This retrospective study was conducted at our hospital from January 2022 to December 2024 to develop and evaluate a nomogram model for predicting malnutrition in patients with colorectal cancer. Patients included in the study met the following criteria: (1) a confirmed pathological diagnosis of colorectal cancer based on histological examination; (2) aged 18 years or older; and (3) a Karnofsky Performance Status (KPS) score of ≥60. Additionally, participants were required to have comprehensive and accurate nutritional assessments available, including body mass index (BMI), serum albumin levels, and Nutritional Risk Screening scores. Exclusion criteria were as follows: (1) a history of secondary malignancies or coexisting cancers unrelated to colorectal cancer; (2) severe comorbidities, including significant cardiovascular, hepatic, renal, or respiratory conditions that could independently affect nutritional status; and (3) incomplete or missing clinical or nutritional data essential for the development of the model. A total of 216 patients with colorectal cancer were included in the study, comprising 56 patients classified as malnourished and 160 patients classified as well-nourished based on established criteria. The study protocol adhered to the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines (15). Informed consent was obtained from all participants or their legal guardians. The study was reviewed and approved by the hospital’s ethics committee (2025-0107-009), and conducted in accordance with relevant guidelines and the Declaration of Helsinki. All data was kept confidential, with personal identifiers removed prior to analysis to ensure participant privacy.

This study adopted the GLIM criteria for the diagnosis of malnutrition (16). The diagnostic process involved a two-step approach: (1) Screening for malnutrition risk: The NRS-2002 tool was utilized as the primary screening instrument. Patients with an NRS-2002 score of ≥3 were identified as being at risk for malnutrition through a review of medical records. (2) Assessment of malnutrition: Patients identified as at risk in the screening step underwent a detailed malnutrition assessment based on the GLIM criteria. The assessment comprised five components grouped into two categories, with patients being classified as malnourished if they fulfilled at least one phenotypic criterion and one etiologic criterion.

Phenotypic criteria: (1) Non-volitional weight loss: A weight loss of >5% within six months or >10% beyond six months. (2) Low body mass index (BMI): Defined according to Asian-specific thresholds—BMI <18.5 kg/m² for individuals under 70 years of age and BMI <20 kg/m² for those aged 70 years or older (17). (3) Reduced muscle mass: Identified as below normal values measured by body composition analysis techniques.

Etiologic criteria: (1) Reduced food intake or absorption: Food intake <50% of normal for over one-week, reduced intake for more than 2 weeks, or chronic gastrointestinal dysfunction impairing digestion or absorption. (2) Disease burden or inflammation: Acute disease- or trauma-related inflammation, or chronic disease-associated inflammation.

Comprehensive patient data were collected using the Hospital Information System and Laboratory Information Management System. The dataset included demographic, clinical, and biochemical parameters to ensure a robust analysis. The collected variables were as follows:

All statistical analyses were performed using SPSS version 27.0. Continuous variables with a normal distribution were expressed as mean ± standard deviation, and comparisons between groups were conducted using the independent t-test. Non-normally distributed data were presented as median (interquartile range) [M(Q₁, Q₃)], with group comparisons performed using the rank-sum test. Categorical variables were summarized as frequencies and percentages [n (%)], and differences between groups were analyzed using the chi-square (χ²) test. Univariate logistic regression analysis was initially conducted to identify potential risk factors with statistical significance. Variables with significant differences were then included in a multivariate logistic regression model. A nomogram was constructed using the R software package based on the significant predictors identified. The predictive performance of the model was evaluated by calculating the area under the receiver operating characteristic (ROC) curve (AUC). The discriminative ability of the model was categorized as follows: AUC values of 0.5–0.7 indicated low discrimination, >0.7–0.9 moderate discrimination, and >0.9–1.0 high discrimination. The concordance index (C-index) was also calculated as a measure of model discrimination, with values closer to 1.0 reflecting better predictive accuracy. The optimal cutoff value for the nomogram was determined using the Youden index, which maximizes the sum of sensitivity and specificity. Model calibration was assessed using the Hosmer–Lemeshow goodness-of-fit test, where a non-significant result (p > 0.05) indicates good agreement between predicted and observed outcomes. For interpretation of the nomogram, each predictor variable corresponds to a point score on the top scale. The individual points assigned to each variable are summed to obtain a total score, which is then projected onto the probability scale to estimate the individualized risk of malnutrition. Calibration curves were plotted to assess the concordance between predicted probabilities and observed outcomes. Clinical utility and applicability of the model were further evaluated using decision curve analysis (DCA). A p-value <0.05 was considered statistically significant.

The baseline characteristics of patients in the malnourished and well-nourished groups are summarized in Table 1. Significant differences were observed between the two groups in several variables. A higher proportion of patients aged 65 years or older was found in the malnourished group (67.9%) compared to the well-nourished group (49.4%) (χ² = 5.707, p = 0.017). In terms of tumor stage, a significantly higher percentage of patients in the malnourished group were at stage IV (73.2%) compared to the well-nourished group (51.2%) (χ² = 8.162, p = 0.004). Additionally, prolonged bed rest, KPS scores, BMI, and NRS-2002 scores all showed significant differences between the two groups. The malnourished group had a higher incidence of prolonged bed rest, lower KPS scores, and lower BMI (all p < 0.05). The malnourished group also had significantly higher NRS-2002 scores, indicating greater nutritional risk. No significant differences were found in sex, marital status, drinking history, or recent gastrointestinal surgery.

There were no significant differences in Cr and AST levels between the malnourished and well-nourished groups (p = 0.568 and p = 0.425, respectively). However, the malnourished group exhibited significantly lower ALT levels compared to the well-nourished group (p = 0.041). BUN levels were significantly higher in the malnourished group (p < 0.001). Regarding blood cell counts, the malnourished group had a similar PLT count compared to the well-nourished group (p = 0.112). However, the malnourished group showed significantly lower RBC count and HGB levels compared to the well-nourished group (p < 0.001 for both). Furthermore, ALB levels were significantly lower in the malnourished group (p < 0.001) (Table 2).

The univariate logistic regression analysis revealed several factors significantly associated with malnutrition. TNM stage was found to be a significant predictor, with an OR of 2.645 (p < 0.001). Bed rest also showed a strong association, with an OR of 17.028 (p < 0.001). BUN levels were positively correlated with malnutrition, exhibiting an OR of 6.83 (p < 0.001). Additionally, ALT levels, age, and KPS score were significantly related to malnutrition, with OR values of 2.392 (p = 0.013), 2.297 (p = 0.001), and 2.22 (p = 0.001), respectively. WBC levels were also significantly associated with malnutrition (OR = 2.10, p = 0.008) (Table 3).

Multivariate logistic regression analysis was conducted using the statistically significant variables identified in the univariate analysis as independent predictors. The results indicated that advanced tumor stage (TNM stage IV) was a significant risk factor for malnutrition, with patients in this stage having a higher likelihood of malnutrition compared to those in stages I-III (p = 0.025). Additionally, advanced age (≥65 years) was found to significantly contribute to malnutrition risk (p = 0.003). Other identified risk factors included a lower KPS score, prolonged bed rest, lower hemoglobin levels, and decreased serum PAB levels. All these factors were found to be independently associated with an increased risk of malnutrition (p < 0.05) (Table 4).

A nomogram for predicting malnutrition in CRC patients was constructed based on the six significant factors identified through multivariate logistic regression analysis. These factors included TNM stage, age, KPS score, bed rest, hemoglobin levels, and serum PAB levels. For each variable, a corresponding score was assigned, and the cumulative score was then used to estimate the probability of malnutrition occurrence in individual patients. The total score, which was determined by summing the individual variable scores, directly corresponds to the likelihood of malnutrition, as reflected on the probability scale at the bottom of the nomogram. Higher total scores indicate a greater risk of malnutrition, thus allowing for more accurate risk stratification and early intervention in clinical practice to improve patient care and outcomes (Figure 1).

The discriminatory ability of the nomogram prediction model for malnutrition risk in patients with CRC was evaluated using the total risk score as the predictor variable and malnutrition occurrence as the outcome variable (Table 5). The model yielded an AUC of 0.819 (95% CI: 0.731–0.895), which indicates moderate discriminative ability according to conventional criteria. At the cutoff point defined by the maximum Youden index, the nomogram achieved a sensitivity of 71.29% and a specificity of 86.58%. The corresponding positive predictive value was 64.9%, and the negative predictive value was 89.6%, as illustrated in Figure 2.

The internal validation of the nomogram prediction model was conducted using the Bootstrap resampling method with 1,000 iterations to reduce potential overfitting and assess model stability. The calibrated C-index was 0.798 (95% CI: 0.715–0.856), suggesting that the model possesses acceptable discriminative capacity in distinguishing between malnourished and well-nourished patients. Model calibration was further evaluated using the Hosmer–Lemeshow goodness-of-fit test, which yielded a χ² value of 2.387 and a p-value of 0.929, indicating no significant deviation between predicted and observed outcomes. In addition, the calibration curve (Figure 3) showed that the bias-corrected predictions were closely aligned with the ideal reference line, while the apparent curve also demonstrated similar trends, confirming that the predicted probabilities were consistent with actual observations. Together, these findings support the reliability and robustness of the model in estimating individual malnutrition risk in CRC patients.

The DCA was performed to assess the clinical effectiveness of the nomogram model for predicting malnutrition in CRC patients. The straight line in the DCA represents the scenario where all patients are assumed to have malnutrition and receive intervention, resulting in a net benefit with a negative slope. The horizontal line reflects the situation where no patients are predicted to have malnutrition and no intervention is provided, leading to a net benefit of zero. The nomogram prediction model demonstrated a net benefit higher than both extreme scenarios across a wide range of threshold probabilities, suggesting potential clinical usefulness (Figure 4).