Abstract
Objectives
In colorectal cancer (CRC) patients, body composition (BC) has been recognized as a patient-specific imaging biomarker associated with prognosis. This study aims to construct a joint prediction model based on computed tomography (CT) quantitative body composition and clinical parameters using the Least Absolute Shrinkage and Selection Operator (LASSO) regression model and evaluate its prognostic value in patients undergoing curative surgery for CRC.
Materials and methods
This multicenter retrospective study analyzed 377 CRC patients who underwent surgical resection from 2017 to 2022. The cohort was split into a training set and a validation set. CT images at the L3 and umbilical levels within 3 months pre-surgery were used to quantify body composition. LASSO regression and COX proportional hazard regression identified independent prognostic factors and developed nomogram prediction models. Model performance was assessed using C-index, receiver operating characteristic (ROC) curves, calibration curves, decision curve analysis (DCA), and external validation.
Results
LASSO regression identified 7 clinical indicators strongly associated with OS and DFS, respectively. Nomogram prediction models were then created incorporating body composition indicators. ROC curves showed the area under the curve (AUC) of 0.915 and 0.882 for 5 years OS and DFS, respectively. The combined model significantly outperformed TNM staging alone, with AUCs of 0.915 vs. 0.788 for 5 years OS in the training set, and 0.910 vs. 0.735 in the validation set. Calibration and DCA curves indicated strong predictive ability and clinical effectiveness.
Conclusion
The combined prediction model based on CT-quantified body composition and clinical indicators effectively predicts CRC prognosis, outperforming the single TNM stage model.
Clinical trial number
Retrospective study, not applicable.
Supplementary Information
The online version contains supplementary material available at https://doi.org/10.1186/s12880-026-02563-1.
Keywords: Colorectal cancer, Body composition, Nomogram, Overall survival, Disease-free survival
Introduction
Colorectal cancer (CRC) represents the third most prevalent malignancy globally and constitutes the second leading cause of cancer-related deaths [1]. Despite substantial progress in radical surgical techniques, adjuvant chemotherapy regimens, and targeted therapeutic interventions [2, 3], the mortality rate among CRC patients remains elevated, particularly in cases involving distant metastasis or postoperative recurrence [4]. The clinical management of CRC hinges on the extent of tumor progression and the overall health status of the patient. The tumor-node-metastasis (TNM) classification and staging system is employed to assess the clinical prognosis of CRC by evaluating factors such as tumor depth, the extent of lymph node involvement, and the presence of distant metastases. However, the actual prognosis of patients is not exclusively determined by the TNM stage. Quantitative body composition assessment via CT imaging, which reflects the patient’s body composition and functional status, has also been shown to predict long-term survival [5–8]. Consequently, there is an imperative clinical need to develop a precise prognostic prediction tool that integrates body composition parameters with clinical indicators.
In previous research endeavors, body weight, body mass index (BMI), and body surface area have been frequently utilized as conventional indicators of body composition. Nonetheless, these measures exhibit limitations when employed as prognostic indicators [9, 10]. Quantitative assessment of body composition through CT scan offers a more precise representation of skeletal muscle and adipose tissue compartments. Several recent investigations have reported a notable correlation between decreased skeletal muscle radiodensity (SMD) and adverse prognosis in patients with CRC [5, 7, 8, 11]. The primary cause of the reduction in skeletal muscle density is attributed to myosteatosis, a condition characterized by the accumulation of both intermuscular and intramuscular adipose tissue [12]. Consequently, the quantification of intermuscular adipose tissue can serve as an indirect indicator of skeletal muscle density. A prior study has addressed these aforementioned points [13]. Previously, the majority of studies have employed the Cox regression model for screening and correlation analysis. Although this methodology is widespread and effective, it may not offer the same level of accuracy or comprehensiveness as the LASSO (Least Absolute Shrinkage and Selection Operator) regression model [14]. Furthermore, in previous studies that combined clinicopathological data with imaging data, both datasets were frequently incorporated into the same Cox regression model for analysis. However, this approach may introduce inaccuracies into the results.
Therefore, the aim of this study was to develop and validate a composite model that incorporates CT-quantified body composition and clinical indicators to predict the prognosis of CRC patients undergoing therapeutic surgery, utilizing LASSO regression models. By amalgamating body composition parameters with clinical indicators, our objective is to augment the precision and clinical applicability of predicting patient outcomes, ultimately facilitating more informed clinical decision-making for the benefit of patients.
Methods
Patient recruitment
This retrospective analysis was conducted on patients with confirmed CRC who underwent surgical treatment at two medical centers from 2017 to 2022. The inclusion criteria were as follows: (1) age ≥ 18 years old; (2) Abdominal CT scan performed within 3 months before surgery. Exclusion criteria were as follows: (1) emergency surgery; (2) multiple colorectal metastasis or recurrence; (3) loss of follow-up or inability to obtain follow-up outcomes; (4) incomplete clinical data; (5) only chemotherapy or radiotherapy before surgery; and (6) poor quality of CT images, such as serious artifacts or unclear boundaries between muscle and adipose tissue due to extreme emaciation of patients. Ultimately, a total of 377 patients were determined to be eligible for inclusion in the analysis. This cohort consisted of 245 patients from the Second People’s Hospital of Hefei City, designated as the training set, and 132 patients from Shihezi University Affiliated Hospital of Xinjiang, serving as the external validation set. The study was approved by the ethics committees of both institutions, and the requirement for informed consent was waived. The research was conducted in strict adherence to the principles outlined in the Declaration of Helsinki, and all patient-related medical data were maintained in strict confidence.
Clinical data collection and follow-up
Patient clinical data were retrieved through querying the electronic medical record system and subsequently analyzed with respect to the following indicators: gender, age, height, weight, BMI, type of surgery, presence of stoma, complications, postoperative chemotherapy administration, intraoperative transfusion, tumor location, tumor size, TNM stage, T stage, N stage, M stage, histological grade, number of lymph nodes retrieved, number of positive lymph nodes, carcinoembryonic antigen, alpha-fetoprotein, CA 12 − 5, CA 19 − 9, and CA 72 − 4. Long-term outcome measures encompassed overall survival (OS), defined as the duration from surgery to the date of death or last follow-up, and disease-free survival (DFS), defined as the period from the date of surgery to the occurrence of tumor recurrence, metastasis, or last follow-up.
The follow-up procedures were established in accordance with the diagnostic and treatment guidelines for colorectal cancer issued by the Chinese Society of Clinical Oncology (CSCO) Collaborative Professional Committee. Patients were subjected to follow-up via telephone or hospital visits, commencing from the day after surgery. Follow-up appointments were scheduled at intervals of 3 months for the initial 3 years, followed by appointments every 6 months for the subsequent 4–5 years. Patients who failed to attend an appointment within 1 year of their last visit were deemed to have been lost to follow-up. Furthermore, based on the patient’s pathological stage, chest and abdominal-pelvic CT imaging was conducted at intervals of 6 or 12 months over a 5 years period to facilitate the detection of recurrence or metastasis. The follow-up information encompassed details regarding adjuvant therapy (such as radiotherapy or chemotherapy), tumor recurrence, duration of recurrence if applicable, patient survival status, and time of death when applicable.
Image analysis and body composition index extraction
Abdominal CT images of CRC patients, acquired within 3 months prior to surgery, were retrieved from the patients’ electronic medical records. For muscle mass and adipose tissue quantification, a single image at the median level of the third lumbar vertebra (L3) was selected (Fig. 1A). This choice was based on previous studies demonstrating that the cross-sectional areas of skeletal muscle and adipose tissue at the L3 vertebral level exhibit a strong correlation with total body tissue volume [15]. Additionally, since previous studies selected the umbilical level to measure visceral adipose tissue [16], we also selected the umbilical level CT images to investigate which level of body composition was more predictive of patient prognosis (Fig. 1B). The SliceOmatic software, version 5.0 (TomoVision), was utilized to analyze the L3 levels and the umbilical horizontal cross-sectional areas of muscle and adipose tissue. The analysis was performed using standard Hounsfield unit (HU) ranges for muscle (-29 HU to 150 HU), visceral adipose tissue (-150 HU to -50 HU), intermuscular adipose tissue (-190 HU to -30 HU), and subcutaneous adipose tissue (-190 HU to -30 HU), as previously described [17]. The adipose tissue indices collected comprised the visceral fat area (VFA), subcutaneous fat area (SFA), and intermuscular fat area (IMFA). The skeletal muscle parameters included the skeletal muscle area (SMA) and SMD. SMD was automatically calculated by the software as the mean CT value of the muscle area within the region of interest. To evaluate the reproducibility of body composition measurements, 30 patients were randomly selected for reliability assessment. All relevant body composition parameters were independently delineated and measured by two trained observers using SliceOmatic software to assess inter-rater reliability. In addition, one observer repeated the delineation and measurement procedure after a two-week interval to evaluate intra-rater reliability. Reproducibility was quantified using intraclass correlation coefficients. The inter-rater and intra-rater ICCs were 0.95 and 0.97, respectively, indicating excellent agreement and high reproducibility of the measurements.
Fig. 1.
Measurements of body composition parameters. Measurements of muscle and fat compartments were performed on a single axial computed tomography (CT) slice at the longitudinal center of the third lumbar vertebra (L3) and umbilical levels. Sagittal reformation of a preoperative CT scan at the L3 level (A), skeletal muscle area (in purple) is 115.9 cm2; intermuscular fat area (in yellow) is 21.39 cm2; visceral fat area (in orange) is 164.7 cm2; subcutaneous fat area (in blue) is 311.9 cm2. Sagittal reformation on preoperative CT scan at the umbilical level (B), skeletal muscle area (in purple) is 108 cm2; intermuscular fat area (in yellow) is 26.39 cm2; visceral fat area (in orange) is 122.1 cm2; subcutaneous fat area (in blue) is 443.8 cm2
Statistical analysis
Statistical analysis was performed utilizing SPSS (version 26.0) and R software (version 4.3.0). Count data, expressed as percentages of the total sample size (N, %), were analyzed using the chi-square test or Fisher’s exact test to evaluate differences in patient characteristics across groups. Acknowledging the gender-specific variations in body composition distribution, receiver operating characteristic (ROC) curve analysis was conducted to determine the optimal cut-off values for each body composition index, separately for men and women, that were associated with OS and DFS. Patients were subsequently stratified based on these determined cut-off values. The LASSO regression model was employed to identify the most predictive indices from the clinical dataset. Univariate and multivariate Cox proportional hazards regression analyses were conducted to identify independent risk factors associated with body composition. A predictor with a coefficient greater than 0 or a hazard ratio (HR) significantly greater than 1 was considered a risk factor for mortality, whereas predictors with coefficients less than 0 or HR values significantly less than 1 were deemed protective factors. Subsequently, nomograms were developed to predict OS and DFS at 1, 3, and 5 years, based on prognostic factors identified through LASSO regression and Cox regression analyses. The predictive performance of the model for patient prognosis was assessed using the C-index and time-dependent ROC curve analysis, and its efficacy was compared to that of the TNM staging system. Calibration plots were utilized to evaluate the concordance between predicted and actual risks. Furthermore, decision curve analysis (DCA) was conducted to assess the clinical benefit and utility of the prediction model. Kaplan-Meier survival curves were generated to compare differences in OS and DFS between patient groups. All statistical tests were two-sided, and a P-value < 0.05 was considered statistically significant.
Results
Patients characteristics
The general clinical characteristics of the patients in both the training and validation sets exhibited no significant differences between the sets (P > 0.05) (Table 1). The cutoffs for each body composition parameter are shown in Table S1. The survival rates for patients in the training and validation sets were 69.0% and 69.7%, respectively. Within the training set, a higher prevalence was observed among males (81.1% vs. 43.9%, P < 0.001) and patients younger than 65 years (56.7% vs. 20.6%, P < 0.001) in the high SMD group. Conversely, patients who received intraoperative blood transfusions (25.8% vs. 14.4%, P = 0.037) and those with moderate histological grades (83.2% vs. 75.6%, P < 0.001) were more prevalent in the low SMD group. Regarding body composition, patients with high SMD had significantly lower IMFA (75.6% vs. 58.7%, P = 0.008) but higher SMA (77.1% vs. 56.1%, P = 0.020) compared to those with low SMD. In terms of long-term prognosis, higher SMD was associated with better OS (78.9% vs. 63.2%, P = 0.011) (Table 2).
Table 1.
Comparison of the baseline clinicopathological characteristics between the training and validation sets
| Characteristic | Training set (n = 245) | Validation set (n = 132) | P-value | |
|---|---|---|---|---|
| Sex | 0.996 | |||
| Male | 141 (57.6%) | 76 (57.6%) | ||
| Female | 104 (42.4%) | 56 (42.4%) | ||
| Age (years) | 0.915 | |||
| < 65 | 83 (33.9%) | 44 (33.3%) | ||
| ≥ 65 | 162 (66.1%) | 88 (66.7%) | ||
| BMI (kg/m2) | 0.760 | |||
| < 18.5 | 25 (10.2%) | 15 (11.4%) | ||
| 18.5–25 | 167 (68.2%) | 85 (64.4%) | ||
| ≥ 25 | 53 (21.6%) | 32 (24.2%) | ||
| Surgery | 0.729 | |||
| Open | 25 (10.2%) | 12 (9.1%) | ||
| Laparoscope | 220 (89.8%) | 120 (90.9%) | ||
| Stoma | 0.648 | |||
| No | 158 (64.5%) | 82 (62.1%) | ||
| Yes | 87 (35.5%) | 50 (37.9%) | ||
| Complications | 0.952 | |||
| No | 157 (64.1%) | 85 (64.4%) | ||
| Yes | 88 (35.9%) | 47 (35.6%) | ||
| Chemotherapy | 0.822 | |||
| No | 101 (41.2%) | 56 (42.4%) | ||
| Yes | 144 (58.8%) | 76 (57.6%) | ||
| Transfusion | 0.563 | |||
| No | 192 (78.4%) | 100 (75.8%) | ||
| Yes | 53 (21.6%) | 32 (24.2%) | ||
| Tumor location | 0.892 | |||
| Ascending colon | 67 (27.3%) | 37 (28.0%) | ||
| Descending colon | 17 (6.9%) | 9 (6.8%) | ||
| Sigmoid colon | 52 (21.2%) | 32 (24.2%) | ||
| Rectal | 109 (44.5%) | 54 (40.9%) | ||
| Tumor size (cm) | 0.758 | |||
| ≤ 2.5 | 20 (8.2%) | 12 (9.1%) | ||
| >2.5 | 225 (91.8%) | 120 (90.9%) | ||
| Stage | 0.962 | |||
| I | 38 (15.5%) | 22 (16.7%) | ||
| II | 104 (42.4%) | 53 (40.2%) | ||
| III | 78 (31.8%) | 42 (31.8%) | ||
| IV | 25 (10.2%) | 15 (11.4%) | ||
| T Stage | 0.955 | |||
| T1 | 4 (1.6%) | 3 (2.3%) | ||
| T2 | 39 (15.9%) | 22 (16.7%) | ||
| T3 | 104 (42.4%) | 57 (43.2%) | ||
| T4 | 98 (40.0%) | 50 (37.9%) | ||
| N Stage | 0.900 | |||
| N0 | 148 (60.4%) | 79 (59.8%) | ||
| N1 | 87 (35.5%) | 47 (35.6%) | ||
| N2 | 10 (4.1%) | 6 (4.5%) | ||
| M Stage | 0.727 | |||
| M0 | 220 (89.8%) | 117 (88.6%) | ||
| M1 | 25 (10.2%) | 15 (11.4%) | ||
| Histological grade | 0.875 | |||
| Well | 27 (11.0%) | 13 (9.8%) | ||
| Moderately | 197 (80.4%) | 106 (80.3%) | ||
| Poorly | 21 (8.6%) | 13 (9.8%) | ||
| VFA | 0.928 | |||
| Low | 194 (79.2%) | 104 (78.8%) | ||
| High | 51 (20.8%) | 28 (21.2%) | ||
| IMFA | 0.844 | |||
| Low | 159 (64.9%) | 87 (65.9%) | ||
| High | 86 (35.1%) | 45 (34.1%) | ||
| SMA | 0.443 | |||
| Low | 94 (38.4%) | 56 (42.4%) | ||
| High | 151 (61.6%) | 76 (57.6%) | ||
| SFA | 0.916 | |||
| Low | 148 (60.4%) | 79 (59.8%) | ||
| High | 97 (39.6%) | 53 (40.2%) | ||
| OS | 0.886 | |||
| Alive | 169 (69.0%) | 92 (69.7%) | ||
| Death | 76 (31.0%) | 40 (30.3%) | ||
| DFS | 0.473 | |||
| Alive | 185 (77.5%) | 104 (78.8%) | ||
| Death | 60 (24.5%) | 28 (21.2%) |
Abbreviations: VFA, visceral fat area; IMFA, intermuscular fat area; SMA, skeletal muscle area; SFA, subcutaneous fat area; SMD, skeletal muscle density; OS, overall survival; DFS, disease-free survival
Table 2.
Patient’s baseline characteristics by SMD
| Characteristic | Low SMD (n = 155) | High SMD (n = 90) | P-value | |
|---|---|---|---|---|
| Sex | < 0.001 | |||
| Male | 68 (43.9%) | 73 (81.1%) | ||
| Female | 87 (56.1%) | 17 (18.9%) | ||
| Age (years) | < 0.001 | |||
| < 65 | 32 (20.6%) | 51 (56.7%) | ||
| ≥ 65 | 123 (79.4%) | 39 (43.3%) | ||
| BMI (kg/m2) | 0.860 | |||
| < 18.5 | 15 (9.7%) | 10 (11.1%) | ||
| 18.5–25 | 105 (67.7%) | 62 (68.9%) | ||
| ≥ 25 | 35 (22.6%) | 18 (20.0%) | ||
| Surgery | 0.067 | |||
| Open | 20 (12.9%) | 5 (5.6%) | ||
| Laparoscope | 135 (87.1%) | 85 (94.4%) | ||
| Stoma | 0.054 | |||
| No | 93 (60.0%) | 65 (72.2%) | ||
| Yes | 62 (40.0%) | 25 (27.8%) | ||
| Complications | 0.358 | |||
| No | 96 (61.9%) | 61 (67.8%) | ||
| Yes | 59 (38.1%) | 29 (32.2%) | ||
| Chemotherapy | 0.170 | |||
| No | 69 (44.5%) | 32 (35.6%) | ||
| Yes | 86 (55.5%) | 58 (64.4%) | ||
| Transfusion | 0.037 | |||
| No | 115 (74.2%) | 77 (85.6%) | ||
| Yes | 40 (25.8%) | 13 (14.4%) | ||
| Tumor location | 0.734 | |||
| Ascending colon | 44 (28.4%) | 23 (25.6%) | ||
| Descending colon | 12 (7.7%) | 5 (5.6%) | ||
| Sigmoid colon | 34 (21.9%) | 18 (20.0%) | ||
| Rectal | 65 (41.9%) | 44 (48.9%) | ||
| Tumor size (cm) | 0.077 | |||
| ≤ 2.5 | 9 (5.8%) | 11 (12.2%) | ||
| >2.5 | 146 (94.2%) | 79 (87.8%) | ||
| Stage | 0.588 | |||
| I | 23 (14.8%) | 15 (16.7%) | ||
| II | 62 (40.0%) | 42 (46.7%) | ||
| III | 52 (33.5%) | 26 (28.9%) | ||
| IV | 18 (11.6%) | 7 (7.8%) | ||
| T Stage | 0.350 | |||
| T1 | 2 (1.3%) | 2 (2.2%) | ||
| T2 | 25 (16.1%) | 14 (15.6%) | ||
| T3 | 60 (38.7%) | 44 (48.9%) | ||
| T4 | 68 (43.9%) | 30 (33.3%) | ||
| N Stage | 0.199 | |||
| N0 | 87 (56.7%) | 61 (67.8%) | ||
| N1 | 61 (39.4%) | 26 (28.9%) | ||
| N2 | 7 (4.5%) | 3 (3.3%) | ||
| M Stage | 0.339 | |||
| M0 | 137 (88.4%) | 83 (92.2%) | ||
| M1 | 18 (11.6%) | 7 (7.8%) | ||
| Histological grade | < 0.001 | |||
| Well | 14 (9.0%) | 13 (14.4%) | ||
| Moderately | 129 (83.2%) | 68 (75.6%) | ||
| Poorly | 12 (7.7%) | 9 (10.0%) | ||
| VFA | 0.164 | |||
| Low | 127 (81.9%) | 67 (74.4%) | ||
| High | 28 (18.1%) | 23 (25.6%) | ||
| IMFA | 0.008 | |||
| Low | 91 (58.7%) | 68 (75.6%) | ||
| High | 64 (41.3%) | 22 (24.4%) | ||
| SMA | 0.020 | |||
| Low | 68 (43.9%) | 26 (28.9%) | ||
| High | 87 (56.1%) | 64 (71.1%) | ||
| SFA | 0.072 | |||
| Low | 87 (56.1%) | 61 (67.8%) | ||
| High | 68 (43.9%) | 29 (32.2%) | ||
| OS | 0.011 | |||
| Alive | 98 (63.2%) | 71 (78.9%) | ||
| Death | 57 (36.8%) | 19 (21.1%) | ||
| DFS | 0.213 | |||
| Alive | 113 (72.9%) | 72 (80.0%) | ||
| Death | 42 (27.1%) | 18 (20.0%) |
Abbreviations: VFA, visceral fat area; IMFA, intermuscular fat area; SMA, skeletal muscle area; SFA, subcutaneous fat area; SMD, skeletal muscle density; OS, overall survival; DFS, disease-free survival. Bold was used to highlight values that were statistically significant (<0.05)
LASSO regression model to screen the clinical indicators
In the training set, the LASSO regression algorithm was utilized to conduct feature selection from the 24 included features. By minimizing the partial likelihood binomial deviation, seven variables were identified as the most significant prognostic factors for OS and DFS, respectively (Fig. 2). These seven predictors were subsequently analyzed using multivariate Cox regression for OS (chemotherapy, surgery, stoma, CA 72 − 4, TNM stage, M stage, and lymph node yield) and DFS (surgery, CA 72 − 4, CA 12 − 5, tumor site, TNM stage, N stage, and lymph node yield). The results of the multivariate Cox regression analysis for OS indicated that chemotherapy (HR = 0.540, 95% CI: 0.331–0.882, P = 0.014), surgery (HR = 2.113, 95% CI: 1.181–3.780, P = 0.012), stoma (HR = 2.090, 95% CI: 1.268–3.444, P = 0.004), CA 72 − 4 (HR = 1.038, 95% CI: 1.023–1.054, P < 0.001), TNM stage (HR = 1.690, 95% CI: 1.120–2.550, P = 0.012), and lymph node yield (HR = 0.947, 95% CI: 0.907–0.998, P = 0.012) were significantly associated with OS in the training set. For DFS, the multivariate Cox regression analysis revealed that surgery (HR = 2.099, 95% CI: 1.108–3.976, P = 0.023), CA 72 − 4 (HR = 1.035, 95% CI: 1.012–1.058, P = 0.002), TNM stage (HR = 1.996, 95% CI: 1.275–3.125, P = 0.003), lymph node yield (HR = 0.954, 95% CI: 0.910–0.999, P = 0.045) and tumor site (HR = 0.743, 95% CI: 0.579-0.955, P = 0.021) were significantly associated with DFS. All of these variables were incorporated into the final prediction models for 1, 3, and 5 years OS and DFS predictions.
Fig. 2.
Clinical risk factors selection using the least absolute shrinkage and selection operator (LASSO) regression model in training sets. OS (A) and DFS (C) prediction models LASSO coefficient plots adjusted for parameter λ by 10-fold cross-validation. The gray line in the figure is the partial likelihood estimate corresponding to the optimal value of lambda. The optimal lambda value of 0.0750655 (OS) and 0.03754004 (DFS) were chosen. The vertical line was plotted at the optimal λ value, and seven factors of OS (B) and eight factors of DFS (D) with non-zero coefficient were selected
Factors related to long-term survival
At the L3 vertebral level, a univariate Cox regression analysis was conducted to assess the association between body composition metrics and OS in patients with CRC. The results indicated that low VFA, low SFA, and low SMD were significantly associated with poorer OS. Furthermore, a multivariate Cox regression analysis revealed that low SMD (HR = 0.423, 95% CI: 0.248–0.721, P = 0.002) was an independent prognostic factor for poor OS in CRC patients (Table 3). A multifactorial Cox regression analysis was also performed to evaluate the relationship between body composition metrics and DFS in CRC patients. The results showed that high IMFA (HR = 1.833, 95% CI: 1.098–3.059, P = 0.020) and low SMA (HR = 0.571, 95% CI: 0.329–0.990, P = 0.046) were independent prognostic factors for poor DFS (Table 4). At the umbilical level, a univariate Cox analysis was conducted to assess the association between body composition metrics and OS in CRC patients. The results indicated that low VFA and low SFA were significantly associated with poorer OS. However, a subsequent multivariate Cox analysis revealed that none of the body composition indices were independent prognostic factors for OS in CRC patients (all P > 0.05) (Tables 3 and 4). The results of univariate multivariate Cox regression analyses of the validation set are shown in Tables S2, S3.
Table 3.
Univariate and multivariate analysis of factors associated with overall survival in the training set
| Variables | Univariate analysis | Multivariate analysis | ||||
|---|---|---|---|---|---|---|
| HR (95%CI) | P Value | HR (95%CI) | P Value | |||
| L3 Level | ||||||
| VFA | ||||||
| Low | 1 | 1 | ||||
| High | 0.478 (0.238–0.959) | 0.038 | 0.762 (0.336–1.730) | 0.517 | ||
| IMFA | ||||||
| Low | 1 | |||||
| High | 1.139 (0.711–1.825) | 0.588 | ||||
| SMA | ||||||
| Low | 1 | |||||
| High | 0.669 (0.425–1.053) | 0.082 | ||||
| SFA | ||||||
| Low | 1 | 1 | ||||
| High | 0.533 (0.324–0.875) | 0.013 | 0.792 (0.419-1.500) | 0.475 | ||
| SMD | ||||||
| Low | 1 | 1 | ||||
| High | 0.482 (0.286–0.813) | 0.006 | 0.423 (0.248–0.721) | 0.002 | ||
| Umbilical Level | ||||||
| VFA | ||||||
| Low | 1 | 1 | ||||
| High | 0.602 (0.376–0.963) | 0.034 | 0.900 (0.510–1.590) | 0.718 | ||
| IMFA | ||||||
| Low | 1 | |||||
| High | 0.646 (0.403–1.033) | 0.068 | ||||
| SMA | ||||||
| Low | 1 | |||||
| High | 0.920 (0.586–1.444) | 0.718 | ||||
| SFA | ||||||
| Low | 1 | 1 | ||||
| High | 0.495 (0.297–0.825) | 0.007 | 0.544 (0.281–1.052) | 0.070 | ||
| SMD | ||||||
| Low | 1 | |||||
| High | 0.646 (0.412–1.012) | 0.057 | ||||
Abbreviations: VFA, visceral fat area; IMFA, intermuscular fat area; SMA, skeletal muscle area; SFA, subcutaneous fat area; SMD, skeletal muscle density. Bold was used to highlight values that were statistically significant (P<0.05)
Table 4.
Univariate and multivariate analysis of factors associated with disease free survival in the training set
| Variables | Univariate analysis | Multivariate analysis | ||||
|---|---|---|---|---|---|---|
| HR (95%CI) | P Value | HR (95%CI) | P Value | |||
| L3 Level | ||||||
| VFA | ||||||
| Low | 1 | |||||
| High | 1.059 (0.638–1.759) | 0.824 | ||||
| IMFA | ||||||
| Low | 1 | 1 | ||||
| High | 1.960 (1.180–3.255) | 0.009 | 1.833 (1.098–3.059) | 0.020 | ||
| SMA | ||||||
| Low | 1 | 1 | ||||
| High | 0.531 (0.307–0.916) | 0.023 | 0.571 (0.329–0.990) | 0.046 | ||
| SFA | ||||||
| Low | 1 | |||||
| High | 0.970 (0.574–1.639) | 0.908 | ||||
| SMD | ||||||
| Low | 1 | |||||
| High | 0.801 (0.481–1.335) | 0.394 | ||||
| Umbilical Level | ||||||
| VFA | ||||||
| Low | 1 | |||||
| High | 0.904 (0.545–1.501) | 0.698 | ||||
| IMFA | ||||||
| Low | 1 | |||||
| High | 0.908 (0.505–1.630) | 0.764 | ||||
| SMA | ||||||
| Low | 1 | |||||
| High | 0.987 (0.571–1.705) | 0.962 | ||||
| SFA | ||||||
| Low | 1 | |||||
| High | 1.405 (0.837–2.357) | 0.198 | ||||
| SMD | ||||||
| Low | 1 | |||||
| High | 0.812 (0.487–1.353) | 0.424 | ||||
Abbreviations: VFA, visceral fat area; IMFA, intermuscular fat area; SMA, skeletal muscle area; SFA, subcutaneous fat area; SMD, skeletal muscle density. Bold was used to highlight values that were statistically significant (P<0.05)
Kaplan-Meier survival curves were generated to visualize the association between various factors and OS as well as DFS in the study population. As depicted in Fig. 3A, a higher SMD was found to be significantly associated with better OS (P = 0.005). Additionally, the absence of a stoma (Fig. 3B), no postoperative chemotherapy (Fig. 3C), the use of laparoscopy as the surgical approach (Fig. 3D), and earlier TNM staging (Fig. 3E) were all significantly associated with improved OS (all P < 0.001). Furthermore, as shown in Fig. 4, low IMFA (P = 0.008) (Fig. 4A), high SMA (P = 0.02) (Fig. 4B), laparoscopy as the surgical approach (P < 0.001) (Fig. 4C), and earlier TNM staging (P < 0.001) (Fig. 4D) were significantly associated with better DFS. To validate the robustness of these findings, the analyses were repeated in a validation set. The results observed in the validation set were consistent with those observed in the training set, indicating the reliability and generalizability of the study findings.
Fig. 3.
Kaplan–Meier analyses of OS based on clinical risk factors and body composition which were used to construct nomogram model of OS in training sets (A, B, C, D, E,) and validation sets (F, G, H, I, J). P Values are shown. The log-rank P values were generated via Kaplan–Meier analysis with the log-rank test. (A, F):SMD, (B, G):Stoma, (C, H):Chemotherapy, (D, I):Surgery, (E, J):TNM stage
Fig. 4.
Kaplan–Meier analyses of DFS based on clinical risk factors and body composition which were used to construct nomogram model of DFS in training sets (A, B, C, D) and validation sets (E, F, G, H). P Values are shown. The log-rank P values were generated via Kaplan–Meier analysis with the log-rank test. (A, E):IMFA, (B, F): SMA, (C, G):Surgery, (D, H):TNM stage
Construction and validation of nomograms
The independent prognostic factors identified through the multifactorial Cox regression analysis of body composition and clinical indicators in the training group were utilized to develop a nomogram prediction model for estimating the 1, 3, and 5 years survival rates of patients (Fig. 5). For the OS -based nomogram model, seven indicators were included: postoperative chemotherapy, type of surgery, presence of a stoma, CA 72 − 4 levels, TNM stage, lymph node yield, and SMD. Similarly, for the DFS -based nomogram model, seven indicators were incorporated: type of surgery, CA 72 − 4 levels, tumor site, TNM stage, lymph node yield, IMFA, and SMA. These nomogram models provide a visual representation of the relative importance of each prognostic factor and facilitate personalized risk assessment and prognosis prediction for patients.
Fig. 5.
Prognostic nomogram constructed combining clinical risk factors and body composition predicting 1-, 3-, and 5-year OS (A) and DFS (B) in training sets
The predictive performance of the developed models for the prognosis of CRC patients was assessed using the C-index and ROC curves. In the training set, the C-indexes for the OS and DFS integrated models were 0.792 (0.736–0.847) and 0.776 (0.718–0.834), respectively. For the validation set, the C-indexes for the OS and DFS integrated models were 0.859 (0.794–0.924) and 0.830 (0.743–0.917), respectively, indicating good discriminative ability. In addition, to quantify the incremental predictive value of body composition, we also compared the C-index between the clinical model with only clinical measures and the body composition model with only body composition measures. In the training set, the C-indexes for the OS and DFS clinical models were 0.785 (0.730–0.840) and 0.772 (0.713–0.832), respectively. For the validation set, the C-indexes for the OS and DFS clinical models were 0.854 (0.793–0.915) and 0.834 (0.756–0.913), respectively. In the training set, the C-indexes for the OS and DFS body composition models were 0.574 (0.520–0.627) and 0.613 (0.547–0.679), respectively. For the validation set, the C-indexes for the OS and DFS body composition models were 0.573 (0.499–0.647) and 0.611 (0.507–0.715), respectively. ROC curves were generated to evaluate the accuracy of the models in predicting survival at different time points. The area under the curves (AUCs) for the 1 year OS prediction were 0.816 and 0.880 in the training and validation sets, respectively. The AUCs for the 3 years OS prediction were 0.847 and 0.909, and for the 5 years OS prediction, they were 0.915 and 0.910 (Fig. 6A, B). Similarly, for DFS prediction, the AUCs for the 1, 3, and 5 years DFS were 0.824, 0.788, and 0.882 in the training set, and 0.868, 0.850, and 0.853 in the validation set, respectively (Fig. 6C, D). Furthermore, DCA curve was performed to compare the clinical utility of the nomogram model with the strategies of treating all patients or not treating any patients. The DCA curves revealed that the nomogram model provided superior net benefit in predicting patient prognosis compared to these alternative strategies (Fig. 6E-H). Finally, nomogram calibration curves were generated to assess the agreement between the model-predicted risk and the observed prognostic status. As shown in Fig. 7, the calibration curves demonstrated good agreement between the predicted and observed outcomes, indicating the reliability of the nomogram models.
Fig. 6.
Time dependent ROC curves of the nomogram at 1, 3, 5-year period of OS in training sets (A), validation sets (B) and DFS in training sets (C), validation sets (D). Decision curves of the nomogram predicting 1, 3, 5-year OS in training sets (E), validation sets (F) and DFS in training sets (G), validation sets (H). The x-axis is the threshold probability, and the y-axis is the net benefit. When the threshold probability is between 10% and 50%, the net benefit of the model exceeds all deaths or no deaths
Fig. 7.
Calibration curves for body composition based nomogram models for 1-, 3-, and 5-year OS in training sets (A, B, C) and validation sets (D, E, F). Calibration curves for 1-, 3-, and 5-year DFS in training sets (G, H, I) and validation sets (J, K, L).The diagonal dashed reference line indicated a perfect performance of the body composition based nomogram by an ideal model. Solid lines indicated the performance of the body composition based nomogram, and the diagonal dashed reference line closer alignment with solid line indicated a better performance
In addition, we conducted a comparative analysis to evaluate the predictive efficacy of the combined model against single TNM staging. The results of this analysis demonstrated that the combined model exhibited superior predictive ability compared to single TNM staging. Specifically, in the training set, the AUCs for the combined model versus single TNM staging were as follows: 0.816 versus 0.647, 0.847 versus 0.666, and 0.915 versus 0.788 for 1, 3, and 5 years OS, respectively; and 0.824 versus 0.748, 0.788 versus 0.709, and 0.882 versus 0.792 for 1, 3, and 5 years DFS, respectively (Fig. 8). Similarly, in the validation set, the AUCs for the combined model versus single TNM staging were: 0.880 versus 0.629, 0.909 versus 0.644, and 0.910 versus 0.735 for 1, 3, and 5 years OS, respectively; and 0.868 versus 0.838, 0.850 versus 0.718, and 0.853 versus 0.741 for 1, 3, and 5 years DFS, respectively (Fig. 8). These results further confirm the superior predictive performance of the combined model over single TNM staging in predicting the prognosis of CRC patients.
Fig. 8.
The Receiver operating characteristics curve (ROC) comparison of nomogram and AJCC TNM stage. 1-, 3-, and 5-year ROC of OS nomogram in training sets (A, B, C) and validation sets (D, E, F). 1-, 3-, and 5-year ROC of DFS nomogram in training sets (G, H, I)) and validation sets (J, K, L)
Discussion
In the present study, we developed a nomogram that integrates CT-quantified body composition and clinical indicators to predict the prognosis of patients who have undergone CRC surgery. When compared to the single TNM staging system, our combined model exhibited improved predictive accuracy and clinical utility. This nomogram can serve as a valuable tool for assessing the prognosis of CRC patients and for guiding the development of postoperative treatment plans and lifestyle intervention programs.
Skeletal muscle mass reduction is indicative of a compromised nutritional and metabolic status in the body, primarily due to the catabolic state that arises from the metabolic demands and systemic pro-inflammatory state associated with cancer [18]. Sarcopenia can be further exacerbated by inadequate dietary intake, particularly in cases of gastrointestinal tumors that impair eating or as a consequence of adverse treatment reactions [19, 20]. Several studies have reported that decreased skeletal muscle mass and density are independently linked to a poor prognosis in CRC patients [6, 21–25]. However, there are also reports suggesting that skeletal muscle mass and density do not significantly impact the prognosis of CRC patients [26]. In our study, we found that low SMD and low SMA were independent risk factors for OS and DFS in CRC patients, which is consistent with the findings of Chai VW et al. [27]. The inconsistency in results among studies may be attributed to variations in ethnicity, the selection of body composition characteristics and thresholds, and the statistical methods used. It is worth noting that Table 2 shows that IMFA and SMA are statistically different between the low SMD group and the high SMD group, which is due to the correlation between IMFA, SMA and SMD. IMFA is the adipose tissue content with low CT value, SMA is the muscle tissue content with high CT value, and SMD is the average CT value of the whole SMA. A high SMD represents less potential fatty infiltration in the SMA and therefore corresponds to more muscle tissue (high SMA) and less adipose tissue (low IMFA), which is consistent with the results in Table 2.
High visceral fat has been well-documented to exert significant pro-inflammatory effects and has been identified as a predictor of poor prognosis in various cancers, including colorectal, esophageal, gastric, renal, bladder, and breast cancers [28–30]. In contrast, our study revealed that high VFA at the L3 and umbilical levels was associated with improved OS only in the univariate Cox regression analysis. This suggests that the survival benefit associated with high fat mass or BMI may be attributed to the reduced prevalence of sarcopenia in patients with elevated BMI, rather than a direct protective effect of increased body weight or adipose tissue. This finding is in agreement with the hypothesis proposed by Caan et al. [31], which posits that the survival advantage associated with high fat mass or BMI may be due to the decreased incidence of sarcopenia in patients with higher BMI. Furthermore, our results indicated that high SFA was associated with better OS in CRC patients, consistent with the findings of Aiya Bimurzayeva et al. [32]. Additionally, Yukina Kusunoki et al. [33] demonstrated that intramuscular adipose tissue content can serve as a marker of malnutrition in patients with gastrointestinal cancers, with potential prognostic value. This finding aligns with our observation that DFS was significantly increased in patients with high IMFA. These results lead us to hypothesize that intermuscular adipose tissue content may be related to the nutritional and metabolic status of the organism and the extent of disease progression, potentially influencing the development of sarcopenia. However, further research is needed to validate this hypothesis.
Most previous studies only used the L3 level to measure body composition indicators, because this level has the greatest correlation with total body tissue composition. Umbilical body composition was also explored in this study, and the results showed that no body composition index was significantly associated with OS and DFS. In previous studies on body composition at the umbilical level, the correlation between it and the prognosis of patients remains controversial and has not reached a unified consensus. Nie et al. [13] pointed out that there is a significant association between the IMFA at the umbilical level and the long-term survival rate of patients, while Heus et al. [16] found that different cut-off values would affect the correlation between the VFA at the umbilical level and the prognosis. This may be related to differences in body composition content at different levels. L3 level is still considered the best level for body composition analysis because of its standardization and high correlation with whole-body muscle mass.
In addition, we utilized the LASSO regression model to screen clinical indicators. This approach is advantageous as it can comprehensively and efficiently identify those indicators that are highly correlated with patient prognosis from a large dataset, compared to traditional methods [34]. While the LASSO regression model has been widely applied in the field of imaging histology [35], its use in screening clinicopathologic indicators is relatively rare. Despite the fact that we applied LASSO regression analysis to only 24 clinicopathological indicators, as opposed to the screening of thousands of indicators in imaging histology, the results obtained from the LASSO regression model were still precise and efficient. This finding is consistent with the study conducted by Zhang et al. [36], which also demonstrated the efficacy of the LASSO regression model in identifying relevant indicators.
In contrast to previous conventional studies [37], our investigation encompassed a diverse array of body composition indicators, which were selected based on clinical indicators. Considering the intricate nature of malignant tumors and the underlying biological processes that involve multiple interacting components, it is imperative to account for the interactions among these various components to comprehensively reflect their complexity [38–40]. To address this, we integrated both body composition and clinical indicators into a nomogram model. This model demonstrated superior efficacy in predicting patients’ 1, 3, and 5 years OS and DFS compared to the TNM staging system alone. Specifically, the AUC for predicting patients’ 5 years OS and DFS reached 0.915 and 0.882, respectively. Furthermore, the reliability and utility of the nomogram model were validated through calibration curve and DCA curve analyses. In the previous study of our team [13, 41], the prediction effects of traditional models were as follows: the best AUC for predicting OS was 0.848; The AUC at 1, 2, and 3 years were 0.811, 0.798, and 0.772, respectively, in the training set, and 0.761, 0.771, and 0.804, respectively, in the validation set. Obviously, the nomogram model constructed by combining LASSO and COX regression analysis in this study has better performance than the traditional model, which indicates that the combination of LASSO and COX regression analysis makes the screening of prognostic indicators more scientific and accurate, and the establishment of the nomogram model allows the prediction of prognosis to be quantitatively calculated by the scores corresponding to each prognostic indicator. Compared with the traditional prognostic model, it is more convenient and easier to generalize in multiple centers. Furthermore, previous studies on body composition mostly constructed models that only included body composition indicators, lacking models that simultaneously incorporated clinical indicators and body composition indicators. This study fills this gap, making the model application more in line with the clinical environment.
In recent years, radiomics, as an emerging medical image post-processing method, has attracted more and more attention in the research of various cancers. Radiomics technology reveals tissue pathophysiological information by extracting high-throughput features from medical images, thereby overcoming the limitations of visual evaluation images. In CRC-related research, radiomics is mainly used to extract morphological and texture features of tumor, and the effect of radiomics based tumor feature model is significantly better than that of traditional clinical indicators and related models [42]. In addition, another study from our group applied radiomics to extract morphological and texture features of body composition, and the body composition model based on this also performed well in predicting the prognosis of patients [41]. These studies demonstrated the great potential of radiomics in CRC prognosis research, which is worthy of further research and exploration.
It is important to acknowledge that the present study is not without limitations. Firstly, as a retrospective study, it is inherently subject to selection bias, which is an unavoidable limitation. To further investigate and validate our findings, additional studies in more independent groups or prospective studies are necessary. Moreover, we utilized ROC curve analysis to determine the optimal cutoff values. Although the cutoff values obtained from the training and validation sets were identical, it is crucial to note that factors such as ethnicity and geography cannot be overlooked. The cutoff values were determined based on Asian populations and may not be universally applicable. Therefore, caution should be exercised when interpreting and applying these cutoff values to other populations.
In conclusion, our study introduces a nomogram prediction model that integrates LASSO and Cox regression analysis. This model is based on CT quantification of body composition and clinical risk factors to predict the prognosis of patients with CRC. By incorporating body composition metrics into the prediction model, we observed a significant enhancement in the predictive capacity for patients’ long-term prognosis. This improvement facilitates the identification of individuals with poor prognosis, enabling clinicians to implement more appropriate and personalized management and treatment strategies.
Supplementary Information
Below is the link to the electronic supplementary material.
Author contributions
Zhihao Liu wrote the main manuscript text, Mingming Song prepared figures and tables. Yixin Heng, Jiaxin Xu, Xiaoyu Wu and Liming Shen proceed validation. Tong Nie proceed formal analysis. Yinghao Cao, Feihong Wu and Le Qin proceed supervision and methodology. All authors reviewed the manuscript.
Funding
This study was supported by the Medjaden Academy & Research Foundation for Young Scientists (No. MJR202409101); The Hubei Province Natural Science Foundation (No. 2024AFB090); the National Natural Science Foundation of China (No.82403144); The Science and Technology Project Fund of the First Affiliated Hospital of Shihezi University(No. BS2024003，2025DLYJ08).
Data availability
The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.
Declarations
Ethics approval and consent to participate
The study was approved by the Ethics Committee of the Second People’s Hospital of Hefei City (No. 2023-S127), and Shihezi University Affiliated Hospital of Xinjiang (KJ2023-351-01). The requirement for informed consent was waived. The research was conducted in strict adherence to the principles outlined in the Declaration of Helsinki, and all patient-related medical data were maintained in strict confidence.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Zhihao Liu, Mingming Song and Yixin Heng contributed equally to this work.
Contributor Information
Yinghao Cao, Email: yinghaocao@hust.edu.cn.
Feihong Wu, Email: wfh_wuhan@163.com.
Le Qin, Email: qinle_xj@163.com.
References
- 1.Sung H, Ferlay J, Siegel RL, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin. 2021;71:209–49. [DOI] [PubMed] [Google Scholar]
- 2.Chen J, Chen K, Yuan X. Advances in the therapeutic study of oncolytic virus in colorectal cancer. Oncol Translational Med. 2024;10:171–7. [Google Scholar]
- 3.Zhou J, Wang B, Su R, et al. Research status of colorectal cancer treatment based on natural killer cell immunotherapy. Oncol Translational Med. 2025;11:271–81. [Google Scholar]
- 4.Yamamoto T, Kawada K, Obama K. Inflammation-related biomarkers for the prediction of prognosis in colorectal cancer patients. Int J Mol Sci. 2021;22. [DOI] [PMC free article] [PubMed]
- 5.Kroenke CH, Prado CM, Meyerhardt JA, et al. Muscle radiodensity and mortality in patients with colorectal cancer. Cancer. 2018;124:3008–15. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Malietzis G, Currie AC, Athanasiou T, et al. Influence of body composition profile on outcomes following colorectal cancer surgery. Br J Surg. 2016;103:572–80. [DOI] [PubMed] [Google Scholar]
- 7.Margadant CC, Bruns ER, Sloothaak DA, et al. Lower muscle density is associated with major postoperative complications in older patients after surgery for colorectal cancer. Eur J Surg Oncol. 2016;42:1654–9. [DOI] [PubMed] [Google Scholar]
- 8.Xiao J, Caan BJ, Cespedes Feliciano EM, et al. Association of Low Muscle Mass and Low Muscle Radiodensity With Morbidity and Mortality for Colon Cancer Surgery. JAMA Surg. 2020;155:942–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Romero-Corral A, Somers VK, Sierra-Johnson J, et al. Accuracy of body mass index in diagnosing obesity in the adult general population. Int J Obes (Lond). 2008;32:959–66. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Ying P, Chen J, Ye Y, Ye J, Cai W. Adipose tissue is a predictor of 30-days mortality in patients with bloodstream infection caused by carbapenem-resistant Klebsiella pneumoniae. BMC Infect Dis. 2022;22:173. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Dolan RD, Almasaudi AS, Dieu LB, Horgan PG, McSorley ST, McMillan DC. The relationship between computed tomography-derived body composition, systemic inflammatory response, and survival in patients undergoing surgery for colorectal cancer. J Cachexia Sarcopenia Muscle. 2019;10:111–22. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Aubrey J, Esfandiari N, Baracos VE, et al. Measurement of skeletal muscle radiation attenuation and basis of its biological variation. Acta Physiol (Oxf). 2014;210:489–97. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Nie T, Wu F, Heng Y, et al. Influence of skeletal muscle and intermuscular fat on postoperative complications and long-term survival in rectal cancer patients. J Cachexia Sarcopenia Muscle. 2024;15:702–17. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Gui J, Li H. Penalized Cox regression analysis in the high-dimensional and low-sample size settings, with applications to microarray gene expression data. Bioinformatics. 2005;21:3001–8. [DOI] [PubMed] [Google Scholar]
- 15.Shen W, Punyanitya M, Wang Z, et al. Total body skeletal muscle and adipose tissue volumes: estimation from a single abdominal cross-sectional image. J Appl Physiol (1985). 2004;97:2333–8. [DOI] [PubMed] [Google Scholar]
- 16.Heus C, Cakir H, Lak A, Doodeman HJ, Houdijk AP. Visceral obesity, muscle mass and outcome in rectal cancer surgery after neo-adjuvant chemo-radiation. Int J Surg. 2016;29:159–64. [DOI] [PubMed] [Google Scholar]
- 17.Mourtzakis M, Prado CM, Lieffers JR, Reiman T, McCargar LJ, Baracos VE. A practical and precise approach to quantification of body composition in cancer patients using computed tomography images acquired during routine care. Appl Physiol Nutr Metab. 2008;33:997–1006. [DOI] [PubMed] [Google Scholar]
- 18.Miao C, Zhang W, Feng L, et al. Cancer-derived exosome miRNAs induce skeletal muscle wasting by Bcl-2-mediated apoptosis in colon cancer cachexia. Mol Ther Nucleic Acids. 2021;24:923–38. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Blum D, Omlin A, Baracos VE, et al. Cancer cachexia: a systematic literature review of items and domains associated with involuntary weight loss in cancer. Crit Rev Oncol Hematol. 2011;80:114–44. [DOI] [PubMed] [Google Scholar]
- 20.Bye A, Jordhoy MS, Skjegstad G, Ledsaak O, Iversen PO, Hjermstad MJ. Symptoms in advanced pancreatic cancer are of importance for energy intake. Support Care Cancer. 2013;21:219–27. [DOI] [PubMed] [Google Scholar]
- 21.Aro R, Makarainen-Uhlback E, Ammala N, et al. The impact of sarcopenia and myosteatosis on postoperative outcomes and 5-year survival in curatively operated colorectal cancer patients - A retrospective register study. Eur J Surg Oncol. 2020;46:1656–62. [DOI] [PubMed] [Google Scholar]
- 22.Best TD, Roeland EJ, Horick NK, et al. Muscle Loss Is Associated with Overall Survival in Patients with Metastatic Colorectal Cancer Independent of Tumor Mutational Status and Weight Loss. Oncologist. 2021;26:e963–70. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Hopkins JJ, Reif RL, Bigam DL, Baracos VE, Eurich DT, Sawyer MB. The Impact of Muscle and Adipose Tissue on Long-term Survival in Patients With Stage I to III Colorectal Cancer. Dis Colon Rectum. 2019;62:549–60. [DOI] [PubMed] [Google Scholar]
- 24.Runkel M, Diallo TD, Lang SA, Bamberg F, Benndorf M, Fichtner-Feigl S. The Role of Visceral Obesity, Sarcopenia and Sarcopenic Obesity on Surgical Outcomes After Liver Resections for Colorectal Metastases. World J Surg. 2021;45:2218–26. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Sueda T, Takahasi H, Nishimura J, et al. Impact of Low Muscularity and Myosteatosis on Long-term Outcome After Curative Colorectal Cancer Surgery: A Propensity Score-Matched Analysis. Dis Colon Rectum. 2018;61:364–74. [DOI] [PubMed] [Google Scholar]
- 26.Carcamo L, Penailillo E, Bellolio F, et al. Computed tomography-measured body composition parameters do not influence survival in non-metastatic colorectal cancer. ANZ J Surg. 2021;91:E298–306. [DOI] [PubMed] [Google Scholar]
- 27.Chai VW, Chia M, Cocco A, Bhamidipaty M, D’Souza B. Sarcopenia is a strong predictive factor of clinical and oncological outcomes following curative colorectal cancer resection. ANZ J Surg. 2021;91:E292–7. [DOI] [PubMed] [Google Scholar]
- 28.Lennon H, Sperrin M, Badrick E, Renehan AG. The Obesity Paradox in Cancer: a Review. Curr Oncol Rep. 2016;18:56. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Malietzis G, Aziz O, Bagnall NM, Johns N, Fearon KC, Jenkins JT. The role of body composition evaluation by computerized tomography in determining colorectal cancer treatment outcomes: a systematic review. Eur J Surg Oncol. 2015;41:186–96. [DOI] [PubMed] [Google Scholar]
- 30.Moore SC, Playdon MC, Sampson JN, et al. A Metabolomics Analysis of Body Mass Index and Postmenopausal Breast Cancer Risk. J Natl Cancer Inst. 2018;110:588–97. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Caan BJ, Meyerhardt JA, Kroenke CH, et al. Explaining the Obesity Paradox: The Association between Body Composition and Colorectal Cancer Survival (C-SCANS Study). Cancer Epidemiol Biomarkers Prev. 2017;26:1008–15. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Bimurzayeva A, Kim MJ, Ahn JS, et al. Three-dimensional body composition parameters using automatic volumetric segmentation allow accurate prediction of colorectal cancer outcomes. J Cachexia Sarcopenia Muscle. 2024;15:281–91. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Kusunoki Y, Okugawa Y, Toiyama Y, et al. Modified intramuscular adipose tissue content as a feasible surrogate marker for malnutrition in gastrointestinal cancer. Clin Nutr. 2021;40:2640–53. [DOI] [PubMed] [Google Scholar]
- 34.Atabaki-Pasdar N, Ohlsson M, Vinuela A, et al. Predicting and elucidating the etiology of fatty liver disease: A machine learning modeling and validation study in the IMI DIRECT cohorts. PLoS Med. 2020;17:e1003149. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Tibshirani R. The lasso method for variable selection in the Cox model. Stat Med. 1997;16:385–95. [DOI] [PubMed] [Google Scholar]
- 36.Zhang Z, Zhang T, Zhang R, et al. Predicting colorectal cancer risk: a novel approach using anemia and blood test markers. Front Oncol. 2024;14:1347058. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Kim JM, Chung E, Cho ES, et al. Impact of subcutaneous and visceral fat adiposity in patients with colorectal cancer. Clin Nutr. 2021;40:5631–8. [DOI] [PubMed] [Google Scholar]
- 38.Jiang Y, Li T, Liang X, et al. Association of Adjuvant Chemotherapy With Survival in Patients With Stage II or III Gastric Cancer. JAMA Surg. 2017;152:e171087. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Jiang Y, Liu W, Li T, et al. Prognostic and Predictive Value of p21-activated Kinase 6 Associated Support Vector Machine Classifier in Gastric Cancer Treated by 5-fluorouracil/Oxaliplatin Chemotherapy. EBioMedicine. 2017;22:78–88. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Jiang Y, Zhang Q, Hu Y, et al. ImmunoScore Signature: A Prognostic and Predictive Tool in Gastric Cancer. Ann Surg. 2018;267:504–13. [DOI] [PubMed] [Google Scholar]
- 41.Zhi X, Nie T, Song M, et al. The prognostic value of CT-measured body composition combined with radiomics in predicting the survival of patients with resectable colon cancer. Radiol Med. 2026;131:350–62. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Sun CH, Wang HD, Sun WH, Gong GW, Deng ZM, Jiang ZW. Construction of a prognostic survival model for colorectal cancer patients using CT image texture analysis: a prospective cohort study. Front Oncol. 2025;15:1738696. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.