Gliomas are the most common malignant primary brain tumors in adults, with glioblastoma (GBM) representing the most aggressive and lethal subtype (1, 2). Despite advancements in multimodal treatment—including maximal surgical resection followed by radiotherapy and temozolomide chemotherapy—the prognosis for glioma, particularly the high-grade glioma and GBM, remains dismal (3). The median overall survival of GBM is typically less than 15 months, and the 5-year survival rate is below 10% (4, 5). Even with optimal therapy, GBM is characterized by rapid progression, frequent recurrence, and marked resistance to conventional treatment (6, 7). Accordingly, identifying prognostic factors that influence survival is essential to refine risk stratification and guide individualized management (8, 9). In addition to tumor-related features, patient-level metabolic comorbidities have emerged as potential modifiers of cancer progression and outcomes (10).

Diabetes, a global health burden with increasing prevalence, has been linked to the prognosis of several malignancies, including pancreatic, breast, and colorectal cancers (11, 12). The mechanisms underlying the potential influence of diabetes on cancer survival are multifactorial and may involve hyperglycemia, insulin resistance, chronic inflammation, and altered cellular metabolism (13, 14). In glioma, preclinical studies suggest that hyperglycemia may promote tumor growth (15), while insulin and insulin-like growth factor signaling pathways may contribute to resistance to apoptosis and radiotherapy (16). However, clinical evidence on the association between diabetes and glioma survival remains inconsistent, with some studies reporting worse outcomes (17–19) and others showing null effects (20–28). To clarify this issue, we conducted a systematic review and meta-analysis to examine the influence of pre-existing diabetes on the overall survival (OS) and progression-free survival (PFS) of patients with glioma. By synthesizing available evidence, this study aims to quantify the prognostic impact of diabetes, explore potential sources of heterogeneity, and provide insights into whether diabetes status should be considered in glioma prognostic models and clinical decision-making.

This meta-analysis provides a comprehensive synthesis of current evidence on the association between pre-existing diabetes mellitus and survival in patients with high-grade glioma. Pooling data from 12 retrospective cohort studies encompassing nearly 8,500 patients, we found that pre-existing diabetes is significantly associated with reduced OS and, to a lesser extent, PFS. These findings indicate that diabetes is a clinically relevant comorbidity that may influence glioma prognosis beyond established tumor- and treatment-related factors.

Several biological and clinical mechanisms may underlie the observed associations between diabetes and adverse glioma outcomes. At the molecular level, chronic hyperglycemia is known to promote cancer cell proliferation, inhibit apoptosis, and enhance resistance to therapy through multiple pathways (34, 35). Elevated glucose levels fuel anaerobic glycolysis in glioma cells, supporting tumor growth and metabolic adaptation—a phenomenon known as the Warburg effect (36, 37). Hyperinsulinemia and insulin resistance, hallmarks of type 2 diabetes, may further activate insulin and insulin-like growth factor (IGF) signaling pathways, stimulating downstream cascades such as PI3K/AKT/mTOR, which are involved in glioma proliferation, angiogenesis, and treatment resistance (38, 39). Additionally, diabetes is a pro-inflammatory state, characterized by elevated levels of cytokines and reactive oxygen species, which may exacerbate glioma aggressiveness and disrupt the tumor immune microenvironment (40). In glioma specifically, preclinical studies have shown that hyperglycemia may accelerate tumor growth by upregulating the expression of the receptor for advanced glycation end products and suppressing antitumor immune responses (41). Moreover, hyperglycemia may actively drive glioblastoma progression by inducing cellular alterations in the tumor microenvironment and activating signaling pathways associated with glioblastoma development (42). From a clinical standpoint, diabetic patients may be more susceptible to treatment-related complications, including poor wound healing, infections, and corticosteroid-induced metabolic disturbances, which could compromise the effectiveness of surgery, radiotherapy, or chemotherapy (43, 44). Future studies are warranted to determine the key molecular pathways underlying the association between diabetes and poor prognosis of patients with high-grade glioma.

The subgroup analyses conducted in this study offer further insights into the robustness and potential modifiers of the association between diabetes and glioma outcomes. The association between preexisting diabetes and poor OS remained consistent across subgroups defined by glioma grade (grade III–IV vs. GBM-only), age, sex distribution, type of diabetes definition (type 2 diabetes vs. overall diabetes), and follow-up duration. Notably, the association appeared stronger in studies including grade III–IV gliomas compared to GBM-only cohorts, although this difference was not statistically significant. This may reflect the possibly more pronounced impact of comorbid conditions in lower-grade tumors with relatively longer survival, whereas in GBM, the rapid disease course may mask some of the effects of systemic factors like diabetes (9). Similarly, the association was slightly stronger in younger patients and those with shorter follow-up durations, which could suggest an early impact of metabolic dysregulation on tumor progression (45). Nonetheless, none of the subgroup differences reached statistical significance, supporting the generalizability of the findings across patient and study characteristics.

The association between pre-existing diabetes and poorer survival in high-grade glioma likely reflects a combination of biological and care-delivery factors. Outcomes in glioma depend on tumor grade and molecular profile, performance status, extent of resection, and delivery of guideline-concordant chemoradiation; however, several cohorts lacked granular molecular data and treatment-adherence details, leaving room for residual confounding. Most studies captured diabetes as a single, pre-existing category without separating prediabetes or newly diagnosed diabetes at admission, and few quantified corticosteroid exposure or stress hyperglycaemia. Such non-differential misclassification (undiagnosed dysglycemia in the comparison group) would be expected to bias estimates toward the null, suggesting the observed effect may be conservative. Future work should stratify by molecular markers and treatment intensity and incorporate standardized measures of glycemic status (HbA1c, peri-treatment glucose) and steroid dosing.

This study has several methodological strengths. First, all included studies were longitudinal cohort designs, which are more appropriate than cross-sectional studies for examining prognostic associations. Second, the meta-analysis focused on multivariate-adjusted hazard ratios for OS, enhancing the credibility of the findings by accounting for key clinical confounders such as age, performance status, extent of resection, and adjuvant treatment. Finally, a series of prespecified subgroup analyses and sensitivity analyses were performed, all of which confirmed the stability of the main findings.

Nonetheless, several limitations should be considered when interpreting the results. First, recurrence status was inconsistently reported, and survival outcomes were rarely stratified by newly diagnosed versus recurrent glioma. Although most cohorts explicitly or implicitly represented newly diagnosed cases, potential differences in treatment history or steroid exposure cannot be entirely excluded. Second, molecular markers such as IDH mutation or O6-methylguanine-DNA methyltransferase (MGMT) methylation were incompletely reported—only three studies provided IDH information, and two restricted analysis to IDH-wild-type tumors. As tumor molecular subtype substantially affects prognosis, the absence of such data in earlier cohorts may contribute to residual heterogeneity and limits molecular-level interpretation of our pooled estimates. Moreover, all included studies were retrospective, making the analyses vulnerable to recall bias, incomplete or inconsistently recorded clinical data, and unmeasured confounding (46). Key factors such as steroid exposure, treatment intensity, glycemic control, and molecular features were variably reported, and such omissions may have biased the observed associations in either direction. Although pooled associations reached statistical significance, the effect sizes were modest (HR = 1.22). In the context of high-grade glioma—where survival is short and shaped by multiple competing determinants—even modest relative increases in mortality risk may carry clinical relevance, especially at a population level. Moreover, such attenuated estimates are likely conservative given potential non-differential misclassification of dysglycemia and residual confounding related to steroid exposure or treatment intensity. In addition, residual confounding (e.g., performance status, extent of resection, molecular profile, treatment intensity, steroid exposure, stress hyperglycaemia) and selective reporting remain plausible. Therefore, these findings should be interpreted as hypothesis-generating rather than causal, and their clinical relevance should be weighed against the modest magnitude and between-study variability. While the studies adjusted for several clinical variables, residual confounding from unmeasured or inaccurately captured factors cannot be excluded. Also, because all included cohorts involved GBM or grade III–IV tumors, the findings primarily apply to high-grade disease. The prognostic role of diabetes in lower-grade (WHO I–II) gliomas remains uncertain, and our results should not be generalized to these populations without dedicated studies designed to evaluate this question. Moreover, although diabetes was generally identified from medical records in these retrospective cohorts and in several studies further validated using ADA criteria or HbA1c thresholds, heterogeneity in diagnostic approaches may have introduced some non-differential misclassification. However, the consistency of results across subgroup analyses comparing T2DM–specific versus overall diabetes definitions suggests that variation in ascertainment methods was unlikely to materially affect the pooled estimates. Furthermore, diabetes was treated as a binary baseline variable in all cohorts. Longitudinal glycemic control, diabetes duration, and treatment changes were rarely reported, and only isolated studies provided limited parameters such as HbA1c or steroid-adjusted glucose. The absence of such time-varying metabolic data precluded assessment of dose–response patterns and likely contributed to residual confounding. Prospective studies incorporating dynamic glycemic measures and detailed diabetes management are therefore warranted. Also, these factors may themselves influence glioma progression and survival, potentially confounding the observed associations. For example, the use of metformin has been associated with improved survival in several cancers, including glioma, and may attenuate the adverse impact of diabetes (47). In addition, we could not separate prediabetes or newly diagnosed diabetes from established diabetes because such strata were not reported. This may introduce non-differential misclassification (e.g., undiagnosed dysglycemia in the reference group), likely attenuating associations and contributing to heterogeneity. Future studies should prospectively distinguish prediabetes, newly diagnosed, and established diabetes, document steroid use and stress hyperglycemia, and evaluate dose–response by glycemic category. Besides, treatment delivery details were inconsistently reported across studies, and no article specified NCCN/EANO compliance or adherence/completion rates. This lack of detail may contribute to residual confounding related to treatment intensity and quality of care. Additionally, the analysis was based on aggregate study-level data rather than individual patient data, precluding more nuanced adjustment and subgrouping. Moreover, although preexisting diabetes was associated with poorer PFS in the pooled analysis, this finding is based on only four studies, limiting assessment of publication bias and overall certainty. The borderline significance and attenuation to non-significance after excluding the single unadjusted, lower-quality study suggest that the PFS association should be viewed as exploratory rather than conclusive. Finally, given the observational nature of the included studies, a causal relationship between diabetes and glioma outcomes cannot be definitively established.

From a clinical perspective, these findings support the consideration of diabetes as a potential adverse prognostic factor in patients with high-grade glioma. While tumor biology remains the dominant determinant of outcome, patient-level comorbidities such as diabetes may further modulate survival trajectories and should not be overlooked. Routine assessment of glycemic status and optimization of metabolic control may be important components of holistic glioma care. Moreover, future studies are warranted to explore whether tighter glycemic control or the use of specific antidiabetic agents, such as metformin, could improve outcomes in this patient population. Prospective studies with individual-level data collection on metabolic parameters, treatment regimens, and inflammatory markers will be essential to elucidate the mechanistic pathways linking diabetes to glioma progression and to identify modifiable targets for intervention. Pending definitive evidence, diabetes status should be routinely documented in neuro-oncology care, with baseline HbA1c and peri-treatment glucose monitoring and steroid-sparing strategies where clinically feasible. Although data on antidiabetic therapies, particularly metformin, were limited and inconsistently reported, their potential to modulate insulin/IGF signaling and tumor metabolism warrants prospective evaluation. Well-designed studies should distinguish prediabetes, newly diagnosed, and established diabetes; predefine glycemic targets; systematically record steroid exposure and stress hyperglycaemia; and assess whether metabolic optimization and specific antidiabetic agents improve survival or treatment tolerance—ideally in molecularly annotated cohorts.

In conclusion, this meta-analysis indicates that pre-existing diabetes is associated with poorer OS and PFS in patients with high-grade glioma. However, the effect sizes are modest and the evidence is entirely retrospective, leaving room for residual confounding. These findings support routine documentation of diabetes in neuro-oncology care and cautious consideration of diabetes status in prognostic modeling where validated. Prospective, molecularly annotated studies with standardized glycemic metrics, explicit reporting of steroid exposure and stress hyperglycaemia, and careful accounting of treatment intensity are needed to establish independence of effect and to test whether metabolic optimization can meaningfully improve outcomes.