All patient data in this study were extracted from the SEER database (https://seer.cancer.gov/). We conducted the subsequent analysis using the most recent database, which is the incidence - SEER Research Plus Data, 17 Registries, Nov 2021 Sub (2000-2019). A total of 5549 patients were enrolled to our study. The criteria for inclusion were as follows: (1) primary rectal cancer, (2) the pathological diagnosis was confirmed as rectal adenocarcinoma (RA) or RMA (ICD-O-3: 8140/2, 8140/3, 8480/3, 8481/3), (3) received curative resection, (4) patients with stage II pathology, (5) patients who received radiotherapy and systemic chemotherapy before surgery. The criteria for exclusion were as follows: (1) did not receive NCR, (2) clinicopathologic information or follow-up data is incomplete. Ultimately, we collected the following information from the enrolled patients: age, gender, race, marital status, household income, pathological T stage, RNE, and tumor size. In addition, a validation cohort comprising 143 RMA patients recruited from the First Affiliated Hospital of Nanchang University from January 2010 to October 2018 was selected. This study was approved by the Ethics Committee of First Affiliated Hospital of Nanchang University. Each participant was signed informed consent forms.

Firstly, patients were divided into RA or RMA groups by the pathological classification. Then, based on whether they had received AC, each group was further divided into two groups, the AC group and the non-AC group. In addition, all RMA patients were divided into low (risk score: < 135, n = 132), moderate (risk score: 136-182, n = 107), and high-risk (risk score: > 182, n = 104) subgroups to evaluate the benefit of AC among different risk subgroups.

The chi-square test was used to assess the relationship between pathological classification, AC and clinicopathological characteristics. Kaplan-Meier (KM) curves and the log-rank test were utilized to analyze differences in OS and cancer-specific survival (CSS) among different groups. Univariate and multivariate Cox regression analysis were employed to identify independent prognostic factors. Nomogram was constructed utilizing the independent prognostic factors identified through multivariate Cox regression analysis. The effectiveness of the nomogram was assessed for its discriminative ability, which was quantified by the concordance index (C-index). A calibration curve was produced to demonstrate the concordance between predicted and actual survival rate. Using X-tile software, all RMA patients were divided into three risk subgroups based on their nomogram risk scores. The statistical analyses in this study were conducted using SPSS software (version 25, IBM, Armonk, NY, USA) and R software (version 4.1.1). P < 0.05 was considered to indicate a statistically significant difference.

Between 2000 and 2019, according to our screening criteria, a total of 5549 patients were enrolled, with 5206 being RA and 343 being RMA. In the RA group, a total of 1589 patients received AC, while 3,617 did not receive AC. In the RMA group, 93 patients received AC, while 250 did not receive AC. Figure 1 displays the flowchart depicting the selection process. Patients with RMA tend to have an advanced T stage (P < 0.001), fewer RNE (P < 0.001), and larger tumor size (P = 0.001) compared to patients with RA ( Table 1 ). There are no differences in terms of gender, race, marital status, household income, pathological T stage and RNE ( Table 1 ).

Then, we performed the survival analysis and found that OS (5-year OS 78.5% vs. 69.6%, P = 0.006, Figure 2A ) and CSS (5-year CSS 84.7% vs. 77.1%, P = 0.002, Figure 2B ) were superior in RA group compared to the RMA group. Further subgroup analysis showed that in the patients without AC, the OS (5-year OS 77.5% vs. 66.3%, P = 0.002, Figure 2C ) and CSS (5-year CSS 84.7% vs. 75.2%, P < 0.001, Figure 2D ) of the RA group were better than those of the RMA group, but in the patients that received AC, there was no difference in OS (5-year OS 80.8% vs. 78.9%, P = 0.95, Figure 2E ) and CSS (5-year CSS 84.7% vs. 82.5%, P = 0.91, Figure 2F ) between the RA group and the RMA group. The above results indicate that RMA patients have a poorer prognosis compared to RA patients. However, AC could bridge the gap in prognosis between the two groups.

Cox regression analysis showed that age (P < 0.001), gender (P < 0.001), household income (P < 0.001), RNE (P < 0.001) and pathological classification (P = 0.017) were independent prognostic factors of OS ( Table 2 ). Age (P < 0.001), marital status (P = 0.002), household income (P = 0.004), RNE (P < 0.001), pathological classification (P = 0.004) were independent prognostic factors of CSS ( Supplementary Table 1 ).

In order to mitigate bias caused by confounding variables, we conducted a repeated analysis using propensity score matching (PSM). After adjustment with PSM, all variables were well balanced and no significant differences were found between the two groups ( Supplementary Table 2 ). KM survival analysis showed that OS (5-year OS 83.1% vs. 69.6%, P < 0.001, Figure 3A ) and CSS (5-year CSS 87.2% vs. 77.1%, P = 0.003, Figure 3B ) were superior in RA group compared to the RMA group, consistent with the results before PSM. Subgroup analysis results also align with the pre-PSM findings. In the patients without AC, RA patients have better OS (5-year OS 83.0% vs. 66.3%, P < 0.001, Figure 3C ) and CSS (5-year CSS 87.4% vs. 75.2%, P = 0.001, Figure 3D ) than RMA patients. However, in the patients that received AC, there are no significant differences in OS (5-year OS 83.4% vs. 78.9%, P = 0.68, Figure 3E ) and CSS (5-year CSS 86.7% vs. 82.5%, P = 0.85, Figure 3F ) between two groups.

After PSM, Cox regression analysis showed that age (P < 0.001), RNE (P < 0.001) and pathological classification (P < 0.001) were independent prognostic factors of OS ( Table 3 ). RNE (P < 0.001), pathological classification (P = 0.003) were independent prognostic factors of CSS ( Supplementary Table 3 ). This further corroborates the reliability of the conclusion that RMA patients have a worse prognosis than RA patients, but AC could bridge the gap in the prognoses of the two groups.

Further stratified analysis revealed that AC could improve the OS of RA patients (5-year OS: 80.8% with AC vs. 77.5% non-AC, P = 0.009), with a 3.3% absolute survival benefit at 5 years ( Figure 4A ). In RMA patients, AC could also improve the OS (5-year OS: 78.9% with AC vs. 66.3% non-AC, P = 0.036). The 5-year absolute survival benefit is as high as 12.6% ( Figure 4C ). Consistent results were observed in RNE < 12 subgroup (P = 0.03, Figure 4E ). However, AC does not improve the CSS for RA patients (P = 0.94, Figure 4B ), RMA patients (P = 0.12, Figure 4D ) and RMA patients with RNE < 12 (P = 0.23, Figure 4F ). The above results indicate that AC could significantly improve the OS of stage II RMA patients. However, the application of AC is likely to introduce toxicity implications for patients. Therefore, selectively administering AC to patients at high risk of distant recurrence could maximize the avoidance of unnecessary toxicity for patients.

Among all stage II RMA patients, those receiving AC were younger (P = 0.048) and had larger tumor sizes (P = 0.042) compared to those without AC ( Table 4 ). Subsequently, to eliminate the impact of AC on RMA patients’ OS, RMA patients without AC were incorporated into the Cox proportional hazard model. In the univariate analysis, age, gender, pathologic T stage, RNE, and tumor size exhibited associations with OS (all P < 0.05, Table 5 ). Meanwhile, in the multivariate analysis, age, gender, pathologic T stage, RNE, and tumor size were all identified as independent prognostic factors (all P < 0.05, Table 5 ).

A nomogram was constructed utilizing the independent prognostic factors identified through multivariate Cox regression analysis to predict 3- and 5-year OS rates, concurrently yielding a risk score for each RMA patient ( Figure 5A ). The C-index of the nomogram is 0.701 (95% CI: 0.652–0.750). The calibration curves of the nomogram for 3- and 5-year OS indicated that the predicted survival probabilities aligned closely with the actual survival probabilities ( Figures 5B, C ). The above results indicate that our nomogram possesses strong predictive potential and a high level of reliability.

Using X-tile software, all RMA patients were divided into three subgroups based on their risk scores: low-risk (risk score: < 135, n = 132), moderate-risk (risk score: 136-182, n = 107), and high-risk (risk score: > 182, n = 104) ( Figures 6A, B ). The 5-year OS rates for the low, moderate, and high-risk groups were 83.4%, 70.4%, and 55.2%, respectively, with significant differences (P < 0.001) ( Figure 6C ). Subgroup analysis showed that in the high-risk group, AC significantly improved 5-year OS (79.5% with AC vs. 48.0% without AC, P = 0.007), yielding a 31.5% absolute survival benefit at 5 years ( Figure 6D ). Similar results were observed in the moderate-risk group, where AC led to a significant improvement in 5-year OS (89.3% with AC vs. 63.6% without AC, P = 0.009), with a 25.7% absolute survival benefit at 5 years ( Figure 6E ). However, in the low-risk group, no notable difference in OS was observed between patients who received AC and those who did not (86.2% with AC vs. 75.4% without AC, P = 0.31) ( Figure 6F ). The above findings suggest that stage II RMA patients in the high and moderate-risk groups could benefit from AC, while those in the low-risk group do not derive a benefit from AC.

Subsequently, the risk stratification system was verified for predictive accuracy in the external validation cohort. Significant differences in 5-year OS rates were observed among the low (84.4%), moderate (71.1%), and high-risk (50.8%) groups (P = 0.001) ( Figure 7A ). In the high-risk subgroup, AC significantly improved 5-year OS (73.3% with AC vs. 39.3% without AC, P = 0.044), with an absolute survival benefit of 34.0% at 5 years ( Figure 7B ). Similarly, in the moderate-risk subgroup, AC led to a significant improvement in 5-year OS (84.7% with AC vs. 58.6% without AC, P = 0.036), with an absolute survival benefit of 26.1% at 5 years ( Figure 7C ). However, in the low-risk subgroup, no notable difference in OS was observed between patients who received AC and those who did not (82.3% with AC vs. 85.7% without AC, P = 0.87) ( Figure 7D ). The above results indicated that our risk stratification system was accurate.