Diabetes as a pan-cancer risk factor: Triangulated evidence from global burden, longitudinal cohorts, and Mendelian randomization

Abstract

BACKGROUND

Diabetes has been implicated in cancer initiation and progression; however, its contribution to cancer burden, risk, and survival remains insufficiently defined across populations, geographical regions, and analytical frameworks.

AIM

To systematically evaluate the impact of diabetes on pan-cancer and delineate shared risk patterns across malignancies.

METHODS

We integrated the Global Burden of Disease, National Health and Nutrition Examination Survey (NHANES), the National Death Index mortality data, China Health and Retirement Longitudinal Study, and genome-wide association study datasets. The Global Burden of Disease estimates were used to assess temporal trends. Multivariable logistic regression was used to examine associations between diabetes and cancer prevalence; Kaplan-Meier, Cox proportional hazards, and competing risk analyses were used to evaluate survival outcomes; and Mendelian randomization was used to assess genetic evidence for causality.

RESULTS

The cancer burden attributable to high fasting plasma glucose levels increased steadily from 1990 to 2021. After multivariable adjustment, diabetes was independently associated with higher cancer prevalence in NHANES (odds ratio = 1.64, 95%CI: 1.18-2.28) and China Health and Retirement Longitudinal Study cohort (odds ratio = 2.00, 95%CI: 1.11-3.59), and with poorer survival (hazard ratio = 1.25, 95%CI: 1.03-1.51) in NHANES. Participants with diabetes generally exhibit higher cancer-specific mortality rates. Mendelian randomization further supported a causal association between diabetes and overall cancer risk, while established diabetes-related markers showed dose-response associations with cancer prevalence.

CONCLUSION

These findings identify diabetes as a systemic pan-cancer risk factor associated with increase cancer burden, elevated cancer risk, and poor survival outcomes.

Key Words: Diabetes; Hyperglycemia; Pan-cancer; Cancer burden; Cancer-specific mortality; Risk stratification

Core Tip: Using multiple independent datasets and Mendelian randomization, this study links diabetes to a rising pan-cancer burden, higher cancer prevalence and poorer survival. Longer diabetes duration and higher insulin-resistance markers showed dose-response associations with cancer prevalence, identifying diabetes-related indicators as quantifiable markers of cancer risk. These findings support diabetes-informed cancer risk stratification, targeted prevention and longitudinal surveillance in high-risk populations.

INTRODUCTION

Diabetes and cancer are the two most pressing global public health issues. According to the International Diabetes Federation, the global prevalence of diabetes continues to increase and, the number of affected individuals is projected to exceed 780 million by 2045[1,2]. Cancer, the second leading cause of death worldwide, exhibits persistent increases in incidence, mortality, and overall disease burden[3,4].

As a prevalent metabolic disorder, diabetes is characterized by disruptions in glucose metabolism, hyperinsulinemia, and lipid metabolism abnormalities, leading to an impaired quality of life and placing a substantial burden on the healthcare system[5-7]. In addition to its direct clinical and economic impacts, growing evidence indicates that diabetes is a key driver of multiple diseases[8-10]. Notably, the potential link between diabetes and cancer has emerged as a focal point at the intersection of metabolic diseases and oncology epidemiology. For example, diabetes increases the risk of colorectal cancer, particularly in individuals with low levels of immune infiltration, and is associated with poor survival outcomes[11]. Moreover, diabetes has been linked to an elevated risk of several other malignancies, including liver, gastric, and pancreatic cancers[12-14].

However, current research has primarily focused on specific cancer types. Consequently, it remains uncertain whether hyperglycemia drives site-specific carcinogenic processes or establishes a systemic, pro-tumorigenic milieu that promotes the overall development of cancer. Additionally, existing studies are constrained by the limited use of large-scale, globally representative population-based datasets and have not fully addressed reverse causality. This undermines the generalizability and robustness of the evidence. Therefore, there is an urgent need to construct a comprehensive evidence chain that integrates macro-level disease burden assessments, large-scale data analyses, and genetic causal validation to systematically elucidate the system-wide, pan-cancer impact of diabetes on cancer risk across multiple anatomical sites.

In this context, we sought to quantify the pan-cancer burden and risk attributable to diabetes using, three major databases to ensure broad geographical representation and methodological consistency. First, we used the Global Burden of Disease (GBD) database to characterize the spatiotemporal dynamics of the overall cancer burden associated with hyperglycemia from 1990 to 2021. We then integrated the U.S. National Health and Nutrition Examination Survey (NHANES), a nationally representative cross-sectional survey, and the China Health and Retirement Longitudinal Study (CHARLS), a prospective cohort study suited to probing temporal and putative causal relationships, to evaluate the association between diabetes status, pan-cancer prevalence, and survival outcomes. Finally, we applied Mendelian randomization (MR) as a genetic causal validation layer to strengthen the causal inference of the relationship between diabetes and pan-cancer. By integrating macro-level disease burden data, multinational cohort and cross-sectional studies, and cutting-edge causal inference methods, we aimed to elucidate the multi-dimensional effect of diabetes on pan-cancer. Together, this study may advance the understanding of the “diabetes-cancer” relationship, enable the translation of diabetes-related markers into risk stratification and prognostic assessment, and inform early detection, tiered prevention, and integrated diabetes-cancer control strategies for high-risk populations across diverse cancer types.

MATERIALS AND METHODS

Data collection

GBD database: GBD database encompasses population-level health data from 232 countries and territories[15]. GBD estimates were extracted spanning 1990-2021, include age-standardized mortality rates (ASMR), disability-adjusted life years (DALYs), years of life lost, and years lived with disability.

NHANES database: NHANES is an ongoing, nationally representative U.S. cross-sectional survey[16]. NHANES cycles from 2007-2018 were analyzed, incorporating sampling weights, strata, and primary sampling units in accordance with official analytic guidance. The National Death Index-linked mortality files were also used to ascertain vital status and follow-up time.

CHARLS database: CHARLS is a national longitudinal cohort of Chinese adults aged ≥ 45 years[17]. CHARLS data collected between 2011 and 2020 were used in this study.

Study design

Participants aged ≥ 18 years from NHANES and CHARLS were included. Individuals with missing data on diabetes status or cancer outcomes, or with prespecified conditions likely to introduce material confounding were excluded according to the predefined criteria summarized in Supplementary Figure 1.

Participants were then categorized into diabetes, prediabetes, or normoglycemia based on standardized laboratory glucose measurements, self-reported physician diagnosis, and self-reported use of glucose-lowering medications. The classification criteria the recommendations of International Diabetes Federation and American Diabetes Association (Supplementary Table 1)[18]. Cancer status was ascertained primarily by self-report using the item, “Has a doctor ever diagnosed you with cancer?”, in both NHANES and CHARLS.

Variable assessment

Exposures and outcomes: As a widely available convenient, and cost-effective test, high fasting plasma glucose (FPG) is commonly employed in national-level studies[19,20]. Therefore, FPG was designated as the exposure in the GBD analyses.

In the NHANES and CHARLS analyses, exposures included diabetes status, diabetes duration, and two insulin-resistance surrogates: The triglyceride-glucose (TyG) index and the metabolic score for insulin resistance (METS-IR) to identify cross-cancer biomarkers. Diabetes status is defined as described above; diabetes duration is calculated as the difference between current age and self-reported age at diagnosis[21]; TyG index was calculated as ln[fasting triglycerides (mg/dL) × FPG (mg/dL)/2][22], and METS-IR as ln[(2 × FPG (mg/dL) + fasting triglycerides (mg/dL)] × body mass index (BMI) (kg/m²)/Ln[high-density lipoprotein cholesterol (mg/dL)][23].

Outcomes include cancer prevalence and survival rates. Cancer prevalence is defined as the proportion of participants reporting a cancer diagnosis, while survival rates were based on survival status and follow-up time. Follow-up began at the baseline interview date and ended at the recorded date of death or the censoring date (last known follow-up). Survival status was categorized into two groups: (1) Endpoint events (all-cause mortality or cancer-specific mortality during follow-up); and (2) Censored data (participants who were still alive or lost to follow-up at the study endpoint).

Covariates: Standardized demographic and health-related covariates were extracted from the NHANES and CHARLS databases, including age, sex, race/ethnicity, education level, household income, BMI, smoking, alcohol use, diet quality, and physical activity level. All variables were uniformly collected through standardized household interviews and questionnaires in both surveys.

In NHANES, education level was classified as (1) “Lower education”, “high school and equivalent“, “postsecondary education“, and “higher education“; (2) Household income was assessed using the family poverty-income ratio; smoking status was classified as “not at all“, “some days“, or “every day“; (3) Alcohol intake was categorized by “heavy drinking“ (women ≥ 3 drinks/day, men ≥ 4 drinks/day), “mild drinking“ (women ≥ 2 drinks/day, men ≥ 3 drinks/day), and “moderate drinking“ (all others); (4) Diet quality was originally rated as poor, fair, good, very good, and excellent; and (5) Physical activity was categorized as none, moderate, or vigorous exercise.

In CHARLS, education level was dichotomized (low vs high); income was measured by annual total household income; smoking and drinking were coded as never or ever; physical activity was categorized as regular exercise or not.

Statistical analysis

GBD data analysis: To quantify the overall cancer burden associated with high FPG, we employed relative risk assessment and population attributable fraction analysis. Additionally, age-standardized rates and ASMR were calculated, which enable direct comparison of cancer incidence and mortality attributable to high FPG across regions and years.

Baseline characteristics analysis: For the NHANES and CHARLS datasets, continuous variables are presented as means ± SD, and discrete variables as n (%). Data were compared using the Kruskal-Wallis test for continuous variables and the χ² test for discrete variables.

Logistic regression and Cox proportional hazards models: To evaluate the independent effects of diabetes status on cancer prevalence and patient survival, we constructed multivariable logistic regression models for cancer prevalence and Cox proportional hazards regression models for survival outcomes. Three models were developed, progressively adjusting for a broader set of covariates.

Survival and competing risk analysis: Survival and competing-risk analyses were performed using R software (version 4.5.1). Kaplan-Meier curves and log-rank tests were used to compare overall survival across groups, whereas cumulative incidence functions and Gray’s test were used to assess cancer-specific mortality, with non-cancer deaths treated as competing events.

Nonlinear relationship analysis: To explore the relationship between the TyG index, METS-IR, diabetes duration, and cancer prevalence, logistic regression models with restricted cubic splines and penalized spline smooth curve fitting were employed. In cases where nonlinear relationships were identified, we estimated threshold values by testing various cutoff points and selecting the one that maximized model likelihood.

MR analysis: We performed MR analysis using diabetes-related single nucleotide polymorphisms from independent genome-wide association study databases as genetic instrumental variables, and pan-cancer summary statistics as outcomes. Causal effects were estimated using five complementary MR approaches: (1) Inverse variance weighting (IVW); (2) MR-Egger regression; (3) The weighted median; (4) Weighted mode; and (5) Simple mode method. Given its efficiency under the valid-instrument assumption, we designated the random-effects IVW estimate as the primary effect measure. Instrument heterogeneity was evaluated using Cochran’s Q statistic.

Ethics considerations

Since the GBD database consists of de-identified public data, and the original studies have received ethical approval, no additional ethical review is required for our study. NHANES protocols were approved by the Centers for Disease Control and Prevention Ethics Review Board, and all participants provided informed consent. The CHARLS study has been approved by the Biomedical Ethics Committee of Peking University (IRB00001052-11015).

RESULTS

High FPG as a key metabolic risk factor for global cancer burden

By 2021, high FPG emerged as a leading metabolic risk factors for the global pan-cancer burden, accounting for 8.0% of cancer-related DALYs worldwide and 5.1% in China (Figure 1A and B). The population attributable fraction of high FPG-attributable cancer DALYs increased steadily from 1990 to 2021 in both settings (Figure 1C and D). This contribution was most pronounced in older adults, peaking in those aged 65-89 (Figure 1E and F).

Figure 1 Global and China patterns of cancer burden attributable to high fasting plasma glucose. A: Relative contributions of risk factors to cancer-related disability-adjusted life years (DALYs) in 2021, globally; B: Relative contributions of risk factors to cancer-related DALYs in 2021, China; C: Proportion of cancer-related DALYs attributable to high fasting plasma glucose (FPG) from 1990 to 2021, globally; D: Proportion of cancer-related DALYs attributable to high FPG from 1990 to 2021, China; E: Age-specific proportions of cancer-related DALYs attributable to high FPG in 2021, globally; F: Age-specific proportions of cancer-related DALYs attributable to high FPG in 2021, China. DALYs: Disability-adjusted life years; FPG: Fasting plasma glucose.

In China, the rising trend of the high FPG-related cancer burden is particularly notable. By 2021, the ASMR for high FPG-related cancer DALYs had increased by 19.80% compared to 1990, and the DALY rates had increased by 16.10% (Supplementary Table 2). Notably, the increase in years lived with disability exceeded that in years of life lost (Supplementary Table 3), indicating the growing impact of hyperglycemia on the morbidity and quality of life of cancer patients.

Population-stratified analyses revealed pronounced age-specific and sex-specific patterns in in patients with a high FPG-attributable cancer burden. Cancer DALYs (Supplementary Figure 2A) and deaths (Supplementary Figure 2B) were relatively low at ages 25-49 years, increased sharply after 50 years, peaked at 65-69 years, and remained elevated through 70-79 years of age. Males consistently showed higher DALYs and mortality rates across age strata (Supplementary Figure 2C and D). From 1990 to 2021, DALY rates associated with high FPG-related cancer consistently increased and, plateaued in the past decade, whereas the increase in ASMR was relatively slower (Supplementary Figure 2E). Geographically, the cancer burden attributable to high FPG is greater in high-income regions (notably North America and Western Europe) and comparatively lower in low-income sub-Saharan Africa (Supplementary Figure 2F and G).

Characteristics stratified by diabetes status

This study included 30372 participants from the NHANES (2007-2018) and 90679 participants from the CHARLS (2011-2020).

In the NHANES, participants were categorized into normoglycemic (n = 14281), prediabetic (n = 10354), and diabetic (n = 5737) groups (Supplementary Table 4). Compared with normoglycemic individuals, those with prediabetes or diabetes were older, more often male, had lower educational attainment and household income, and exhibited a higher BMI and a less favorable lifestyle profile.

In the CHARLS, participants were classified as normoglycemic (n = 73162), prediabetic (n = 6127), or diabetic (n = 11390) (Supplementary Table 5). The diabetes group was more likely to be older, female, and have a higher education and household income; they also had a higher BMI and poorer overall lifestyle.

Association between diabetes and cancer

Association between diabetes status and cancer prevalence: In the NHANES, diabetes status was significantly associated with cancer prevalence, which increased stepwise from normoglycemia to prediabetes and was highest in the diabetic groups (Supplementary Table 6). In the CHARLS cohort, the cancer prevalence varied across groups, reaching 2.9% in participants with diabetes and 1.6% in those with normoglycemia (Supplementary Figure 3).

In the NHANES, after full adjustment, diabetes remained associated with a higher cancer prevalence than did normoglycemia [odds ratio (OR) = 1.64, 95%CI: 1.18-2.28]. In the CHARLS, this association persisted in the fully adjusted model (OR = 2.00, 95%CI: 1.11-3.59; Table 1).

Table 1 Association between diabetic status and cancer prevalence.

Case		Model 1			Model 2			Model 3
		OR	95%CI	P value	OR	95%CI	P value	OR	95%CI	P value
1NHANES	Prediabetes	1.66	1.48-1.86	< 0.001	1.26	1.12-1.41	< 0.001	1.25	0.99-1.59	0.068
	Diabetes	2.46	2.16-2.81	< 0.001	1.59	1.36-1.85	< 0.001	1.64	1.18-2.28	0.004
2CHARLS	Prediabetes	0.95	0.77-1.17	0.632	0.85	0.67-1.08	0.179	1.69	0.98-2.89	0.057
	Diabetes	1.79	1.58-2.03	< 0.001	1.65	1.44-1.89	< 0.001	2.00	1.11-3.59	0.020

¹Normoglycemia as reference group in National Health and Nutrition Examination Survey (NHANES). Model 2 for NHANES adjusted for age, gender, and ethnicity. Model 3 for NHANES adjusted for age, gender, ethnicity, education level, family income, alcohol consumption, cigarette use, physical activity, hypertension, and body mass index.

²Normoglycemia as reference group in China Health and Retirement Longitudinal Study (CHARLS). Model 2 for CHARLS adjusted for age, gender, and race. Model 3 for CHARLS adjusted for age, gender, race, education level, family income, alcohol consumption, cigarette use, physical activity, dietary patterns, hypertension, and body mass index.

CHARLS: China Health and Retirement Longitudinal Study; NHANES: National Health and Nutrition Examination Survey; OR: Odds ratio.

Association between diabetes status and cancer patient survival outcomes: Integrating National Death Index-linked mortality data, we performed survival analyses among the overall NHANES cohort and among the 3101 patients with cancer. Throughout the follow-up period, the diabetic group consistently showed lower overall survival than the normoglycemic group, both in the overall cohort and among participants with cancer, with the gap widening over time (Figure 2A and B). Specifically, participants with diabetes showed a higher mortality proportion (32.5% in unweighted cohort and 26.2% in weighted cohort) and a shorter mean follow-up duration (65.59 ± 40.70 person-months unweighted; 64.67 ± 39.87 person-months weighted) (Supplementary Table 7). In the fully adjusted model, diabetes was associated with a higher risk of all-cause mortality in participants with cancer (hazard ratio = 1.25, 95%CI: 1.03-1.51; Table 2).

Figure 2 Associations of diabetes status with all-cause and cancer-specific mortality in the National Health and Nutrition Examination Survey. A: Survival probability of all participants stratified by diabetic status; B: Survival probability of patients with cancer stratified by diabetic status; C: Cancer-specific survival probability of patients with cancer stratified by diabetic status; D: Cancer-specific competing risk of patients with cancer stratified by diabetic status.

Table 2 Association between diabetic status and survival in patients with cancer from the National Health and Nutrition Examination Survey.

Case	Model 1			Model 2			Model 3
	HR	95%CI	P value	HR	95%CI	P value	HR	95%CI	P value
Prediabetes	0.93	0.78-1.10	0.387	0.89	0.75-1.06	0.187	0.91	0.76-1.09	0.312
Diabetes	1.37	1.15-1.63	< 0.001	1.28	1.07-1.53	0.007	1.25	1.03-1.51	0.023

Normoglycemia as reference group. Model 2 adjusted for age, gender and race. Model 3 adjusted for age, gender, race, education level, family income, alcohol consumption, cigarette use, physical activity, dietary patterns, hypertension, and body mass index. HR: Hazard ratio.

Cancer-specific survival also differed across the glycemic groups, with the participants with diabetes showing poorer survival (Figure 2C). Consistently, competing risk analysis further revealed significant differences in the cumulative incidence of cancer-specific mortality across groups among participants with cancer, with the diabetes group generally exhibiting a higher cumulative incidence during mid-to-late follow-up (Figure 2D).

Causal association between diabetes and cancer: The MR results revealed a robust positive causal association between genetically predicted diabetes and the overall cancer risk. All eight genetic instruments were strong, with F-statistics indicating a low likelihood of weak-instrument bias (Supplementary Table 8). The IVW analysis showed an increase in cancer risk (OR = 1.50, 95%CI: 1.09-2.06), consistent with the findings from the weighted median (OR = 1.57, 95%CI: 1.21-2.05), simple mode (OR = 1.76, 95%CI: 1.10-2.81) and weighted mode methods (OR = 1.59, 95%CI: 1.23-2.04; Table 3). The scatter plots show broadly consistent regression slopes across the MR methods, with no evident influential outliers (Supplementary Figure 4). Sensitivity analysis revealed no evidence of directional horizontal pleiotropy.

Table 3 Mendelian randomization analysis of the causal association between diabetes and cancer.

Diabetes (exposure)	Cancer (outcome)	Method	Single nucleotide polymorphisms	P value	OR (95%CI)	P value for pleiotropy
finn-b-E4_DMNAS	ukb-e-2453_CSA	IVW	8	0.012	1.50 (1.09, 2.06)	0.685
		Weighted median	8	< 0.001	1.57 (1.21, 2.05)
		Simple mode	8	0.049	1.76 (1.10, 2.81)
		Weighted mode	8	0.009	1.59 (1.23, 2.04)
		MR Egger	8	0.121	1.65 (0.96, 2.82)

IVW: Inverse variance weighting; OR: Odds ratio; MR: Mendelian randomization.

Association between diabetes-related indicator and cancer prevalence

To identify high-risk subgroups for diabetes-related cancer using quantifiable indicators, we examined the association between diabetes duration (Figure 3A), TyG index (Figure 3B), METS-IR (Figure 3C), and cancer risk. Restricted cubic splines analysis revealed a significant dose-response relationship between diabetes duration and cancer risk, with cancer prevalence steadily increasing, particularly after 10 years of disease duration. Moreover, both indices demonstrated a roughly linear positive association with cancer prevalence, with risk increasing more markedly at TyG ≥ 8.8 and METS-IR ≥ 35.

Figure 3 Nonlinear relationships of diabetes-related markers and cancer prevalence. A: Nonlinear relationships of diabetes duration and cancer prevalence in the National Health and Nutrition Examination Survey; B: Nonlinear relationships of triglyceride-glucose index and cancer prevalence in the China Health and Retirement Longitudinal Study; C: Nonlinear relationships of metabolic score for insulin resistance and cancer prevalence in the China Health and Retirement Longitudinal Study. OR: Odds ratio; TyG: Triglyceride-glucose; METS-IR: Metabolic score for insulin resistance.

DISCUSSION

Principal findings

By integrating GBD estimates, large-scale population-based datasets, and genetic instrumental-variable analyses, our study provides, to our knowledge, the first pan-cancer evaluation of diabetes across three complementary layers: (1) Population-level burden; (2) Individual risk; and (3) Causal inference. We showed that high FPG levels are a key metabolic risk factor contributing to global pan-cancer burden. After comprehensive multivariable adjustment, diabetes was significantly associated with higher cancer prevalence and remained independently associated with poorer survival among patients with cancer. MR analyses provide further genetic evidence to support a positive causal relationship between diabetes and cancer. Moreover, longer diabetes duration, and higher insulin resistance are also associated with an increased risk of cancer.

Comparisons and interpretations

FPG is commonly used as a diagnostic biomarker of diabetes in database studies, providing both practicality and consistency in large-scale epidemiological research. Previous studies have linked high FPG levels to several cancer types, including colorectal, breast and bladder cancers[24-26]. Building on this evidence, we conceptualized high FPG as a pivotal bridge between macro-level and micro-level analyses, serving as a reliable proxy indicator of diabetes-related cancer burden in the GBD framework, and as an objective criterion for defining diabetes status in cohort and cross-sectional analyses. We constructed an integrated evidence chain that connects population attribution, individual risk, and causal inference.

Upon comparing the CHARLS and NHANES participants, we observed systematic differences in population characteristics, which is consistent with previous reports[22,27]. In the CHARLS, individuals with diabetes were more likely to have a relatively higher income and educational attainment, whereas in the NHANES, diabetes was more prevalent among individuals with lower socioeconomic status and lower education levels. These differences may reflect the different stages of the diabetes epidemic in each setting. The CHARLS cohort, which represents middle-aged and older adults in China undergoing rapid socioeconomic transition, reflects the early stages of the disease epidemic[28]. The relatively higher levels of education and income in this group likely expose them to lifestyle changes associated with urbanization, such as dietary transitions and reduced physical activity, thus making them susceptible to the early effects of diabetes[29]. Concurrently, the population benefits from better healthcare access, earlier diagnosis, milder disease progression at detection, and more standardized management[30]. In contrast, the NHANES, which represents a high-income setting in the United States, reflects a more mature phase of the diabetes epidemic, with a higher prevalence of diabetes in socioeconomically disadvantaged groups[31]. Limited access to resources, poor availability of healthy foods, and lower health literacy amplify the disease burden in these populations, resulting in delayed diagnosis, more advanced disease stages, and higher comorbidity burdens[32]. These differences underscore the generalizability and robustness of our findings, supporting the imperative for targeted interventions that address the unique needs of diverse populations while acknowledging the universal threat posed by diabetes to pan-cancer risk.

Building upon previous studies, our findings further indicate that diabetes is not restricted to promoting a narrow subset of malignancies; rather, it may elevate overall cancer susceptibility by shifting the systemic host physiology toward a carcinogenesis-permissive state. By integrating our results with those of the previous studies, several interrelated mechanisms may explain these observations. First, persistent hyperglycemia creates a metabolic environment that drives tumor growth, overcomes tumor suppression, and facilitates metastasis by enhancing glycolytic flux, oxidative stress, and AGE-RAGE-driven inflammation[33-35]. Second, hyperinsulinemia and insulin resistance represent a complementary oncogenic axis: Elevated insulin/insulin-like growth factor bioactivity that can cross-activate insulin and insulin-like growth factor-1 receptors, thereby co-engaging the canonical growth and survival pathways, including PI3K-AKT-mTOR and RAS-ERK, to promote cancer cell proliferation, survival, and metabolic plasticity[36,37]. Third, diabetes-associated chronic low-grade inflammation and immune dysfunction can erode antitumor immunosurveillance. Nutrient excess and tumor-immune metabolic competition may skew myeloid polarization and constrain T-cell effector programs, whereas tumor-intrinsic glycolytic rewiring can further facilitate immune evasion through defined molecular checkpoints[38,39].

However, these systemic mechanisms are unlikely to fully explain the marked heterogeneity across cancer sites, indicating the importance of tissue-specific processes. Hyperglycemia-induced barrier disruption and microbial translocation in the stomach and colorectum sustain mucosal inflammation, epithelial injury, and compensatory proliferation, thereby promoting gastric and colorectal carcinogenesis[40]. In the liver, insulin resistance and ectopic lipid accumulation drive metabolic dysfunction-associated steatotic liver disease/metabolic dysfunction-associated steatohepatitis, fibrosis, and immune dysfunction, creating a pro-tumor microenvironment that favors hepatocellular carcinoma[41]. In the pancreas, hyperinsulinemia and local inflammatory signaling activate the acinar, myeloid, and stellate cells, promoting acinar-to-ductal metaplasia, desmoplasia, and KRAS-driven pancreatic tumorigenesis[37]. Collectively, these findings suggest that the diabetes-associated cancer risk arises from the interplay between systemic metabolic dysfunction and organ-specific microenvironmental changes.

Implications and innovation

Moving beyond prior work largely confined to single cancer types, we implemented a pan-cancer analytical framework to integrate signals across malignancies, delineating diabetes-associated shared risk architectures and putative convergent mechanisms at the cross-cancer level. This approach enhances data efficiency and the generalizability of inference and establishes a foundation for identifying cross-cancer biomarkers, convergent pathway perturbations, and more precise, individualized strategies for prevention and therapy.

At the population level, we identified the pivotal role of high FPG levels in pan-cancer burden. At the individual level, diabetes is consistently associated with increased cancer prevalence and poor survival outcomes. Beyond confirming the diabetes-cancer association, our analyses quantified the entire continuum of diabetes from onset to outcome. This approach elevates the role of diabetes in the cancer continuum from a mere “risk factor” to a potential “process driver”, providing critical theoretical support for the implementation of precise, full-course cancer prevention strategies in clinical practice.

From a clinical perspective, our study highlights the relationships between diabetes duration, TyG index, METS-IR, and cancer risk, thereby providing quantifiable markers for risk stratification among individuals with diabetes. The TyG cutoff of 8.8 is broadly consistent with previous studies, which have reported values of approximately 8.5-8.8 in association with higher cancer prevalence and mortality[42-44]. Although no universally accepted cancer-specific METS-IR threshold has been established, a cutoff value of 35 is supported by recent studies linking comparable values to increased risks of thyroid and gastric cancers[45,46]. Both indices can be readily derived from routine clinical measurements, and their accessibility and low cost support their potential use as adjunct tools for identifying metabolically high-risk individuals who may benefit from intensified metabolic management and more individualized cancer surveillance. Based on these findings, we propose that patients with a TyG index ≥ 8.8, METS-IR ≥ 35, and diabetes duration ≥ 10 years should be considered a high-risk subgroup for diabetes-related cancers. For this population, we advocate establishing integrated health records that combine metabolic profiles and cancer risk indicators and developing cross-cancer, system-level screening, and long-term dynamic monitoring strategies. Specifically, we advocate evaluations every 3-6 months, including the assessment of FPG, HbA1c, and insulin resistance-related indices, with the prompt initiation of targeted cancer screening and imaging in response to rapidly worsening metabolic abnormalities. Furthermore, we recommend a structured, multidisciplinary follow-up model involving quarterly endocrinology, oncology, and nutrition services. This comprehensive integrated health management approach has shifted the focus from isolated glycemic control to holistic metabolic and oncological prevention, offering a feasible clinical pathway to reduce cancer risk and improve long-term outcomes in patients with diabetes.

Lastly, to further minimize confounding and reverse causation, we applied MR to genetically investigate the causal nature of the diabetes-cancer relationship. This is the first study to provide genetic evidence confirming the causal effect of diabetes on the overall cancer risk. We obtained highly consistent causal estimates, indicating that genetic variations primarily influence cancer risk through diabetes-related pathways.

Limitation and future direction

Despite the strengths of this study, it has several limitations. First, our analyses were based on existing public datasets and not directly linked to prospective interventional clinical studies. Future research should incorporate experimental models and rigorously designed clinical studies to further validate the relationship between diabetes and specific cancer outcomes. Second, cancer history in the NHANES and CHARLS was based on self-reported physician diagnoses, which may therefore have been subject to recall bias. In addition, the cross-sectional design may have underrepresented individuals with diabetes who rapidly developed fatal cancers, thereby potentially attenuating the observed associations through survivor bias. Future prospective studies with registry or medical record linkages are needed to validate the diagnoses and ascertain incident cancers and post-diagnosis outcomes. Third, although we adjusted for a broad range of covariates, unmeasured or incompletely captured health-related factors might have influenced our estimates. To gauge the sensitivity of our estimates to unmeasured confounding factors, we calculated E values, which indicated moderate robustness of the fully adjusted associations. Future studies incorporating deeper phenotyping are critical to strengthen causal attribution. Finally, although the causal estimates were directionally consistent across complementary MR methods, no evidence of directional horizontal pleiotropy or residual pleiotropy through diabetes-independent pathways such as insulin-like growth factor signaling could be fully excluded. Future studies should systematically evaluate all instruments and their linkage-disequilibrium proxies, and apply multivariable MR and colocalization approaches to distinguish diabetes-mediated effects from alternative biological pathways.

Elucidation of the intricate links between diabetes and cancer requires an interdisciplinary framework that integrates metabolic research, molecular oncology, genomics, and clinical epidemiology. Large-scale genomic studies, standardized clinical trials, and longitudinal biomarker-based monitoring should be leveraged to refine the identification of high-risk subgroups without diabetes for cancer and to define actionable biological targets. It is also essential to further explore the dynamic evolution of the diabetes-cancer relationship and validate the quantitative association between blood glucose control and cancer risk and prognosis to clarify the precise impact of blood glucose management on reducing cancer incidence and mortality. This positions diabetes as a critical “modifiable target” for cancer prevention and control. Ultimately, these efforts will drive a shift in global cancer control from a predominantly reactive, treatment-centered model to a proactive, prevention-focused approach, while guiding the development of novel cancer treatment strategies to benefit a broader patient population.

CONCLUSION

Our study further highlights the multidimensional impact of diabetes on escalating the global overall cancer burden, elevated disease risk, and poor survival outcomes. These findings highlight the need to integrate glycemic control and diabetes screening into cancer prevention and treatment strategies. Future research should elucidate the biological mechanisms linking diabetes and cancer and develop personalized cancer risk and prognostic models based on diabetes-related markers to inform precision prevention and therapy.