Abstract
Background
Triglyceride-to-glucose index (TyG), a key diagnostic marker for insulin resistance, has been linked to colorectal cancer (CRC). Nevertheless, TyG prognostic significance in CRC survival has not been established. This study seeks to develop and validate a robust machine learning (ML) based predictive model combining TyG and inflammatory markers for predicting long-term survival outcomes in CRC patients.
Methods
A retrospective study was performed on (n=1,893) CRC patients who underwent radical surgery at The First Affiliated Hospital of Kunming Medical University. The patients were randomly assigned to training cohort (70%) and an internal validation cohort (30%). An external validation cohort (n=493) from another hospital was used to test model generalizability. Independent prognostic factors were identified via multivariate Cox regression. An integrative predictive model was constructed using various ML algorithms [random survival forest (RSF), eXtreme Gradient Boosting (XGBoost), gradient boosting machines (GBM)], evaluated by concordance index (C-index), receiver operating characteristic (ROC) curve, calibration plots, and decision curve analyses (DCAs).
Results
The final model integrated eight independent prognostic factors: age, lymphocyte-to-monocyte ratio (LMR), TyG, carcinoembryonic antigen (CEA), carbohydrate antigen 19-9 (CA199), Union for International Cancer Control (UICC) stage, tumor differentiation, and perineural invasion. The model achieved good predictive accuracy (C-index: training =0.742, internal validation =0.735, external validation =0.752). ROC curves demonstrated robust predictive accuracy for 1-, 3-, and 5-year survival [area under the curve (AUC): 0.79, 0.76, 0.74, respectively]. SHapley Additive exPlanations (SHAP) analysis ranked TyG as the second-most influential prognostic indicator.
Conclusions
Elevated TyG index independently predicts favorable long-term outcomes in CRC patients. Our validated ML model, combining TyG, inflammatory markers, and clinicopathological features, provides a reliable, economical, and practical clinical tool for prognosis assessment. Further multicenter, prospective studies are necessary to confirm the widespread applicability of this model.
Keywords: Triglyceride-to-glucose index (TyG), lymphocyte-to-monocyte ratio (LMR), colorectal cancer (CRC), prognosis, machine learning (ML)
Highlight box.
Key findings
• A machine learning (ML) based prognostic model integrating triglyceride-to-glucose index (TyG), inflammatory markers, and clinicopathological variables accurately predicted 1-, 3-, and 5-year survival in colorectal cancer (CRC).
• TyG was an independent prognostic factor and ranked as the second-most influential variable by SHapley Additive exPlanations (SHAP) analysis.
• The model showed good and consistent performance across training, internal, and external validation cohorts.
What is known and what is new?
• TyG is a surrogate marker of insulin resistance and has been associated with CRC risk, while inflammatory markers and tumor stage influence CRC prognosis.
• This study demonstrates that TyG has independent prognostic value for long-term survival in CRC and introduces a validated ML model that integrates metabolic, inflammatory, and clinical factors, outperforming conventional approaches.
What is the implication, and what should change now?
• TyG, an inexpensive and routinely available index, can be incorporated into prognostic assessment for CRC patients.
• ML-based integrative models may improve risk stratification and personalized follow-up strategies beyond traditional staging alone.
• Future practice should move toward prospective, multicenter validation and consider metabolic-inflammatory profiles in CRC prognostic evaluation.
Introduction
Colorectal cancer (CRC) remains a major cause of cancer-related morbidity and mortality worldwide (1). Its prognosis is influenced by multiple factors, including metabolic status (2), inflammatory responses (3), and genetic mutations such as microsatellite instability (MSI), and Kirsten rat sarcoma (KRAS), v-raf murine sarcoma viral oncogene homolog B1 (BRAF), and neuroblastoma RAS viral oncogene homolog (NRAS) (4). Although genetic profiling provides valuable prognostic information, its high-cost limits accessibility for many patients. Therefore, it is essential to establish cost-effective and widely applicable prognostic markers derivable from routine clinical assessments.
Multiple studies have demonstrated associations between inflammation-based biomarkers and CRC outcomes (5-7). Among these biomarkers, the neutrophil-to-lymphocyte ratio (NLR) (8) and lymphocyte-to-monocyte ratio (LMR) (9) have been linked to patient prognosis. However, the predictive power of any single inflammatory marker remains limited. Therefore, combining multiple markers may enhance the accuracy of prognostic assessment in CRC.
The triglyceride-to-glucose index (TyG), a surrogate marker of insulin resistance (10), is calculated using fasting triglycerides and fasting blood glucose levels with the formula: TyG = ln [fasting triglycerides (mg/dL) × fasting glucose (mg/dL)/2] (11). The TyG index is an established prognostic factor in several diseases, including cardiovascular (12) and endocrine disorders (13), as well as various cancers such as pancreatic (14), gastric (15), and renal malignancies (16). Furthermore, elevated TyG index levels have been associated with increased CRC incidence (17). Despite these observations, the prognostic significance of TyG index in CRC remains unclear.
This study aims to integrate the TyG index, inflammatory biomarkers, and clinicopathological features to predict long-term survival in patients with CRC. By constructing a visualized prognostic model, we seek to provide a convenient and cost-effective tool for clinical application. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2893/rc).
Methods
Patients
This study was conducted in accordance with the principles of the Declaration of Helsinki and its subsequent amendments. This study was a retrospective analysis approved by the Ethics Committee of the First Affiliated Hospital of Kunming Medical University (No. 2024-L-124) and the Ethics Committee of The Third People’s Hospital of Honghe Prefecture (No. 2024-KYXM-29), which exempted the requirement for informed consent from the patients. A retrospective review of (n=1,893) patients who underwent radical resection for CRC at the First Affiliated Hospital of Kunming Medical University from January 2014 to December 2023 was performed. The patients were divided into a training set (n=1,326) and an internal validation set (n=567) in a 70:30 ratio, with (n=493) cases from the Third People’s Hospital of Honghe Prefecture serving as an external validation set. The study workflow is shown in Figure 1.
Figure 1.
Flow chart of the study. Illustrates the selection process of patients included in the study and the final number of patients analyzed. AUC, area under the curve.
Inclusion criteria: (I) patients diagnosed with adenocarcinoma after radical resection for CRC; (II) complete preoperative clinical and laboratory examination data; (III) complete follow-up information on postoperative outcomes.
Exclusion criteria: (I) patients with preoperative fever or hematologic disorders; (II) patients who underwent preoperative chemotherapy or radiotherapy.
Data collected
Basic clinical information [age, gender and body mass index (BMI)], laboratory test indicators [absolute neutrophil count (ANC), absolute lymphocyte count (ALC), absolute monocyte count (AMC), albumin (Alb), fasting blood glucose, fasting triglycerides, carcinoembryonic antigen (CEA), and carbohydrate antigen 19-9 (CA19-9)], and pathological variables [tumor size, tumor differentiation, pathological tumor (pT) stage, pathological node (pN) stage, metastasis (M) stage, Union for International Cancer Control (UICC) stage, perineural invasion, and vascular invasion], diabetes were retrieved from the electronic medical record system. The following indices were calculated: TyG-BMI = TyG × BMI, prognostic nutritional index (PNI) = Alb (g/L) + 5 × lymphocyte count (109/L), NLR = ANC/ALC, and LMR = ALC/AMC.
Follow-up
Patients from The First Affiliated Hospital of Kunming Medical University were followed up until September 2024 or until their death; patients from The Third People’s Hospital of Honghe Prefecture were followed up until March 2025 or until their death.
Machine learning (ML) analysis and prognostic model construction
We implemented an integrative ML framework comprising nine complementary algorithms and 89 combinatorial modeling strategies to develop a robust reproducible predictive model. The multimodal architecture consisted of regularized regression models [least absolute shrinkage and selection operator (LASSO), Ridge, and ElasticNet (Enet) with α = 0.1–0.9], supervised survival models (Stepwise Cox regression with forward, backward, and both selection), ensemble algorithms [random survival forest (RSF), gradient boosting machine (GBM), CoxBoost, and extreme gradient boosting (XGBoost)], and their hybrid combinations. Model development followed four sequential stages:
Predictor selection: univariate logistic regression analysis was used to identify clinically relevant variables in the training cohort.
Ensemble modeling: eighty-nine algorithmic combinations (e.g., CoxBoost + Enet, RSF + LASSO, and StepCox + GBM hybrids) were evaluated using leave-one-out cross-validation (LOOCV).
Multicenter validation: performance was assessed across two independent external validation cohorts.
Optimal model selection: the final signature was chosen according to the highest mean area under the curve (AUC) through receiver operating characteristic (ROC) curve analysis across all cohorts.
To support clinical translation, we developed an interactive web-based calculator using the R Shiny framework to facilitate clinical implementation after selecting the optimal prognostic model based on performance metrics [e.g., concordance Index (C-index), time-dependent AUC, and calibration plots]. The application was deployed on the shinyapps.io platform (https://www.shinyapps.io/) to ensure global accessibility without requiring local software installation. The tool is freely available and compatible with all modern web browsers, enabling clinicians and researchers to obtain real-time prognostic estimates.
Statistical analysis
Statistical analysis and visualization were performed using SPSS 22.0 and R (V4.1.3). Survival curves were generated in GraphPad Prism 8.0 software. Continuous variables (NLR, LMR, tumor size, TyG, and TyG-BMI) were evaluated using X-tile version 3.6.1 to determine optimal cutoff points, yielding thresholds of 3.3, 3.8, 2.1, 5.8, and 144.7, respectively (Figure S1). Based on the cohorts median Alb (41.9 g/L), median PNI (50.5) and median age (63 years), and prior literature (18), the age was dichotomized at 60 years old (Table S1). Overall survival (OS) was estimated using the Kaplan-Meier (KM) method, and compared using the log-rank test. Variables with P<0.05 in univariate analysis were entered into a multivariate Cox regression model using conditional forward stepwise selection, excluding variables that had no significant impact on OS to determine the final independent prognostic factors.
Results
Clinical characteristics
This study included (n=1,893) patients with CRC, comprising 1,075 (56.8%) men and 818 (43.2%) women. Among them, 1,127 patients (59.5%) were diagnosed with stage I–II disease, and 766 (40.5%) with stage III–IV disease. A TyG index ≥5.8 was recorded in 1,622 patients (85.7%), whereas 271 (14.3%) had a TyG index <5.8. Additionally, 1,098 patients (58.0%) presented with a LMR ≥3.8, while 795 (42.0%) had a LMR <3.8. Comprehensive clinicopathological characteristics are summarized in Table S1.
Association between clinicopathological features and long-term prognosis
Univariate survival analysis showed that gender, age, NLR, LMR, TyG, TyG-BMI, CEA, CA19-9, tumor size, tumor differentiation, UICC stage, perineural invasion, and vascular invasion were significantly associated with OS (P<0.05). After adjustment for potential confounders, multivariate Cox analysis identified eight independent prognostic factors for CRC: age, LMR, TyG, CEA, CA19-9, tumor differentiation, UICC stage, and perineural invasion (P<0.05; Figure 2). Detailed results are summarized in Table 1 and Table S2.
Figure 2.
Survival prognosis of colorectal cancer. KM curves of OS according to age (A), TyG (B), LMR (C), CEA (D), CA19-9 (E), UICC stage (F), tumor differentiation (G), and perineural invasion (H) in the training set from the cohort of First Affiliated Hospital of Kunming Medical University (P<0.05). CA19-9, carbohydrate antigen 19-9; CEA, carcinoembryonic antigen; KM, Kaplan-Meier; LMR, lymphocyte-to-monocyte ratio; OS, overall survival; TyG, triglyceride-to-glucose index; UICC, Union for International Cancer Control.
Table 1. Univariate and multivariate analysis for overall survival of colorectal cancer.
| Clinicopathological features | Univariate analysis | Multivariate analysis | |||
|---|---|---|---|---|---|
| HR (95% CI) | P | HR (95% CI) | P | ||
| TyG | |||||
| <5.8 | 1.455 (1.101–1.922) | 0.008 | 1.365 (1.029–1.81) | 0.03 | |
| ≥5.8 | Reference | ||||
| Age (years) | |||||
| <60 | 0.715 (0.571–0.895) | 0.003 | 0.703 (0.56–0.883) | 0.002 | |
| ≥60 | Reference | ||||
| LMR | |||||
| <3.8 | 1.59 (1.279–1.976) | <0.001 | 1.46 (1.172–1.819) | 0.001 | |
| ≥3.8 | Reference | ||||
| CEA (ng/mL) | |||||
| <5 | 0.423 (0.339–0.527) | <0.001 | 0.623 (0.493–0.788) | <0.001 | |
| ≥5 | Reference | ||||
| CA19-9 (U/mL) | |||||
| <30 | 0.384 (0.305–0.484) | <0.001 | 0.575 (0.45–0.735) | <0.001 | |
| ≥30 | Reference | ||||
| Tumor differentiation | |||||
| Medium | 1.207 (0.951–1.531) | 0.12 | 1.074 (0.844–1.366) | 0.56 | |
| Low/undifferentiated | 3.112 (2.142–4.522) | <0.001 | 2.333 (1.593–3.417) | <0.001 | |
| High | Reference | ||||
| UICC | |||||
| I–II stage | 0.312 (0.248–0.393) | <0.001 | 0.404 (0.319–0.513) | <0.001 | |
| III–IV stage | Reference | ||||
| Perineural invasion | |||||
| Negative | 0.484 (0.389–0.603) | <0.001 | 0.595 (0.476–0.744) | <0.001 | |
| Positive | Reference | ||||
CA19-9, carbohydrate antigen 19-9; CEA, carcinoembryonic antigen; CI, confidence interval; HR, hazard ratio; LMR, lymphocyte-to-monocyte ratio; TyG, triglyceride-to-glucose index; UICC, Union for International Cancer Control.
Association between clinicopathological features and TyG index
Univariate analysis revealed that the TyG index was correlated with LMR, BMI, PNI, and Alb (Table 2). Subsequent multivariate analysis identified BMI ≥18.6 kg/m2 and PNI ≥50.5 as factors independently associated with a TyG index of ≥5.8 (Table 3). Given that the binary logistic regression confirmed BMI and PNI as the sole significant correlates of TyG, we proceeded with subgroup analyses based on these two factors.
Table 2. Correlation analysis of low-/high-TyG with clinicopathological features in colorectal cancer.
| Clinicopathological features | Low-TyG (n=271), n (%) | High-TyG (n=1,622), n (%) | χ2 | P |
|---|---|---|---|---|
| Gender | 0.457 | 0.50 | ||
| Male | 159 (14.8) | 916 (85.2) | ||
| Female | 112 (13.7) | 706 (86.3) | ||
| Age (years) | 2.467 | 0.12 | ||
| ≥60 | 139 (13.2) | 915 (86.8) | ||
| <60 | 132 (15.7) | 707 (84.3) | ||
| BMI (kg/m2) | 68.065 | <0.001* | ||
| ≤18.5 | 52 (29.4) | 125 (70.6) | ||
| 18.6–23.9 | 175 (16.5) | 888 (83.5) | ||
| 24–27.9 | 39 (7.3) | 495 (92.7) | ||
| ≥28 | 5 (4.2) | 114 (95.8) | ||
| Diabetes | 3.014 | 0.08 | ||
| Yes | 49 (11.7) | 370 (88.3) | ||
| No | 222 (15.1) | 1,252 (84.9) | ||
| NLR | 2.718 | 0.10 | ||
| <3.3 | 215 (13.7) | 1,353 (86.3) | ||
| ≥3.3 | 56 (17.2) | 269 (82.8) | ||
| LMR | 4.633 | 0.03* | ||
| <3.8 | 130 (16.4) | 665 (83.6) | ||
| ≥3.8 | 141 (12.8) | 957 (87.2) | ||
| PNI | 37.59 | <0.001* | ||
| <50.5 | 183 (19.2) | 769 (80.8) | ||
| ≥50.5 | 88 (9.4) | 853 (90.6) | ||
| Alb (g/L) | 16.251 | <0.001* | ||
| <41.9 | 169 (17.5) | 797 (82.5) | ||
| ≥41.9 | 102 (11) | 825 (89) | ||
| CEA (ng/mL) | 1.736 | 0.19 | ||
| <5 | 175 (15.2) | 979 (84.8) | ||
| ≥5 | 96 (13) | 643 (87) | ||
| CA19-9 (U/mL) | 0.327 | 0.57 | ||
| <30 | 224 (14.5) | 1,317 (85.5) | ||
| ≥30 | 47 (13.4) | 305 (86.6) | ||
| Tumor size (cm) | 0.586 | 0.44 | ||
| <2.1 | 38 (12.9) | 257 (87.1) | ||
| ≥2.1 | 233 (14.6) | 1,365 (85.4) | ||
| Tumor differentiation | 3.234 | 0.20 | ||
| High | 105 (14.7) | 610 (85.3) | ||
| Medium | 145 (13.5) | 927 (86.5) | ||
| Low/undifferentiated | 21 (19.8) | 85 (80.2) | ||
| pT stage | 0.588 | 0.44 | ||
| pT1–2 | 89 (15.2) | 495 (84.8) | ||
| pT3–4 | 182 (13.9) | 1,127 (86.1) | ||
| pN stage | 3.959 | 0.14 | ||
| pN0 | 160 (13.8) | 998 (86.2) | ||
| pN1 | 62 (13.3) | 404 (86.7) | ||
| pN2 | 49 (18.2) | 220 (81.8) | ||
| M stage | 0.724 | 0.40 | ||
| M0 | 253 (14.1) | 1,535 (85.9) | ||
| M1 | 18 (17.1) | 87 (82.9) | ||
| UICC | 0.51 | 0.48 | ||
| I–II stage | 156 (13.8) | 971 (86.2) | ||
| III–IV stage | 115 (15) | 651 (85) | ||
| Perineural invasion | 0.007 | 0.93 | ||
| Negative | 198 (14.4) | 1,181 (85.6) | ||
| positive | 73 (14.2) | 441 (85.8) | ||
| Vascular invasion | 3.653 | 0.056 | ||
| Negative | 202 (13.5) | 1,292 (86.5) | ||
| Positive | 69 (17.3) | 330 (82.7) |
*, P<0.05. Low-TyG represents TyG <5.8 and high-TyG represents TyG ≥5.8. Alb, albumin; BMI, body mass index; CA19-9, carbohydrate antigen 19-9; CEA, carcinoembryonic antigen; LMR, lymphocyte-to-monocyte ratio; M, metastasis; NLR, neutrophil-to-lymphocyte ratio; pN, pathological node; PNI, prognostic nutritional index; pT, pathological tumor; TyG, triglyceride-to-glucose index; UICC, Union for International Cancer Control.
Table 3. Correlation multivariate analysis of TyG with clinicopathological features in colorectal cancer.
| Clinicopathological features | OR (95% CI) | P |
|---|---|---|
| BMI (kg/m2) | ||
| 18.6–23.9 | 1.991 (1.380–2.872) | <0.001* |
| 24–27.9 | 4.706 (2.958–7.485) | <0.001* |
| ≥28 | 8.637 (3.320–22.468) | <0.001* |
| ≤18.5 | Reference | |
| PNI | ||
| <50.5 | 0.475 (0.360–0.627) | <0.001* |
| ≥50.5 | Reference |
*, P<0.05. BMI, body mass index; CI, confidence interval; OR, odds ratio; PNI, prognostic nutritional index; TyG, triglyceride-to-glucose index.
In the subgroup with PNI <50.5, as well as in the subgroup with a normal-range BMI (18.6–23.9 kg/m2), patients with a TyG index ≥5.8 demonstrated a significantly better OS compared to those with TyG <5.8 (Table S3, Figure S2). In all other subgroups, TyG showed no significant association with OS.
Development a prognostic model for CRC survival
Using ML, we developed a prognostic model based on the identified independent prognostic variables. The cohort from the First Affiliated Hospital of Kunming Medical University was randomly divided into a training set (70%) and an internal validation set (30%). The resulting model demonstrated excellent predictive performance, with C-index values of 0.742 and 0.735 in the training and internal validation cohorts, respectively (Figure 3A). A forest plot was then drawn based on independent prognostic variables (Figure 3B).
Figure 3.
Risk models for training sets. (A) List of prediction models generated via machine learning and the calculation of the C-index of each model across all validation datasets. (B) The forest map showed the results of multivariate COX regression analysis with eight independent prognostic factors. (C) The SHAP chart shows the extent to which the above eight factors contribute to the prognostic effect of CRC in this model. (D) Nomogram model for predicting the 1-, 3-, and 5-year OS rate of CRC patients. (E) ROC curve of CRC prognosis in the training set. (F) KM curves of OS according to the median score in the training set (P<0.05). (G) DCA of the 1-, 3-, and 5-year CRC model. (H) The calibration curve of the training set for predicting the 1-, 3-, and 5-year OS rates in CRC patients. AUC, area under the curve; C-index, concordance index; CA19-9, carbohydrate antigen 19-9; CEA, carcinoembryonic antigen; CI, confidence interval; CRC, colorectal cancer; DCA, decision curve analysis; KM, Kaplan-Meier; LMR, lymphocyte-to-monocyte ratio; OS, overall survival; ROC, receiver operating characteristic; SHAP, SHapley Additive exPlanations; TyG, triglyceride-to-glucose index; UICC, Union for International Cancer Control.
SHapley Additive exPlanations (SHAP) analysis ranked the contribution of each variable to the prognostic outcome, with TyG identified as the second most impactful factor in CRC patients (Figure 3C). A nomogram was generated to facilitate clinical interpretation (Figure 3D). The model’s performance, assessed using the area under the curve (AUC), was 0.79, 0.76, and 0.74 for the prediction of 1-, 3-, and 5-year OS, respectively (Figure 3E). Using a median risk score of −0.026 as the cutoff, CRC patients were stratified into high- and low-risk groups, with a significant survival difference between groups, where the high-risk group demonstrated significantly poorer survival than the low-risk group (P<0.001; Figure 3F). Decision curve analysis (DCA) confirmed the model’s superior clinical utility for CRC patients, demonstrating a greater net benefit across a wider threshold probability range compared with UICC stage alone (Figure 3G). The calibration curves demonstrated excellent concordance between predicted and observed survival for CRC patients (Figure 3H).
Internal validation
A total of 567 CRC patients were randomly selected for internal validation. The model demonstrated a C-index of 0.735 (Figure 3A). The time-dependent AUCs for 1-, 3-, and 5-year survival were 0.86, 0.73, and 0.72, respectively (Figure 4A). KM analysis confirmed significant survival differences between the high- and low-risk groups, where the low-risk group exhibited significantly better survival than in the high-risk group (median risk score: −0.104; P<0.001; Figure 4B). The calibration curves for 1-, 3-, and 5-year survival demonstrated excellent agreement between the model’s predictions and observed outcomes in CRC patients (Figure 4C).
Figure 4.
Validate the risk model of the internal validation set. (A) ROC curve of CRC prognosis in the internal validation set from the cohort of the first affiliated hospital of Kunming medical university. (B) Kaplan-Meier curves of OS according to the model median score in the internal validation set (P<0.001). (C) The calibration curve of the internal validation set for predicting the 1-, 3-, and 5-year OS rates in CRC patients. AUC, area under the curve; CRC, colorectal cancer; OS, overall survival; ROC, receiver operating characteristic.
External validation
An independent external validation cohort comprising 493 CRC patients from the Third People’s Hospital of Honghe Prefecture was used to evaluate the model’s generalizability. The model achieved a C-index of 0.752 (Figure 3A). The time-dependent ROC analysis showed AUCs of 0.78, 0.80, and 0.74 for predicting 1-, 3-, and 5-year survival, respectively (Figure 5A). KM curves demonstrated that patients in the high-risk group had significantly worse OS than those in the low-risk group (median risk score: 0.322; P<0.001; Figure 5B). The calibration curves confirmed good consistency between the predicted and observed survival across all time points (Figure 5C).
Figure 5.
Validate the risk model of the external validation set. (A) ROC curve of CRC prognosis in the internal validation set from the cohort of The Third People’s Hospital of Honghe Prefecture. (B) Kaplan-Meier curves of OS according to the model median score in the external validation set (P<0.001). (C) The calibration curve of the external validation set for predicting the 1-, 3-, and 5-year OS rates in CRC patients. AUC, area under the curve; CRC, colorectal cancer; OS, overall survival; ROC, receiver operating characteristic.
Clinical application of the model
To enhance clinical usability, we developed a web-based application using the R Shiny framework for real-time prognostic estimation. This interactive tool allows clinicians to input patient-specific parameters and generate individualized risk scores and survival predictions. The application is publicly accessible at https://github.com/OrigamiSheep/CRC_TyG_shiny, and is compatible with all major browsers (Figure S3).
Discussion
CRC is one of the most prevalent malignancies of the digestive tract. Owing to its high incidence and mortality rates, there is substantial interest in developing accurate and efficient prognostic tools (8). However, simple and reliable prediction models remain limited in clinical practice. In this study, we identified eight independent prognostic factors in CRC patients: age, CEA, CA19-9, LMR, TyG, tumor differentiation, UICC stage, and perineural invasion and integrated them to develop an ML-based prognostic model. The model demonstrated consistently strong performance in both internal and external validation cohorts. Calibration plots showed excellent agreement between predicted and observed survival outcomes in CRC patients, and the 1-, 3-, and 5-year AUC values all exceeded 0.70, supporting the model’s predictive accuracy and clinical applicability.
Age, CEA, CA19-9, tumor differentiation, and UICC stage are well-established prognostic indicators for CRC patients, consistent with findings from previous studies (19). Perineural invasion, increasingly recognized as a potential “fifth pathway” for CRC metastasis, has also been strongly associated with poor prognosis (20). This association may be driven by interactions between enteric neurons and tumor cells, which modulate intracellular signaling pathways and promote tumor proliferation (21). Furthermore, tumor-induced neurogenesis and reciprocal activation between tumor cells and infiltrated nerve fibers may further contribute to tumor progression and dissemination (22).
The prognostic significance of LMR observed in our study aligns with previous findings (23,24). Low LMR levels have been associated with poorer clinical outcomes (25). Notably, SHAP analysis showed that LMR exerted a greater influence on prognosis than CEA (Figure 3C). Because lymphocyte counts are the primary determinant of LMR value, this observation highlights the importance of host immune response, particularly lymphocyte infiltration in CRC progression. Previous research has demonstrated that peripheral lymphocytosis correlates with tumor-infiltrating lymphocytes (TILs), and patients with elevated TIL levels tend to respond better to immune checkpoint inhibitors, leading to improved prognosis (26-28).
The TyG index, calculated from fasting glucose and triglyceride levels, is a well-established surrogate marker of insulin resistance. In our model, a lower TyG index was significantly associated with poorer postoperative outcomes. Patients with low TyG values exhibited a 1.365-fold higher risk of mortality compared with those with elevated TyG values (Table 1). Although previous studies have reported that higher TyG levels are associated with an increased incidence of CRC and other malignancies (29,30). several studies have also identified the TyG index as an independent prognostic factor in multiple cancer types, including renal, prostate, and breast cancers (16,31-33).
Despite both BMI and PNI are correlated with TyG status, low-TyG (<5.8) and high-TyG (≥5.8), subgroup survival analyses revealed that a high TyG (≥5.8) was associated with improved OS only in specific subgroups: patients with a PNI <50.5 and those with a BMI in the normal range (18.6–23.9 kg/m2). Within these subgroups, a TyG ≥5.8 was confirmed as an independent protective factor for OS following CRC surgery, a finding consistent with the prognostic trend observed in the overall study population.
Cai et al. (15) and Yao et al. (34) similarly reported that higher TyG values are associated with better prognosis in gastric cancer (35). Our findings are consistent with these observations and may be partly explained by interactions between insulin resistance and the gut microbiome. Studies have shown that insulin resistance may reduce the abundance of Bacteroides, which promote intestinal inflammation and contribute to tumorigenesis (36,37). Consequently, the TyG index may influence CRC prognosis through microbiota-mediated immune modulation.
Meanwhile, previous studies have reported that TyG as a marker of systemic glucose metabolism. An elevated TyG may indicate reduced glucose utilization by tumor cells, potentially blocking their primary means of nutrient acquisition via glycolysis. This restriction can limit tumor growth, which might contribute to a better prognosis in CRC patients (38). Interestingly, this relationship may seem paradoxical, as immune cell activation also relies on glycolysis to generate adenosine triphosphate (ATP) for proliferation and anti-tumor activity (39). This highlights the dual role of glucose metabolism in cancer.
The prognostic model developed in this study integrates clinical, pathological, and laboratory parameters, all of which are routinely collected in clinical practice, making it both cost-effective and feasible for widespread implementation. The model achieved a high C-index (0.742) and was further strengthened by the development of an online calculator that provides automated risk estimation. Clinicians can input patient-specific variables to obtain individualized prognostic information, facilitating more precise risk stratification and clinical decision-making.
Limitations
There are several limitations in this retrospective study. First, the external validation dataset was limited. Second, multicenter data were not included, which may affect the generalizability of the findings. Third, the follow-up duration was relatively short. Fourth, molecular biomarker data (e.g., KRAS, BRAF, NRAS, and MSI status) were not available. Fifth, lacked data to adjust for potential confounders such as patient comorbidities, glycated hemoglobin (HbA1c), performance status (PS), American Society of Anesthesiologists (ASA) physical status classification system and detailed treatment regimens, which may have influenced the outcomes. Finally, because only preoperative indicators were analyzed, the complex interplay between tumor prognosis and inflammation metabolism dynamics may not have been fully captured.
Future studies should adopt prospective, multicenter designs with longer follow-up durations and more extensive external validation to enhance the accuracy of long-term prognostic predictions. Incorporating dynamic biomarkers will also be important for clarifying the temporal effects of immune and metabolic changes on CRC prognosis.
Conclusions
This study presents a novel prognostic model for CRC that incorporates the TyG index, LMR, and key clinicopathological features. The model demonstrated strong predictive performance and external validity, providing a practical, cost-effective tool for long-term survival prediction in patients with CRC. Because it relies solely on routinely collected clinical and laboratory data, the model has substantial potential for widespread clinical adoption. Future prospective, multicenter studies are warranted to further validate its performance and evaluate its applicability in real-world clinical settings.
Supplementary
The article’s supplementary files as
Acknowledgments
We thank The First Affiliated Hospital of Kunming Medical University, and the Third People’s Hospital of Honghe Prefecture Gejiu, for providing the data for this study. We also acknowledge the contributions of the research team, data analysts, and medical staff in patient care and data collection.
Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was a retrospective analysis approved by the Ethics Committee of The First Affiliated Hospital of Kunming Medical University (No. 2024-L-124) and the Ethics Committee of The Third People’s Hospital of Honghe Prefecture (No. 2024-KYXM-29), which exempted the requirement for informed consent from the patients.
Footnotes
Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2893/rc
Funding: This work was supported by the National Natural Science Foundation of China (No. 82160533), Applied Basic Foundation of Yunnan Province (No. 202501AY070001-015), Science and Technology Projects of Yunnan Universities Serving Key Industries (No. FWCY-BSPY2025076), 535 Talent Project of First Affiliated Hospital of Kunming Medical University (No. 2022535D07), Yunnan Revitalization Talent Support Program (Nos. RLQB20200004 and RLMY20220013), Reserve Talents of Young and Middle-aged Academic and Technical Leaders in Yunnan Province (No. 202305AC160018) and Education teaching research project of The First Affiliated Hospital of Kunming Medical University (No. 2024-JY-19).
Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2893/coif). The authors have no conflicts of interest to declare.
Data Sharing Statement
Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2893/dss
References
- 1.Siegel RL, Kratzer TB, Giaquinto AN, et al. Cancer statistics, 2025. CA Cancer J Clin 2025;75:10-45. 10.3322/caac.21871 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Yarla N, Madka V, Rao C. Targeting Triglyceride Metabolism for Colorectal Cancer Prevention and Therapy. Curr Drug Targets 2022;23:628-35. 10.2174/1389450122666210824150012 [DOI] [PubMed] [Google Scholar]
- 3.Yamamoto T, Kawada K, Obama K. Inflammation-Related Biomarkers for the Prediction of Prognosis in Colorectal Cancer Patients. Int J Mol Sci 2021;22:8002. 10.3390/ijms22158002 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Li Y, Xiao J, Zhang T, et al. Analysis of KRAS, NRAS, and BRAF Mutations, Microsatellite Instability, and Relevant Prognosis Effects in Patients With Early Colorectal Cancer: A Cohort Study in East Asia. Front Oncol 2022;12:897548. 10.3389/fonc.2022.897548 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Negrut RL, Cote A, Feder B, et al. Comparative Prognostic Role of PLR and NLR in Colon Cancer: A Retrospective Analysis of Preoperative Inflammatory Markers. Medicina (Kaunas) 2025;61:1580. 10.3390/medicina61091580 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Sun J, Xiong W, Fang Z, et al. Prognostic role of preoperative inflammatory markers in postoperative lung metastasis of colorectal cancer: a retrospective study. BMC Gastroenterol 2025;25:493. 10.1186/s12876-025-04091-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Xiao Z, Wang X, Chen X, et al. Prognostic role of preoperative inflammatory markers in postoperative patients with colorectal cancer. Front Oncol 2023;13:1064343. 10.3389/fonc.2023.1064343 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Wen X, Sun H, Du S, et al. A nomogram of inflammatory indexes for preoperatively predicting the risk of lymph node metastasis in colorectal cancer. Tech Coloproctol 2024;28:148. 10.1007/s10151-024-03010-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Mei P, Huang L, Lin L, et al. The prognostic utility of the ratio of lymphocyte to monocyte in patients with metastatic colorectal cancer: a systematic review and meta-analysis. Front Oncol 2025;15:1394154. 10.3389/fonc.2025.1394154 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Omer HFE, Alghazali M, Ibrahim MY, et al. Association between the triglyceride-glucose index (TyG Index) and risk of colorectal cancer: a systematic review and meta-analysis. World J Surg Oncol 2025;23:280. 10.1186/s12957-025-03930-y [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Ramdas Nayak VK, Satheesh P, Shenoy MT, et al. Triglyceride Glucose (TyG) Index: A surrogate biomarker of insulin resistance. J Pak Med Assoc 2022;72:986-8. 10.47391/JPMA.22-63 [DOI] [PubMed] [Google Scholar]
- 12.Tao LC, Xu JN, Wang TT, et al. Triglyceride-glucose index as a marker in cardiovascular diseases: landscape and limitations. Cardiovasc Diabetol 2022;21:68. 10.1186/s12933-022-01511-x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Selvi NMK, Nandhini S, Sakthivadivel V, et al. Association of Triglyceride-Glucose Index (TyG index) with HbA1c and Insulin Resistance in Type 2 Diabetes Mellitus. Maedica (Bucur) 2021;16:375-81. 10.26574/maedica.2021.16.3.375 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Yi T, Lin Z, Mai Z, et al. The triglyceride-glucose index associated with reduced risk of liver metastasis in pancreatic cancer. Front Endocrinol (Lausanne) 2025;16:1592788. 10.3389/fendo.2025.1592788 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Cai C, Chen C, Lin X, et al. An analysis of the relationship of triglyceride glucose index with gastric cancer prognosis: A retrospective study. Cancer Med 2024;13:e6837. 10.1002/cam4.6837 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Qin G, Sun Z, Jin Y, et al. The association between the triglyceride-glucose index and prognosis in postoperative renal cell carcinoma patients: a retrospective cohort study. Front Endocrinol (Lausanne) 2024;15:1301703. 10.3389/fendo.2024.1301703 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Liu T, Zhang Q, Wang Y, et al. Association between the TyG index and TG/HDL-C ratio as insulin resistance markers and the risk of colorectal cancer. BMC Cancer 2022;22:1007. 10.1186/s12885-022-10100-w [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Guo HL, Chen JY, Tang YZ, et al. Minimally invasive surgery versus laparotomy of nonmetastatic pT4a colorectal cancer: a propensity score analysis. Int J Surg 2023;109:3294-302. 10.1097/JS9.0000000000000627 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Liang L, Xu N, Ding L, et al. Combined inflammation-related biomarkers and clinicopathological features for the prognosis of stage II/III colorectal cancer by machine learning. BMC Cancer 2024;24:1548. 10.1186/s12885-024-13331-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Que Y, Wu R, Li H, et al. A prediction nomogram for perineural invasion in colorectal cancer patients: a retrospective study. BMC Surg 2024;24:80. 10.1186/s12893-024-02364-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Godlewski J, Kmiec Z. Colorectal Cancer Invasion and Atrophy of the Enteric Nervous System: Potential Feedback and Impact on Cancer Progression. Int J Mol Sci 2020;21:3391. 10.3390/ijms21093391 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Jobling P, Pundavela J, Oliveira SM, et al. Nerve-Cancer Cell Cross-talk: A Novel Promoter of Tumor Progression. Cancer Res 2015;75:1777-81. 10.1158/0008-5472.CAN-14-3180 [DOI] [PubMed] [Google Scholar]
- 23.Li Q, Chen L, Jin H, et al. Pretreatment Inflammatory Markers Predict Outcomes and Prognosis in Colorectal Cancer Patients With Synchronous Liver Metastasis. Clin Med Insights Oncol 2022;16:11795549221084851. 10.1177/11795549221084851 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Constantin GB, Firescu D, Mihailov R, et al. A Novel Clinical Nomogram for Predicting Overall Survival in Patients with Emergency Surgery for Colorectal Cancer. J Pers Med 2023;13:575. 10.3390/jpm13040575 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Zhou M, Gu Q, Zhou M, et al. Extensive study on the associations of 12 composite inflammatory indices with colorectal cancer risk and mortality: a cross-sectional analysis of NHANES 2001-2020. Int J Surg 2025;111:7559-75. 10.1097/JS9.0000000000002996 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Suzuki Y, Okabayashi K, Hasegawa H, et al. Comparison of Preoperative Inflammation-based Prognostic Scores in Patients With Colorectal Cancer. Ann Surg 2018;267:527-31. 10.1097/SLA.0000000000002115 [DOI] [PubMed] [Google Scholar]
- 27.Bai Z, Zhou Y, Ye Z, et al. Tumor-Infiltrating Lymphocytes in Colorectal Cancer: The Fundamental Indication and Application on Immunotherapy. Front Immunol 2021;12:808964. 10.3389/fimmu.2021.808964 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Qin M, Chen G, Hou J, et al. Tumor-infiltrating lymphocyte: features and prognosis of lymphocytes infiltration on colorectal cancer. Bioengineered 2022;13:14872-88. 10.1080/21655979.2022.2162660 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Kityo A, Lee SA. Triglyceride-Glucose Index, Modifiable Lifestyle, and Risk of Colorectal Cancer: A Prospective Analysis of the Korean Genome and Epidemiology Study. J Epidemiol Glob Health 2024;14:1249-56. 10.1007/s44197-024-00282-w [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Wu Z, Yang J, Ma Z, et al. Associations between triglyceride-glucose related indices and the risk of incident pancreatic cancer: a large-scale prospective cohort study in the UK Biobank. BMC Cancer 2025;25:327. 10.1186/s12885-025-13718-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Liu XY, Zhang Q, Zhang X, et al. Prognostic value of insulin resistance in patients with female reproductive system malignancies: A multicenter cohort study. Immun Inflamm Dis 2023;11:e1107. 10.1002/iid3.1107 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Horsanali MO, Eren H, Dıl E, et al. A novel prognostic risk factor for patients undergoing radical prostatectomy: Triglyseride-glucose index. Int J Clin Pract 2021;75:e13978. 10.1111/ijcp.13978 [DOI] [PubMed] [Google Scholar]
- 33.Li F, Gao T, Li Z, et al. Triglyceride-glucose index and triglyceride-glucose-body mass index as prognostic factors for early stage breast cancer patients receiving neoadjuvant chemotherapy. Transl Oncol 2025;53:102292. 10.1016/j.tranon.2025.102292 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Yao ZY, Ma X, Cui YZ, et al. Impact of triglyceride-glucose index on the long-term prognosis of advanced gastric cancer patients receiving immunotherapy combined with chemotherapy. World J Gastroenterol 2025;31:102249. 10.3748/wjg.v31.i5.102249 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Liang GZ, Li XS, Hu ZH, et al. Development and validation of a nomogram model for predicting overall survival in patients with gastric carcinoma. World J Gastrointest Oncol 2025;17:95423. 10.4251/wjgo.v17.i2.95423 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.He Z, Xiong H, Liu L, et al. Influence of Postoperative Insulin Resistance on Short-Term Outcomes of Radical Gastrectomy for Gastric Cancer: A Microbiome and Metabolome-Based Prospective Cohort Study. Ann Surg Oncol 2024;31:8638-50. 10.1245/s10434-024-16125-8 [DOI] [PubMed] [Google Scholar]
- 37.Yu LC. Microbiota dysbiosis and barrier dysfunction in inflammatory bowel disease and colorectal cancers: exploring a common ground hypothesis. J Biomed Sci 2018;25:79. 10.1186/s12929-018-0483-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Pascale RM, Calvisi DF, Simile MM, et al. The Warburg Effect 97 Years after Its Discovery. Cancers (Basel) 2020;12:2819. 10.3390/cancers12102819 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Alwarawrah Y, Kiernan K, MacIver NJ. Changes in Nutritional Status Impact Immune Cell Metabolism and Function. Front Immunol 2018;9:1055. 10.3389/fimmu.2018.01055 [DOI] [PMC free article] [PubMed] [Google Scholar]