-rank P<0.001). (F) Tumor size: tumors ≥3 cm were associated with a greater risk of recurrence compared with those <3 cm (HR =1.576, 95% CI: 1.228–2.022, log-rank P<0.001). CEA, carcinoembryonic antigen; CI, confidence interval; HR, hazard ratio; N, node; NOS, not otherwise specified; SEER, Surveillance, Epidemiology, and End Results; T, tumor.

Discussion

In this study, we utilized the SEER database and an institutional cohort to model and validate the 10-year recurrence risk after curative resection for stage III CRC. The results showed that the 10-year cumulative recurrence rate in the SEER training cohort was 32.6%, which is consistent with previously reported ranges for postoperative recurrence risk in stage III CRC (9). A recent study has indicated that, with advances in treatment strategies, the recurrence risk in stage III colon cancer has declined, with the 5-year recurrence rate decreasing from approximately 35% to around 25% (6). Nevertheless, a considerable proportion of patients still eventually experience recurrence (4). In particular, patients with high tumor burden continue to have poor prognoses; for example, nearly 70% of patients in the T4N2b subgroup experience disease recurrence or progression within 5 years (4). Our findings similarly demonstrated a continuous accumulation of recurrence risk over time, with the first 5 years representing the peak period for recurrence. Although the recurrence curve flattened after 5 years, it did not completely stabilize, and by 10 years, the cumulative recurrence rate in both cohorts approached 50%. These results suggest that for patients with stage III CRC, follow-up duration should be extended to 10 years after surgery to enable timely detection and management of late recurrences.

Through analysis of risk factors associated with recurrence, we found that tumor burden–related indicators (T stage and N stage) were the predominant drivers of postoperative recurrence in stage III CRC. This was particularly evident in our data: recurrence rates increased significantly with greater tumor invasion depth and extent of lymph node involvement. The 10-year recurrence rate reached 53.8% in T4 patients, compared with only about 10.0% in T1 patients; similarly, patients with N2 lymph node metastasis had a recurrence rate of 45.3%, markedly higher than 26.6% for N1. Multivariate Cox regression further confirmed that T4 and N2 stages conferred significantly elevated recurrence risks compared with the reference groups (T1 and N1). These findings are consistent with the well-established prognostic role of TNM staging: the degree of local tumor invasion and regional nodal involvement directly influence recurrence probability (10). Indeed, even within stage III patients, substantial heterogeneity in disease-free survival exists across different T/N combinations (7). Therefore, it is unsurprising that tumor stage-related variables played a central role in our prediction model. Similarly, another recent study has also identified T stage and N stage as indispensable factors in CRC prognostic prediction (2).

Another notable finding of this study was the impact of advanced age on recurrence risk. Patients aged ≥80 years had a 10-year recurrence rate of 48.1%, significantly higher than that observed in younger groups (e.g., approximately 28% in those aged 25–59 years). In multivariate analysis, age over 80 remained an independent risk factor (HR ≈3.9), indicating that elderly patients face several-fold higher postoperative recurrence risk compared with younger counterparts. Several mechanisms may underlie this association. First, elderly patients often present with multiple comorbidities and reduced physiological reserve, which lower the likelihood of receiving and completing standard adjuvant chemotherapy; undertreatment in turn may increase recurrence risk (11). Although no statistically significant association between chemotherapy and recurrence-related outcomes was observed in this registry-based analysis, this finding should be interpreted with caution, given the presence of treatment selection bias (i.e., confounding by indication) and the limited granularity of chemotherapy regimen information in the SEER database, while adjuvant chemotherapy remains the standard of care for most patients with stage III CRC (12). Similarly, owing to limited sample size and the lack of detailed treatment sequencing in SEER, patients who received radiotherapy in either the neoadjuvant or adjuvant setting were grouped together for analysis. Although no significant association between radiotherapy and recurrence-related outcomes was observed, this result warrants cautious interpretation, especially in light of the routine application of neoadjuvant chemoradiotherapy in rectal cancer (13). Second, age-related decline in immune function, together with potentially more aggressive tumor biology in the elderly, may further contribute to higher recurrence rates (14). Therefore, in stage III patients of advanced age, postoperative follow-up and therapeutic decision-making should carefully account for their substantially elevated recurrence risk.

Histological subtype was also identified as an important prognostic factor. We observed that patients with MAC and other poorly differentiated or special histological types had significantly higher postoperative recurrence rates compared with those with conventional adenocarcinoma. In the SEER cohort, the 10-year recurrence rate in MAC patients was approximately 43.2%, significantly higher than 31.2% in non-MAC cases. Even after multivariable adjustment, MAC still showed a trend toward increased recurrence risk (HR ≈1.41, P=0.058), while other rarer unfavorable subtypes (e.g., signet-ring cell carcinoma) demonstrated an even greater risk compared with conventional adenocarcinoma (HR ≈1.92, P=0.03). These findings are consistent with previous reports indicating that the mucinous subtype of CRC is associated with poorer outcomes: MAC patients are often diagnosed at a later stage with greater tumor burden, exhibit reduced sensitivity to standard chemotherapy, and generally experience worse disease-free survival and overall survival compared with conventional adenocarcinoma patients (15). Subgroup analyses stratified by tumor location have similarly shown that within stage III disease, survival in MAC is significantly inferior to that of non-MAC (16). Therefore, for stage III patients with MAC, intensified postoperative surveillance and consideration of more aggressive adjuvant treatment strategies are warranted to mitigate recurrence risk.

Serum tumor marker CEA plays an important role in prognostic evaluation of CRC. In our study, patients with positive CEA had a 10-year recurrence rate of 43.1%, markedly higher than the 27.5% observed in CEA-negative patients. Although CEA did not reach statistical significance in the multivariable model (adjusted HR ≈1.45, P=0.10), its prognostic relevance should not be overlooked. CEA levels are closely correlated with tumor burden, and elevated preoperative CEA often indicates higher disease risk (17). Indeed, international guidelines have incorporated abnormal preoperative CEA as a high-risk factor for stage II colon cancer (18). We speculate that within our model, which already includes T and N staging, the predictive strength of CEA may have been partially “diluted”. Nevertheless, in stage III patients with elevated CEA, clinicians should remain vigilant and ensure closer surveillance to facilitate early detection of recurrence. Tumor size also emerged as a potentially relevant factor. Our univariate analysis demonstrated that patients with primary tumor diameter >3 cm had a recurrence rate of 35.6%, compared with 24.9% in those with tumors ≤3 cm. Although this factor did not remain independently significant in multivariable analysis, the trend suggests that tumor volume may partially reflect recurrence risk. Larger tumors are often associated with deeper invasion and more extensive lymph node involvement, thereby increasing the likelihood of recurrence (19).

Based on a multivariable Cox model, we developed a nomogram to predict postoperative recurrence in stage III CRC, incorporating six variables: age, histological type, T stage, N stage, CEA status, and tumor size. The model demonstrated good discrimination and calibration in both the SEER training cohort and the institutional validation cohort. In the training set, the AUCs were 0.74 and 0.73 at 5 and 10 years, respectively, while in the external validation set, the corresponding AUCs were approximately 0.77, indicating stable predictive performance. The discriminative ability of our model is comparable to that of similar models reported in the literature (most with a concordance index of around 0.70–0.75), though some models targeting specific endpoints have reported higher discriminatory power (20). Moreover, the nomogram provides an intuitive means of quantifying individualized recurrence probability at the 5- and 10-year timepoints, thereby supporting clinicians in tailoring follow-up strategies according to different risk levels. DCA further confirmed the clinical utility of the model: within a relatively broad range of threshold probabilities (10–50%), the use of this model for guiding intervention decisions yielded higher net benefit than either the “treat-all” or “treat-none” strategies. In other words, the model can help identify higher-risk patients for whom more intensive surveillance or preventive interventions may be warranted. Such risk-stratified management strategies align with the current paradigm of individualized follow-up; as noted by other scholars, intensified monitoring and screening for high-risk patients may contribute to improved overall outcomes (21). Importantly, the novelty of this study lies in translating well-established risk indicators into a long-term, externally validated, and clinically actionable prediction tool, thereby providing real-world reference value for individualized postoperative risk stratification in Chinese patients with stage III CRC.

There are several limitations in this study. First, we used retrospective cohort data. Although rigorous inclusion and exclusion criteria were applied (e.g., restricting to first primary tumors and excluding perioperative events), unavoidable selection bias and information bias may still have influenced model construction. Second, the sample size of the external validation cohort was relatively small (N=92), which limited statistical power; as a result, some clinical variables did not reach significance in the validation set, and the calibration curves showed wider CIs. Future studies with larger, prospective, and independent populations are needed to further validate the generalizability of this model. Third, the predictors included in our model were all routine clinicopathological variables, without integration of molecular biomarkers. With the development of precision medicine, molecular detection of postoperative minimal residual disease—such as circulating tumor DNA (ctDNA)—is expected to provide additional prognostic value (4). A previous study has reported that postoperative ctDNA positivity is a powerful prognostic indicator of recurrence in CRC (HR up to approximately 10) (22) and may help identify stage II/III patients who are most likely to benefit from adjuvant chemotherapy. Therefore, future predictive models could incorporate molecular indicators such as ctDNA alongside clinical variables to improve the accuracy of recurrence risk assessment and enhance guidance for therapeutic decision-making. In addition, detailed information on neoadjuvant chemotherapy was not available in the SEER database, and the number of patients receiving neoadjuvant radiotherapy was very limited; therefore, the potential impact of neoadjuvant treatment on recurrence risk could not be reliably evaluated in this study.

Conclusions

Using data from the SEER database restricted to the Chinese population, we developed a prediction model for 5- and 10-year recurrence probabilities after curative resection of stage III CRC and validated its reliability in an independent institutional cohort. The model incorporated multiple independent prognostic factors and provided relatively accurate individualized assessments of recurrence risk in stage III CRC patients. Through risk stratification, high-risk patients can be identified after surgery, thereby enabling more intensive follow-up monitoring and tailored therapeutic interventions. This model offers valuable decision support for postoperative management of stage III CRC and, following further prospective validation, holds promise for clinical application to improve long-term patient outcomes.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2402/rc

Data Sharing Statement: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2402/dss

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2402/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2402/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of The Ninth Hospital of Hangzhou (No. 2025-024). Because of the retrospective nature of the study and the use of de-identified data, the requirement for informed consent was waived by the ethics committee.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Morgan E, Arnold M, Gini A, et al. Global burden of colorectal cancer in 2020 and 2040: incidence and mortality estimates from GLOBOCAN. Gut 2023;72:338-44. [Crossref] [PubMed]
2. Zeng H, Zhong X, Liu W, et al. Predicting treatment failure in stage III colon cancer patients after radical surgery. Front Oncol 2024;14:1397468. [Crossref] [PubMed]
3. Gao Z, Huang D, Chen H, et al. Development and validation of postoperative circulating tumor DNA combined with clinicopathological risk factors for recurrence prediction in patients with stage I-III colorectal cancer. J Transl Med 2023;21:63. [Crossref] [PubMed]
4. Kotani D, Oki E, Nakamura Y, et al. Molecular residual disease and efficacy of adjuvant chemotherapy in patients with colorectal cancer. Nat Med 2023;29:127-34. [Crossref] [PubMed]
5. Nors J, Gotschalck KA, Erichsen R, et al. Incidence of late recurrence and second primary cancers 5-10 years after non-metastatic colorectal cancer. Int J Cancer 2024;154:1890-9. [Crossref] [PubMed]
6. Salem ME, Yin J, Goldberg RM, et al. Evaluation of the change of outcomes over a 10-year period in patients with stage III colon cancer: pooled analysis of 6501 patients treated with fluorouracil, leucovorin, and oxaliplatin in the ACCENT database. Ann Oncol 2020;31:480-6. [Crossref] [PubMed]
7. Sobrero AF, Puccini A, Shi Q, et al. A new prognostic and predictive tool for shared decision making in stage III colon cancer. Eur J Cancer 2020;138:182-8. [Crossref] [PubMed]
8. Kunst N, Alarid-Escudero F, Aas E, et al. Estimating Population-Based Recurrence Rates of Colorectal Cancer over Time in the United States. Cancer Epidemiol Biomarkers Prev 2020;29:2710-8. [Crossref] [PubMed]
9. Boute TC, Swartjes H, Greuter MJE, et al. Cumulative Incidence, Risk Factors, and Overall Survival of Disease Recurrence after Curative Resection of Stage II-III Colorectal Cancer: A Population-based Study. Cancer Res Commun 2024;4:607-16. [Crossref] [PubMed]
10. Sueda T, Tei M, Yasuyama A, et al. Impact of regional lymph node metastasis on pulmonary metastasis as the first recurrence site. Updates Surg 2023;75:1843-55. [Crossref] [PubMed]
11. Piercey O, Wong HL, Leung C, et al. Adjuvant Chemotherapy for Older Patients With Stage III Colorectal Cancer: A Real-World Analysis of Treatment Recommendations, Treatment Administered and Impact on Cancer Recurrence. Clin Colorectal Cancer 2024;23:95-103.e3. [Crossref] [PubMed]
12. Taieb J, Gallois C. Adjuvant Chemotherapy for Stage III Colon Cancer. Cancers (Basel) 2020;12:2679. [Crossref] [PubMed]
13. Liu S, Jiang T, Xiao L, et al. Total Neoadjuvant Therapy (TNT) versus Standard Neoadjuvant Chemoradiotherapy for Locally Advanced Rectal Cancer: A Systematic Review and Meta-Analysis. Oncologist 2021;26:e1555-66. [Crossref] [PubMed]
14. Lian J, Yue Y, Yu W, et al. Immunosenescence: a key player in cancer development. J Hematol Oncol 2020;13:151. [Crossref] [PubMed]
15. Han F, Xu Y, Li X, et al. Clinicopathological features and prognosis analysis of proximal colonic mucinous adenocarcinoma. Sci Rep 2024;14:18682. [Crossref] [PubMed]
16. Yu F, Huang L, Shen F, et al. Prognostic implications of mucinous histology in stage III colon cancer with the receipt of adjuvant chemotherapy. J Gastrointest Oncol 2020;11:858-69. [Crossref] [PubMed]
17. Khan MA, Maken RN, Nisar H, et al. THE Role of Preoperative carcinoembryonic antigen in recurrence of resectable colorectal carcinoma. Acta Clin Croat 2020;59:216-22. [Crossref] [PubMed]
18. Ogata S, Fujita F, Fujiyoshi K, et al. Prognostic Value of Preoperative Carcinoembryonic Antigen and Carbohydrate Antigen 19-9 Levels for Adjuvant Chemotherapy in Stage II Colorectal Cancer: A Nationwide Multicenter Retrospective Study. J Anus Rectum Colon 2022;6:249-58. [Crossref] [PubMed]
19. Naito A, Iwamoto K, Ohtsuka M, et al. Risk Factors for Lymph Node Metastasis in Pathological T1b Colorectal Cancer. In Vivo 2021;35:987-91. [Crossref] [PubMed]
20. Huang F, Wei R, Mei S, et al. Clinical calculator based on clinicopathological characteristics predicts local recurrence and overall survival following radical resection of stage II-III colorectal cancer. Front Oncol 2025;15:1494255. [Crossref] [PubMed]
21. Park MY, Park IJ, Ryu HS, et al. Optimal Postoperative Surveillance Strategies for Colorectal Cancer: A Retrospective Observational Study. Cancers (Basel) 2021;13:3502. [Crossref] [PubMed]
22. Chen G, Peng J, Xiao Q, et al. Postoperative circulating tumor DNA as markers of recurrence risk in stages II to III colorectal cancer. J Hematol Oncol 2021;14:80. [Crossref] [PubMed]