Highlights
-
•Summarise the established knowledge on this subject.
-
1Middle-aged and older patients are at high risk of rectal adenocarcinoma; however, studies comprehensively analysing its predictors and the construction of visual nomogram models are limited.
-
2Most studies that reported on the prediction of colorectal cancer-related survival models had limited samples and included data from a single centre. The included predictors were limited, or the evaluation indicators were not easy to obtain, greatly limiting clinical application.
-
3With the advancement of medical care, the clinical outcomes of patients with rectal adenocarcinoma have changed. Therefore, new, more comprehensive, and practical indicators are required for constructing clinical prediction models to effectively determine the prognosis of patients.
-
1
-
•What are the significant and/or new findings of this study?
-
1We included demographic and clinicopathological data from thousands of middle-aged and elderly patients with rectal adenocarcinoma to find relevant prognostic factors. New cut-offs were developed and used for the construction of nomograms.
-
2The nomogram constructed this time has excellent predictive ability and clinical decision-making ability, and has good clinical practicability.
-
3The nomogram survival prediction model constructed this time can effectively help evaluate the prognosis of middle-aged and elderly patients with rectal adenocarcinoma and guide the selection of clinical treatment measures.
-
1
Keywords: Rectum adenocarcinoma, Nomogram, Prognostic model
Abstract
Objective
To develop a new nomogram tool for predicting survival in middle-aged and elderly patients with rectal adenocarcinoma.
Methods
A total of 6,116 patients were randomly assigned in a 7:3 ratio to training and validation cohorts. Univariate and multivariate Cox proportional hazards regression analyses were used to identify independent prognostic factors associated with overall survival (OS) and cancer-specific survival (CSS) in the training set, and two nomogram prognostic models were constructed. The validity, accuracy, discrimination, predictive ability, and clinical utility of the models were assessed based on the concordance index (C-index), area under the receiver operating characteristics (ROC) curve, time-dependent area under the ROC curve (AUC), Kaplan-Meier survival curve, and decision curve analyses.
Results
Predictors of OS and CSS were identified, and nomograms were successfully constructed. The calibration discrimination for both the OS and CSS nomogram prediction models was good (C-index: 0.763 and 0.787, respectively). The AUC showed excellent predictive performance, and the calibration curve exhibited significant predictive power for both nomograms. The time-dependent AUC showed that the predictive ability of the predictor-based nomogram was better than that of the TNM stage. The nomograms successfully discriminated high-, medium-, and low-risk patients for all-cause and cancer-specific mortality. The decision curve demonstrated that the nomograms are useful with respect to good decision power.
Conclusion
Our nomogram survival prediction models may aid in evaluating the prognosis of middle-aged and older patients with rectal adenocarcinoma and guiding the selection of the clinical treatment measures.
Introduction
The incidence of rectal cancer has exceeded that of gastric cancer and liver cancer, becoming the second most common solid malignancy [1]. The most frequently observed site of colorectal cancer is the rectum, and the main pathological type is adenocarcinoma. Rectal adenocarcinomas mainly affect middle-aged and older patients (aged >45 years) [2]. Therefore, it is important to establish a prognostic model of rectal adenocarcinoma for this population in order to develop effective methods for diagnosis and treatment, as well as to assess prognosis.
The Surveillance, Epidemiology, and End Results (SEER) database, a tumor-related registry database established by the National Cancer Institute in 1973, currently covers 28% of cancer patients. Data are derived from clinical sources, including patients’ clinically relevant information, treatment costs, and social information, which provide evidence support and important data for medical research [3]. Nomograms are used to construct survival prediction models that can comprehensively incorporate multiple prognostic indicators and quantify risk with intuitive graphs. They are used as tools for assessing risks and benefits, and they aid in clinical diagnoses and decision-making regarding treatment strategies [4].
Previous clinical survival model studies have used limited numbers of samples and relied on restricted evaluations, reducing their clinical application. At the same time, recent advances in medical technology have affected prognostic outcomes. Thus, it is important to establish a method to evaluate the prognostic outcome of middle-aged and older patients with rectal adenocarcinoma in a reasonable and accurate way. Here, we analyzed relevant clinical data of this population registered in the SEER database between 2010 and 2015 to construct a nomogram survival model for predicting patient 3- and 5-year overall survival (OS) and cancer-specific survival (CSS). The nomogram provides a good prediction tool that may help guide physicians in generating more accurate individual diagnoses and treatment plans.
Materials and methods
Patient data collection
Data were obtained from the SEER database (https://seer.cancer.gov/data/). SEER*Stat version 8.3.6 software was used. Permission to access the SEER database was obtained (accession number: 12285-Nov2019).
The inclusion criteria were as follows: (1) registration in the SEER database from 2010 to 2015, (2) diagnosis of rectal adenocarcinoma, (3) surgical treatment, and (4) availability of complete follow-up information. The exclusion criteria were as follows: (1) unclear diagnostic methods; (2) unknown ethnicity; (3) unknown TNM stage; (4) unknown tumor size; (5) unknown histological grade; (6) unknown carcinoembryonic antigen (CEA), circumferential resection margin (CRM), tumor implantation (TD), and perineural invasion (PNI) data; or (7) unknown number of positive lymph nodes and examined lymph nodes.
We collected the following information for each patient: year of diagnosis, age, ethnicity, sex, tumor location, histological grade, clinical stage, CEA, TD, CRM, PNI, tumor size, number of positive lymph nodes, number of examined lymph nodes, metastasis status, histopathological type (malignant behavior based on ICD-O-3), survival time, cause of death, and survival status. Clinical staging was based on the 7th edition of the American Joint Committee on Cancer (AJCC) staging system. The CEA level was determined according to the highest value in the preoperative test results.
Clinical data were obtained from the SEER database for 260,833 patients. A total of 6116 eligible patients were enrolled. The flowchart for the inclusion of patients is shown in Fig. 1. Seventy percent of eligible patients were randomly divided into a training cohort and the remaining 30% into a validation cohort using R software.
Fig. 1.
Flow chart of patient selection.
From the 260,882 patients with rectal cancer in the seer database, a total of 6116 eligible patients were screened.
Study endpoints
The study endpoints were OS and CSS. Moreover, the 3-year and 5-year survival outcomes were assessed. The validity, accuracy, discrimination, predictive ability, and clinical utility of the nomogram were assessed based on the C-index, receiver operating characteristics (ROC) curve, time-dependent area under the ROC curve (AUC), decision curve, and calibration curve.
OS was defined as the time from diagnosis to death or follow-up [5]. CSS was defined as the time from diagnosis to death or follow-up for rectal cancer [6].
Statistical analysis
The X-tile software was used to divide variables into different basins based on changes in markers and to visualize the optimal cut-points for creating such segmentations [7]. SPSS software (version 24.0, SPSS Inc., Chicago, USA) and R software (www.r-project.org, version 3.63) were used for statistical analysis. Cox regression analysis was performed using the R package “rms,” “foreign,” and “survival.” The concordance indexes (C-index) and risk score were calculated, and Kaplan-Meier survival curves, decision curves, and calibration curves were plotted. A nomogram was drawn by the R package “regplot.” ROC curve and time-dependent ROC-based AUC were plotted by the R package “timeROC.” Statistical tests involved two-way analyses. P values <0.05 were considered to indicate statistical significance.
Nomogram construction and performance evaluation
The X-tile software was used to assess the optimal cut-off values for age, tumor size, number of positive lymph nodes, and number of examined lymph nodes. The optimal cut-off values were as follows: age, 60 and 75 years; tumor size, 30 and 60 mm; number of positive lymph nodes, 1 and 4; and number of examined lymph nodes, 10 (Supplementary Fig. 1).
Frequencies and percentages were used to describe the clinical data of the validation cohort and training cohort. The chi-square test was used to determine the difference between the two groups. P <0.05 was considered to indicate statistical significance. Univariate and multivariate Cox regression analyses were used to identify the prognostic factors and calculate hazard ratios (HRs) with 95% confidence intervals (CIs).
A nomogram was constructed using the R software and indicators with statistical significance in the multivariate Cox analysis as predictors. The C-index and AUC were used to assess the predictive effect of the nomogram. The calibration curve was used to evaluate the agreement between actual and predicted results.
Verification of nomogram discrimination
A risk score was calculated for each patient using the “predicted” function of the R software. The X-tile software was used to identify patients in the training set, and the patients were then divided into the low-, medium-, and high-risk groups according to the cut-off optimal risk score. The log-rank test was used to assess the differences in survival among the three groups. The Kaplan–Meier survival curves were plotted for OS and CSS based on the risk scores for the validation and training sets.
Evaluation of the predictive power of the nomogram
The time-dependent AUC shows the values of different prediction models of patients at time points of change to further evaluate the accuracy of the constructed nomogram prediction model and that of the TNM stage. Therefore, we used time-dependent AUC to assess the predictive power of the constructed nomogram with that of the TNM stage.
Evaluation of the clinical efficacy of nomograms
Decision curves analytically(DCA) assess clinical utility and net benefit [8]. To test the clinical efficacy of the nomogram, we used the decision curve of the training group versus the validation group.
Results
Baseline patient characteristics
Except for the CRM, no statistically significant differences were found in the remaining variables between the two groups (P > 0.05; Table 1). The results of randomization were satisfactory.
Table 1.
Baseline demographic and clinical characteristics of middle-aged and elderly patients with rectal adenocarcinoma.
| Training Cohort (4284) |
Validation Cohort (1832) |
Overall (6116) |
|||||
|---|---|---|---|---|---|---|---|
| Variable | Quantity | SCALE | Quantity | SCALE | Quantity | SCALE | P value |
| Age | 0.219 | ||||||
| 45-60 | 1957 | 45.68% | 854 | 46.62% | 2811 | 45.96% | |
| 61-75 | 1656 | 38.66% | 723 | 39.47% | 2379 | 38.90% | |
| > 75 | 671 | 15.66% | 255 | 13.92% | 926 | 15.14% | |
| Race | 0.245 | ||||||
| Black | 317 | 7.40% | 158 | 8.62% | 475 | 7.77% | |
| White | 3410 | 79.60% | 1433 | 78.22% | 4843 | 79.19% | |
| Other | 557 | 13.00% | 241 | 13.16% | 798 | 13.05% | |
| Sex | 0.18 | ||||||
| F | 1713 | 39.99% | 699 | 38.16% | 2412 | 39.44% | |
| M | 2571 | 60.01% | 1133 | 61.84% | 3704 | 60.56% | |
| Grade | 0.933 | ||||||
| I/II | 3778 | 88.19% | 1617 | 88.26% | 5395 | 88.21% | |
| III/IV | 506 | 11.81% | 215 | 11.74% | 721 | 11.79% | |
| Site | 0.134 | ||||||
| Rectosigmoid Junction | 1495 | 34.90% | 676 | 36.90% | 2171 | 35.50% | |
| Rectum | 2789 | 65.10% | 1156 | 63.10% | 3945 | 64.50% | |
| Stage | 0.951 | ||||||
| I | 675 | 15.76% | 284 | 15.50% | 959 | 15.68% | |
| II | 1276 | 29.79% | 547 | 29.86% | 1823 | 29.81% | |
| III | 1874 | 43.74% | 812 | 44.32% | 2686 | 43.92% | |
| IV | 459 | 10.71% | 189 | 10.32% | 648 | 10.60% | |
| Stage_T | 0.277 | ||||||
| T1 | 226 | 5.28% | 93 | 5.08% | 319 | 5.22% | |
| T2 | 721 | 16.83% | 296 | 16.16% | 1017 | 16.63% | |
| T3 | 2870 | 66.99% | 1212 | 66.16% | 4082 | 66.74% | |
| T4 | 467 | 10.90% | 231 | 12.61% | 698 | 11.41% | |
| Stage_N | 0.974 | ||||||
| 1 | 2056 | 47.99% | 876 | 47.82% | 2932 | 47.94% | |
| 2 | 1582 | 36.93% | 682 | 37.23% | 2264 | 37.02% | |
| 3 | 646 | 15.08% | 274 | 14.96% | 920 | 15.04% | |
| Stage_M | 0.643 | ||||||
| M1 | 3825 | 89.29% | 1643 | 89.68% | 5468 | 89.40% | |
| M2 | 459 | 10.71% | 189 | 10.32% | 648 | 10.60% | |
| Tumor size | 0.784 | ||||||
| < = 30 | 1192 | 27.82% | 525 | 28.66% | 1717 | 28.07% | |
| < = 60 | 2233 | 52.12% | 948 | 51.75% | 3181 | 52.01% | |
| > 60 | 859 | 20.05% | 359 | 19.60% | 1218 | 19.91% | |
| CEA | 0.673 | ||||||
| Low | 2414 | 56.35% | 1043 | 56.93% | 3457 | 56.52% | |
| High | 1870 | 43.65% | 789 | 43.07% | 2659 | 43.48% | |
| TD | 0.262 | ||||||
| Neg | 3821 | 89.19% | 1616 | 88.21% | 5437 | 88.90% | |
| Pos | 463 | 10.81% | 216 | 11.79% | 679 | 11.10% | |
| CRM | 0.023 | ||||||
| Neg | 1940 | 45.28% | 771 | 42.09% | 2711 | 44.33% | |
| Pos | 2344 | 54.72% | 1061 | 57.91% | 3405 | 55.67% | |
| PNI | 0.15 | ||||||
| Neg | 3685 | 86.02% | 1550 | 84.61% | 5235 | 85.60% | |
| Pos | 599 | 13.98% | 282 | 15.39% | 881 | 14.40% | |
| Number of positive lymph nodes | 0.498 | ||||||
| 0 | 2605 | 60.81% | 1106 | 60.37% | 3711 | 60.68% | |
| < = 4 | 1235 | 28.83% | 550 | 30.02% | 1785 | 29.19% | |
| > 4 | 444 | 10.36% | 176 | 9.61% | 620 | 10.14% | |
| Number of lymph nodes examined | 0.183 | ||||||
| < = 10 | 626 | 14.61% | 292 | 15.94% | 918 | 15.01% | |
| > 10 | 3658 | 85.39% | 1540 | 84.06% | 5198 | 84.99% | |
In the univariate Cox regression analysis of OS, other factors, except the tumor site, showed statistical significance (P < 0.05). In the univariate Cox regression analysis of CSS, other factors, except sex, race, and tumor location, showed statistical significance (P < 0.05; Table 2). Multivariate Cox regression analysis was performed for variables with statistically significant differences. The results of the multivariate Cox regression analysis of OS showed significant statistical differences in all variables, except “White” for race and “< = 60” in “II” and “III” tumor sizes (P < 0.05). The results of the multivariate Cox regression analysis of CSS showed significant statistical differences in all variables, except “< = 60” of tumor size (P < 0.05; Table 3).
Table 2.
Univariate cox regression analysis of cancer-specific survival and overall survival in the training cohort.
| OS |
CSS |
|||||
|---|---|---|---|---|---|---|
| Variable | HR | 95% CI | P | HR | 95% CI | P |
| Age | < 0.001 | < 0.001 | ||||
| 45-60 | ||||||
| 61-75 | 1.63 | 1.36 ∼ 1.95 | < 0.001 | 1.38 | 1.13 ∼ 1.69 | 0.0014 |
| > 75 | 3.56 | 2.95 ∼ 4.31 | < 0.001 | 2.51 | 2.01 ∼ 3.15 | < 0.001 |
| Race | 0.02 | 0.2 | ||||
| Black | ||||||
| White | 0.78 | 0.61 ∼ 1.01 | 0.0636 | 0.80 | 0.59 ∼ 1.09 | 0.1592 |
| Other | 0.63 | 0.45 ∼ 0.88 | 0.0067 | 0.72 | 0.49 ∼ 1.06 | 0.0971 |
| Sex | 0.01 | 0.2 | ||||
| F | ||||||
| M | 1.21 | 1.04 ∼ 1.41 | 0.0148 | 1.13 | 0.95 ∼ 1.35 | 0.1680 |
| Grade | < 0.001 | < 0.001 | ||||
| I/II | ||||||
| III/IV | 1.85 | 1.53 ∼ 2.23 | < 0.001 | 2.23 | 1.81 ∼ 2.74 | < 0.001 |
| Site | 0.5 | 0.6 | ||||
| Rectosigmoid Junction | ||||||
| Rectum | 0.95 | 0.82 ∼ 1.11 | 0.5340 | 0.95 | 0.79 ∼ 1.14 | 0.5690 |
| Stage | < 0.001 | < 0.001 | ||||
| I | ||||||
| II | 1.35 | 1.01 ∼ 1.81 | 0.0443 | 2.37 | 1.49 ∼ 3.76 | < 0.001 |
| III | 1.87 | 1.42 ∼ 2.46 | < 0.001 | 3.91 | 2.52 ∼ 6.06 | < 0.001 |
| Ⅳ | 6.10 | 4.58 ∼ 8.14 | < 0.001 | 16.15 | 10.36 ∼ 25.17 | < 0.001 |
| Tumor size | < 0.001 | < 0.001 | ||||
| < = 30 | ||||||
| < = 60 | 1.50 | 1.24-1.82 | < 0.001 | 1.66 | 1.31 ∼ 2.1 | < 0.001 |
| > 60 | 2.31 | 1.86 ∼ 2.87 | < 0.001 | 2.70 | 2.08 ∼ 3.49 | < 0.001 |
| CEA | < 0.001 | < 0.001 | ||||
| Low | ||||||
| High | 2.22 | 1.91 ∼ 2.58 | < 0.001 | 2.37 | 1.98 ∼ 2.83 | < 0.001 |
| TD | < 0.001 | < 0.001 | ||||
| Neg | ||||||
| Pos | 2.10 | 1.72 ∼ 2.56 | < 0.001 | 2.71 | 2.19 ∼ 3.37 | < 0.001 |
| CRM | < 0.001 | < 0.001 | ||||
| Neg | ||||||
| Pos | 1.36 | 1.17-1.58 | < 0.001 | 1.47 | 1.23 ∼ 1.76 | < 0.001 |
| PNI | < 0.001 | < 0.001 | ||||
| Neg | ||||||
| Pos | 2.19 | 1.83 ∼ 2.61 | < 0.001 | 2.62 | 2.15-3.2 | < 0.001 |
| Number of positive lymph nodes | < 0.001 | < 0.001 | ||||
| 0 | ||||||
| < = 4 | 1.70 | 1.44 ∼ 2.01 | < 0.001 | 2.19 | 1.79 ∼ 2.67 | < 0.001 |
| > 4 | 3.27 | 2.68 ∼ 3.97 | < 0.001 | 4.82 | 3.86 ∼ 6.01 | < 0.001 |
| Number of lymph nodes examined | 0.001 | 0.002 | ||||
| < = 10 | ||||||
| > 10 | 0.74 | 0.62 ∼ 0.89 | 0.0015 | 0.72 | 0.58 ∼ 0.89 | 0.0023 |
Table 3.
Multivariate cox regression analysis of cancer-specific survival and overall survival in the training cohort.
| OS | CSS | |||||
|---|---|---|---|---|---|---|
| Variable | HR | 95% CI | P | HR | 95% CI | P |
| Age | ||||||
| 45-60 | ||||||
| 61-75 | 1.83 | 1.52 ∼ 2.18 | < 0.001* | 1.57 | 1.28-1.92 | < 0.001* |
| > 75 | 4.79 | 3.93 ∼ 5.83 | < 0.001* | 3.54 | 2.8 ∼ 4.46 | < 0.001* |
| Race | ||||||
| Black | ||||||
| White | 0.82 | 0.64 ∼ 1.07 | 0.1429 | |||
| Other | 0.62 | 0.44 ∼ 0.86 | < 0.001* | |||
| Sex | ||||||
| F | ||||||
| M | 1.26 | 1.08 ∼ 1.47 | 0.0036* | |||
| Grade | ||||||
| I/II | ||||||
| III/IV | 1.52 | 1.25-1.84 | < 0.001* | 1.71 | 1.38 ∼ 2.12 | < 0.001* |
| Stage | ||||||
| I | ||||||
| II | 1.10 | 0.81 ∼ 1.49 | 0.5358 | 1.94 | 1.21 ∼ 3.09 | 0.0056* |
| III | 1.19 | 0.84 ∼ 1.67 | 0.3321 | 2.24 | 1.36 ∼ 3.68 | 0.0015* |
| Ⅳ | 3.30 | 2.3-4.72 | < 0.001* | 7.82 | 4.7 ∼ 13 | < 0.001* |
| Tumor size | ||||||
| < = 30 | ||||||
| < = 60 | 1.20 | 0.98 ∼ 1.46 | 0.0785 | 1.27 | 1 ∼ 1.62 | 0.0520 |
| > 60 | 1.73 | 1.38 ∼ 2.17 | < 0.001* | 1.91 | 1.46 ∼ 2.49 | < 0.001* |
| CEA | ||||||
| Low | ||||||
| High | 1.60 | 1.36 ∼ 1.88 | < 0.001* | 1.44 | 1.19 ∼ 1.74 | < 0.001* |
| TD | ||||||
| Neg | ||||||
| Pos | 1.36 | 1.09 ∼ 1.69 | 0.0058* | 1.50 | 1.19 ∼ 1.89 | < 0.001* |
| CRM | ||||||
| Neg | ||||||
| Pos | 1.26 | 1.08 ∼ 1.47 | 0.0029* | 1.35 | 1.12-1.61 | 0.0013* |
| PNI | ||||||
| Neg | ||||||
| Pos | 1.43 | 1.18 ∼ 1.74 | < 0.001* | 1.43 | 1.15 ∼ 1.78 | 0.0013* |
| Number of positive lymph nodes | ||||||
| 0 | ||||||
| < = 4 | 1.35 | 1.05 ∼ 1.73 | 0.0185* | 1.42 | 1.08 ∼ 1.89 | 0.0132* |
| > 4 | 1.91 | 1.44 ∼ 2.53 | < 0.001* | 2.14 | 1.57 ∼ 2.92 | < 0.001* |
| Number of lymph nodes examined | ||||||
| < = 10 | ||||||
| > 10 | 0.70 | 0.58 ∼ 0.84 | < 0.001* | 0.61 | 0.49 ∼ 0.76 | < 0.001* |
Note:
indicates P < 0.05.
Determination of predictors and construction of nomogram models
Variables that were not significant or had mild effects were excluded. We used age, ethnicity, sex, histological grade, clinical stage, CEA, TD, CRM, PNI, tumor size, number of positive lymph nodes, and number of examined lymph nodes as predictors of OS models; constructed nomograms; and then plotted the corresponding training set calibration curves. Age, histological grade, clinical stage, tumor size, CEA, TD, CRM, PNI, number of positive lymph nodes, and number of examined lymph nodes were used as predictors of the CSS model; nomograms were constructed; and the corresponding training set calibration curves were plotted (Fig. 2). Specific scores for each predictor in the nomogram are provided in Supplementary Table 2. In the nomogram, the 3-year and 5-year OS/CSS probabilities of middle-aged and older patients with rectal adenocarcinoma could be predicted according to the total score of predictors. The validation cohort's medium-, 3-, and 5-year OS and the calibration curve of the CSS also showed agreement between the actual and predicted clinical outcomes.
Fig. 2.
OS and CSS nomograms and calibration curves.
The total score obtained by summing the individual scores of the predictors was used to predict the 3- and 5-year survival rates of the patients. The calibration curves showed a high degree of agreement between the predicted and actual values of the OS nomogram and the CSS nomogram. Example: A 50-year-old male Caucasian patient with rectal adenocarcinoma, tumor size 50 mm, CRM (+), TD(+), CEA (+), PNI (+), clinical stage V, grade: poorly differentiated, number of positive postoperative lymph nodes: 3, number detected: 11. The red indicator line in the figure represents the score of patients: OS total score 434, 3-year OS survival: 0.325, 5-year OS survival: 0.142; CSS total score 356, 3-year CSS survival: 0.26, 5-year CSS survival: 0.12.
Evaluation of the predictive power and usefulness of the model
The C-indexes of the OS and CSS nomograms of the training set were 0.763 (95% CI 0.745–0.781) and 0.787 (95% CI 0.765–0.80), respectively. The AUCs of the 3- and 5-year survival rate in the OS prediction model of the training set were 0.773 and 0.768, respectively, and those of the CSS prediction model were 0.802 and 0.790, respectively. The prediction model showed good predictive power. Time-dependent AUC curves were not only used to evaluate the predictive power of the nomogram, but also used to compare the predictive ability of TNM stage. The curve in Fig. 3 shows that the AUC value of the OS/CSS nomogram is significantly higher than that of the TNM stage at 0–60 months, and the predictive ability of the nomogram is better than that of the TNM stage.
Fig. 3.
ROC and time-dependent AUC.
ROC values for the training cohort are shown in the figure. AUC values of nomogram versus TNM stage based on temporal changes are shown. In the training and validation cohorts, the AUC of nomogram was higher than that of TNM stage.
Validation of the discrimination capability of the predictive model
A risk score was calculated for each patient using the “predicted” function of the R software. The optimal OS and the CSS risk score cut-off values obtained by the X-tile software were 1.2 and 3.5, 1.4 and 4.1, respectively. Patients were divided into three groups according to the optimal cut-off value: low-, intermediate-, and high-risk groups. Kaplan–Meier survival curves were plotted. We found significant survival differences between the training cohort and all three groups of patients in the validation cohort (Fig. 4). Therefore, the OS nomogram can successfully distinguish all-cause mortality in high- and medium-risk patients, and the CSS nomogram can successfully distinguish cancer-specific mortality in high-, medium-, and low-risk patients.
Fig. 4.
Kaplan–Meier survival curves for low-, medium-, and high-risk groups based on risk scores. The optimal OS risk score cut-off was 1.2, 3.5. CSS risk score cutoffs were 1.4, 4.1. Significant differences in OS, CSS were observed among the low-risk, intermediate-risk, and high-risk patients in the training and validation cohorts.
Evaluation clinical efficacy of nomograms
Decision curves of the OS and CSS nomograms were constructed at a threshold probability of <88% and <80% at 3 years, respectively (Fig. 5). The OS and CSS nomograms provided a net benefit over the “all treatment” or “no treatment” strategy. In addition, similar results were obtained in the validation set, when the threshold probabilities were <88% and <82% at 3 years, respectively. Therefore, the presented nomogram displays good, clinically relevant decision power.
Fig. 5.
Decision curve analysis. Plot net benefit versus threshold probability.
The net benefit was calculated by subtracting the proportion of all false-positive cases from the proportion of true-positive cases, weighing the relative harm of abandoning treatment against the adverse consequences of unnecessary treatment. The gray and black lines indicate the net benefit of treating all patients and no patient strategies, respectively. Dashed lines represent nomograms. The results showed that the nomogram had good decision power in the training cohort and the validation cohort.
Discussion
In this study, we constructed a nomogram based on the data of 6,116 middle-aged and older patients with rectal adenocarcinoma. We included patient demographic and clinicopathologic data and sought a suitable new cut-off for our model. We found significant prognostic factors associated with the OS and CSS and constructed nomograms for middle-aged and older patients with rectal adenocarcinoma.
Most previous studies presenting prediction models for colorectal cancer-related survival had limited samples and included data from a single center. The included predictors were limited, or the evaluation indicators were not easy to obtain, greatly limiting the clinical application of these models [9,10]. In addition, the study endpoint was limited to a single prediction of CSS or OS for some studies [11,12], and few studies have been constructed to predict both. Middle-aged and older patients are at a high risk of rectal adenocarcinoma; however, studies comprehensively analyzing its predictors and the construction of visual nomogram models are limited. With the advancement of medical care, the clinical outcomes of patients with rectal adenocarcinoma have changed. Therefore, new, more comprehensive, and practical indicators are required for constructing clinical prediction models to effectively determine the prognosis of patients.
Our constructed nomogram prediction models for OS and CSS clearly displayed the effects of various predictors on middle-aged and older patients with rectal adenocarcinoma and provided accurate scores. We found 12 independent prognostic variables associated with OS and 10 with CSS. Of note, sex and ethnicity were predictive of OS in the study population, with men having a poorer prognosis than women, and black and white participants having poorer prognosis than other ethnic groups. These findings are similar to those of Brenner et al. [13] and Wen et al. [14].
We found that the age of the included population was associated with a poorer prognosis. The pathological grade was “poorly differentiated,” and the prognosis of patients with undifferentiated grade was poorer than those with “moderately differentiated” and “well-differentiated” grades, which was also confirmed by Julien et al. [15]. Among the clinicopathological features, although no significant differences were found in the scores of stage II and III patients, the scores of stage I and IV patients were significantly different. We believe that lymph node metastasis and distant metastasis are important factors in the prognosis of patients with rectal adenocarcinoma.
Several factors, including CEA, TD, CRM, and PNI have been demonstrated to be associated with poor prognosis [16], [17], [18], [19]. CEA is a tumor marker used in the differential diagnosis and detection of colorectal cancer [20]. TD is defined as one or more satellite peritumoral nodules in the adipose tissue surrounding the colorectum of the primary cancer, and histological evidence does not support residual lymph nodes or identifiable vascular or neural structures [21]. CRM positivity is defined as the presence of tumor cells within 1 mm of the inner CRM [22]. PNI is a metastatic modality associated with an aggressive cancer phenotype that exhibits poor survival as well as an increased risk of local and distant recurrence, occurring through invasion of the intraneural or extramural plexus independent of lymphatic invasion [23]. These critical predictors provided important predictive power and proportion of scores in our nomogram prediction model. The nomogram showed that preoperative CEA levels were higher than normal, and that TD, PNI, and CRM were positive, all of which indicate a poor prognosis.
Lymph node dissection is the focus of surgery, and the prognostic survival of patients is closely related to the degree of lymph node dissection [24]. An increased number of positive lymph nodes often indicates a high chance of recurrence and metastasis and a poor prognosis [25]. The AJCC proposes that the detection of ≥12 lymph nodes can improve the accuracy of postoperative staging of colorectal cancer and help determine the presence or absence of lymph node metastasis [26]. Therefore, the number of lymph nodes should be accurately assessed to provide a basis for clinical staging and diagnosis, as well as treatment [27]. Chang et al. [28] found that with the increase in the number of detected lymph nodes, the 5-year survival rate of patients increases. Although the SEER database did not have accurate information on the number of lymph nodes, our survival model did evaluate patients based on lymph node number. The cut-off number for node positivity and examined nodes was 1 or 4, and 10, respectively. We found that the number of positive lymph nodes was negatively correlated with patient prognosis, while the number of examined lymph nodes was positively correlated with patient prognosis.
Here, we comprehensively analyzed the effect of predictive factors on the prognosis of patients with rectal cancer and constructed a prognostic factor-based nomogram to assess the 3-year and 5-year OS and CSS of patients. Both the C-index and AUC of ROC suggest that the model has excellent predictive ability. Both the training and validation cohort survival curves showed good discrimination ability of the prediction model. The use of time-dependent AUC in the training and validation cohorts confirmed that our prediction model was consistently superior to traditional TNM staging for 5-year survival prediction. The DCA indicated that the prediction model may have good clinical decision power, which needs further validation in clinical practice, and the correction curves of both cohorts showed that the prediction value was highly consistent with the actual value. OS has been considered the gold standard primary endpoint for assessing the effect of cancer treatment, and it provides meaningful evidence for clinical benefit [29,30]. Differences in CSS can reflect changes in treatment quality and are influenced by patient characteristics [31,32]. The nomogram constructed in this study can simultaneously predict the survival rates for CSS and OS using patient-independent clinical data, and may, therefore, be beneficial in guiding the clinical decision-making of physicians.
This study included many patients, had a long overall observation time, and evaluated the effect of multiple factors on rectal adenocarcinoma in middle-aged and older patients. With the validation cohort, the constructed model may provide significant value for clinical diagnosis and treatment. Our model has some evaluation advantages. First, the proposed prognostic model is indicated for middle-aged and older patients with rectal adenocarcinoma and can better reflect the characteristics of this population. Second, our nomogram is based on patient demographic characteristics and clinicopathological related data, which are key indicators easily obtained in clinical practice, and the nomogram has good clinical utility. Third, studies have reported that the tumor size, number of dissected lymph nodes, and number of detected lymph nodes are important predictors. We found that the optimal cut-off values of these three factors were different from those defined in the most current data, which reminded us to establish the cut-off values of relevant indicators in middle-aged and older patients with rectal adenocarcinoma, rather than using the universal cut-off values for patients with colorectal cancer.
There are some limitations to our study. First, this study was retrospective, and patients were screened based on strict inclusion and exclusion criteria; therefore, potential selection bias may have occurred. Second, the SEER database does not specify procedures, operators, and other such factors, and bias may exist owing to different experience levels of operators and pathologists. Third, we could not analyze the 8- or 10-year survival rates due to the length of follow-up for the included population data. Additional databases could be used for further evaluation. Finally, unrecorded clinical characteristics may affect patient outcomes, such as complications, nutritional status, and detailed chemoradiotherapy information.
In conclusion, we constructed a nomogram model of CSS and OS to determine the 3- and 5-year survival rates of middle-aged and older patients with rectal adenocarcinoma. The results showed that the prediction model had satisfactory prognostic discrimination ability and survival prediction ability, as well as good clinical decision-making power. This nomogram can be used to individualize the survival prediction for middle-aged and older patients with rectal adenocarcinoma, providing a good tool for gastrointestinal surgeons to accurately assess a patient's condition.
Author contribution
Please specify the contribution of each author to the paper, e.g. study design, data collections, data analysis, writing, others, who have contributed in other ways should be listed as contributors.
| Author | study design | data collections | data analysis | writing | final approval of manuscript |
| Hao Liu | √ | √ | √ | √ | √ |
| Liang Lv | √ | √ | √ | √ | √ |
| Yi Dan Qu | √ | √ | √ | √ | √ |
| Zi Wen Zheng | √ | √ | √ | ||
| Jun Jiang Zhao | √ | √ | √ | ||
| Bo Liu | √ | √ | √ | ||
| Da Sen Zhang | √ | √ | √ | ||
| He Xiang Wang | √ | √ | √ | √ | √ |
| Jian Zhang | √ | √ | √ | √ | √ |
Declaration of Competing Interest
The authors declare that there is no conflict of interest.
Acknowledgments
Acknowledgments
The authors thank the members of the Department of General Surgery of the Affiliated Hospital of Qingdao University for assistance with data analysis.
Funding
This study was supported by the National Natural Science Foundation of China (grant No. 81770631).
Footnotes
Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.tranon.2020.100938.
Contributor Information
Hexiang Wang, Email: 18661808669@126.com.
Jian Zhang, Email: drjianzhang@126.com.
Appendix. Supplementary materials
References
- 1.Artaş G., Ozercan H.I. The expression of STAT3, BCL-XL and MMP-2 proteins in colon adenocarcinomas and their relationship with prognostic factors. Turk Patoloji Derg. 2014;30(3):178–183. doi: 10.5146/tjpath.2014.01269. [DOI] [PubMed] [Google Scholar]
- 2.Kim J.Y., Jang E.S., Park S.Y. Distinct characteristics of colorectal cancer and frequency of right colon cancer in elderly patients of Korea. J. Korean Geriatrics Soc. 2010 [Google Scholar]
- 3.Duggan M.A., Anderson W.F., Altekruse S., Penberthy L., Sherman M.E. The surveillance, epidemiology, and end results (SEER) program and pathology: toward strengthening the critical relationship. Am. J. Surg. Pathol. 2016;40(12):e94–e102. doi: 10.1097/PAS.0000000000000749. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Balachandran V.P., Gonen M., Smith J.J., DeMatteo R.P. Nomograms in oncology: more than meets the eye. Lancet Oncol. 2015;16(4):e173–e180. doi: 10.1016/S1470-2045(14)71116-7. PMID: 25846097PMCID: PMC4465353. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Polley M.Y., Lamborn K.R., Chang S.M., Butowski N., Clarke J.L., Prados M. Conditional probability of survival in patients with newly diagnosed glioblastoma [published correction appears in J Clin Oncol. 2011 Dec 20;29(36):4847] J. Clin. Oncol. 2011;29(31):4175–4180. doi: 10.1200/JCO.2010.32.4343. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Guan X., Chen W., Jiang Z. Exploration of the optimal minimum lymph node count after colon cancer resection for patients aged 80 years and older. Sci. Rep. 2016;6:38901. doi: 10.1038/srep38901. Published 2016 Dec 12. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Camp R.L., Dolled-Filhart M., Rimm D.L. X-tile: a new bio-informatics tool for biomarker assessment and outcome-based cut-point optimization. Clin. Cancer Res. 2004;10(21):7252–7259. doi: 10.1158/1078-0432.CCR-04-0713. [DOI] [PubMed] [Google Scholar]
- 8.Hijazi Z., Oldgren J., Lindbäck J. A biomarker-based risk score to predict death in patients with atrial fibrillation: the ABC (age, biomarkers, clinical history) death risk score. Eur. Heart J. 2018;39(6):477–485. doi: 10.1093/eurheartj/ehx584. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Pan F., Chen T., Sun X. Prognosis prediction of colorectal cancer using gene expression profiles. Front Oncol. 2019;9:252. doi: 10.3389/fonc.2019.00252. Published 2019 Apr 9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Liu J., Huang X., Yang W. Nomogram for predicting overall survival in stage II-III colorectal cancer. Cancer Med. 2020;9(7):2363–2371. doi: 10.1002/cam4.2896. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Zheng P., Lai C., Yang W., Guo J., Xiao S., Chen Z. Nomogram predicting cancer-specific survival in elderly patients with stages I-III colon cancer. Scand. J. Gastroenterol. 2020;55(2):202–208. doi: 10.1080/00365521.2020.1720280. [DOI] [PubMed] [Google Scholar]
- 12.Pei J.P., Zhang C.D., Liang Y. Novel nomograms individually predicting overall survival of non-metastatic colon cancer patients. Front Oncol. 2020;10:733. doi: 10.3389/fonc.2020.00733. Published 2020 May 6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Purim O., Gordon N., Brenner B. Cancer of the colon and rectum: potential effects of sex-age interactions on incidence and outcome. Med. Sci. Monit. 2013;19:203–209. doi: 10.12659/MSM.883842. Published 2013 Mar 20. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Fu W.J. Racial-sex disparities–a challenging battle against cancer mortality in the USA. J. Racial. Ethn. Health Disparities. 2015;2(2):158–166. doi: 10.1007/s40615-014-0059-6. [DOI] [PubMed] [Google Scholar]
- 15.Langrand-Escure J., Diao P., Garcia M.A. Outcome and prognostic factors in 593 non-metastatic rectal cancer patients: a mono-institutional survey. Sci. Rep. 2018;8(1):10708. doi: 10.1038/s41598-018-29040-2. Published 2018 Jul 16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Gao Y., Wang J., Zhou Y., Sheng S., Qian S.Y., Huo X. Evaluation of serum CEA, CA19-9, CA72-4, CA125 and ferritin as diagnostic markers and factors of clinical parameters for colorectal cancer. Sci. Rep. 2018;8(1):2732. doi: 10.1038/s41598-018-21048-y. Published 2018 Feb 9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Liu F., Zhao J., Li C. The unique prognostic characteristics of tumor deposits in colorectal cancer patients. Ann. Transl. Med. 2019;7(23):769. doi: 10.21037/atm.2019.11.69. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Baik S.H., Kim N.K., Lee Y.C. Prognostic significance of circumferential resection margin following total mesorectal excision and adjuvant chemoradiotherapy in patients with rectal cancer. Ann. Surg. Oncol. 2007;14(2):462–469. doi: 10.1245/s10434-006-9171-0. [DOI] [PubMed] [Google Scholar]
- 19.Kazutsugu I., Toshihiro S., Shin T. Evaluation of perineural invasion in rectal cancer as a prognostic factor. Jpn. J. Gastroenterol. Surg. 2013;46(9):635–646. [Google Scholar]
- 20.Hu H., Huang J., Lan P. CEA clearance pattern as a predictor of tumor response to neoadjuvant treatment in rectal cancer: a post-hoc analysis of FOWARC trial. BMC Cancer. 2018;18(1):1145. doi: 10.1186/s12885-018-4997-y. Published 2018 Nov 20. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Mirkin K.A., Kulaylat A.S., Hollenbeak C.S., Messaris E. Prognostic significance of tumor deposits in stage III colon cancer. Ann. Surg. Oncol. 2018;25(11):3179–3184. doi: 10.1245/s10434-018-6661-9. [DOI] [PubMed] [Google Scholar]
- 22.O'Farrell N.J., Donohoe C.L., Muldoon C. Lack of independent significance of a close (<1 mm) circumferential resection margin involvement in esophageal and junctional cancer. Ann. Surg. Oncol. 2013;20(8):2727–2733. doi: 10.1245/s10434-013-2899-4. [DOI] [PubMed] [Google Scholar]
- 23.Kim C.H., Yeom S.S., Lee S.Y. Prognostic impact of perineural invasion in rectal cancer after neoadjuvant chemoradiotherapy. World J. Surg. 2019;43(1):260–272. doi: 10.1007/s00268-018-4774-8. [DOI] [PubMed] [Google Scholar]
- 24.Numata M., Yamaguchi T., Kinugasa Y. Index of estimated benefit from lateral lymph node dissection for middle and lower rectal cancer. Anticancer Res. 2017;37(5):2549–2555. doi: 10.21873/anticanres.11598. [DOI] [PubMed] [Google Scholar]
- 25.Yang J., Chen Q., Li J., Song Z., Cheng Y. Short-term clinical and oncological outcome of prolonging operation interval after neoadjuvant chemoradiotherapy for locally advanced middle and low rectal cancer. Cancer Manag. Res. 2020;12:2315–2325. doi: 10.2147/CMAR.S245794. Published 2020 Mar 27. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Compton C.C., Greene F.L. The staging of colorectal cancer: 2004 and beyond. CA Cancer J. Clin. 2004;54(6):295–308. doi: 10.3322/canjclin.54.6.295. [DOI] [PubMed] [Google Scholar]
- 27.Kelder W., Inberg B., Schaapveld M. Impact of the number of histologically examined lymph nodes on prognosis in colon cancer: a population-based study in the Netherlands. Dis. Colon. Rectum. 2009;52(2):260–267. doi: 10.1007/DCR.0b013e3181979164. [DOI] [PubMed] [Google Scholar]
- 28.Chang G.J., Rodriguez-Bigas M.A., Skibber J.M., Moyer V.A. Lymph node evaluation and survival after curative resection of colon cancer: systematic review. J. Natl. Cancer Inst. 2007;99(6):433–441. doi: 10.1093/jnci/djk092. [DOI] [PubMed] [Google Scholar]
- 29.Shimokawa M., Kogawa T., Shimada T. Overall survival and post-progression survival are potent endpoint in phase III trials of second/third-line chemotherapy for advanced or recurrent epithelial ovarian cancer. J. Cancer. 2018;9(5):872–879. doi: 10.7150/jca.17664. Published 2018 Feb 16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Nie R.C., Zou X.B., Yuan S.Q. Disease-free survival as a surrogate endpoint for overall survival in adjuvant trials of pancreatic cancer: a meta-analysis of 20 randomized controlled trials. BMC Cancer. 2020;20(1):421. doi: 10.1186/s12885-020-06910-5. Published 2020 May 14. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.McArdle C.S., Hole D.J. Outcome following surgery for colorectal cancer: analysis by hospital after adjustment for case-mix and deprivation. Br. J. Cancer. 2002;86(3):331–335. doi: 10.1038/sj.bjc.6600120. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Ling C.R., Wang R., Wang M.J., Ping J., Zhuang W. Prognosis and value of preoperative radiotherapy in locally advanced rectal signet-ring cell carcinoma. Sci. Rep. 2017;7:45334. doi: 10.1038/srep45334. Published 2017 Mar 27. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.