Abstract

Background:

Diabetes mellitus is increasingly recognized as a risk factor for malignancy, partly through chronic low-grade inflammation and insulin resistance. The composite C-reactive protein–triglyceride glucose (CRP–TyG) index integrates inflammatory and metabolic dimensions and may therefore provide a more informative marker for cancer risk stratification than either component alone, particularly in hospitalized patients with diabetes. However, its association with incident malignancy and its potential nonlinear pattern have not been well characterized in this population.

Methods:

This retrospective cohort study included 5,500 adult inpatients with diabetes between 2014 and 2025. The CRP–TyG composite index was calculated using standardized laboratory measurements obtained within 24–48 hours of admission. Incident malignancies were identified primarily through pathology-confirmed diagnoses and tumor registry records, with diagnostic coding used as supplementary evidence when necessary. Associations between the CRP–TyG index and malignancy risk were evaluated using multivariable Cox proportional hazards models, restricted cubic spline analyses for nonlinear dose–response assessment, and segmented regression to identify threshold effects. Stepwise covariate adjustment, subgroup analyses, landmark analyses, and additional sensitivity analyses were performed to test the robustness of the findings.

Results:

During follow-up, 344 incident malignancies were documented. Malignancy risk increased progressively across CRP–TyG quartiles, with the highest quartile showing a significantly elevated risk compared with the lowest quartile in the fully adjusted model (HR = 1.92, 95% CI: 1.42–2.60; P for trend < 0.001). A significant nonlinear association was observed, with an inflection point at a CRP–TyG z-score of 0.62, above which the risk increased more steeply. Compared with CRP or TyG alone, the composite index showed better discriminatory performance and demonstrated the strongest association with gastrointestinal cancers. The overall findings remained directionally consistent across subgroup and landmark analyses.

Conclusions:

The CRP–TyG composite index was independently associated with a higher risk of incident malignancy in hospitalized patients with diabetes and showed a clear nonlinear dose–response pattern. As a simple biomarker derived from routine laboratory tests, it may help support early malignancy risk stratification in this high-risk population.

1 Introduction

With global population aging and the continued shift toward modern sedentary lifestyles, the burden of diabetes mellitus has increased substantially worldwide. Beyond its role as a major metabolic disorder, diabetes is now increasingly recognized as a condition that contributes to a broader spectrum of chronic complications, including malignancy. Multiple large-scale epidemiological studies have shown that diabetes is independently associated with the risk of several cancer types (1, 2). This association is biologically plausible because diabetes and cancer share several pathophysiological features, particularly chronic low-grade inflammation, insulin resistance, and sustained metabolic dysregulation. Prolonged hyperglycemia can promote systemic inflammatory activation and the release of pro-inflammatory cytokines, thereby increasing oxidative stress, inducing DNA damage, and creating a microenvironment that may favor malignant transformation (3). In parallel, hyperinsulinemia secondary to insulin resistance may activate insulin and insulin-like growth factor signaling pathways, promoting abnormal cellular proliferation and clonal expansion. Taken together, inflammation and metabolic dysfunction provide an important biological basis for the excess malignancy risk observed in patients with diabetes.

Among patients with diabetes, hospitalized individuals represent a particularly heterogeneous yet risk-enriched subgroup. Compared with community-based or outpatient populations, hospitalized patients with diabetes often have poorer glycemic control, a heavier burden of complications, and more complex treatment regimens (4). They are also more likely to experience acute physiological stress during admission, which may further intensify both inflammatory activation and metabolic disturbance. At the same time, the inpatient setting provides a unique opportunity for clinical research because standardized and relatively high-density laboratory testing is routinely performed during the early stage of hospitalization. This allows more precise capture of baseline inflammatory and metabolic profiles at cohort entry. Studying malignancy risk in this specific population is therefore meaningful not only because it addresses an underexplored area in diabetes-associated cancer prevention, but also because it may help clinicians identify higher-risk individuals in complex inpatient settings (5).

With respect to candidate biomarkers, C-reactive protein (CRP) is one of the most widely used indicators of systemic inflammation and has been linked to both cancer occurrence and prognosis (6). The triglyceride-glucose (TyG) index, in contrast, has gained increasing attention as a practical surrogate for insulin resistance because it is inexpensive, easy to calculate, and does not require direct insulin measurement. However, carcinogenesis in patients with diabetes is unlikely to be driven by a single inflammatory or metabolic pathway alone. A marker that reflects only inflammation or only insulin resistance may therefore be insufficient to capture the broader biological burden relevant to malignancy development.

Against this background, the composite CRP–TyG index may offer a more integrated way to characterize the inflammatory–metabolic axis. Nevertheless, evidence on its association with incident malignancy in hospitalized patients with diabetes remains limited, and it is still unclear whether this relationship is linear across the exposure range or whether a threshold-like pattern exists. Therefore, in this study, we analyzed a large inpatient cohort of adults with diabetes to evaluate the association between the CRP–TyG composite index and the risk of incident malignancy (7, 8). We further used restricted cubic spline models to explore potential nonlinear dose–response patterns, examined whether a meaningful threshold effect could be identified, and assessed the robustness of the observed association through subgroup and sensitivity analyses. We hypothesized that the CRP–TyG composite index would show a positive but non-simple linear relationship with malignancy risk, with risk increasing more sharply once the composite burden exceeded a clinically relevant threshold.

2 Methods2.1 Study design

This retrospective, real-world cohort study was based on electronic medical record data from hospitalized adults with diabetes at Tongren Hospital, Shanghai Jiao Tong University School of Medicine. The index hospitalization was defined as the first eligible diabetes-related inpatient admission for each individual during the study period from January 1, 2014, to January 31, 2025. The primary objective was to evaluate the association between the composite C-reactive protein–triglyceride glucose (CRP–TyG) index and subsequent incident malignancy, with particular attention to potential nonlinear dose–response patterns.

Clinical data were extracted from the hospital information system, laboratory information system, discharge diagnosis database, pathology database, and institutional tumor registration records. These data sources were linked at the individual level using unique patient identifiers to ensure consistency in exposure ascertainment, covariate definition, follow-up, and outcome verification. Baseline demographic characteristics, comorbidities, medication use, and laboratory indicators were collected from the index hospitalization. The overall study design and analytical framework are presented in Figure 1.

Study design and analytical strategy for assessing the CRP-TyG composite index and malignancy risk in hospitalized patients with diabetes.

To ensure temporal consistency of exposure measurement, baseline CRP (or hsCRP), fasting plasma glucose, and triglyceride values used to construct the CRP–TyG index were obtained within 24–48 hours after admission. This fixed ascertainment window was chosen to reduce measurement heterogeneity related to prolonged hospitalization, treatment modification, or late in-hospital complications. Because occult malignancy present at admission might influence inflammatory or metabolic biomarkers, prespecified landmark analyses were incorporated to reduce the likelihood of reverse causation. In these analyses, patients diagnosed with malignancy within 90 days and 180 days after the index hospitalization were excluded in separate sensitivity models.

The primary endpoint was first incident malignancy occurring after cohort entry. Malignancy events were identified primarily through pathology-confirmed diagnoses and institutional tumor registry records, with standardized diagnostic coding used as supplementary evidence when direct pathological documentation or registry confirmation was unavailable. For site-specific analyses, malignant tumors were further classified by major organ systems. The main analytical strategy included multivariable time-to-event modeling, restricted cubic spline analyses to assess nonlinear associations, threshold analyses using segmented regression, subgroup interaction analyses, and multiple sensitivity analyses. Because death from non-cancer causes could preclude the occurrence of the primary endpoint, competing-risk analyses were additionally performed to examine the robustness of the main findings (9).

2.2 Participants

Adult inpatients with diabetes who were admitted between January 2014 and January 2025 were screened for eligibility. Diabetes was identified according to documented clinical diagnosis, discharge coding, antidiabetic treatment records, or laboratory criteria consistent with routine clinical practice at the study institution. To avoid within-person correlation caused by repeated hospitalizations, only the first eligible admission for each patient was retained as the index hospitalization.

Patients were included if they were aged 18 years or older and had available baseline measurements of CRP (or hsCRP), fasting plasma glucose, and triglycerides within 24–48 hours after admission. These measurements were required to calculate the TyG index and construct the composite CRP–TyG exposure. The participant selection process is summarized in Figure 2.

Participant selection flowchart: inclusion/exclusion, index hospitalization, and malignancy ascertainment.

Patients were excluded for the following reasons. First, individuals with any documented malignant tumor before or at the time of the index hospitalization were excluded, including patients with pre-existing pathology-confirmed cancer, tumor registry records, or prior diagnostic documentation indicating malignancy. This step ensured that the outcome represented newly identified malignancy after cohort entry rather than prevalent cancer. Second, patients with missing core exposure variables were excluded, including those lacking CRP/hsCRP, fasting plasma glucose, or triglyceride measurements during the baseline assessment window. Third, patients younger than 18 years, those with duplicate records, invalid identifiers, implausible laboratory values after quality-control review, or incomplete follow-up information were excluded.

After applying these criteria, 5,500 patients were included in the primary analysis cohort. During follow-up, 344 patients developed incident malignancy and 5,156 remained free of malignancy. For site-specific analyses, malignancies were classified into major organ-system categories, including gastrointestinal, respiratory, urinary, reproductive, and hematological cancers. Given the limited number of events in some site-specific categories, these analyses were treated as exploratory and were modeled using a reduced set of covariates to avoid overfitting and to better respect the events-per-variable principle.

2.3 Exposure definition

The primary exposure was the composite CRP–TyG index, designed to capture the combined burden of systemic inflammation and insulin resistance in hospitalized patients with diabetes. All exposure variables were derived from the first available laboratory measurements obtained within 24–48 hours after the index hospitalization.

The TyG index was calculated using the conventional formula:

When triglycerides or fasting plasma glucose were reported in SI units, they were converted to mg/dL before calculation using standard conversion factors. Because CRP typically shows a right-skewed distribution, CRP values were natural log-transformed after adding a small constant when needed to accommodate values at or near zero, using ln(CRP + 0.1).

To construct the composite exposure, the TyG index and log-transformed CRP were standardized to z scores within the analytic cohort, and the primary CRP–TyG index was defined as the sum of these two standardized components:

This additive approach was prespecified to avoid scale imbalance between the inflammatory and metabolic components and to facilitate interpretation across models. In the main analyses, the composite index was modeled both as a continuous variable and as quartiles. Quartile-based analyses were used for descriptive comparisons and trend testing, whereas the continuous form was used to assess nonlinear associations.

To examine whether the observed associations depended on the specific method used to construct the composite index, alternative exposure definitions were evaluated in sensitivity analyses. These included a multiplicative form based on standardized ln(CRP) × standardized TyG and an interaction-term model including ln(CRP), TyG, and their product term. These analyses were used to assess the robustness of the primary exposure definition rather than to replace it.

2.4 Outcome definition

The primary outcome was first incident malignant tumor diagnosed after the index hospitalization. Outcome ascertainment followed a hierarchical verification strategy. Pathology-confirmed malignancy and institutional tumor registry records were used as the primary basis for event confirmation. When these sources were unavailable, standardized diagnostic codes consistent with malignant neoplasms were used as supplementary evidence, provided that the diagnostic information was clinically coherent and traceable within the medical record system.

For patients with multiple malignancy-related records during follow-up, only the earliest confirmed event date was retained as the incident event. Malignant tumors were additionally grouped by major organ systems for exploratory site-specific analyses, including gastrointestinal, respiratory, urinary, reproductive, and hematological malignancies.

Follow-up began on the date of index hospitalization and continued until the first occurrence of one of the following: incident malignancy, death from a non-cancer cause, last available clinical record, or administrative end of follow-up on January 31, 2025, whichever came first. Patients who did not develop malignancy were censored at the date of death, loss to follow-up, or administrative study end. The median follow-up duration for the cohort was 4.2 years (interquartile range, 2.1–6.8 years), and the total follow-up time was 22,950 person-years.

Because death from causes other than malignancy could prevent the occurrence of the primary endpoint, non-cancer death was treated as a competing event in secondary analyses using Fine–Gray subdistribution hazard models. In addition, prespecified landmark analyses were conducted by excluding malignancies diagnosed within 90 days and 180 days after index hospitalization to reduce potential reverse causation arising from occult cancers already present at baseline.

2.5 Covariates and confounding control plan

Covariates were selected a priori based on clinical relevance, biological plausibility, and prior literature regarding factors that could be associated with both CRP–TyG levels and malignancy risk. These covariates were grouped into demographic characteristics, metabolic status, organ function indicators, inflammatory burden, comorbidities, and medication use.

Demographic variables included age and sex. Anthropometric and metabolic indicators included body mass index (BMI), glycated hemoglobin (HbA1c), and mean fasting glucose during the first 72 hours when needed to supplement assessment of in-hospital glycemic status. Organ function markers included serum creatinine, estimated glomerular filtration rate, alanine aminotransferase, aspartate aminotransferase, and albumin. To reduce confounding by acute inflammatory states that might distort CRP levels independently of long-term metabolic status, white blood cell count was included as an inflammatory surrogate, and hospital-recorded infection diagnoses and/or anti-infective treatment during the index admission were incorporated when available.

Comorbidity adjustment included chronic kidney disease, cardiovascular disease, hypertension, dyslipidemia, and other clinically relevant chronic conditions documented before or during the index hospitalization. Medication variables included metformin, insulin, statins, and other major glucose-lowering therapies when available. Because some of these medications may influence both metabolic parameters and cancer-related pathways, they were treated as important confounders rather than downstream intermediates.

Missing covariate data were handled using a structured approach. Complete-case analysis was used in the primary model when the proportion of missingness was low, defined as less than 5% for each covariate. For variables with greater missingness but judged to be clinically important, multiple imputation by chained equations was used as a sensitivity analysis under the missing-at-random assumption, based on 10 imputed datasets with 20 iterations each. Results from imputed models were compared with complete-case estimates to assess the robustness of the findings.

To reduce the risk of model instability, collinearity among covariates was examined before multivariable modeling using variance inflation factors and correlation matrices. A variance inflation factor threshold of 5.0 was prespecified to indicate potentially problematic multicollinearity. If two variables were highly collinear, the clinically more informative variable was retained in the final model.

2.6 Statistical analysis

Baseline characteristics were summarized according to quartiles of the CRP–TyG index. Continuous variables were reported as mean ± standard deviation for approximately normally distributed data or as median and interquartile range for skewed data. Categorical variables were described as counts and percentages. Group differences were assessed using one-way analysis of variance or the Kruskal–Wallis test for continuous variables, and the chi-square test or Fisher’s exact test for categorical variables, as appropriate.

The primary association between the CRP–TyG index and incident malignancy was evaluated using Cox proportional hazards regression models, with results reported as hazard ratios and 95% confidence intervals. A series of progressively adjusted models was prespecified. Model 1 adjusted for age and sex. Model 2 further adjusted for BMI, HbA1c, and key organ function indicators. Model 3 additionally incorporated inflammatory markers, comorbidities, and medication use. Tests for linear trend across quartiles were performed by assigning the median value of each quartile and modeling this variable continuously.

The proportional hazards assumption was formally assessed using Schoenfeld residuals and graphical inspection of log-minus-log survival plots. If minor deviations from proportionality were detected for selected covariates, sensitivity analyses using time-stratified or alternative specifications were performed. No material violation was observed for the primary exposure. To further evaluate model stability, covariate collinearity was examined using variance inflation factors before fitting the final model.

To explore possible nonlinear associations, the CRP–TyG index was entered into restricted cubic spline models as a continuous variable. Knot placement was prespecified at the 5th, 35th, 65th, and 95th percentiles of the exposure distribution, balancing flexibility with model parsimony. The reference value for spline plots was set at the cohort median. Overall and nonlinear P values were reported. When the spline curves suggested a threshold effect, two-piecewise Cox regression models were fitted to estimate effect sizes on either side of the inflection point. The threshold value was identified by searching for the model with the maximum likelihood and then confirmed using segmented regression procedures.

Because non-cancer death could compete with the occurrence of malignancy, Fine–Gray competing-risk regression was performed as a secondary analysis, treating non-cancer death as the competing event. Subdistribution hazard ratios and 95% confidence intervals were reported to examine whether the main findings remained materially unchanged under this framework.

Prespecified subgroup analyses were conducted according to sex, age group, BMI category, glycemic control level, renal function status, and use of metformin or statins. Interaction terms between the CRP–TyG index and subgroup variables were tested to assess effect heterogeneity. Given the limited number of events for some specific cancer types, site-specific analyses were considered exploratory and were conducted using reduced adjustment sets chosen on the basis of clinical importance and events-per-variable constraints.

Several sensitivity analyses were performed to test the robustness of the findings. These included landmark analyses excluding malignancy events occurring within 90 days and 180 days after the index hospitalization, analyses using alternative CRP–TyG construction methods, complete-case versus multiply imputed models, and analyses excluding patients with acute infection-related diagnoses during the index admission when such data were available. Discriminative performance was compared between the composite CRP–TyG index and its individual components using receiver operating characteristic analysis within a predefined 3-year incident malignancy window, as appropriate to the final analytic framework.

All statistical tests were two-sided, and a P value < 0.05 was considered statistically significant. Analyses were performed using R version 4.3.2 (R Foundation for Statistical Computing, Vienna, Austria) and SPSS version 27.0 (IBM Corp., Armonk, NY, USA).

3 Results3.1 Cohort description and baseline characteristics by CRP–TyG quartiles

After applying the predefined inclusion and exclusion criteria, 5,500 hospitalized adults with diabetes were included in the final analytic cohort. During follow-up, 344 incident malignancies were identified. Participants were categorized into quartiles according to the CRP–TyG composite index, with 1,375 patients in each quartile. The participant selection process is shown in Figure 2.

Baseline characteristics across CRP–TyG quartiles are summarized in Table 1. Overall, patients in higher quartiles showed a less favorable metabolic and inflammatory profile. Mean age increased progressively from Q1 to Q4, whereas the sex distribution did not differ significantly across groups. Markers of metabolic burden also worsened across quartiles. Compared with patients in Q1, those in Q4 had higher body mass index, systolic blood pressure, fasting plasma glucose, HbA1c, and triglyceride levels, while HDL-C decreased significantly across quartiles.

CharacteristicQ1 Q2 Q3 Q4 P valueCRP–TyG composite index-1.12 ± 0.39-0.28 ± 0.220.41 ± 0.241.19 ± 0.43<0.001Age, years63.4 ± 10.864.7 ± 10.166.1 ± 9.667.5 ± 9.20.018Male, n (%)785 (57.1)832 (60.5)880 (64.0)927 (67.4)0.231BMI, kg/m²24.7 ± 3.425.4 ± 3.626.1 ± 3.727.0 ± 3.9<0.001Systolic BP, mmHg131.6 ± 16.9134.2 ± 17.4136.8 ± 18.1140.3 ± 18.70.012Diastolic BP, mmHg76.4 ± 10.577.2 ± 10.178.6 ± 10.780.1 ± 10.80.083Current smoker, n (%)224 (16.3)256 (18.6)287 (20.9)336 (24.4)0.468Current alcohol use, n (%)144 (10.5)176 (12.8)193 (14.0)224 (16.3)0.742Diabetes duration, years7.2 (3.6–11.4)8.1 (4.1–12.3)9.0 (4.8–13.8)10.2 (5.6–15.4)0.006HbA1c, %7.4 ± 1.27.7 ± 1.38.1 ± 1.48.5 ± 1.5<0.001Fasting plasma glucose, mmol/L7.2 ± 1.67.9 ± 1.78.6 ± 1.99.4 ± 2.1<0.001Triglycerides, mmol/L1.34 (1.02–1.71)1.62 (1.22–2.06)1.98 (1.52–2.52)2.52 (1.96–3.21)<0.001Total cholesterol, mmol/L4.33 ± 0.944.41 ± 0.974.52 ± 1.014.67 ± 1.050.156LDL-C, mmol/L2.43 ± 0.722.46 ± 0.742.54 ± 0.772.61 ± 0.790.338HDL-C, mmol/L1.18 ± 0.291.14 ± 0.271.10 ± 0.261.05 ± 0.250.004CRP, mg/L2.1 (1.0–4.3)3.8 (2.0–6.9)6.7 (3.9–11.6)12.9 (7.8–22.4)<0.001White blood cells, ×109/L7.1 ± 2.17.5 ± 2.28.0 ± 2.58.7 ± 2.8<0.001Neutrophil-to-lymphocyte ratio2.4 (1.7–3.4)2.8 (2.0–3.9)3.3 (2.3–4.6)4.1 (2.9–5.9)<0.001Albumin, g/L39.8 ± 4.138.9 ± 4.337.8 ± 4.636.5 ± 4.9<0.001ALT, U/L22 (16–32)24 (17–36)26 (18–40)28 (19–45)0.041eGFR, mL/min/1.73m²82.6 ± 19.878.9 ± 21.473.2 ± 23.667.8 ± 25.9<0.001Hypertension, n (%)880 (64.0)960 (69.8)1,008 (73.3)1,103 (80.2)0.071Chronic kidney disease, n (%)193 (14.0)256 (18.6)336 (24.4)448 (32.6)0.010Coronary artery disease, n (%)239 (17.4)287 (20.9)320 (23.3)400 (29.1)0.278Prior stroke, n (%)112 (8.1)128 (9.3)160 (11.6)193 (14.0)0.641NAFLD, n (%)287 (20.9)352 (25.6)415 (30.2)512 (37.2)0.049COPD, n (%)96 (7.0)112 (8.1)128 (9.3)160 (11.6)0.734Metformin use, n (%)736 (53.5)704 (51.2)656 (47.7)576 (41.9)0.392Insulin use, n (%)384 (27.9)432 (31.4)512 (37.2)640 (46.5)0.041Statin use, n (%)608 (44.2)640 (46.5)688 (50.0)720 (52.3)0.693ACEI/ARB use, n (%)560 (40.7)592 (43.0)623 (45.3)671 (48.8)0.771SGLT2 inhibitor use, n (%)160 (11.6)176 (12.8)193 (14.0)224 (16.3)0.872GLP-1RA use, n (%)96 (7.0)112 (8.1)128 (9.3)144 (10.5)0.896Aspirin use, n (%)304 (22.1)336 (24.4)367 (26.7)416 (30.2)0.628

Baseline characteristics of hospitalized patients with diabetes across CRP–TyG quartiles.

Indicators of inflammatory activation and nutritional status also varied systematically according to CRP–TyG level. Median CRP, white blood cell count, and neutrophil-to-lymphocyte ratio increased stepwise from the lowest to the highest quartile, whereas albumin levels declined progressively. In parallel, renal function was less favorable in higher quartiles, with lower mean estimated glomerular filtration rate and a higher prevalence of chronic kidney disease. The prevalence of nonalcoholic fatty liver disease also increased across quartiles. With respect to medication use, insulin therapy was more common in higher quartiles, whereas the distribution of metformin, statin, ACEI/ARB, SGLT2 inhibitor, GLP-1 receptor agonist, and aspirin use did not differ materially between groups.

Taken together, these baseline patterns indicate that a higher CRP–TyG composite index was associated with greater metabolic dysregulation, a more pronounced inflammatory state, and a higher burden of selected cardiometabolic comorbidities at cohort entry.

3.2 Malignancy burden: overall and system-specific distribution

In the final cohort of 5,500 hospitalized patients with diabetes, 344 incident malignancies were identified during follow-up. Event counts increased progressively across CRP–TyG quartiles, from 58 events in Q1 to 120 events in Q4, indicating a graded increase in malignancy burden with higher levels of the composite index. The corresponding quartile-specific associations are shown in Table 2.

Outcome
(incident malignancy)Events, n
(Q1/Q2/Q3/Q4)Adjusted HR
(95% CI), Q2 vs Q1Adjusted HR
(95% CI), Q3 vs Q1Adjusted HR
(95% CI), Q4 vs Q1P for trendOverall malignancy (primary)344 (58/75/91/120)1.21 (0.87–1.67)1.46 (1.07–1.99)1.92 (1.42–2.60)<0.001Gastrointestinal cancers122 (18/25/32/47)1.29 (0.74–2.24)1.62 (0.96–2.73)2.05 (1.25–3.36)0.001Respiratory cancers76 (13/16/21/26)1.18 (0.58–2.38)1.45 (0.75–2.81)1.78 (0.95–3.34)0.019Urinary system cancers51 (9/11/14/17)1.14 (0.49–2.63)1.44 (0.66–3.14)1.69 (0.79–3.62)0.033Reproductive system cancers44 (7/9/12/16)1.24 (0.48–3.18)1.53 (0.64–3.64)1.83 (0.79–4.22)0.028Hematologic malignancies26 (5/6/7/8)1.09 (0.34–3.47)1.28 (0.42–3.89)1.58 (0.54–4.67)0.176Other cancers25 (6/8/5/6)1.22 (0.44–3.36)1.02 (0.33–3.16)1.62 (0.56–4.66)0.210

Association between CRP–TyG quartiles and the risk of malignancy, stratified by major cancer system.

For the primary endpoint of overall incident malignancy, the fully adjusted hazard ratios increased across quartiles, with the strongest association observed in Q4 relative to Q1 (HR 1.92, 95% CI 1.42–2.60; P for trend < 0.001). This pattern supported a positive dose–response relationship between the CRP–TyG composite index and malignancy risk.

In site-specific analyses, gastrointestinal cancers accounted for the largest proportion of events (n = 122) and showed the most pronounced gradient across quartiles. Compared with Q1, the fully adjusted hazard ratio for gastrointestinal cancer in Q4 was 2.05 (95% CI 1.25–3.36; P for trend = 0.001). Respiratory, urinary, and reproductive system cancers also showed positive trend associations across quartiles, although the confidence intervals were wider and some quartile-specific estimates were less precise than those observed for the overall endpoint.

By contrast, hematologic malignancies and the heterogeneous category of other cancers had relatively few events. Although point estimates generally trended upward across quartiles, these analyses were characterized by limited precision and wider confidence intervals. Accordingly, site-specific findings for lower-frequency cancer categories should be interpreted as exploratory rather than definitive.

3.3 Primary association: CRP–TyG and malignancy risk

The primary time-to-event analysis showed a stepwise increase in malignancy risk across quartiles of the CRP–TyG composite index. In crude Cox regression, compared with patients in the lowest quartile, those in Q3 and Q4 had significantly higher risks of incident malignancy, with hazard ratios of 1.58 (95% CI 1.15–2.17) and 2.10 (95% CI 1.56–2.82), respectively (Table 3).

CRP–TyG quartileEvents/NIncidence
(%)Model 0 (Crude) HR
(95% CI)Model 1 HR
(95% CI)Model 2 HR
(95% CI)Model 3 HR
(95% CI)Q1 (lowest)58/1,3754.221.00 (ref)1.00 (ref)1.00 (ref)1.00 (ref)Q275/1,3755.451.29 (0.92–1.80)1.24 (0.89–1.73)1.22 (0.88–1.70)1.21 (0.87–1.67)Q391/1,3756.621.58 (1.15–2.17)1.49 (1.09–2.05)1.47 (1.07–2.01)1.46 (1.07–1.99)Q4 (highest)120/1,3758.732.10 (1.56–2.82)2.01 (1.49–2.71)1.96 (1.45–2.65)1.92 (1.42–2.60)P for trend<0.001<0.001<0.001<0.001Per 1-SD increase344/5,5006.25