The Geriatric Nutritional Risk Index predicts postoperative complications and prognosis in elderly patients with colorectal cancer after curative surgery

Malnutrition has been considered to be associated with the prognosis of cancer. The Geriatric Nutritional Risk Index (GNRI), based on serum albumin levels, present body weight, and ideal body weight, is a simple screening tool to predict the risk of nutrition-related morbidity and mortality in elderly patients. We aimed to evaluate whether preoperative GNRI was associated with postoperative complications and prognosis in elderly patients with colorectal cancer (CRC). We retrospectively enrolled 313 CRC patients aged ≥65 years after curative surgery and classified them into an all-risk GNRI (≤98) group and a no-risk GNRI (>98) group. Kaplan-Meier analysis showed overall survival was significantly worse in the all-risk GNRI group than in the no-risk GNRI group (P = 0.009). Multivariable analyses showed low GNRI (≤98) was an independent risk factor for postoperative complications (P = 0.048) and overall survival (P = 0.001) in the patients. Among the complications, the incidence of surgical site infection, in particular, was significantly higher in the all-risk GNRI group (P = 0.008). In conclusion, low preoperative GNRI (≤98) was associated with increased postoperative complications and poor prognosis. Preoperative GNRI can be used as an identifier for potential high-risk group of morbidity and mortality in elderly CRC patients.

To date, there have been no reports on the relationship between GNRI and short-or long-term outcomes for elderly patients with CRC after surgery. Therefore, in this study, we investigated whether preoperative GNRI was associated with postoperative complications and prognosis for elderly patients with CRC who underwent curative surgery.
Methods patients and datasets. This study retrospectively enrolled 313 patients with CRC aged ≥65 years who underwent curative resection at Osaka University Hospital from August 2007 to December 2012. Patients who underwent curative resection for distant metastases were also included. Exclusion criteria for patients were as follows: (1) aged <65 years, (2) surgery for recurrence, (3) multiple primaries, (4) colitic cancer, (5) received neoadjuvant chemotherapy, (6) underwent transanal endoscopic microsurgery, (7) cases which lacked any of preoperative laboratory data or pathological findings described in Table 1. Two hundred and eighteen elderly CRC patients who underwent curative surgery at Osaka International Cancer Institute from January 2007 to December 2013 were enrolled according to the same criteria as described above, and analyzed as another dataset.
Clinicopathological factors such as age, sex, body mass index (BMI), serum albumin level (ALB), white blood cells, C-reactive protein (CRP), carcinoembryonic antigen (CEA), carbohydrate antigen 19-9 (CA19-9), primary tumor location, distant metastases, pathological findings, and postoperative complications were collected from patients' medical records. Clinicopathological factors were classified according to the eighth edition of the Union for International Cancer Control (UICC) tumor-node-metastasis (TNM) classification 25 . Preoperative blood samples, height, and weight data were obtained within 7 days before surgery. Postoperative complications were classified according to the Clavien-Dindo (CD) grade 26 . In the present study, we examined those of CD grade ≥II 27 .
After surgery, all patients were followed up according to the Japanese guidelines 28 . They were regularly examined using tumor markers, such as CEA and CA19-9, and screened using computed tomography every 3-6 months and colonoscopy every 1-2 years. nutritional assessment by GnRi. The GNRI is a simple and objective screening tool for elderly patients' nutrition-related risk calculated using ALB, present body weight (PBW), and ideal body weight (IBW). IBW in this study was calculated as follows: IBW = height 2 (m) × 22. The GNRI formula is: GNRI = 1.487 × ALB (g/L) + 41.7 × PBW/IBW (kg) 17 .
Statistical analysis. Continuous variables were expressed as means ± standard deviation (SD) values.
Differences between the classified GNRI groups and clinicopathological factors were analysed using chi-squared test or Fisher's exact test. The relationships between GNRI and each complication were also analysed by the same tests. Continuous variables with parametric distribution were analysed by Student's t-test or analysis of variance www.nature.com/scientificreports www.nature.com/scientificreports/ (ANOVA). Overall survival (OS) curves were plotted using the Kaplan-Meier method and compared using the generalised log-rank test. Univariate and multivariate analyses were performed using a logistic regression model to identify independent risk factors for postoperative complications and using a Cox proportional hazards regression model for OS. Receiver operating characteristic (ROC) curve analysis was used to predict the optimal cut-off value of GNRI for OS 29 . In this study, we used the patients who were followed for at least one year as evaluable for the prognostic outcome to perform the ROC analysis. Then, the value was provided based on Youden's index 30 . Two-sided P < 0.05 was considered to denote statistical significance. All statistical analyses were performed using JMP software version 13 (SAS Institute Inc., Cary, NC, USA). compliance with ethical review. This study was performed in accordance with the principles of Declaration of Helsinki. This study was approved by the Institutional Review Boards of Osaka University and Osaka International Cancer Institute, and informed consent was obtained from all patients according to the guideline.  Table 1. The median age was 73 years (range, 65-94 years). There were 29 (9.3%) patients with stage 0, 115 (36.7%) patients with stage I, 77 (24.6%) patients with stage II, 83 (26.5%) patients with stage III, and 9 (2.9%) patients with stage IV. The stage IV cases included liver metastasis (6 cases), extra-regional lymph node metastasis (1 case), and peritoneal dissemination (2 cases). Sixty-four (20.4%) patients had postoperative complications and 41 (13.1%) patients had those of CD grade ≥II.

Results
Distribution and classification of GNRI. The mean preoperative GNRI in 313 patients with CRC was 98.2 ± 9.6. Differences in the distribution of preoperative GNRI according to postoperative complications (CD grade ≥II) and TNM stages are shown in Fig. 1. The mean GNRI was 98.9 ± 9.2 in patients who had postoperative complications and 93.8 ± 11.0 in those without complications. There was a significant difference in preoperative GNRI between the two groups (P = 0.002) (Fig. 1a). The mean GNRI was 99.9 ± 7.1 in stage 0, 98.9 ± 8.8 in stage I, 97.3 ± 9.6 in stage II, 97.2 ± 11.3 in stage III, and 101.4 ± 9.1 in stage IV. There were no significant differences in preoperative GNRI among these stages (P = 0.390) (Fig. 1b).
A previous study showed that a good sensitivity for risk prediction was found only for a GNRI cut-off value of 98 31 . ROC curve analysis for OS also showed that the optimal cut-off value of GNRI was 98.082 (area under the curve = 0.574, sensitivity = 0.591, and specificity = 0.569) (Fig. 2).
According to previous studies 23,24,31 and the ROC analysis, we classified patients more simply into an all-risk GNRI (≤98) group (137 patients, 43.8%) and a no-risk GNRI (>98) group (176 patients, 56.2%), instead of the four classifications of Bouillanne et al. 17 . The relationship between GNRI status and clinicopathological factors in all patients is shown in Table 2. Between the all-and no-risk GNRI groups, there were no significant differences in age, white blood cells, preoperative CEA, preoperative CA19-9, tumor location, degree of differentiation, depth of tumor invasion, lymph node metastasis, lymphatic vessel invasion, venous invasion, distant metastasis, or TNM stage. However, there were significant differences in sex, BMI, ALB, CRP, and postoperative complications (CD grade ≥II) between the two groups. postoperative complications (cD grade ≥ ii). A total of 41 patients had postoperative complications defined CD grade ≥II. These were surgical site infection (11 cases), ileus (8 cases), anastomotic leakage (7 cases), intra-abdominal abscess (5 cases), colitis (4 cases), pneumonia (3 cases), and urinary infection (3 cases). More patients had postoperative complications in the all-risk GNRI group (18.2%) than in the no-risk GNRI group (9.1%) (P = 0.018). The relationship between GNRI status and each complication was examined, and surgical site infection occurrence was higher in the all-risk GNRI group than in the no-risk GNRI group (P = 0.008) ( Table 3). www.nature.com/scientificreports www.nature.com/scientificreports/ Univariate and multivariate analyses of clinicopathological factors for postoperative complications (CD grade ≥II) are shown in Table 4. According to the univariate analysis, high CRP (P = 0.032), tumor location (rectum) (P = 0.005), and low GNRI (P = 0.019) were significantly correlated with the complications. The multivariate analysis showed that tumor location (rectum) (P = 0.005) and low GNRI (P = 0.048) were independent risk factors for postoperative complications.
Survival analysis and risk factors for mortality. The median follow-up was 60.5 months (range, 1-137 months). Thirty-two death events and 105 censoring cases were recorded in the all-risk GNRI group, and 26 death events and 150 censoring cases were recorded in the no-risk GNRI group. OS rate was significantly worse in the    www.nature.com/scientificreports www.nature.com/scientificreports/ all-risk GNRI group than in the no-risk GNRI group (P = 0.009) (Fig. 3). The 3-and 5-year OS rates in the all-risk GNRI group were 89.0% and 79.6%, and those in the no-risk GNRI group were 92.2% and 86.0%, respectively.

Analyses of the complications and prognosis in the other dataset based on GNRI.
To verify whether GNRI could be used for the prediction, we performed the other center study using the patient data in Osaka International Cancer Institute. Characteristics of all the patients in the other dataset are listed in Supplementary Table 1. The median age was 72 years (range, 65-88 years). Fifty-three (24.3%) patients had postoperative complications of CD grade ≥II. The mean preoperative GNRI in the patients was 101.9 ± 9.2.
www.nature.com/scientificreports www.nature.com/scientificreports/ Furthermore, Kaplan-Meyer curve analysis in the other center study also showed that OS rate was significantly worse in the all-risk GNRI group than in the no-risk GNRI group (P = 0.002) (Supplementary Fig. 1). The 3-and 5-year OS rates in the all-risk GNRI group were 83.7% and 77.6%, and those in the no-risk GNRI group were 96.7% and 91.1%, respectively. The univariate and multivariate analyses for OS are shown in Supplementary  Table 3. According to the univariate analysis, lymph node metastasis (P = 0.004), distant metastasis (P = 0.001), and low GNRI (P = 0.005) were significantly related to OS. The multivariate analysis also showed that lymph node metastasis (P = 0.035), distant metastasis (P = 0.042), and low GNRI (P = 0.048) were independent prognostic risk factors for OS.

Discussion
Our results showed that GNRI was associated with increased postoperative complications and poor prognosis of CRC in elderly patients. Malnutrition has been found to be an important risk factor for postoperative morbidity and mortality in malignant tumors 32,33 . The Nutritional Risk Index (NRI), calculated by ALB, PBW, and usual body weight, was proposed by Buzby et al. to evaluate the association between nutrition and postoperative complications 34,35 . However, the NRI is often difficult to use in elderly patients 36 because half of them do not remember their own usual body weight 37 . Thus, Bouillanne et al. replaced usual body weight with IBW in the formula of NRI and developed a simple screening tool specific for elderly patients to predict nutrition-related risk of morbidity and mortality 17 . GNRI was developed in the population of which elderly patients aged ≥65 years were admitted into a geriatric rehabilitation care hospital due to rehabilitation after fractures, neurologic diseases, cardiovascular diseases, and postinfectious diseases and also reported to be significantly correlated with ALB, prealbumin, weight, and BMI 17 .
There are several methods for assessing nutritional status, such as BMI, prognostic nutritional index, skeletal muscle mass index, and subjective global assessment. While these measures are relevant for the prognosis of   Table 5. The univariate and multivariate analyses of prognostic factors for overall survival. HR = hazard ratio, CI = confidence interval, BMI = body mass index, CEA = carcinoembryonic antigen, CA19-9 = carbohydrate antigen 19-9, por = poorly differentiated adenocarcinoma, pap = papillary adenocarcinoma, muc = mucinous adenocarcinoma, tub1 = well differentiated adenocarcinoma, tub2 = moderately differentiated adenocarcinoma, GNRI = geriatric nutritional risk index, Asterisk values indicate P-values < 0.05.
In contrast, the advantage of GNRI is that it is an objective and easily available predicting tool. The classification value of GNRI has already been proposed 17 . Moreover, this index is calculated using ALB, height, and body weight, which are usually measured on admission.
Previously, GNRI was considered as a prognostic predictor for length of stay in hospital 31 and chronic diseases in elderly patients, such as those with heart failure 18 or chronic obstructive pulmonary disease 19 , or those undergoing haemodialysis 20 . Recently, GNRI has been reported to be useful as a predictor for morbidity and mortality in patients with cancer. Li et al. reported that lower GNRI value was associated with severe postoperative complications, including liver failure, and poor OS in elderly patients with hepatocellular carcinoma 21 . Kushiyama et al. suggested that GNRI < 92 was a risk factor for postoperative complications in elderly patients with gastric cancer 22 . Bo et al. indicated that GNRI ≤ 98 could be an indicator of poor survival in elderly patients with oesophageal cancer treated with radiotherapy 23 . Miyake et al. also reported that GNRI could be a prognostic predictor in elderly patients with non-metastatic renal cell carcinoma, and those with GNRI ≤ 98 had significantly worse cancer-specific survival (CSS) than those with GNRI > 98 24 .
Some reports used the modified GNRI classification according to the complications 22,42 , OS 23 , CSS 24 and length of hospital stay 31 , and some reports used the four-group classification proposed by Bouillanne et al. 21,43 . In this study, we divided the patients more simply into two groups by the GNRI value 98 based on the ROC analysis and these previous studies 23,24 . Cereda et al. also suggested that only a GNRI cut-off value of 98 had good sensitivity for risk prediction 31 . Our results showed that GNRI was related to the complications and prognosis of CRC, and it was considered that our classification was appropriate.
Postoperative complications after CRC resection have been reported to be associated with poor oncologic outcomes, even if they are mild or moderate (CD grade II) 27 . For this reason, we considered not only severe complications (CD grade ≥III) but all complications of CD grade ≥II in the present study.
To the best of our knowledge, this is the first study to investigate the relationship between GNRI and outcomes in elderly patients with CRC. Our study demonstrated that low preoperative GNRI (≤98) was correlated with increased postoperative complications (CD grade ≥II) and worse OS compared with high GNRI (>98) and that low GNRI was an independent risk factor for morbidity and mortality. In addition, although we examined the relationship between GNRI and TNM stages, no significant correlation between them was found. Therefore, we considered that GNRI was also an independent prognostic factor that did not depend on TNM stage.
Several studies have suggested that preoperative nutritional status is an independent risk factor for anastomotic leakage and wound infection in patients with CRC 44,45 . Our study also showed that GNRI was a significant risk factor for wound infection, but it was not a risk factor for anastomotic leakage. Our result obtained for anastomotic leakage might be due to the small number of cases, and these may have been more influenced by tumor location and surgical procedure. Some studies also showed that enhanced recovery after surgery protocol was associated with decreased postoperative complications 46 and improved survival in CRC 47 . Appropriate management of nutritional status before and after surgery may be important to improve surgical risk and prognosis.
Low ALB is correlated with poor prognosis of cancer 48 . ALB is a known indicator of nutritional status 49 , and malnutrition impairs various functions, such as immunity, digestive tract function, and wound healing 50 . Deficiency of these functions increases the risk of infection and postoperative complications 51,52 , and an immunosuppressed condition leads to inadequate anti-tumor immunological reaction 53,54 . Furthermore, ALB is also influenced by inflammation 49 , and systemic inflammation is associated with poor prognosis of cancer 55 . On the other hand, the PBW/IBW ratio used in GNRI, which replaces the PBW/usual body weight ratio indicating weight loss, might be interpreted as reflecting the degree of frailty and cachexia associated with poor prognosis in elderly patients 56 . Thus, the GNRI, which combines factors of ALB and body weight, may predict nutrition-related risk better than ALB alone.
There are some limitations to our study. First, this study was a retrospective study evaluated only a small number of patients and institutes, and also affected by some selection and information bias. Prospective multicenter studies should be performed. Second, there is no single definition of elderly patients. While we defined elderly patients as those aged ≥ 65 years in the present study, the life span has extended and the number of patients aged>80 years has been increasing. Similar analyses may also have to be performed in patients aged >80 years. Third, our study did not assess the influence of smoking behavior because of lack of the information. Smoking is well known as a risk factor of malnutrition, postoperative complications, and poor cancer prognosis [57][58][59] . In contrast, there is no consensus on the association between smoking and BMI or body weight 59,60 . How smoking actually influences on GNRI status and our findings is not clear, and further research including smoking status is necessary to make it more meaningful and accurate.
In conclusion, our study demonstrated that low preoperative GNRI value (≤98) was associated with increased postoperative complications and poor prognosis in patients with CRC aged ≥65 years after curative surgery. Preoperative GNRI can be a useful tool to identify high-risk population of morbidity and mortality in elderly patients with CRC.

Data availability
The dataset used and analysed in the present study is available from the corresponding author on reasonable request. www.nature.com/scientificreports www.nature.com/scientificreports/ www.nature.com/scientificreports www.nature.com/scientificreports/