A prognostic nomogram and web-based real-time calculator 1 for predicting CSS and OS of primary rectal 2 adenocarcinoma patients

Background: The purpose of this study was to construct a clinical prognostic model of primary 17 RAC to assist in clinical diagnosis and treatment. Methods: Primary RAC patients from 2010 to 2015 were selected from the surveillance, 20 epidemiology and end results (SEER). The relevant significant variables were used to 21 develop the prognosis model. A score was determined for each prognostic factor in the 22 visualized by the nomogram for 3 and 5 years and the web-based real-time calculator 45 are helpful to optimize clinical work, facilitate patient consultation and clinical individualized treatment.

, and peritoneal invasion (Shepherd et al., 1995). There are 57 also other factors strongly influencing RAC prognosis : For example, individual 58 patient differences, including age, sex, marital status, and insurance; treatment 59 measures including surgery, radiation, chemotherapy and specific biological markers; 60 and other factors associated with the tumor, including tumor size and perineural 61 invasion (Soto-Ferrari et al., 2017;Liu et al., 2018;Fan et al., 2019). It is significant to 62 predict the prognosis of the patient with primary RAC by combining all 63 above-mentioned factors, thus enabling to improve the quality of individualized 64 survival and the strategy of treatment, according to the risk such as health disparities, 65 therapy method. 66 For the moment, it is most common to guide treatment strategy and predict the 67 prognosis of primary RAC with the AJCC TNM staging system, which is for a widely 68 applied tool of tumor assessment (Cong et al., 2018;Fujima et al., 2019). Though the 69 TNM stage is vital and useful, it may be inadequate that it takes into account only 70 some of the factors associated with the tumor itself, such as the depth of tumor 71 invasion, numbers of lymph nodes and whether it has distant metastasis. As is known 72 to us all, when patients are at the same TNM stage, their prognosis usually varies 73 widely (Zhao and Chen, 2015;Gao et al., 2018). The nomogram or a web-based 74 calculator is a multivariable model that considers multiple prognostic factors 75 simultaneously, which is more suitable for the study of multivariable effects in large 76 data sets to accurately predict the prognosis of cancer patients (Ullman et al., 2015;77 Jorissen et al., 2019;Hu et al., 2020;Jehi et al., 2020;Kim et al., 2020).

78
Approximately 90% of colorectal cancer is pathologically diagnosed as 79 adenocarcinoma (Evdokimova et al., 2018). However, the nomogram or a web-based 80 calculator for evaluating the prognosis of patients with primary RAC has not been 81 constructed.

82
In this study, we used the SEER database combining tumor-related variables, 83 patient-specific variational factors, therapeutic measures, and tumor biomarkers to 84 analyze a series of factors that may affect primary RAC prognosis aiming to establish 85 a primary RAC prognosis prediction model shown by a predicted nomogram for 3 or 86 5 years and a real-time web-based calculator.

88
Case collection from SEER database 89 The patients diagnosed with the primary RAC between 2010 and 2015 were 90 retrieved from the newest SEER database (version 8.3.5; https://seer.cancer.gov/).

91
Since the database is publicly supported for use, a local ethics approval or a statement Variables analyzed in this study were extracted including age, sex, race, marital 104 status, insurance, tumor size, Grade, T stage, N stage, M stage, SEER stage, tumor 105 deposits, regional LN examined, perineural invasion, Mets at dx-bone, Mets at 106 dx-brain, Mets at dx-liver, Mets at dx-lung, surgery, chemotherapy, radiation sequence 107 with surgery, tumor biomarker CEA, survival related information, etc. Overall 108 survival (OS) was considered as the primary endpoint and cancer-specific survival 109 (CSS) was considered as the secondary endpoint. The optimal cut-off value of age, 110 tumor size and regional LN examined was obtained by X-tile software which can 111 quickly find the optimal truncation value when searching for truncation value 112 of survival statistics (Guan et al., 2016) and can transform these three continuous 113 variables into component type variables.

114
The construction of prognostic model for the OS and CSS of the primary RAC 115 The Kaplan-Meier method and log-rank test were applied to conduct survival 116 analysis. Univariate Cox analysis was applied to select significant variables (p<0.05), 117 which were further submitted to multivariate COX analysis. Independent prognostic 118 factors filtered by multivariable Cox proportional hazards models were used to 119 construct a prognosis prediction model of primary RAC based on the training set. In 120 this model, a score was defined for each risk factor. Accordingly, a new risk     Table 1. 156 We used the X-tile software to find the optimal cutoff values for age, tumor size, 157 and number of lymph nodes examined, which were at 66 and 77 years, 33 mm and 61 158 mm, nine and fourteen, respectively ( Figure 2). Univariate and multivariate analyses 159 of overall survival in the training set were described in Table 2. According to the 160 results, fourteen variables including age at diagnosis, sex, marital status, insurance, 161 tumor size, T stage, N stage, CEA, regional LN examined, perineural invasion, Mets 162 at dx-liver, Mets at dx-lung, surgery and radiation sequence with surgery were 163 associated with OS of the primary RAC. Results of univariate and multivariate 164 analyses of CSS in the training set were presented in Table 3. To more vividly depict  variables. In this model, a score was defined for each risk factor. (Table4) Then, 175 according to the total score of each patient in the training set, the cutoff value was 176 selected using X-tile software, and a risk classification system was established. The 177 patients were divided into three groups: low-risk group, medium-risk group and 178 high-risk group. (Figure 3) Moreover, the Kaplan-Meier method and the log-rank test  The predictive effectiveness analysis of the prognosis models 225 To test whether the model has better clinical applicability, the DCA curve was 226 plotted. (Figure 9) From the diagram, we can see that the nomogram has higher net 227 benefits than the   The authors report no conflicts of interest.        Tis  0  0  T1  77  73  T2  82  72  T3  88  86  T4a  100  100  T4b  95  94  N stage  N0  0  0  N1  6  8  N2 13