Data for these patients come from the SEER database, which provide a good representation of the epidemiology and cancer statistics of the US population. Using SEER*Stat 8.4.0.1 software, we extracted clinically relevant data of patients diagnosed with rectal cancer from 2004 to 2015 from SEER Research Plus Data.17 Registries, Nov 2021 Sub (2000–2019). Data included baseline demographics, tumor characteristics, treatment information, diagnostic staging, and survival time.

Data inclusion relies on the following inclusion criteria: (a) the disease was diagnosed between 2004 and 2014; (b) the number of positive lymph nodes is not 0; (c) surgical procedures to determine positive lymph nodes include anterior resection, Hartmann’s operation, low anterior resection (LAR), and trans-sacral rectosigmoidectomy (Code 30); and (d) histology behavior is adenocarcinoma of the rectum (code as 8140/3). Our study included the following exclusion criteria: (a) race unknown (n=22); (b) marital status unknown (n=420); (c) T0, Tis, Tx (n=232); (d) Nx (n=136); (e) M1, Mx (n=1,705); (f) multiple primary tumor (n=1,989); (g) cause of death unknown (n=44); (h) size unknown (n=841); (i) number of lymph nodes unknown (n=43); and (j) number of positive lymph nodes unknown (n=1,577). Ultimately, 4,310 patients with lymph- node- positive rectal cancer from the SEER database were included in our study based on inclusion and exclusion criteria and analyzed ( Figure 1 ).

The SEER database included several components of variables namely, characteristics of the population (age at diagnosis, sex, race, and marital status), characteristics of tumors (histology, T stage, N stage, tumor size, number of lymph nodes, and number of positive lymph nodes), information on treatment (chemotherapy, radiotherapy, and surgical treatment), and survival information (months of survival and CSS).

In the column “RX Summ-Surg Prim Site (1998 +),” the surgical method was chosen. Anterior resection, Hartmann’s operation, low anterior resection (LAR), and trans-sacral rectosigmoidectomy were coded as 30. We converted continuous variables such as age, tumor size, number of lymph nodes examined, and number of positive lymph nodes examined into categorical variables: age (≤60, 61–78, and ≥78), tumor size (<3 cm, 3–5cm, and >5 cm), lymph nodes examined (<11, 11–20, 21–30, and >30), and positive lymph nodes examined (1, 2–10, and >10). Other variables included (1) sex (male, female), (2) race [American Indian/Alaska Native (AIAN), Asian or Pacific Islander (API), black, white], (3) marital status [married, others (divorced/separated, single, and widowed)], (4) T stage (T1, T2, T3, T4), (5) N stage (N1, N2), (6) radiotherapy (no/unknown, yes), and (7) chemotherapy (no/unknown, yes).

In SEER, CSS is classified as a potentially fatal cause of rectal cancer, according to the cause -specific death classification column. In this study, CSS is defined as the interval between initial cancer diagnosis and specific death from rectal cancer.

Of the patients, 70% were randomly assigned to the development cohort and 30% to the validation cohort. The selected variables in the development cohort were evaluated by univariate Cox regression analysis, and variables that had statistical difference (p-value<0.05) were included in multivariate Cox regression analysis. In Cox regression analysis, all results were presented as hazard ratios (HR) and their 95% confidence intervals (CIs). It was applied to the development cohort to build nomograms and a system of risk classification. The validation cohort underwent internal validation.

We used the concordance index (C-index), calibration curves, and receiver operating characteristics curve (ROC) to verify the nomogram. The C-index was used to reflect the performance and prediction accuracy of the nomogram. In our study, calibration plots (1,000 bootstrap resamples) were plotted for 3 and 5 years to compare predicted and observed CSS. In the calibration diagram, the 45-degree line represents the actual results of the model. Generated ROC curves were based on specificity and sensitivity of the nomogram. Based on the cutoff values of the development cohort’s total nomogram scores, patients were divided into low-, intermediate-, and high-risk groups. In order to compare the CSS of patients in different risk groups, Kaplan–Meier curves and log-rank tests were used.

We extracted the data using SEER*Stat software version 8.4.0.1. A comparison of baseline data between the development cohort and validation cohort was conducted using SPSS 25.0 (IBM Corp, Armonk, NY). Cox regression analyses based on univariate and multivariate variables, plotting nomogram, C-index, calibration plots, ROC curves, and Kaplan–Meier curves were generated using R version 4.2.0 and related packages. In the X-Tile version 3.6.1, the cutoff value of age and total score were calculated. When p-value < 0.05, the differences were statistically significant.

We included a total of 4,310 patients with lymph- node- positive rectal cancer in our study, including 3,017 in the development cohort and 1,293 in the validation cohort. Of the patients included in the study, 2,503 (58.1%) were male, 3,419 (79.3%) were white, 2,684 (62.3%) were married, and 3,202 (74.3%) were had stage T3. A total of 2,152 (49.9%) patients had tumors with a diameter of 3–5 cm. Most patients (n=3,414, 79.2%) received chemotherapy and 2,999 (69.6%) received radiotherapy. The following graph shows the baseline demographic and clinical characteristics of patients in both the development cohort and the internal validation cohort. There was no significant difference between the two groups in terms of baseline data ( Table 1 ). Upon completion of follow-up, a total of 1,600 (37.1%) patients died of rectal cancer, including 1,147 (38.0%) in the development cohort and 453 (35.0%) in the validation cohort.

The predictive power of the factors was evaluated using Cox regression analysis. In the univariate Cox regression analysis of the development cohort, age, race, marital status, T stage, N stage, chemotherapy, radiation, tumor size, lymph nodes, and positive lymph nodes were significantly different (p < 0.05). As a result of the multivariate Cox regression analysis, we obtained the following results: age, race, T stage, tumor size, lymph nodes, and positive lymph nodes had a great influence on prognosis. In terms of age, younger age is associated with better prognosis; the age>78 contributed to worse survival. In terms of race, American Indian/Alaska Native (AIAN) people have better survival, and Asian or Pacific Islander (API) people have worse survival. For T stage, T1 contributes to better survival, and T4 contributes to worse survival. As for lymph nodes and positive lymph nodes, a positive correlation was found between the number of lymph nodes examined and patient prognosis, while a negative correlation was found between the number of positive lymph nodes and patient prognosis. In the final analysis, nine predictors, including age, race, marital status, T stage, N stage, chemotherapy, tumor size, lymph nodes, and positive lymph nodes, were identified as independent predictors of CSS ( Table 2 ).

Based on these CSS prognostic factors, an algorithm for predicting 3- and 5-year CSS in lymph- node- positive rectal cancer patients had been developed and presented virtually as a nomogram ( Figure 2 ). According to the nomogram, we observed that young age, receiving chemoradiotherapy, small tumor size, more lymph nodes (positive lymph nodes and negative lymph nodes), and fewer positive lymph nodes were associated with better prognosis of rectal cancer. Patients with stage T4 had the worst prognosis. The results obtained by the nomogram were in agreement with those obtained by previous studies (14–18).

The nomogram was validated in the internal validation cohort. The nomogram C-index was 0.702 (95% CI, 0.687–0.717) in the development cohort and 0.690 (95% CI, 0.665–0.715) in the validation cohort. In both the development cohort and the validation cohort, the calibration curves showed good agreement between predicted results and actual observations for 3- and 5-year CSS ( Figure 3 ). The AUCs were 0.758 and 0.740 for 3- and 5-year CSS in the development cohort, respectively ( Figures 4A, B ), and 0.735 and 0.730 for 3- and 5-year CSS in the validation cohort, respectively ( Figures 4C, D ), respectively. All these results demonstrate the good performance and application of our nomogram.

All factors in the nomogram that we built in the study were scored between 0 and 100 based on how much they contributed to the nomogram ( Table 3 ). As part of the nomogram generation, based on the cutoff value of the total score in the development cohort, we developed a risk stratification system ( Figure 5 ). All patients in the development cohort were divided into low-risk group (1,264/3,017, score<143.54), intermediate-risk group (1,352/3,017, score = 143.54–214.75), and high-risk group (401/3,017, score >214.75). Kaplan–Meier analysis showed a significant difference in CSS between the low-, intermediate-, and high-risk groups (p<0.05) ( Figure 6 ).