Published online Dec 21, 2026. doi: 10.3748/wjg.124207
Revised: July 14, 2026
Accepted: July 27, 2026
Published online: December 21, 2026
Processing time: 140 Days and 15.1 Hours
Metabolomics may provide complementary information for colorectal cancer (CRC) risk prediction beyond clinical factors and polygenic risk scores (PRS), but its incremental value remains uncertain. We hypothesized that integrating plasma nuclear magnetic resonance metabolomics with clinical and genetic factors would improve CRC risk stratification and prediction.
To evaluate the incremental predictive value of plasma metabolomics beyond clinical factors and PRS for CRC.
This study included 468095 UK Biobank participants. Cox models examined associations between 325 baseline nuclear magnetic resonance-based plasma metabolic markers and incident CRC. A metabolomic risk score (MetRS) was developed using Light Gradient Boosting Machine-based feature selection. We evaluated its risk stratification ability and compared the predictive performance of a clinical model adapted from QCancer (hereafter referred to as QCancer), QCancer + MetRS, QCancer + PRS, and QCancer + PRS + MetRS. Performance was assessed by internal-external cross-validation using Harrell’s C-index, calibration, time-dependent areas under the curve, decision-curve analysis, and reclassification metrics.
During a median follow-up of 14.7 years, 7150 participants developed CRC. Among 325 metabolic markers, 115 were significantly associated with CRC risk after false discovery rate correction. The MetRS stratified long-term CRC risk. Compared with QCancer + PRS, adding MetRS modestly improved discrimination, with Harrell’s C-index increasing from 0.705 (95% confidence interval: 0.696-0.714) to 0.711 (95% confidence interval: 0.703-0.719). The integrated model showed generally acceptable calibration, the greatest net benefit across most examined thresholds, and modest improvement in reclassification.
Plasma metabolomics complements clinical and genetic factors for CRC risk assessment, providing modest improvements in risk stratification and predictive performance.
Core Tip: Using data from 468095 UK Biobank participants, we evaluated 325 nuclear magnetic resonance-based plasma metabolic markers and developed a seven-feature metabolomic risk score for incident colorectal cancer. The score stratified long-term risk and modestly improved prediction when added to an adapted QCancer clinical model and a polygenic risk score. Internal-external cross-validation across 10 geographical clusters supported consistency and geographical transportability within the UK Biobank. These findings suggest that metabolomics may complement, rather than replace, established clinical and genetic risk assessment. Independent external validation and cost-effectiveness evaluation are required before clinical implementation.
- Citation: Zeng LT, Hu JN, Pang MH, Cao QX, Hou NY, Liu F. Integrating metabolomics, clinical factors, and genetic risk for colorectal cancer risk prediction: Model development and internal-external cross-validation. World J Gastroenterol 2026; 32(47): 124207
- URL: https://www.wjgnet.com/1007-9327/full/v32/i47/124207.htm
- DOI: https://dx.doi.org/10.3748/wjg.124207
Colorectal cancer (CRC) is the third most commonly diagnosed malignancy worldwide and the second leading cause of cancer-related death, imposing a substantial public health and economic burden[1]. Although the incidence and mortality of CRC have declined in some high-income countries with the implementation of screening programmes, its overall disease burden remains considerable[2,3]. Colonoscopy remains one of the cornerstone modalities for CRC screening because it enables the direct detection and removal of precancerous lesions[4]. However, its invasive nature, the need for bowel preparation and skilled operators, its relatively high time and resource requirements, and suboptimal adherence in the general population all limit its widespread implementation in large-scale population screening[5-7]. Therefore, developing generalizable and noninvasive risk assessment tools to support risk stratification and personalized screening strategies is of substantial clinical and public health importance.
Among the many candidate biomarkers, blood-based metabolites have attracted considerable attention because of their minimally invasive nature and high clinical accessibility[8,9]. Circulating metabolites may capture the integrated effects of multiple biological processes, including genetic background, lifestyle, the gut microbiome, inflammatory status, and potential tumour-related metabolic reprogramming, and may therefore provide valuable information for CRC risk assessment[10-13]. Previous studies have reported associations of various blood metabolites with CRC risk or early detection, but the findings have not been entirely consistent[14-17]. Possible reasons include limited sample sizes, differences in metabolite detection platforms and feature coverage, and substantial heterogeneity in study design and validation strategies[8,16]. Previous studies have suggested that metabolomics may hold promise for CRC risk assessment or early detection, but existing evidence remains heterogeneous[8,16]. In particular, a recent UK Biobank and ESTHER cohort analysis found that metabolomics had limited incremental value for improving established CRC risk models[13]. However, whether metabolomics provides complementary predictive information when integrated with an adapted QCancer model and polygenic risk score (PRS) in a large-scale internal-external cross-validation (IECV) framework remains insufficiently explored. Therefore, the true added value of metabolomics for CRC risk prediction remains to be further clarified[13,16].
In this study, using metabolomics data from one of the largest prospective cohorts currently available, including 468095 participants, we systematically evaluated the associations between 325 plasma metabolic markers and incident CRC risk. First, we comprehensively analyzed the associations between individual metabolic markers and CRC risk. Next, we developed and evaluated risk prediction models integrating metabolomic features, clinical factors, and PRS, and assessed their discrimination and generalizability within an IECV framework. Finally, we quantified the incremental predictive value of metabolomics over existing risk models and explored its potential utility for CRC risk stratification.
Participants in this study were drawn from the UK Biobank, a prospective cohort of over 500000 individuals aged 40-69 years recruited through 22 assessment centres in the United Kingdom between 2006 and 2010. At baseline, participants completed questionnaires, attended in-person interviews and physical examinations, and provided blood, urine, and saliva samples.
For this study, 501936 participants were initially available for screening. We further excluded individuals with missing data for predictors required to construct QCancer or the PRS (n = 25969), those with prevalent CRC at baseline (n = 2176), and those without nuclear magnetic resonance (NMR) metabolomics data (n = 5696). Therefore, all clinical and genetic predictors included in the Cox regression models were complete. For metabolomic risk score (MetRS) development, missing values in metabolomic markers were not imputed; instead, they were retained and processed internally by Light Gradient Boosting Machine (LightGBM), a missingness-tolerant tree-based algorithm, during model training and prediction. The information for all metabolites are provided in Supplementary Table 1. A total of 468095 participants were included in the final analysis (Supplementary Figure 1).
This study was conducted in accordance with the ethical principles outlined in the Declaration of Helsinki. All participants provided written informed consent before data collection. The UK Biobank has ethical approval from the North West–Haydock Research Ethics Committee as a research tissue bank, approval No. 21/NW/0157. This study was conducted under UK Biobank application No. 354686.
CRC outcomes were ascertained through linkage to national cancer registry records. The outcome was defined as incident CRC occurring after the baseline assessment and was identified using International Classification of Diseases, Tenth Revision codes C18 to C20. Follow-up time was calculated from the date of the baseline assessment until the earliest of the date of first recorded CRC diagnosis, death, or the end of available follow-up. The date of death was obtained through linkage to death registry data. National cancer registry data were available through May 31, 2023 for England, September 30, 2023 for Scotland, and December 31, 2016 for Wales.
Clinical predictors were constructed from baseline data in the UK Biobank, with variable definitions and coding aligned as closely as possible with previously published QCancer-related models[18-20]. In this study, QCancer refers to an adapted model refitted using available QCancer-related predictors in the UK Biobank. The predictors included age, sex, ethnicity, Townsend deprivation index, body mass index, smoking status, alcohol intake, family history of CRC, and medical history (Supplementary Table 2)[18-20]. For improved generalizability to other populations, we excluded the geographical region variable, which was tailored to the United Kingdom population, from the QCancer predictors. Instead, geographical region was used to define the IECV clusters, allowing the assessment of geographical transportability. To provide a unified CRC risk prediction framework for integrating clinical factors, PRS, and MetRS, we fitted a pooled model with sex included as a covariate. This approach is consistent with recent CRC risk prediction work using a combined modelling framework that integrates clinical information, PRS, and omics-derived scores[19]. For genetic susceptibility, we used the UK Biobank “Standard PRS for bowel cancer (CRC)” (Data Field 26218) as the PRS[21].
Metabolomics data were obtained from the UK Biobank NMR metabolomics dataset. In the version used in this study, 249 metabolic biomarkers were available, including 168 absolute concentration measures and 81 derived ratio measures, quantified from baseline ethylenediaminetetraacetic acid plasma samples from approximately 491000 participants, with data updated to November 2025. We used the “ukbnmr” R package to remove technical variation. In the course of this procedure, 76 additional biomarker ratios with potential biological relevance, absent from the original dataset, were generated, yielding 325 NMR metabolic indicators for analysis in total. Details on the derivation of the additional ratios and metabolomics data preprocessing are provided in the Supplementary Methods. The abbreviations, full names, units, and grouping information for all metabolites are provided in Supplementary Table 1, while the missing percentages for individual metabolic markers are reported in Supplementary Table 3.
We first fitted Cox proportional hazards models for all 325 metabolic indicators in the full analytic population, adjusting for age at recruitment, sex, geographical region, ethnic background, and body mass index. We further conducted subgroup analyses stratified by age (< 60 years vs ≥ 60 years) and sex (female vs male), with the results presented in Supplementary Tables 4 and 5. False discovery rate correction was applied for multiple testing, and the 115 indicators that remained statistically significant were entered into a LightGBM-based sequential feature-selection procedure to construct the MetRS. Details of feature selection, hyperparameter tuning, model optimization, overfitting control, model interpretation, and missing-data handling are provided in the Supplementary Methods and Supplementary Table 6.
To evaluate the risk stratification capacity of the MetRS, participants were categorized according to MetRS tertiles, and Kaplan-Meier curves were plotted to compare cumulative CRC risk across groups. Absolute CRC risks at 5 years, 10 years, and 15 years were estimated from the Kaplan-Meier curves. Cox models were used to estimate hazard ratios (HR) for different MetRS groups. In addition, stratified analyses were conducted according to age, sex, and genetic risk to further assess the consistency of the association between the MetRS and CRC risk.
To evaluate the incremental value of metabolomic information for CRC risk prediction, we developed and validated prediction models within an IECV framework[22]. According to the geographical locations of the UK Biobank assessment centres, the study population was divided into 10 predefined geographical clusters (Table 1, Geographical region). In each iteration, data from 9 clusters were used for model development and the remaining cluster was used for external validation, so that each participant contributed one out-of-fold prediction. We prespecified four Cox proportional hazards models, including QCancer, QCancer + MetRS, QCancer + PRS, and QCancer + PRS + MetRS. Details of the geographical clustering scheme and IECV implementation are provided in the Supplementary Methods.
| Characteristic | All participants | Incident CRC cases |
| Number of participants | 468095 | 7150 |
| Follow-up (years), median (IQR) | 14.7 (13.89, 15.42) | 7.43 (3.98, 10.6) |
| Sex | ||
| Female | 256073 (54.7) | 3199 (44.7) |
| Male | 212022 (45.3) | 3951 (55.3) |
| Age (years), median (IQR) | 58 (50, 63) | 62 (57, 66) |
| Geographical region | ||
| East Midlands | 32180 (6.9) | 471 (6.6) |
| London | 63101 (13.5) | 850 (11.9) |
| North East | 54810 (11.7) | 886 (12.4) |
| North West | 70869 (15.1) | 1130 (15.8) |
| Scotland | 33488 (7.2) | 671 (9.4) |
| South East | 41559 (8.9) | 710 (9.9) |
| South West | 40899 (8.7) | 575 (8.0) |
| Wales | 19633 (4.2) | 180 (2.5) |
| West Midlands | 41594 (8.9) | 630 (8.8) |
| Yorkshire and Humber | 69962 (14.9) | 1047 (14.6) |
| Ethnic group | ||
| Indian | 5257 (1.1) | 33 (0.5) |
| Pakistani | 1608 (0.3) | 10 (0.1) |
| Bangladeshi | 202 (0.0) | 1 (0.0) |
| Other Asian | 1669 (0.4) | 10 (0.1) |
| Black African | 2952 (0.6) | 17 (0.2) |
| Caribbean | 4046 (0.9) | 38 (0.5) |
| Chinese | 1438 (0.3) | 14 (0.2) |
| White/not recorded | 443995 (94.9) | 6963 (97.4) |
| Other | 6928 (1.5) | 64 (0.9) |
| Townsend deprivation index, median (IQR) | -2.17 (-3.66, 0.48) | -2.35 (-3.72, 0.26) |
| Body mass index (kg/m2), median (IQR) | 26.7 (24.1, 29.9) | 27.3 (24.7, 30.4) |
| Smoking status | ||
| Non-smoker | 258176 (55.2) | 3311 (46.3) |
| Ex-smoker | 163422 (34.9) | 3178 (44.4) |
| Light smoker | 19670 (4.2) | 280 (3.9) |
| Moderate smoker | 14346 (3.1) | 203 (2.8) |
| Heavy smoker | 12481 (2.7) | 178 (2.5) |
| Alcohol intake | ||
| Non-drinker | 37082 (7.9) | 508 (7.1) |
| Trivial drinker | 149364 (31.9) | 2010 (28.1) |
| Light drinker | 142660 (30.5) | 2048 (28.6) |
| Moderate drinker | 107153 (22.9) | 1840 (25.7) |
| Heavy drinker | 20759 (4.4) | 465 (6.5) |
| Very heavy drinker | 11077 (2.4) | 279 (3.9) |
| Medical history | ||
| Ulcerative colitis | 2431 (0.5) | 57 (0.8) |
| Colorectal polyps | 1347 (0.3) | 39 (0.5) |
| Diabetes | 22696 (4.8) | 527 (7.4) |
| Breast cancer | 10184 (2.2) | 151 (2.1) |
| Uterine cancer | 1130 (0.2) | 27 (0.4) |
| Ovarian cancer | 772 (0.2) | 19 (0.3) |
| Cervical cancer | 1873 (0.4) | 16 (0.2) |
| Lung cancer | 237 (0.1) | 7 (0.1) |
| Blood cancers | 2204 (0.5) | 40 (0.6) |
| Oral cancer | 777 (0.2) | 24 (0.3) |
| Family history of CRC | ||
| No | 417109 (89.1) | 6054 (84.7) |
| Yes | 50986 (10.9) | 1096 (15.3) |
We evaluated model performance across discrimination, calibration, clinical utility, and reclassification improvement, including Harrell’s C-index, calibration slope, time-dependent areas under the curve (AUCs) at 5 years, 10 years, and 15 years with 95% confidence intervals (95%CI), calibration plots, decision-curve analysis (DCA), and the continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI). Detailed definitions and computation of performance metrics are described in the Supplementary Methods.
The baseline characteristics of the participants were summarized by variable type. Continuous variables were reported as medians with interquartile ranges, whereas categorical variables were summarized as counts and percentages. All statistical analyses and visualizations were performed using R software (version 4.4.0). This study was reported in accordance with the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis Plus Artificial Intelligence (TRIPOD + AI) statement[23].
To assess potential reverse causation, we excluded all participants with an observed follow-up time of less than 2 years, including CRC diagnoses occurring within the first 2 years after baseline, and repeated the MetRS association and model-discrimination analyses (Supplementary Table 7).
A total of 468095 participants were ultimately included in this study, and their baseline characteristics are presented in Table 1. The median age of the participants was 58 years (interquartile range: 50-63 years), and 256073 (54.7%) were women. Over a median follow-up of 14.7 years (interquartile range, 13.89-15.42 years), 7150 participants (1.53%) were diagnosed with CRC.
Among the 325 metabolites examined, 115 were significantly associated with CRC (Figure 1 and Supplementary Table 3). Among the plasma metabolites examined, significant associations with incident CRC were mainly observed for markers related to fatty acid composition, inflammation, glycolysis and ketone body metabolism, and lipoprotein-related measures. In contrast, conventional lipid traits, including total cholesterol, low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), apolipoprotein A1, and apolipoprotein B, showed no clear overall associations. Overall, glycoprotein acetyls (GlycA) showed the strongest positive association with CRC risk (HR = 1.102, 95%CI: 1.075-1.131, Q = 9.51 × 10-12), highlighting systemic inflammation as the most prominent metabolic feature associated with higher CRC risk, whereas polyunsaturated fatty acids to total fatty acids percentage showed the strongest inverse association (HR = 0.924, 95%CI: 0.902-0.946, Q = 1.22 × 10-8). Fatty acid-related measures displayed the clearest and most consistent pattern. Higher fatty acid unsaturation and higher proportions of polyunsaturated fatty acids were generally associated with lower CRC risk, whereas monounsaturated and saturated fatty acid-related measures tended to be associated with higher risk. Among amino acids, the overall associations were relatively modest, with histidine (His) showing the strongest inverse association (HR = 0.936, 95%CI: 0.913-0.960, Q = 9.70 × 10-6). For energy metabolism-related traits, metabolites related to glycolysis and ketone body metabolism generally showed positive associations with risk. Among these, 3-hydroxybutyrate (bOHbutyrate) showed the strongest positive association (HR = 1.077, 95%CI: 1.055-1.099, Q = 3.04 × 10-10), and positive associations were also observed for glucose, acetoacetate, and acetone. Lipoprotein-related analyses further revealed marked heterogeneity. Overall, triglyceride-rich very-low-density lipoprotein (VLDL)-related measures and concentrations of very large or extremely large VLDL particles tended to be positively associated with CRC risk, whereas several relative measures reflecting cholesterol composition within VLDL particles showed inverse associations. For example, concentration of chylomicrons and extremely large VLDL particles was significantly positively associated with CRC risk (HR = 1.043, 95%CI: 1.019-1.068, Q = 2.54 × 10-3).
Subgroup analyses showed different association patterns across sex and age groups (Supplementary Tables 4 and 5 and Supplementary Figures 2 and 3). In sex-stratified analyses, the most pronounced difference was observed for creatinine, which was significantly associated with a higher CRC risk in women (HR = 1.093, 95%CI: 1.062-1.126, Q = 7.61 × 10-7) but showed an inverse association in men (HR = 0.957, 95%CI: 0.923-0.992, P = 0.015). In addition, positive associations for triglyceride-rich VLDL subclasses and their lipid components were more evident in women, whereas inverse associations for amino acid-related traits were more apparent in men. In age-stratified analyses, the most evident differences were concentrated in cholesterol-related measures. In particular, total cholesterol minus HDL-C (non_HDL_C) was positively associated with CRC risk in younger participants (HR = 1.049, 95%CI: 1.006-1.093, Q = 0.043), but showed an inverse association in older participants (HR = 0.969, 95%CI: 0.942-0.997, P = 0.031). Similar patterns were also observed for total cholesterol, LDL-C, and remnant cholesterol, which tended to be positively associated with CRC risk in younger participants but were attenuated or reversed in older participants, suggesting substantial heterogeneity in CRC-related metabolic profiles across age groups.
Among the 115 candidate metabolic indicators identified in the preliminary screening, LightGBM ultimately selected seven metabolic features for construction of the MetRS: Linoleic acid to total fatty acids percentage (LA_pct), 3-bOHbutyrate, albumin, glucose, cholesteryl esters to cholesterol in very small VLDL percentage (XS_VLDL_CE_pct_C), His, and omega-6 fatty acids to total fatty acids percentage (Omega_6_pct) (Supplementary Figure 4A). As the metabolic features were sequentially incorporated according to their importance, model performance gradually improved and reached a plateau after inclusion of these seven features. The resulting MetRS was used for subsequent risk stratification and predictive performance evaluation. Stability analysis across folds showed that these seven features were selected consistently across the 10 IECV folds: LA_pct, bOHbutyrate, albumin, and glucose were selected in all 10 folds; His was selected in 9 folds; and XS_VLDL_CE_pct_C and Omega_6_pct were each selected in 8 folds (Supplementary Table 8).
We further used a Shapley additive explanations summary plot to visualize the contributions of the selected metabolic features to the model (Supplementary Figure 4B). Among them, LA_pct ranked first in feature importance. The Shapley additive explanations plot showed that higher levels of LA_pct were associated with a lower predicted risk of CRC, whereas lower levels were associated with a higher predicted risk. The remaining metabolic features also exhibited distinct contribution patterns, suggesting that the direction and magnitude of their effects on predicted risk varied across features.
We next evaluated the ability of the MetRS to stratify CRC risk and examined its performance across different subgroups. Overall, a higher baseline MetRS percentile was associated with a higher CRC event rate. In addition, men and older participants had higher overall event rates (Figure 2A). The Kaplan-Meier curves showed clear separation in cumulative CRC risk across MetRS tertiles, with the high MetRS group exhibiting the highest cumulative risk (Figure 2B). To improve clinical interpretation, we also reported absolute CRC risk estimates at 5 years, 10 years, and 15 years. Cumulative CRC risk increased progressively from the low to the high MetRS tertile. The 15-year cumulative CRC risk was 0.98% in the low MetRS tertile, 1.50% in the middle tertile, and 2.29% in the high tertile (Supplementary Table 9). The results of the subgroup analyses are shown in the forest plot (Figure 2C). Compared with the low MetRS group, the HR point estimate for the high MetRS group was higher among participants younger than 60 years (HR = 1.80, 95%CI: 1.62-2.00) than among those aged 60 years or older (HR = 1.62, 95%CI: 1.49-1.77). The association between MetRS and CRC risk was generally similar in men and women. After stratification by genetic risk, the HR point estimate was highest in the high PRS subgroup, with an HR of 1.59 (95%CI: 1.45-1.75) for the high MetRS group compared with the low MetRS group.
We next evaluated the predictive value of the MetRS for future CRC risk. Figure 3 shows the region-specific and pooled estimates of Harrell’s C-index and calibration slope for each model. QCancer + PRS + MetRS showed the highest, although modestly improved, discrimination, with a Harrell’s C-index of 0.711 (95%CI: 0.703-0.719), followed by QCancer + PRS at 0.705 (95%CI: 0.696-0.714), QCancer + MetRS at 0.683 (95%CI: 0.674-0.692), and QCancer at 0.676 (95%CI: 0.667-0.684). Compared with QCancer + PRS, QCancer + PRS + MetRS showed a mean paired regional increase in Harrell’s C-index of 0.006 (95%CI: 0.004-0.009), indicating a small but consistent improvement (Supplementary Table 10).
Time-dependent AUC analysis further showed that the addition of PRS or MetRS improved model performance. Among all models, the QCancer + PRS + MetRS model achieved the highest AUC at 5 years, 10 years, and 15 years, with values of 0.732 (95%CI: 0.719-0.745), 0.724 (95%CI: 0.711-0.736), and 0.732 (95%CI: 0.721-0.743), respectively (Supplementary Figure 5). The calibration slopes of all models were close to 1, and calibration plots generally showed acceptable agreement between predicted and observed risks across the four models (Supplementary Figure 6). DCA showed that, across most examined threshold probabilities, the QCancer + PRS + MetRS model yielded the greatest net benefit and outperformed the “treat all” strategy (Figure 4). In practical terms, this suggests that the integrated model may support risk-adapted CRC screening by prioritizing individuals at higher predicted risk for further screening or colonoscopy referral, while potentially reducing unnecessary procedures among those at lower predicted risk.
To further evaluate the improvement in predictive performance after adding PRS or MetRS, we compared the continuous NRI and IDI of each model at 5 years, 10 years, and 15 years (Supplementary Figure 7). Compared with the QCancer model, the addition of PRS or MetRS improved both the continuous NRI and IDI at 5 years, 10 years, and 15 years, with a more pronounced improvement observed after adding PRS. The greatest improvement was observed when both PRS and MetRS were incorporated into the model.
Further stratified analyses showed that the addition of PRS improved reclassification for both event and non-event individuals. In contrast, the improvement associated with MetRS was mainly observed in the reclassification of non-event individuals. Compared with the QCancer + PRS model, the QCancer + PRS + MetRS model showed modest additional improvement in reclassification at 5 years, 10 years, and 15 years. This improvement was mainly driven by better classification of non-event individuals, suggesting that MetRS may contribute to refining risk estimates among individuals with lower predicted risk.
To assess the potential influence of reverse causation, we performed a sensitivity analysis excluding participants with less than 2 years of follow-up. After exclusion, 465194 participants and 6312 incident CRC cases remained in the analysis. The association between MetRS and incident CRC remained materially unchanged. Compared with participants in the low MetRS tertile, those in the middle and high MetRS tertiles had higher risks of CRC, with HRs of 1.21 (95%CI: 1.13-1.30) and 1.51 (95%CI: 1.41-1.62), respectively. Each 1-SD increment in MetRS was also associated with a higher risk of CRC (HR = 1.21, 95%CI: 1.18-1.25; Supplementary Table 7). Model discrimination was also similar to that observed in the main analysis, with Harrell’s C-index values of 0.672 for QCancer, 0.679 for QCancer + MetRS, 0.702 for QCancer + PRS, and 0.707 for QCancer + PRS + MetRS (Supplementary Figure 8).
In this study, based on data from 468095 UK Biobank participants, we systematically evaluated the associations between 325 NMR-based plasma metabolic measures and the risk of incident CRC, and further examined their potential value in CRC risk prediction. Our findings showed widespread associations between plasma metabolic measures and CRC risk, with 115 metabolic measures significantly associated with disease incidence. These associations were mainly related to pathways involving fatty acid composition, inflammation, glycolysis and ketone body metabolism, and lipoprotein remodeling. In contrast, conventional lipid traits, including total cholesterol, LDL-C, HDL-C, apolipoprotein A1, and apolipoprotein B, did not show clear overall associations. More importantly, integrating metabolomic information into prediction models based on clinical factors and genetic risk further improved the identification of high-risk individuals and provided additional information for CRC risk stratification.
Earlier work suggested that metabolite panels may have potential for detecting colorectal neoplasms[8], whereas prospective pre-diagnostic metabolomics studies have shown substantial heterogeneity across populations, platforms, and validation strategies[16]. More recently, a UK Biobank and ESTHER cohort analysis reported limited incremental value of metabolomics beyond established environmental and genetic risk factors for CRC prediction[13]. Consistent with this cautious view, MetRS showed clear risk stratification ability in our study, but its incremental improvement over the QCancer and PRS was modest. Notably, one of the additionally derived ratio measures, XS_VLDL_CE_pct_C, was selected into the final MetRS, suggesting that refined metabolic composition indicators may provide complementary information for CRC risk stratification.
Inflammation-related metabolic disturbance was one of the most prominent signals in this study. GlycA showed the strongest positive association with CRC risk, and albumin was also included in the final MetRS, suggesting that low-grade systemic inflammation and nutritional status may be altered before CRC diagnosis. This finding is consistent with previous prospective evidence linking higher GlycA levels to CRC incidence and mortality[24], as well as studies showing associations of albumin and inflammation-based markers with CRC outcomes[25,26].
Fatty acid composition also showed a clear and consistent pattern. polyunsaturated fatty acids to total fatty acids percentage and fatty acid unsaturation measures were inversely associated with CRC risk, whereas monounsaturated and saturated fatty acid-related measures tended to be positively associated. LA_pct ranked first in feature importance, and Omega_6_pct was also included in the final MetRS. These findings suggest that CRC risk may be related to the overall circulating fatty acid profile rather than a single fatty acid measure. Importantly, LA_pct and Omega_6_pct reflect relative circulating fatty acid composition rather than dietary omega-6 intake alone. Previous studies have reported inconsistent associations between omega-6 fatty acids and CRC risk, and dietary intake and circulating or tissue fatty acid levels may show different relationships with CRC[27,28]. Mechanistically, linoleic acid can serve as a precursor for arachidonic acid-derived eicosanoids involved in inflammation and tumor microenvironment remodeling[29,30]. However, human evidence does not support a simple pro-inflammatory or pro-tumorigenic interpretation of omega-6 fatty acids[31,32]. Therefore, the inverse associations of LA_pct and Omega_6_pct observed in our study may reflect a more favorable circulating fatty acid composition and systemic metabolic status rather than a direct causal protective effect.
Glucose and bOHbutyrate were both included in the MetRS, further highlighting the potential role of altered energy metabolism in CRC development. Previous studies have shown that diabetes and higher blood glucose levels are associated with increased CRC risk[33,34]. In contrast, the biological role of bOHbutyrate appears to be more complex. Some studies suggest that it may suppress intestinal tumor growth[35], whereas others indicate that it may promote CRC cell proliferation and metastasis[36]. Therefore, in the prospective population setting of this study, the positive association of bOHbutyrate is more likely to reflect changes in substrate use and systemic metabolic stress before the onset of CRC. Meanwhile, the protective association of His was consistent with previous prospective findings[37]. The inclusion of XS_VLDL_CE_pct_C in the final model may further reflect lipoprotein remodeling in relation to CRC risk. XS_VLDL_CE_pct_C denotes the cholesteryl ester proportion within very small VLDL particles. Triglyceride-rich lipoproteins and their remnants undergo dynamic intravascular remodeling, during which remnant particles may become relatively enriched in cholesteryl esters and interact with inflammatory and oxidative pathways[38]. In CRC-related epidemiological studies, dyslipidemia and hypertriglyceridemia have been associated with CRC risk, although findings for individual lipid fractions remain heterogeneous[39-41]. Therefore, the selection of XS_VLDL_CE_pct_C should be interpreted as a marker of altered lipoprotein composition and metabolic remodeling rather than evidence for a direct biological effect of this single VLDL fraction. In addition, the subgroup analyses suggested that these metabolic associations may differ across population subgroups. For example, creatinine was positively associated with CRC risk in women but showed an inverse association in men. Total cholesterol minus HDL-C and several other cholesterol-related measures tended to be associated with higher risk in younger participants, whereas these associations were weaker or even reversed in older participants. These findings suggest that CRC-related metabolic features may vary across popu
By integrating these key metabolic measures, we constructed the MetRS and found that it could reliably distinguish different levels of CRC risk. Higher MetRS percentiles were associated with a higher cumulative incidence of CRC, and the Kaplan-Meier curves showed clear separation across MetRS groups. In subgroup analyses, the HR point estimate for high vs low MetRS was higher among participants younger than 60 years and was numerically highest in the high PRS subgroup, whereas the associations were generally similar in men and women. Although the improvement in model discrimination after adding MetRS to QCancer and PRS was overall modest, the direction of improvement was consistent across Harrell’s C-index, time-dependent AUC, NRI, and IDI. Therefore, the incremental value of MetRS should be interpreted as complementary rather than transformative. Its potential clinical utility may lie in refining risk stratification among individuals already assessed by clinical and genetic factors, rather than replacing existing screening strategies. This interpretation is also supported by the DCA results, which suggest that the integrated model may be useful for risk-adapted screening decisions rather than universal intensification of screening. Given the additional cost and logistical requirements of metabolomic profiling, future studies should evaluate its cost-effectiveness and feasibility before routine implementation. Further gains were still observed when MetRS was added on top of QCancer and PRS, suggesting that circulating metabolic features do provide information beyond traditional risk factors and genetic susceptibility. Notably, the reclassification improvement provided by MetRS was mainly seen in non-event individuals, which suggests that its potential value may lie more in improving the identification of individuals at low or intermediate risk.
This study also has several strengths. First, NMR metabolomics enabled us to systematically assess a large number of circulating metabolic measures at the same time. Second, the large sample size and long follow-up increased statistical power and reduced the possibility of reverse causation. Third, we combined machine-learning-based feature selection with the construction of the MetRS, and further evaluated its risk stratification and predictive performance across different subgroups, providing a more complete assessment of the potential value of metabolomics in CRC risk pre
This study also has several limitations. First, the current NMR platform mainly captures lipid- and lipoprotein-related measures, and future studies should combine complementary metabolomics technologies to expand metabolite coverage. Second, although IECV assessed geographical transportability within the UK Biobank, it cannot replace validation in fully independent external cohorts. Because UK Biobank participants are generally healthier, less socioeconomically diverse, and predominantly White, and because the number of CRC events among non-White participants was small, external validation in more diverse populations is required before clinical implementation. Third, the main prediction models used a cause-specific Cox framework, with death treated as censoring. Although this approach was appropriate for evaluating relative risk ranking and was consistent with the adapted QCancer framework, competing mortality may influence absolute risk estimation, calibration, and DCA. Future validation should compare the present findings with competing-risk approaches, including Fine-Gray models. Fourth, the LightGBM feature-selection step used a binary incident CRC outcome and did not explicitly account for censoring or differences in follow-up time; future studies should compare this two-stage framework with survival-aware machine-learning approaches. Fifth, the feature-selection process was not fully nested within the IECV, and some degree of optimism in the performance estimates cannot be excluded. Although fold-wise analyses supported the stability of the seven-metabolite panel, fully nested evaluation and independent external validation remain warranted. Finally, this observational study cannot establish causality. Although the results remained similar after excluding participants with less than 2 years of follow-up, reverse causation cannot be completely excluded.
In this large prospective cohort, plasma metabolic features related to fatty acid composition, inflammation, energy metabolism, and lipoprotein remodeling were associated with incident CRC. The MetRS provided complementary information beyond clinical factors and polygenic risk, with modest improvements in risk stratification and predictive performance. Independent external validation, prospective clinical-impact assessment, and cost-effectiveness evaluation are required before the score can be considered for routine screening practice.
The authors sincerely thank all the participants and professionals involved in the UK Biobank.
| 1. | Bray F, Laversanne M, Sung H, Ferlay J, Siegel RL, Soerjomataram I, Jemal A. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2024;74:229-263. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 16785] [Cited by in RCA: 17925] [Article Influence: 8962.5] [Reference Citation Analysis (41)] |
| 2. | Morgan E, Arnold M, Gini A, Lorenzoni V, Cabasag CJ, Laversanne M, Vignat J, Ferlay J, Murphy N, Bray F. Global burden of colorectal cancer in 2020 and 2040: incidence and mortality estimates from GLOBOCAN. Gut. 2023;72:338-344. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 1885] [Cited by in RCA: 1757] [Article Influence: 585.7] [Reference Citation Analysis (15)] |
| 3. | Cardoso R, Guo F, Heisser T, Hackl M, Ihle P, De Schutter H, Van Damme N, Valerianova Z, Atanasov T, Májek O, Mužík J, Nilbert MC, Tybjerg AJ, Innos K, Mägi M, Malila N, Bouvier AM, Bouvier V, Launoy G, Woronoff AS, Cariou M, Robaszkiewicz M, Delafosse P, Poncet F, Katalinic A, Walsh PM, Senore C, Rosso S, Vincerževskienė I, Lemmens VEPP, Elferink MAG, Johannesen TB, Kørner H, Pfeffer F, Bento MJ, Rodrigues J, Alves da Costa F, Miranda A, Zadnik V, Žagar T, Lopez de Munain Marques A, Marcos-Gragera R, Puigdemont M, Galceran J, Carulla M, Chirlaque MD, Ballesta M, Sundquist K, Sundquist J, Weber M, Jordan A, Herrmann C, Mousavi M, Ryzhov A, Hoffmeister M, Brenner H. Colorectal cancer incidence, mortality, and stage distribution in European countries in the colorectal cancer screening era: an international population-based study. Lancet Oncol. 2021;22:1002-1013. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 52] [Cited by in RCA: 353] [Article Influence: 70.6] [Reference Citation Analysis (4)] |
| 4. | Nishihara R, Wu K, Lochhead P, Morikawa T, Liao X, Qian ZR, Inamura K, Kim SA, Kuchiba A, Yamauchi M, Imamura Y, Willett WC, Rosner BA, Fuchs CS, Giovannucci E, Ogino S, Chan AT. Long-term colorectal-cancer incidence and mortality after lower endoscopy. N Engl J Med. 2013;369:1095-1105. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 1237] [Cited by in RCA: 1214] [Article Influence: 93.4] [Reference Citation Analysis (7)] |
| 5. | Rex DK, Imperiale TF, Latinovich DR, Bratcher LL. Impact of bowel preparation on efficiency and cost of colonoscopy. Am J Gastroenterol. 2002;97:1696-1700. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 483] [Cited by in RCA: 473] [Article Influence: 19.7] [Reference Citation Analysis (2)] |
| 6. | Stoop EM, de Haan MC, de Wijkerslooth TR, Bossuyt PM, van Ballegooijen M, Nio CY, van de Vijver MJ, Biermann K, Thomeer M, van Leerdam ME, Fockens P, Stoker J, Kuipers EJ, Dekker E. Participation and yield of colonoscopy versus non-cathartic CT colonography in population-based screening for colorectal cancer: a randomised controlled trial. Lancet Oncol. 2012;13:55-64. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 265] [Cited by in RCA: 279] [Article Influence: 19.9] [Reference Citation Analysis (3)] |
| 7. | Keswani RN, Crockett SD, Calderwood AH. AGA Clinical Practice Update on Strategies to Improve Quality of Screening and Surveillance Colonoscopy: Expert Review. Gastroenterology. 2021;161:701-711. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 137] [Cited by in RCA: 125] [Article Influence: 25.0] [Reference Citation Analysis (4)] |
| 8. | Erben V, Bhardwaj M, Schrotz-King P, Brenner H. Metabolomics Biomarkers for Detection of Colorectal Neoplasms: A Systematic Review. Cancers (Basel). 2018;10:246. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 38] [Cited by in RCA: 55] [Article Influence: 6.9] [Reference Citation Analysis (1)] |
| 9. | Gold A, Choueiry F, Jin N, Mo X, Zhu J. The Application of Metabolomics in Recent Colorectal Cancer Studies: A State-of-the-Art Review. Cancers (Basel). 2022;14:725. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 9] [Cited by in RCA: 48] [Article Influence: 12.0] [Reference Citation Analysis (1)] |
| 10. | Chen L, Zhernakova DV, Kurilshikov A, Andreu-Sánchez S, Wang D, Augustijn HE, Vich Vila A; Lifelines Cohort Study, Weersma RK, Medema MH, Netea MG, Kuipers F, Wijmenga C, Zhernakova A, Fu J. Influence of the microbiome, diet and genetics on inter-individual variation in the human plasma metabolome. Nat Med. 2022;28:2333-2343. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 127] [Cited by in RCA: 256] [Article Influence: 64.0] [Reference Citation Analysis (5)] |
| 11. | Vojinovic D, Radjabzadeh D, Kurilshikov A, Amin N, Wijmenga C, Franke L, Ikram MA, Uitterlinden AG, Zhernakova A, Fu J, Kraaij R, van Duijn CM. Relationship between gut microbiota and circulating metabolites in population-based cohorts. Nat Commun. 2019;10:5813. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 232] [Cited by in RCA: 208] [Article Influence: 29.7] [Reference Citation Analysis (1)] |
| 12. | Schmidt DR, Patel R, Kirsch DG, Lewis CA, Vander Heiden MG, Locasale JW. Metabolomics in cancer research and emerging applications in clinical oncology. CA Cancer J Clin. 2021;71:333-358. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 491] [Cited by in RCA: 494] [Article Influence: 98.8] [Reference Citation Analysis (2)] |
| 13. | Seum T, Cardoso R, Stevenson-Hoare J, Holleczek B, Schöttker B, Hoffmeister M, Brenner H. Exploring metabolomics for colorectal cancer risk prediction: evidence from the UK Biobank and ESTHER cohorts. BMC Med. 2025;23:283. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 5] [Reference Citation Analysis (0)] |
| 14. | Cross AJ, Moore SC, Boca S, Huang WY, Xiong X, Stolzenberg-Solomon R, Sinha R, Sampson JN. A prospective study of serum metabolites and colorectal cancer risk. Cancer. 2014;120:3049-3057. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 88] [Cited by in RCA: 97] [Article Influence: 8.1] [Reference Citation Analysis (4)] |
| 15. | Shu X, Xiang YB, Rothman N, Yu D, Li HL, Yang G, Cai H, Ma X, Lan Q, Gao YT, Jia W, Shu XO, Zheng W. Prospective study of blood metabolites associated with colorectal cancer risk. Int J Cancer. 2018;143:527-534. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 24] [Cited by in RCA: 51] [Article Influence: 6.4] [Reference Citation Analysis (3)] |
| 16. | Seum T, Frick C, Cardoso R, Bhardwaj M, Hoffmeister M, Brenner H. Potential of pre-diagnostic metabolomics for colorectal cancer risk assessment or early detection. NPJ Precis Oncol. 2024;8:244. [RCA] [PubMed] [DOI] [Full Text] [Cited by in RCA: 6] [Reference Citation Analysis (0)] |
| 17. | Yuan F, Jia G, Wen W, Xu S, Gunchick V, Deng K, Long J, Yu D, Shu XO, Zheng W. Blood metabolic biomarkers and colorectal cancer risk: results from large prospective cohort and Mendelian randomisation analyses. Br J Cancer. 2025;133:94-103. [RCA] [PubMed] [DOI] [Full Text] [Cited by in RCA: 9] [Reference Citation Analysis (0)] |
| 18. | Hippisley-Cox J, Coupland C. Development and validation of risk prediction algorithms to estimate future risk of common cancers in men and women: prospective cohort study. BMJ Open. 2015;5:e007825. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 84] [Cited by in RCA: 115] [Article Influence: 10.5] [Reference Citation Analysis (0)] |
| 19. | Sun J, Liu Y, Zhao J, Lu B, Zhou S, Lu W, Wei J, Hu Y, Kong X, Gao J, Guan H, Gao J, Xiao Q, Li X. Plasma proteomic and polygenic profiling improve risk stratification and personalized screening for colorectal cancer. Nat Commun. 2024;15:8873. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 26] [Cited by in RCA: 38] [Article Influence: 19.0] [Reference Citation Analysis (1)] |
| 20. | Briggs SEW, Law P, East JE, Wordsworth S, Dunlop M, Houlston R, Hippisley-Cox J, Tomlinson I. Integrating genome-wide polygenic risk scores and non-genetic risk to predict colorectal cancer diagnosis using UK Biobank data: population based cohort study. BMJ. 2022;379:e071707. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 60] [Cited by in RCA: 41] [Article Influence: 10.3] [Reference Citation Analysis (2)] |
| 21. | Thompson DJ, Wells D, Selzam S, Peneva I, Moore R, Sharp K, Tarran WA, Beard EJ, Riveros-Mckay F, Giner-Delgado C, Palmer D, Seth P, Harrison J, Futema M; Genomics England Research Consortium, McVean G, Plagnol V, Donnelly P, Weale ME. A systematic evaluation of the performance and properties of the UK Biobank Polygenic Risk Score (PRS) Release. PLoS One. 2024;19:e0307270. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 132] [Reference Citation Analysis (0)] |
| 22. | Steyerberg EW, Harrell FE Jr. Prediction models need appropriate internal, internal-external, and external validation. J Clin Epidemiol. 2016;69:245-247. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 427] [Cited by in RCA: 952] [Article Influence: 86.5] [Reference Citation Analysis (1)] |
| 23. | Collins GS, Moons KGM, Dhiman P, Riley RD, Beam AL, Van Calster B, Ghassemi M, Liu X, Reitsma JB, van Smeden M, Boulesteix AL, Camaradou JC, Celi LA, Denaxas S, Denniston AK, Glocker B, Golub RM, Harvey H, Heinze G, Hoffman MM, Kengne AP, Lam E, Lee N, Loder EW, Maier-Hein L, Mateen BA, McCradden MD, Oakden-Rayner L, Ordish J, Parnell R, Rose S, Singh K, Wynants L, Logullo P. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 2024;385:e078378. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 1587] [Cited by in RCA: 2548] [Article Influence: 1274.0] [Reference Citation Analysis (12)] |
| 24. | Chandler PD, Akinkuolie AO, Tobias DK, Lawler PR, Li C, Moorthy MV, Wang L, Duprez DA, Jacobs DR, Glynn RJ, Otvos J, Connelly MA, Post WS, Ridker PM, Manson JE, Buring JE, Lee IM, Mora S. Association of N-Linked Glycoprotein Acetyls and Colorectal Cancer Incidence and Mortality. PLoS One. 2016;11:e0165615. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 25] [Cited by in RCA: 34] [Article Influence: 3.4] [Reference Citation Analysis (0)] |
| 25. | Sheinenzon A, Shehadeh M, Michelis R, Shaoul E, Ronen O. Serum albumin levels and inflammation. Int J Biol Macromol. 2021;184:857-862. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 337] [Cited by in RCA: 286] [Article Influence: 57.2] [Reference Citation Analysis (2)] |
| 26. | Zhou QP, Li XJ. C-Reactive Protein to Albumin Ratio in Colorectal Cancer: A Meta-Analysis of Prognostic Value. Dose Response. 2019;17:1559325819889814. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 20] [Cited by in RCA: 33] [Article Influence: 4.7] [Reference Citation Analysis (1)] |
| 27. | Atashi N, Eshaghian N, Anjom-Shoae J, Askari G, Asadi M, Sadeghi O. Dietary intake and tissue biomarkers of omega-6 fatty acids and risk of colorectal cancer in adults: a systematic review and dose-response meta-analysis of prospective cohort studies. Nutr Diabetes. 2025;15:17. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 4] [Cited by in RCA: 5] [Article Influence: 5.0] [Reference Citation Analysis (4)] |
| 28. | Lu Y, Li D, Wang L, Zhang H, Jiang F, Zhang R, Xu L, Yang N, Dai S, Xu X, Theodoratou E, Li X. Comprehensive Investigation on Associations between Dietary Intake and Blood Levels of Fatty Acids and Colorectal Cancer Risk. Nutrients. 2023;15:730. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 47] [Cited by in RCA: 43] [Article Influence: 14.3] [Reference Citation Analysis (1)] |
| 29. | Tuncer S, Banerjee S. Eicosanoid pathway in colorectal cancer: Recent updates. World J Gastroenterol. 2015;21:11748-11766. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in CrossRef: 38] [Cited by in RCA: 48] [Article Influence: 4.4] [Reference Citation Analysis (0)] |
| 30. | Wang Y, Wang W, Sanidad KZ, Shih PA, Zhao X, Zhang G. Eicosanoid signaling in carcinogenesis of colorectal cancer. Cancer Metastasis Rev. 2018;37:257-267. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 23] [Cited by in RCA: 29] [Article Influence: 3.6] [Reference Citation Analysis (0)] |
| 31. | Innes JK, Calder PC. Omega-6 fatty acids and inflammation. Prostaglandins Leukot Essent Fatty Acids. 2018;132:41-48. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 803] [Cited by in RCA: 682] [Article Influence: 85.3] [Reference Citation Analysis (12)] |
| 32. | Kim Y, Kim J. N-6 Polyunsaturated Fatty Acids and Risk of Cancer: Accumulating Evidence from Prospective Studies. Nutrients. 2020;12:2523. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 37] [Cited by in RCA: 33] [Article Influence: 5.5] [Reference Citation Analysis (2)] |
| 33. | Nilsen TI, Vatten LJ. Prospective study of colorectal cancer risk and physical activity, diabetes, blood glucose and BMI: exploring the hyperinsulinaemia hypothesis. Br J Cancer. 2001;84:417-422. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 193] [Cited by in RCA: 190] [Article Influence: 7.6] [Reference Citation Analysis (0)] |
| 34. | Cheng HC, Chang TK, Su WC, Tsai HL, Wang JY. Narrative review of the influence of diabetes mellitus and hyperglycemia on colorectal cancer risk and oncological outcomes. Transl Oncol. 2021;14:101089. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 48] [Cited by in RCA: 57] [Article Influence: 11.4] [Reference Citation Analysis (0)] |
| 35. | Dmitrieva-Posocco O, Wong AC, Lundgren P, Golos AM, Descamps HC, Dohnalová L, Cramer Z, Tian Y, Yueh B, Eskiocak O, Egervari G, Lan Y, Liu J, Fan J, Kim J, Madhu B, Schneider KM, Khoziainova S, Andreeva N, Wang Q, Li N, Furth EE, Bailis W, Kelsen JR, Hamilton KE, Kaestner KH, Berger SL, Epstein JA, Jain R, Li M, Beyaz S, Lengner CJ, Katona BW, Grivennikov SI, Thaiss CA, Levy M. β-Hydroxybutyrate suppresses colorectal cancer. Nature. 2022;605:160-165. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 285] [Cited by in RCA: 264] [Article Influence: 66.0] [Reference Citation Analysis (1)] |
| 36. | Mao T, Qin F, Zhang M, Li J, Li J, Lai M. Elevated serum β-hydroxybutyrate, a circulating ketone metabolite, accelerates colorectal cancer proliferation and metastasis via ACAT1. Oncogene. 2023;42:1889-1899. [RCA] [PubMed] [DOI] [Full Text] [Cited by in RCA: 26] [Reference Citation Analysis (0)] |
| 37. | Rothwell JA, Bešević J, Dimou N, Breeur M, Murphy N, Jenab M, Wedekind R, Viallon V, Ferrari P, Achaintre D, Gicquiau A, Rinaldi S, Scalbert A, Huybrechts I, Prehn C, Adamski J, Cross AJ, Keun H, Chadeau-Hyam M, Boutron-Ruault MC, Overvad K, Dahm CC, Nøst TH, Sandanger TM, Skeie G, Zamora-Ros R, Tsilidis KK, Eichelmann F, Schulze MB, van Guelpen B, Vidman L, Sánchez MJ, Amiano P, Ardanaz E, Smith-Byrne K, Travis R, Katzke V, Kaaks R, Derksen JWG, Colorado-Yohar S, Tumino R, Bueno-de-Mesquita B, Vineis P, Palli D, Pasanisi F, Eriksen AK, Tjønneland A, Severi G, Gunter MJ. Circulating amino acid levels and colorectal cancer risk in the European Prospective Investigation into Cancer and Nutrition and UK Biobank cohorts. BMC Med. 2023;21:80. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 17] [Cited by in RCA: 40] [Article Influence: 13.3] [Reference Citation Analysis (1)] |
| 38. | Ginsberg HN, Packard CJ, Chapman MJ, Borén J, Aguilar-Salinas CA, Averna M, Ference BA, Gaudet D, Hegele RA, Kersten S, Lewis GF, Lichtenstein AH, Moulin P, Nordestgaard BG, Remaley AT, Staels B, Stroes ESG, Taskinen MR, Tokgözoğlu LS, Tybjaerg-Hansen A, Stock JK, Catapano AL. Triglyceride-rich lipoproteins and their remnants: metabolic insights, role in atherosclerotic cardiovascular disease, and emerging therapeutic strategies-a consensus statement from the European Atherosclerosis Society. Eur Heart J. 2021;42:4791-4806. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 668] [Cited by in RCA: 738] [Article Influence: 147.6] [Reference Citation Analysis (7)] |
| 39. | Yao X, Tian Z. Dyslipidemia and colorectal cancer risk: a meta-analysis of prospective studies. Cancer Causes Control. 2015;26:257-268. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 68] [Cited by in RCA: 102] [Article Influence: 8.5] [Reference Citation Analysis (0)] |
| 40. | Hsu SH, Syu DK, Chen YC, Liu CK, Sun CA, Chen M. The Association between Hypertriglyceridemia and Colorectal Cancer: A Long-Term Community Cohort Study in Taiwan. Int J Environ Res Public Health. 2022;19:7804. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 1] [Cited by in RCA: 20] [Article Influence: 5.0] [Reference Citation Analysis (0)] |
| 41. | Yang Z, Tang H, Lu S, Sun X, Rao B. Relationship between serum lipid level and colorectal cancer: a systemic review and meta-analysis. BMJ Open. 2022;12:e052373. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 48] [Cited by in RCA: 45] [Article Influence: 11.3] [Reference Citation Analysis (1)] |