Glioblastoma and Blood Microenvironment Predictive Model for Life Expectancy of Patients

Abstract

Background: Glioblastoma multiforme (GBM) is a very malignant brain tumor. GBM exhibits cellular and molecular heterogeneity that can be exploited to improve patient outcomes by individually tailoring chemotherapy regimens. Objective: Our objective was to develop a predictive model of the life expectancy of GBM patients using data on tumor cells' sensitivity to chemotherapy drugs, as well as the levels of blood cells and proteins forming the tumor microenvironment. Methods: The investigation included 31 GBM patients from the Almazov Medical Research Centre (Saint Petersburg, Russia). The cytotoxic effects of chemotherapy drugs on GBM cells were studied by an MTT test using a 50% inhibitory concentration (IC₅₀). We analyzed the data with life expectancy by a one-way ANOVA, principal component analysis (PCA), ROC, and Kaplan-Meier survival tests using GraphPad Prism and Statistica 10 software. Results: We determined in vitro the IC₅₀ of six chemotherapy drugs for GBM and 32 clinical and biochemical blood indicators for these patients. This model includes an assessment of only three parameters: IC₅₀ of tumor cells to carboplatin (CARB) higher than 4.115 μg/mL, as well as levels of band neutrophils (NEUT-B) below 2.5% and total protein (TP) above 64.5 g/L in the blood analysis, which allows predicting with 83.3% probability (sensitivity) the life expectancy of patients for 15 months or more. In opposite, a change in these parameters-CARB above 4115 μg/mL, NEUT-B below 2.5%, and TP above 64.5 g/L-predict with 83.3% probability (specificity) no survival rate of GBM patients for more than 15 months. The relative risk for CARB was 6.41 (95 CI: 4.37-8.47, p = 0.01); for NEUT-B, the RR was 0.40 (95 CI: 0.26-0.87, p = 0.09); and for TP, it was 2.88 (95 CI: 1.57-4.19, p = 0.09). Overall, the model predicted the risk of developing a positive event (an outcome with a life expectancy more than 10 months) eight times (95 CI 6.34-9.66, p < 0.01). Cross k-means validation on three clusters (n = 10) of the model showed that its average accuracy (sensitivity and specificity) for cluster 1 was 74.98%; for cluster 2, it was 66.7%; and for cluster 3, it was 60.0%. At the same time, the differences between clusters 1, 2, and 3 were not significant. The results of the Sobel test show that there are no interactions between the components of the model, and each component is an independent factor influencing the event (life expectancy, survival) of GBM patients. Conclusions: A simple predictive model for GBM patients' life expectancy has been developed using statistical analysis methods.

Keywords: band neutrophils; blood cells; blood proteins; carboplatin; chemotherapy; glioblastoma; life expectancy of patients; predictive model; total protein; tumor microenvironment.

Figure 1

Principal component analysis for detection of main factors from (A) chemotherapy and NGF, (B) blood cells, (C) blood proteins, and (D) blood parameters in lifespan of GBM patients. The values on the X and Y axes represent the level of dispersion for each drug. The higher the value, the higher the dispersion (spread) of values for the drug.

Figure 2

A heat map and a false color image with a dendrogram added to the left side and to the top. The intensity of the color determines the magnitude of the correlation coefficient. The darker the color, the greater the correlation. The brackets at the top and bottom indicate the blood parameters between which the correlation was calculated.

Figure 3

PCA for predictive life expectancy model. The observations in the space of the first three principal components. Colors indicate the following: green—life expectancy higher than 10 months; red—life expectancy less than or equal to 10 months; black—life expectancy unknown. The values on each axis represent the level of dispersion for each patient.

Figure 4

Comparison of levels of (A) the IC₅₀ of CARB, (B) band neutrophils, and (C) total protein in the blood with the lifespan of GBM patients. The vertical dotted line separates the groups of patients with low and high life expectancy.

Figure 5

ROC analysis for AUC, sensitivity, and specificity of (A) IC₅₀ of CARB, (B) band neutrophils, and (C) total protein for prediction of GBM patients’ lifespan. Se—sensitivity; Sp—specificity.

Figure 6

Kaplan–Meier survival analysis for GBM patients based on levels of (A) IC₅₀ of CARB, (B) band neutrophils, and (C) total protein. Threshold values: CARB 4115 μg/mL, NEUT-B 2.5%, TP 64.5 g/L. Sign * indicates statistically significant differences at p < 0.05.

Figure 7

Three-component model, CARB-NEUT-B-TP: (A) ROC-curve, (B) survival analysis, and (C) PCA analysis. Threshold values: CARB 4115 μg/mL, NEUT-B 2.5%, TP 64.5 g/L.

Figure 8

Relative risks of the three main predictors (CARB, NEUT-B, TP) and the whole model.

Figure 9

Multivariate linear regression analysis of the predictive model, CARB-NEUT-B-TP, of the life expectancy of GBM patients: (A) actual vs. predicted plot, (B) residual vs. order plot, (C) homoscedasticity plot, and (D) QQ plot. Each point on the graphs represents a patient in whom the three factors we studied influenced life expectancy.

Figure 10

Cross k-means validation between three clusters (n = 10). ac—accuracy.

Figure 11

Comparison of patient survival prognosis in the predictive model with the theoretical (ideal) model for CARB, band neutrophils, and total protein as predictors and their threshold values. C—carboplatin; N—neutrophil; T—total protein. Threshold values: CARB 4115 μg/mL, NEUT-B 2.5%, and TP 64.5 g/L. C-N-T—whole model carboplatin-band neutrophils-total protein.