This study ultimately enrolled 328 stage II-III colon cancer patients, who underwent curative surgical resection at the Department of General Surgery, the First Affiliated Hospital of Xi’an Jiaotong University from January 2011 to December 2016 ( Figure 1 ). All subjects were diagnosed with preoperative biopsy or postoperative pathological examination. All patients were diagnosed with colon cancer and not in combination with other cancers, and none of them underwent neoadjuvant chemotherapy before surgery. The following were the exclusion criteria: definite inflammatory or hematologic diseases affecting immune or nutritional status; unknown prognostic information due to lost to follow-up; missing blood test results within one week before surgery. Patients were followed up via outpatient services or telephone interviews to obtain their health condition. The follow-up concluded in December 2022. The outcome was specified as either OS or CSS. OS was described as the interval between the surgical procedure and mortality resulting from any cause. The period from surgery to mortality as a result of recurrence or metastasis of the primary malignancy was referred to as CSS. For CSS, a patient’s death was defined as a non-endpoint event if the cause of death was not due to cancer. Cases lacking an outcome event were censored in the analysis.

The clinicopathological indicators extracted from the patient’s electronic health data include age, sex, underlying disease, anemia, tumor location, American Society of Anesthesiologists (ASA) classification, surgical approach, ascites, intestinal perforation, obstruction, tumor size, gross tumor type, pathology type, differentiation, N stage, T stage, TNM stage, lymph node dissection, perineural invasion, vessel invasion and chemotherapy. Laboratory examinations within one week before surgery include ALB, CHOL, total counts of neutrophils, lymphocytes, and monocytes.

M-NPS is a scoring system calculated from four factors: ALB, CHOL, LMR and NLR. Given that traditional NPS failed to exhibit substantial clinical utility in the study population, X-Tile 3.6.1 software (Yale University, New Haven, USA) (33) was employed to identify the most suitable threshold value for the four indicators according to the maximum chi-square and the lowest P values of log-rank tests ( Figure 2 ). Table 1 provides the comprehensive scoring criteria for the M-NPS. Patients were classified into three groups according to their total M-NPS scores: group 1 for scores of 0, group 2 for scores of 1-2, and group 3 for scores of 3-4.

All curative surgeries for patients enrolled in the study were conducted by a skilled surgical team. They underwent open resection or laparoscopic operation, subject to the location of the disease and the surgical team’s preference. Adjuvant chemotherapy was implemented within two months postoperatively according to pathologic TNM staging results and the patient’s condition. Some patients discontinue chemotherapy midway because they cannot tolerate the side effects.

The medians and interquartile ranges (IQRs) of continuous variables are shown, while categorical variables are described as percentages and numbers. Continuous variables were compared by the Kruskal-Waills test or the Mann-Whitney U test. For comparison of categorical variables between two or more groups, chi-square or Fisher’s exact tests were implemented. The survival and survminer R packages were employed to plot KM survival curves and compared their survival rates with a log-rank test. Independent prognostic analysis for OS or CSS was determined by multivariate Cox regression. A multivariate analysis was performed only on parameters with P<0.05 in a univariate analysis. The dataset was randomly divided in 6.5:3.5 into training and validation sets using the sample function. Independent prognostic variables were employed to develop nomograms in the training set using the rms R package. The model discrimination can be assessed with the C-index and the ROC curve. The calibration curve was generated with the rms R package to evaluate the prediction ability of the model. Additionally, the DCA was employed to evaluate the clinical efficacy of the model using the ggDCA R package. Version 4.3.2 of the R software was used to conduct all statistical analyses. Statistical tests with P< 0.05 are significantly difference.

Figure 1 displays the inclusion of 328 patients in the ultimate statistical analysis. Of the 328 patients, the age at diagnosis had a median value of 62 years. The research comprised 165 male and 163 female participants. The study subjects consisted of 113 patients with underlying diseases and 215 patients without underlying diseases. There were 148 patients with left hemicolon cancers, while 180 patients had right hemicolon cancers. 199 patients received open resection and 129 underwent laparoscopic surgery ( Table 2 ). The median OS and CSS were both 86 months. Based on M-NPS scores, the cohort was stratified into three groups: 153 patients in group 0, 145 in group 1, and 30 in group 2. Of all clinicopathologic features, male, no underlying disease, open surgery and larger tumor size were associated with higher M-NPS scores. Lower ASA classification, ascites, no intestinal perforation, no obstruction and chemotherapy were related to lower M-NPS scores ( Table 2 ). In addition, ALB, CHOL and LMR were all negatively correlated with M-NPS scores, whereas NLR was positively related to M-NPS scores ( Table 2 ).

Survival analysis of OS and CSS based on M-NPS grouping was performed and survival curves were plotted in Figure 3 . For OS in stage II-III patients, the median survival for group 2 was 95 months. Group 0 had more favorable survival than both group 1 and group 2 ( Figure 3A ). Further stratified survival analysis using staging showed that in comparison to group 0 and group 1, group 2 had the poorest prognosis for stage II patients ( Figure 3B ). In stage III patients, group 0 had superior survival than group 1 ( Figure 3C ), and a significant survival difference was not observed between group 1 and 2, nor between group 0 and 2. In terms of CSS for stage II-III patients, the median survival time in group 2 was 119 months. Group 0 had the best survival compared to group 1 and group 2 ( Figure 3D ). Similarly, among stage II patients, group 2 had poorer CSS compared to those in group 1 and 0 ( Figure 3E ). For stage III patients, group 0 had the most favorable CSS, and the remaining two groups did not appear to differ markedly in survival ( Figure 3F ).

We conducted univariate and multivariate Cox regression analysis on OS for stage II-III patients. The results are displayed in Table 3 . OS had a strong association with the M-NPS group, as indicated by the univariate analysis. Additional multivariate analysis revealed that the M-NPS group was an independent risk indicator for OS. Group 1 and group 2 had worse OS compared to group 0 [G1 vs G0: 1.73 (1.05–2.86); G2 vs G0: 2.55 (1.29–5.03)]. Other independent prognostic risk indicators included age, obstruction, gross tumor type, and N stage.

In addition, the same analytical methods as for OS were executed for CSS ( Table 4 ). CSS was notably correlated with the M-NPS group in the univariate analysis. CSS was independently influenced by the M-NPS group in multivariate analysis. Group 2 had worse CSS compared to group 0 [G2 vs G0: 2.52 (1.24–5.14)]. Obstruction, gross tumor type, and N stage were also independent prognostic risk parameters of CSS.

The total of 328 patients finally included were stochastically allocated into two cohorts: 213 in the training set and 115 in the validation set. All clinicopathological characteristics in two sets are listed in Table 5 . No statistical difference in clinicopathological characteristics between the two cohorts proved that the allocation was completely randomized ( Table 5 ).

OS and CSS nomograms were created to forecast the survival rates at 3, 5, and 10 years, using independent parameters that influence OS and CSS, respectively ( Figure 4 ). Nomograms can personalize the forecast of OS and CSS for each patient by calculating the sum of the corresponding score for each variable in the models.

The C-indexes for OS and CSS nomograms in the training cohort were 0.802 (0.747–0.857) and 0.783 (0.722–0.844), respectively. In the validation cohort, the C-indexes for OS and CSS nomograms were 0.742 (0.656–0.828) and 0.745 (0.649–0.841), respectively.

ROC analysis indicated that the nomogram for OS in the training cohort achieved AUC values of 0.82, 0.82, and 0.87 for the 3-, 5-, and 10-year predictions, respectively ( Figure 5A ). In the validation cohort, the predictive model for OS reached AUC values of 0.75, 0.77, and 0.82, respectively ( Figure 5B ). The nomogram’s calibration curves demonstrated that excellent alignment between predicted and actual 3-, 5-, and 10-year OS in both the training ( Figure 5C ) and validation sets ( Figure 5D ). In addition, DCA curves showed that the predictive model of OS had excellent clinical efficacy at 3, 5, and 10 years in clinical practice ( Figures 5E, F ).

At the same time, the CSS nomogram in the training set attained AUC values of 0.81, 0.80, and 0.80 for the 3-, 5-, and 10-year predictions, respectively ( Figure 6A ). In the validation set, the CSS nomogram got AUC values of 0.73, 0.75, and 0.80 for the 3-, 5-, and 10-year predictions, respectively ( Figure 6B ). In the training ( Figure 6C ) and validation cohort ( Figure 6D ), the nomogram’s calibration curves exhibited an excellent level of consistency between the model-forecasted CSS and the actual survival outcomes. Additionally, DCA curves for the nomogram of CSS indicated that the model had favorable clinical efficacy in both the training ( Figure 6E ) and validation cohort ( Figure 6F ).