Development and Validation of a Nomogram for Predicting Overall Survival In Synchronous Peritoneal Metastasis of Colorectal Cance

Wenle Chen Sun Yat-sen University Sixth A liated Hospital https://orcid.org/0000-0002-1016-7205 Zixu Yuan Sun Yat-sen University Sixth A liated Hospital Aiwen Wu Peking University Cancer Hospital: Beijing Cancer Hospital Ming Cui Peking University Cancer Hospital: Beijing Cancer Hospital Zhongyi Yue Xinxiang Medical University First A liated Hospital Qianxin Luo Sun Yat-sen University Sixth A liated Hospital Huaiming Wang A liated Gastrointestinal Hospital of Sun Yat-sen University: Sun Yat-sen University Sixth A liated Hospital Li Yang Sun Yat-sen University Sixth A liated Hospital Qiushi Dong Peking University School of Oncology: Beijing Cancer Hospital Duo Liu Sun Yat-sen University Sixth A liated Hospital Michael Hu Wang IBloomington High School South. Ying Huang Xiehe A liated Hospital of Fujian Medical University Hui Wang Sun Yat-sen University Sixth A liated Hospital Jian Cai (  caij29@mail.sysu.edu.cn ) Sun Yat-sen University Sixth A liated Hospital


Background
Colorectal cancer (CRC) is one of the major cancer-related death in world, and peritoneal metastasis (PM) were presented in approximately 11-20% of CRC patients 1 . Synchronous PM was de ned as colorectal cancer patients who have PM within 6 months at diagnosis of primary tumor, which is considered to be the last stage of CRC and is associated with poor prognosis in many studies [2][3] . Due to worse prognosis of PM compared to other organ metastasis, PM is classi ed as M1c of c stage (metastasize to the peritoneal surface with or without other organ) the National Comprehensive Cancer Network (NCCN) guideline (Version 1, 2020) 4 . The mean OS of M1c patients were only 5.2 months, which was lower than M1a or M1b 5 .
Sugarbaker, who created cytoreductive surgery (CRS) for PM, reported 50% of survival rate in colon cancer with peritoneal cancer index (PCI) < 10 6 . However, early diagnosis of PM is often inaccurate or even overdiagnosis. For the last decade, there were many reports on the treatments for PM, including CRS, hyperthermic intraperitoneal chemotherapy (HIPEC), neoadjuvant chemotherapy combined with targeted drugs, systemic chemotherapy, and the combination treatment plan [7][8][9] . Complete CRS requires resection of both the tumor and the entire peritoneum (peritoneal stripping surgery). CRS can remove all macroscopic tumor and prolong the median survival time of patients with PM. The operation is extensive and traumatized, so the morbidity and mortality are relatively high, and recurrences are common. The long-term survival rate has not increased as expected 10 . Most PM treatment targets are palliative rather than curative according to NCCN guidelines (Version 1, 2020), mainly including systemic treatment and no standard scheme. An experienced center may consider surgical R0 removal of isolated peritoneal disease 4 .
Peritoneal Surface Disease Severity Score (PSDSS), a prognostic score was introduced by Pelz at cl. to assess the extent of carcinomatosis, which based on clinical symptoms, PCI and primary tumor pathology. Patients with stage -of PSDSS was obtain survival bene ts from CRS and HIPEC treatment.
PSDS IV re ects biologically invasive cancer, with a 2.6-fold increase in mortality 11 . Another prognostic score named Colorectal Peritoneal Metastases Prognostic Surgical Score (COMPASS) was a new score of prognoses based on age, PCI, lymph nodes and the histology of signet ring cells, which shows better discriminative ability than PSDSS 12 .
Synchronous PM is di cult to be diagnosed preoperatively by current imaging tools including CT scans, MRI imaging or even PET-CT scans 13 . The tools and models to predict survival of PM are also lacking.
Nomogram is a novel model through combining many prognostic factors by mathematical algorithm to predict survival of PM in recent years. It has been used in a variety of malignancies, including lung, stomach, esophageal carcinoma [14][15][16] . In this study, we have included some of factors that were identi ed in the COMPASS model. We have constructed a new nomogram to predict the survival of PM in CRC to help clinicians to make decisions in clinical practice and cancer surveillance.

Patients cohorts
We have enrolled patients of synchronous PM from February 2007 to February 2018 at the hospital where the researchers were based. Those patients were taken as the training cohort. Patients originated of appendiceal cancer or lacking of follow up were excluded. Patients with the missing data of major variables are excluded. The clinicopathological data of all patients were retrieved from a prospective CRC database. An external independent validation cohort composed of 81 patients from other two hospitals were utilized to verify the nomogram. Overall survival (OS) was de ned as the time from the diagnosis of synchronous peritoneal metastasis of colorectal cancer to the death. Our study was approved by the local Ethics Committee.

Variables and De nitions
Refer to risk factors that may be related to PM in previous literature and reported models, we examined the following variables for associations with OS by univariate analysis: age (continuous), gender, body mass index (BMI; continuous) 17 , PCI(continuous) 12 , CRS, HIPEC (yes, no) 18 , Location of primary tumor (left-sided, right-sided), differentiation 19 , presence of liver or lung metastasis 20 , and laboratory markers including hemoglobin, ALB, CA19-9, CA125 and CEA 21 . All variables in this study were converted into binary variables based on the commonly used clinical cutoffs or the cutoffs provided by relevant literature. Among them, PCI was divided by cut-off point of 16 that was analyzed by the X-tile software(chi-sq Hi/Lo = 22.85, P < 0.01). CRS was converted into two groups according to the completeness of cytoreductive surgery (CC): CC 0-1 represents complete CRS of PM, and CC 2-3 was de ned as those patients underwent incomplete CRS or received no CRS. The differentiation of tumor was also converted into two groups: poor-differentiated group represents patients with mucinous adenocarcinoma or signet ring cell carcinoma or poorly differentiated cancer, good-differentiated group consists of patients with moderate to well differentiated, or other pathological types of cancer.

Tests for Two-Way Interactions
To determine whether the effects of any covariates were dependent on other covariates, all pairs of variables with statistical signi cance in univariate analysis were tested for two-way interaction.
Signi cant interaction (P < 0.001) and clinically differentiable effect mediation were required for subsequent consideration in nal models. Higher ordered interactions were not tested for interpretability and reproducibility.

Construction of the Nomogram
Based on the results of multivariable analysis of Cox regression model, the nomogram is constructed with the "rms" and "survival ROC" package in R software studio to visualize Cox regression model. The individual prognosis score and 0.5-, 1-, 2-, and 3-year survival probability were calculated directly based on the nomogram. The predictive performance of the nomogram was evaluated using the concordance index (C-index) and calibration curves. The value of the C-index ranges from 0.5 to 1.0. Brie y, 0.5 indicated a random chance and 1.0 indicated a perfect ability to correctly discriminate the outcome with the model 22 .

Validation and Calibration Curve
The nomogram was subjected to 1,000 bootstrap resamples for validation of the training cohort and the validation cohorts. Calibration curves of the nomogram for 0.5-, 1-, 2-, and 3-year OS were performed with R software by comparing the mean predicted survival rate with the mean actual survival rate determined using a Kaplan-Meier analysis after grouping the nomogram-predicted OS. In a perfectly calibrated model, predictive rates would fall on a 45-degree diagonal line.

Statistical Analysis
All binary variables were calculated by the Kaplan-Meier method and the log-rank test. Log-rank tests were applied to determine univariate prognostic factors. A multivariate Cox proportional hazards model was applied to estimate the independent effects of prognostic factors of OS. The independent prognostic factors by the multivariate analysis were used to construct the nomogram for OS. SPSS v23.0 (IBM, USA) statistical software and R software were used for all statistical analysis in this study. All P values were two-sided, and those variables with P < 0.05 were considered to be statistically signi cant.

Results
A total of 332 patients were analyzed in this retrospective study. All patients underwent surgical exploration and were scored of PCI intraoperatively. There were 127 women (38,2%) and 205 men (61.8%). The median age was 56 (interquartile range = 45.25-67) years. The preoperative CA19-9 was > 37 g/L in 140 patients (42.2%), and preoperative albumin was < 35 g/L in 84 patients (25.3%). There were 221 PM patients (66.6%) combined with liver or lung metastasis. The median PCI score was 12 (interquartile range = 6-26) points, and there were 150 cases (45.2%) with PCI score > 16. In the patients who underwent CRS, 49 patients (14.8%) received complete CRS of PM with CC 0-1, and 85.2% of 283 patients (85.2%) underwent incomplete CRS of CC 2-3. There were 63 patients (19%) who underwent HIPEC plus postoperative adjuvant chemotherapy. The characteristics in the training and validation cohorts were shown in Table 1. The median follow-up time was 13.6 months (range from 1 to 60 months). The median OS was 15.6 months (95% CI: 12.4-19.6 months). Overall, the baseline characteristics were similar between the training and validation cohort. In univariate analysis (Table 2) (Table 2). All variables were analyzed for potential inclusion in the nal multivariable models by forward regression. However, variables with no signi cance and clinically relevant interactions were identi ed for either end point, where clinical relevance was checked via examination of spline plots for continuous variables and hazard ratios for categorical variables across subgroups (data not shown). A model by incorporating these independent prognostic factors was constructed and shown as a nomogram (Fig. 1). Each variable at the topic line located on the relevant axis. A straight line is drawn up to the point axis. Each variable was scored and the nal score of individuals was the sum of all variables, which was matched to the survival rate of 0.5-,1-, 2-, and 3-year for individual patient.
All of these shown a good discrimination of the nomogram. The calibration curves of 0.5-, 1-, 2-and 3years OS in the validation cohort were shown in Fig. 3. These results indicated the predicted OS of nomogram had a good correlation with the actual OS. In addition, the Hosmer-Lemeshow test was not statistically signi cant (P = 0.881), which suggested the good tting of the nomogram. In validation cohort, the model displayed good discrimination with a C-index of 0.642 (95% CI: 0.563-0.720). A nonstatistical signi cance was also observed in the Hosmer-Lemeshow test (P = 0.317). The nomogram was validated by calibration curves of 0.5-, 1-, 2-and 3-years OS with good correlations with actual OS (Fig. 4).

Discussion
In this study, we have constructed a nomogram and validated it with satis ed performances in a validation cohort of PM patients. This nomogram can individually predict 0.5-,1-,2-and 3-year OS of PM patients. As the prognosis of synchronous peritoneal metastasis of colorectal cancer is extremely poor, and the median survival time of its natural outcome is about 0.5 years. The 3-year survival rate is very low after CRS plus HIPEC treatment. Therefore, we have adopted multiple survival time within three years, which is more precise and more suitable in clinical practice. By incorporating demographic and clinical characteristics, the nomogram showed good discrimination and calibration performance by C-index and calibration curve, and is an easy-to-use noninvasive tool for clinicians to estimate the OS and personalize subsequent treatments for individuals. More e cient chemotherapy regime is needed for PM patients with low probability of survival.
PM often is considered to be the terminal state, and obtains worst survival, comparing to other organ metastases. However, the median survival of CRS plus systemic chemotherapy can reach 16 months, which was consistent with the survival of 15.6 months in this study. PCI, CRS and CA19-9 are related to the prognosis of PM patients both in our study and previous studies [23][24][25] . CA19-9 is one important predictor of disease recurrence and associated with postoperative peritoneal recurrence 26 . In our nomogram, PCI score and CRS act the important roles in the model, especially PCI score. PCI score can be obtained by radiological examination and surgical evaluation. The sensitivity of CT scan in diagnosis of PM nodules < 5 mm is only 11-48%. The limited sensitivity of CT scans can in uence the diagnosis of PM, the diagnostic power of PM can be improved with the development of arti cial intelligence to assist the imaging diagnosis 27 . Therefore, radiological PCI is less sensitive than surgical PCI [28][29] . In PM with PCI score > 17, CRS + HIPEC does not offer additional survival bene t 30 . CRS can resolute bowel obstruction and remit bleeding to improve quality of life and create good physical condition to receive systematic chemotherapy and prolong the OS as possible [31][32] . The survival bene t can exceed the disadvantage of high risks of surgical complications for CRS.
It is generally believed that the prognosis of multiple organ metastasis will be worse than single organ metastasis. However, one study suggested that colorectal cancer with liver/lung metastasis obtained similar survival to those with peritoneal-only involvement 33 . We thought the survival of synchronous peritoneal metastasis is extremely poor than multiple organ metastasis including liver/lung metastasis, which can survive for more than 2 years in many patients. Therefore, the presence of liver/lung metastasis may not be a separate factor affecting the prognosis for synchronous peritoneal metastasis of colorectal cancer patients.
Sugarbaker rstly proposed CRS procedure in PM and reported the median operation time was 8 h (5-10 h) with 3.9L (0.5-30L) of blood loss during operation. The overall morbidity rate varied from 23 to 44%, and the mortality rate ranged from 0 to 12% in nine different institutions 34

Conclusion
In summary, we had developed an easy-to-use non-invasive model to predict overall survival of synchronous PM in CRC. The main objective of this predictive model is achieved with the methodology exposed, and the model will help to predict the prognosis of synchronous PM of CRC, but also can give information for the treatment decision.

Consent for publication
Not applicable Postoperative prognostic nomogram for patients with synchronous colorectal peritoneal metastasis.