The preoperative peripheral blood routine tests were conducted and retrospectively analyzed. All patients were confirmed to have glioma by the pathology department of the First Affiliated Hospital of Nanjing Medical University (Nanjing, China), and were classified according to the 2021 WHO classification of central nervous system tumors (4). The inclusion criteria were as follows: i) Patients had no previous antitumor treatments, including surgery, radiotherapy or chemotherapy; and ii) they underwent routine blood tests and biochemical examinations within 1 week prior to surgery. The exclusion criteria were: i) Patients with a history of other malignant tumors; ii) patients with acute or chronic inflammatory diseases; iii) individuals with diabetes, heart disease, hypertension, thyroid dysfunction or autoimmune diseases; iv) patients who received steroid treatment within the last 6 months; and v) patients with incomplete case and follow-up data.

The present study included clinical data from 571 patients with glioma who visited The First Affiliated Hospital of Nanjing Medical University between December 2014 and September 2021, along with 1,899 HCs from the hospital's Health Examination Center during the same period (Table I). Retrospective analysis compared a total of 22 indicators between the HC group and glioma group, including: Hemoglobin concentration (HGB); mean corpuscular hemoglobin concentration (MCHC); hematocrit (HCT); monocyte count (MONO#); white blood cell count (WBC); monocyte percentage (MONO%); red blood cell count (RBC); red cell distribution width-coefficient of variation (RDW-CV); lymphocyte percentage (LYMPH%); lymphocyte count (LYMPH#); mean corpuscular volume (MCV); mean corpuscular hemoglobin (MCH); mean platelet volume (MPV); basophil percentage (BASO%); basophil count (BASO#); eosinophil count (EOS#); eosinophil percentage (EOS%); platelet count (PLT); platelet distribution width (PDW); plateletcrit (PCT); neutrophil percentage (NEUT%); and neutrophil count (NEUT#). Hematological indicators were measured using 3 ml of fasting venous blood collected in the morning that was subsequently mixed in a sodium citrate anticoagulant tube and analyzed with a Sysmex AC-500 (Sysmex Corporation) fully automated immunoassay analyzer. All data were obtained from the last test conducted within the week prior to surgery. Enrolled patients were followed up via outpatient visits or phone calls. Follow-up occurred every 3 months for the first 2 years post-surgery, every 6 months from the third to the fifth year, and then twice annually thereafter until death or loss to follow-up. The survival endpoint for the present study was tumor-related death or the date of the last follow-up, with the last follow-up date being May 1, 2023. Overall survival (OS) was defined as the time from the first day after surgery until death or the date of the last follow-up.

Categorical variables were presented as case numbers, while continuous variables were expressed as the mean ± standard deviation. One-way ANOVA tests were performed to compare continuous variables across multiple groups, followed by Tukey's post-hoc test for pairwise comparisons. To address multicollinearity, the correlations among 22 preoperative peripheral blood routine indicators were analyzed using Spearman correlation analysis, and biomarkers with a correlation coefficient |r|<0.35 were retained to reduce redundancy. The optimal thresholds for preoperative blood indicators were determined using receiver operating characteristic (ROC) curves and binary logistic regression was applied. Survival analysis was performed using the Kaplan-Meier (K-M) method and differences between groups were assessed with log-rank tests. Biomarkers with multicollinearity were excluded based on variance inflation factor (VIF) calculations and the remaining biomarkers were included in the Cox proportional hazards model. The proportional hazards assumption was verified by plotting Schoenfeld residuals. The model's performance was evaluated using the concordance index (C-index) and calibration curves for 3-year survival rates. All statistical analyses were conducted using R software (version 4.2.2; R Foundation for Statistical Computing). P<0.05 was considered to indicate a statistically significant difference.

The HC group comprised 1,034 males (54.45%) and 865 females (45.55%), with an age range of 18 to 84 years and a median age of 48 years. The glioma group included 332 males (58.14%) and 239 females (41.86%), with an age range of 19 to 84 years and a median age of 56 years (Table I). Age was higher in the glioma group compared with the HC group (mean ± SD: 54.19±13.77 vs. 47.54±13.97 years, P=0.002). The sex distribution did not differ between groups (male, 58.14 vs. 54.45%; χ²=3.21, P=0.073). The distribution of patient data was as follows: i) Patients were separated into the HC group (n=1,899) and the glioma group (n=571); and ii) within the glioma group, patients were divided into three molecular subtypes: Astrocytoma (n=92), oligodendroglioma (n=77) and glioblastoma (n=402). Peripheral blood routine data (comprising 22 indicators) from 1,899 HCs and 571 patients with glioma were analyzed using one-way ANOVA, with results presented as the mean ± standard deviation (Table SI).

Due to the presence of multiple interrelated preoperative peripheral blood routine indicators, such as the NEUT% and NEUT#, EOS% and EOS#, retaining all indicators would affect the accuracy of subsequent analyses. Therefore, the present study first performed a Spearman correlation analysis of the 22 biomarkers and visualized the results in a heatmap (Fig. 1A). A total of 10 indicators with low correlation coefficients (|r|<0.35) (Fig. 1B) were retained for further study. The present study assessed the distribution differences of the same biomarkers between HCs and different glioma molecular subtypes (astrocytoma, oligodendroglioma and glioblastoma) (Fig. 1C-L, Table SI). HCs displayed significantly higher levels in HCT, RDW-CV, MCH, BASO#, EOS# and PDW compared with all glioma subtypes (P<0.0001). Conversely, NEUT% demonstrated a significant increase in value with the progression of glioma grades when compared with the HC group, and patients diagnosed with glioblastoma showed the highest levels compared with lower-grade gliomas and HCs (P<0.0001) (Fig. 1K). Despite significant differences between HCs and oligodendroglioma (P=0.0364) and between oligodendroglioma and glioblastoma (P=0.0181), no statistically significant difference was observed in MONO% with the progression of glioma grades (Fig. 1E).

To construct the binary logistic regression model, all of the biomarkers were first transformed into binary classification data based on their optimal cut-off values. Data greater than or equal to the optimal threshold were recorded as high-level, while the remaining data were classified as low-level (Table SII). Subsequently, a binary logistic regression model was used to test the classification performance of the remaining 10 biomarkers between the HC group and glioma group. The results indicated that, except for EOS#, all other biomarkers showed significant associations with the classification outcomes (Fig. 2A). Among them, the odds ratios (OR) for WBC, MONO% and NEUT% were 2.20 (95% CI: 1.73–2.80; P<0.001), 1.71 (95% CI: 1.37–2.14; P<0.001) and 3.26 (95% CI: 2.60–4.09; P<0.001), respectively, which means that as the levels of these biomarkers increased, the risk of developing glioma also increased significantly. By contrast, a decrease in the levels of the remaining six biomarkers indicated that the likelihood of developing glioma in healthy individuals significantly increased. This suggested that lower levels of these biomarkers were negatively associated with health. The ROC curve analysis for the classification model showed an area under the curve (AUC) of 0.9293 (P<0.0001), indicating excellent classification performance in differentiating the HC group from the glioma group (Fig. 2B).

A similar model was also constructed for the classification of the low-grade glioma (LGG) and high-grade glioma (HGG) groups using the nine biomarkers that showed significant effects in the aforementioned binary logistic regression model. Astrocytoma and oligodendroglioma were classified as LGGs, while glioblastoma was classified as a HGG. The results showed that the PLT (OR, 0.63; P=0.025) and NEUT% (OR, 3.17; P<0.001) were identified as key biomarkers for distinguishing between LGG and HGG. A decrease in PLT was associated with a significantly higher risk of developing HGG in patients with glioma, while an increase in NEUT% indicated a significantly elevated risk of HGG progression. However, the remaining biomarkers did not show any significant differences (P>0.05) (Fig. 2C). The ROC analysis yielded an AUC of 0.6900 (P<0.0001), suggesting some discriminative potential; however, the sub-0.70 AUC indicates limited performance and should not be taken as evidence of a strong diagnostic marker (Fig. 2D).

To evaluate whether preoperative blood routine indicators can predict survival (OS) in patients with glioma, patients with complete prognostic information were selected in the present study based on the WHO 2021 glioma grading guidelines (4). The group included 67 patients with astrocytoma, 63 patients with oligodendroglioma and 267 patients with glioblastoma. Survival analysis was first performed using the K-M method, with log-rank tests assessing intergroup differences. K-M survival curves were plotted for all 22 preoperative blood routine indicators and 6 biomarkers were statistically significant, as shown in Fig. 3 (all P<0.05). The results showed that these 6 biomarkers significantly affected the OS of the patients with glioma. Since LYMPH% and LYMPH#, NEUT% and NEUT# were highly clinically correlated (Fig. 1A), it was decided to include LYMPH#, MONO%, PCT and NEUT% in the multivariate Cox regression analysis. Previously, the VIF was calculated for these indicators and plotted in Table II, which showed that all Variance Inflation Factor (VIF) values were low (MONO%=1.34, LYMPH#=1.88, PCT=1.09, NEUT%=1.16), indicating that multicollinearity was not a concern in the present model.

Subsequently, a multivariable Cox regression analysis was performed, and residual plots were generated using the Schoenfeld residuals method (Fig. 4A). These plots indicated that the residual distributions of the remaining 4 indicators were uniform, suggesting independence from time and their suitability for inclusion in the multivariable Cox regression analysis (Table II). PCT showed a protective trend but did not reach statistical significance (P=0.116). NEUT% was the only independent prognostic factor with statistical significance (P<0.001). Although the associated hazard ratio of 1.035 indicated a relatively small effect size, NEUT% elevation was consistently and significantly associated with poor prognosis.

The model showed good overall fit (likelihood ratio χ²=22.45, df=4, P=<0.0001), acceptable discrimination (C-index=0.62, SD=0.02) (data not shown) and good 3-year calibration (Fig. 4B). The predicted probabilities, indicated by the red curve in Fig. 4B, align well with the actual observed outcomes, depicted by the gray dashed line, indicating that the model has relatively high predictive accuracy.