A New Risk Model Predicting Overall Survival Using Preoperative Nutritional and Inammation Status (NIS) in Patients After Curative Resection of Colorectal Cancer

It has been shown that nutritional status correlates with survival in patients with various kinds of cancers. Besides, cancer causes inammation which has been suggested to stimulate cancer progression. Therefore, inammation status also has shown to reect prognosis of cancers. In this study, we evaluated several kinds of nutritional and inammation parameters in preoperative blood samples and constructed new risk model predicting a survival in patients with CRC (colorectal cancers). We retrospectively examined 286 patients with stage I-III CRC who had undergone curative resection in Teikyo University Hospital between 2013 to 2017. The association between overall survival (OS) and preoperative body mass index, albumin (Alb), cholesterol (Chol), and lymphocyte count, white blood cell count (WBC), neutrophil count (Neu), platelet count (Plt), C-reactive protein (CRP) were examined using Kaplan-Meier curve and log rank test. and eventually Alb, Chol, Neu, Plt, and CRP were shown to correlate with OS. Alb, Chol, Neu, Plt, and CRP were shown to correlate with OS. We constructed a new risk model (NIS: nutrition inammation status) using these factors, and compared its usefulness with known models such as CRP-albumin ratio (CAR), Glasgow prognostic score (GPS), prognostic nutritional index (PNI), and neutrophil lymphocyte ratio (NLR). NIS prepared using nutritional indicators and inammatory ndings was useful as a new model for predicting overall survival in patients undergoing curative resection for CRC, compared with known models.


Introduction
In recent years, as typi ed by the term Precision medicine, identi cation of recurrence risk factors according to individual conditions and formulation of recurrence risk classi cation, which exceeds the TNM classi cation, have been sought. Recent studies have reported that preoperative nutritional status and internal in ammatory ndings as patient factors correlate with postoperative recurrence risk, and that diets that cause colorectal in ammation increase the risk of CRC (colorectal cancer). It suggests a relationship between in ammation and colorectal cancer (1)(2)(3). We clinicians need a simple prognostic tool for CRC. Simple prognostic tools have been developed using in ammatory markers and nutrition index. Control of nutritional status (CONUT), which consists of albumin (ALB), total lymphocyte count (TLC), and total cholesterol serum levels, GPS (Glasgow prognosis score), and PNI (prognostic nutritional index) is a widely established predictor for cancer patients (4)(5)(6). However, its accuracy is still insu cient and there is no classi cation ability that exceeds the TNM classi cation. This study, we aimed to develop a tool using nutritional indicators and in ammatory markers that indicate excellent OS.

Patient selection
Stage I-III CRC diagnosed based on the 8th edition (27) of the United States Joint Commission on Cancer (AJCC) staging system and undergoing radical resection at Teikyo University Hospital in Japan between 2013and 2017. We enrolled 286 consecutive patients. This study has been approved by Teikyo University comittee (Registration Number; . A written informed consent was obtained from all participants, the reporting of our research is in accordance with the STROBE guidelines (28).

PNI
The PNI calculated using serum albumin and the peripheral lymphocyte count is a simple and useful score for predicting the prognosis for various cancer patients (30), where PNI = serum albumin level (g/dL) + 5 × total lymphocyte count (31). Onodera reported that this index provided an accurate, quantitative estimate of operative risk (32). In general, resection and anastomosis of the gastrointestinal tract can be safely practiced when the index is >45. The same procedure may be dangerous when the index is between 45 and 40. When the PNI is <40, this surgery may be contraindicated (32).
Indices of general condition: NLR, CAR NLR is the neutrophil / lymphocyte ratio. CAR is the C-reactive protein / albumin ratio. NLR and CAR were obtained by taking blood within 1-2 week preoperatively. Follow-up Surgical resection was de ned as curative when there was no evidence of tumor recurrence and the distant metastases were histologically and macroscopically complete. Patients were followed up every 3 months for the rst 3 years, every 6 months for the next 2 years. At each follow-up, all patients underwent physical examination and measurements of CEA (serum carcinoembryonic antigen) and CA19-9 (carbohydrate antigen . They also underwent colonoscopy 1-2 years after surgery (rectal cancer was every year after surgery). Thoracoabdominal computed tomography scans were usually taken every 6 months. Recurrence was de ned as the appearance of a radiological, clinical, and / or pathological diagnosis of cancer cells that were local or distant from their original location.

Statistical analysis
Overall survival (OS) were calculated from the date of the patient underwent surgery to that of death, using the Kaplan-Meier method. A Cox regression analysis was performed to identify factors that are signi cantly associated with OS. Probability (p)-values ≤0.05 were considered signi cant. All statistical analyses were performed using JMP 15 software (SAS, Cary, NC, USA).

Determination of cut-off values
The continuous values are based on the Receiver operating characteristics (ROC) curve, and Youdenindex is the Cut-off value. The comparison was made by dividing into two groups, a high-value group and a low-value group. For each value, a ROC curve was created by plotting the sensitivity and speci city of each result under investigation. The score closest to the point with both maximum sensitivity and speci city was selected as the cutoff value, and the largest number of tumors were correctly classi ed for clinical outcome.
The interpretation of an index of probability of concordance (C-index) between predicted probability and actual outcome was used so as to evaluate the predictive ability and discrimination of the model. The value of the C-index should fall between 0.5 and 1.0, with 0.5 indicating random chance and 1.0 indicating a perfect discriminative ability (R version 4.0.3 software).
The AIC (Akaike information criterion ) is a popular method for comparing the adequacy of multiple, possibly nonnested models (33). A statistic that evaluates the predictability of a statistical model using the difference between the observed value and the theoretical value. The smaller the value, the better the t (34).

Determination of cut-off values
The ROC curve analysis results in 3yer-OS indicated that the most appropriate cutoff value for WBC was 6000. All patients were categorized into the high WBC group (WBC ≥6000; n=140, 49.0%) and a low WBC group( WBC 6000; n=146, 51.0%). Neu were classi ed into a high Neu group (Neu ≥4307; n=102, 35.6%) and a low Neu group ( Relationship between nutritional ndings and overall survival BMI (body mass index), lym (lymphocyte count), Alb (albumin), and Chol (cholesterol ) were used as nutritional indicators. The Kaplan-Meier curve of the OS in each item is shown in Fig.1. BMI and lymphocyte count were not signi cant factors. On the other hand, the low Alb group had a signi cantly worse overall survival rate than the high Alb group (p <0.0002). Similarly, Chol had signi cantly worse overall survival in the low Chol group than in the high Chol group (p <0.0001) (Fig.1).
Relationship between In ammatory ndings and overall survival WBC (White blood cell count) , Neu (neutrophil count), Plt (platelets), and CRP (C-reactive protein) were used as in ammatory ndings. The Kaplan-Meier curve of the OS in each item is shown in Fig.2. WBC were not a signi cant factor in the relationship for overall survival (p = 0.234). On the other hand, in the high Neu group, high Plt group, and high CRP group, the overall survival rate is signi cantly worse than that of the low Neu group, low Plt group, and low CRP group, respectively. It was (p = 0.017, 0.006, 0.0003) (Fig. 2).
Relationship between a combination model of known nutritional and in ammatory ndings (NLR, CAR, PNI) tumor marker and overall survival As known models, NLR and CAR were used as in ammation indicators, and PNI and GPS were used as nutritional indicators. NLR was not a signi cant factor in known models and overall survival (p = 0.051).
CAR, PNI, and GPS were the indicators signi cantly associated with overall survival (p = 0.0001, p <0.0001, p = 0.006) (Fig. 3). CEA and CA19-9 were used as tumor markers, but CA19-9 was not a signi cant factor (p = 0.427). On the other hand, the overall survival rate was signi cantly worse in the high CEA group than in the low CEA group (p = 0.002).

Strati cation using patient factors
Univariate analysis was performed using the COX proportional hazard model for the factors of BMI, Alb, Lym, Chol, WBC, Neu, Plt, and CRP in the OS. We decided to create a new risk model using ve factors that were signi cant: Alb, Chol, Neu, Plt, and CRP ( Table 3). The risk model was calculated by adding up the above-mentioned number of risk factors. The factors high group is positive. NIS counted that one positive was 1, 2 was 2, 3 was 3, 4 was 4, and 5 was 5. Kaplan-Meier survival curve was drawn by dividing into positive numbers (Fig. S1). As a result, 3 or more were de ned as the high value group and 2 or less as the low value group. We named this new risk model NIS (nutrition in ammation status).

Kaplan-Meier Curve of NIS
Survival analyses were performed between low NIS group and high NIS group according to cutoff value of NIS. Statistically signi cant differences between the two groups were revealed by Kaplan-Meier curves on 3-year OS (P < 0.0001), indicating a potential prognostic value of NIS. The 3-year OS were 97.1% for the low NIS group, 77.3% for the high NIS group, respectively. A survival curve comparison of 3-year OS between the low NIS and high NIS groups showed a signi cantly poorer prognosis in the high NIS group (Fig.4).
Evaluation of NIS using C-index and AIC compared with T factor and N factor, CAR, PNI, GPS, CEA T factor and N factor were used as oncological factors. Univariate and multivariate analysis was performed using the Cox proportional hazard model to determine the effects of T factor and N factor, CAR, PNI, GPS, CEA, and NIS on overall survival. As a result of multivariate analysis, only three signi cant factors, N factor PNI and NIS, were extracted (HR = 2.9, 95% con dence interval (CI):1.3-6.5, HR = 4.8, 95%CI:1.50-15.5, HR = 4.4, 95% CI: 1.5-12.9, respectively). As shown in Table2, C-indexes for the score (N factor, PNI, NIS) to predict OS were 0.636, 0.726 and 0.747, respectively. The AIC values of each index for OS were 277.85 for N factor, 265.32 for the PNI, 257.34 for the NIS. According to this comparison, NIS was the best goodness-of-t, followed by N factor and GPS. NIS was the best model to re ect the prognosis after colorectal cancer surgery compared to CAR, PNI and GPS (table.2 The items used in the new risk model were ve factors that were signi cantly associated with OS: alb, chol, Neu, plate, and CRP. Each of the ve factors has been reported to be associated with OS. Albumin is produced by hepatocytes, and re ects the nutritional status and amount of skeletal muscle, and is decreased by in ammatory mediators (7) (8). Hypoalbuminemia often may re ect the presence of progressive disease and poor performance status caused by tumor cachexia. The role of pretreatment serum albumin as a prognostic factor was demonstrated by many studies. Cholesterol is widely distributed not only in the blood but also throughout the body such as the brain, internal organs, and muscles, and is a material for bile that helps digestion and absorption of cell membranes, sex hormones, corticosteroids, and fat (9). Total cholesterol levels are known as an indicator of a patient's reserve calories (10). Other studies have reported that low serum cholesterol levels have a poor prognosis for various cancer patients (11)(12)(13). The relationship between low total blood cholesterol and cancer mortality has been reported in many previous studies. It is thought that this is mainly because the total cholesterol level decreases with the development of cancer, especially with the progression of colorectal cancer and advanced cancer (14). Therefore, hypocholesterolemia is not considered to be the cause of cancer, but it is thought to be caused by cancer (13). In this study, it was a poor prognosis factor in the group with a large number of neutrophils.
The mechanism underlying the association between a large number of neutrophils and worse outcomes has not yet been completely clari ed. However, it could be attributed to the association of neutrophils with in ammation. Neutrophils can inhibit the immune system, thereby eliminating the cytolytic activity of immune cells (15) (16) . Therefor both tumor and host cells including neutrophils can simultaneously produce chemokines and cytokines, thus contributing to tumor progression (17).
Rao X et al reported that an elevated platelet count is a negative predictor of survival in both primary CRC and resectable colorectal liver metastases(18). Many studies have shown that increased platelet count promotes the growth, in ltration, and metastasis of cancer (19,20). Platelets can secrete a variety of growth factors and angiogenesis regulatory proteins (IL-6, vascular endothelial growth factor, plateletderived growth factor, platelet factor 4) to promote tumor formation and its metastasis (21) (22).
Allin et al reported that patients with invasive breast cancer and diagnostic CRP levels >3mg/L had a 1.7fold increased risk of death from breast cancer compared to patients with diagnostic CRP levels <1mg/L (30). Biological mechanisms associated with elevated CRP lebels and poor cancer prognosis may explain in ammation in the tumor microenvironment. In ammation of the tumor microenvironment induces DNA damage, promotes angiogenesis, and promotes tumor spread and distant metastasis. The result is an attractive tumorigenesis promoting environment for tumor growth (23)(24)(25).
Currently, TNM staging is most used to predict survival outcomes and treatment choices. However, because TNM staging was done postoperatively, survival cannot be predicted preoperatively. Also, no further treatment strategy can be determined. In addition, the TNM stage can only re ect the biological behavior of the tumor. The prognosis of cancer was associated with nutritional status not only with the clinicopathological features of the tumor, but also with the in ammatory response of the host (1) (26). NIS is based on peripheral neutrophils, platelet count, CRP, blood albumin levels, cholesterol levels, re ects the state of the tumor microenvironment and the preoperative host in ammatory response and nutritional status. Our ndings show that preoperative NIS has stronger prognostic discrimination capabilities compared to NLR, CAR, GPS, and PNI. Therefore, the use of a combination of parameters that re ects the host of both the nutritional status and the systemic in ammatory status may be important for accurately predicting survival outcomes in CRC patients.
This study has its limitations. First, this study was retrospective in design and included patients from a single institution. Second, this study was small sample size. Third, there are no consensus regarding the nutrition and in ammatory markers cut-off value. We selected nutrition and in ammatory markers cut-off value by performing a ROC analysis. Our current ndings require further review and validation in more CRC patients due to the smaller number of cases. Fourth, the results of this study do not apply to stage IV patients, as they included stage I-III CRC patients who underwent curative surgery.