Prognostic value of preoperative hematological markers in patients with glioblastoma multiforme and construction of random survival forest model

Patients with GBM have a poor overall prognosis, with an overall survival time of only 12–15 months after surgery plus STUPP [1]. As a result, accurately predicting the prognosis of GBM patients is critical. According to research, the prognosis of GBM patients is affected by their age, tumor characteristics, treatment plan, and other factors [2]. However, the accuracy of predicting the prognosis of GBM is still limited, and more prognostic factors are needed to evaluate the prognosis of GBM patients. Preoperative systemic inflammatory response, coagulation function, and nutritional status of patients have all been shown in studies to influence the anti-tumor effect [3‐6]. Some preoperative hematological markers, such as NLR, MLR, PLR, FPR, SII (SII = NLR* platelet count), and PNI (PNI = albumin + 5* lymphocytes), have been linked to the prognosis of certain malignant tumors, including gastric cancer, esophageal cancer, colorectal cancer, and breast cancer [7‐12]. However, there is no agreement on the role of preoperative hematological markers in GBM, and more research is required.

The overall survival time of GBM patients can be improved after surgery plus the STUPP protocol, but the difference in prognosis and survival time between patients remains significant [13]. It is critical to screen out these patients with poor prognoses and accurately predict patient survival time after treatment.

In this study, the prognostic factors of 281 GBM patients treated at the first affiliated hospital of Zhengzhou University were systematically examined. Our goals are as follows: 1. to investigate the predictive value of preoperative peripheral blood inflammatory response, coagulation function, and nutritional status in patients with GBM. 2. Develop a GBM random forest prognosis model that incorporates clinical fundamentals, molecular pathology, imaging features, and preoperative peripheral blood markers. At the same time, we collected data on 115 GBM patients treated at the People’s Hospital of Henan Province to further validate the forecast model.

An accurate prognosis evaluation is critical for GBM treatment design and clinical management. However, current prognosis prediction models are unable to meet the medical needs of GBM patients. This study systematically examined the clinical features, molecular pathological features, magnetic resonance spectroscopy imaging features, and prognostic value of the preoperative peripheral blood inflammation index, coagulation index, and nutritional status index of 281 patients with GBM. A random forest prognosis model with peripheral blood markers was developed. This model accurately predicts GBM patients’ 3-year survival after surgical resection and the + STUPP regimen.

Cancer-related inflammation has emerged as a new cancer indicator in recent decades. Nutritional status, inflammatory status, and immune function are frequently thought to be related to the prognosis in patients with malignant tumors [14, 15]. Inflammation, as part of the tumor microenvironment, can promote tumor occurrence, development, angiogenesis, and metastasis, as well as affect tumor patients’ clinical outcomes. As a result, inflammatory cells in peripheral blood can interact with tumor cells directly or indirectly, promoting tumors’ malignant biological behavior [16]. Inflammatory conditions, such as pro-inflammatory cytokines, growth factors, and chemokines, directly contribute to cancer progression [17]. The tumor microenvironment also generates inflammatory mediators that contribute to cell apoptosis and angiogenesis [6, 18]. We discovered that preoperative NLR, MLR, PLR, FPR, and SII were associated with the prognosis of GBM patients who received surgical treatment plus the STUPP protocol. As a result, a high level of SII was an independent prognostic factor of GBM. Reasons for analysis: Neutrophils have the potential to alter the tumor microenvironment by secreting angiogenic growth factors (such as vascular endothelial growth factors and matrix metalloproteinases) and inhibiting the cytotoxic activity of other immune cells (such as activated T cells, natural killer cells, and so on) [19]. Lymphocytes, on the other hand, inhibit tumor progression and eliminate new tumor cells as part of anti-cancer immunity [20]. Monocytes and their highly specialized macrophages inhibit tumor cell migration, invasion, metastasis, tumor-related angiogenesis, and the anti-tumor immune response [21]. Monocytes will be recruited into the brain parenchyma as a source of tumor-associated macrophages under any pathological condition, particularly glioma [22]. Platelet-tumor aggregates are responsible for the recruitment of neutrophils and the release of factors related to tumor growth, metastasis, and angiogenesis during the first few hours of cancer cell colonization. Some studies, on the other hand, have found that platelet-transmitted factors can have cytotoxic effects on proliferating tumor cells and even enhance apoptosis [23]. As a result, inflammatory cells in the peripheral blood can interact with tumor cells directly or indirectly, promoting tumors’ malignant biological behavior. The SII is a comprehensive index that combines neutrophil, lymphocyte, and platelet counts and can indicate the level of inflammation in cancer patients [24]. Thrombocytopenia, neutropenia, and/or lymphopenia are frequently present in glioma patients, along with an increase in SII in peripheral blood, all of which can promote tumor cell differentiation, proliferation, and metastasis [25, 26].

A prognostic model is a model that predicts the outcome of diseases. It is commonly used in the clinic to predict disease progression, patient survival time, and the likelihood of developing a specific disease stage. It employs a variety of predictive factors and methods. Many researchers have attempted similar studies on the prognosis evaluation model for glioma patients. Rathore et al. [27] determined the likelihood of predicting the early recurrence of GBM after surgery using the radiochemical characteristics of edema around the tumor. This model can help guide super-total resection and/or postoperative intensive radiotherapy. Wang et al. [28] performed univariate and multivariate Cox regression analysis on GBM and human autophagy-related genes in the American Cancer Gene Atlas Database (TCGA). The genes neuroregulatory protein 1, integrin subunit 3, and microtubule-binding protein 1 light chain 3 were chosen to create a prognostic risk scoring model, and a prognostic nomogram was created that included autophagy characteristics, age, drug therapy, radiotherapy, and the IDH mutation. This model was proven useful. Gorlia et al. [29] developed a prognosis model of GBM patients treated with temozolomide based on age, surgical scope, the mini-mental state evaluation score, and MGMT methylation status, which was used to assess the factors influencing GBM patient survival. Some researchers developed a glioma prognosis model based on non-coding RNA and divided patients into high-risk and low-risk groups [30]. Peng et al. [31] conducted a thorough examination of PDI family members and discovered several new potential signal pathways involved in the progression of glioma. A glioma prognosis nomogram was developed based on the survival risk score of PDIs and other clinical factors. These studies have fully validated the prognostic evaluation model’s positive role in predicting patient prognosis. In recent years, an increasing number of molecular markers have been used to predict the prognosis of gliomas, but the research and development cycle for these markers is lengthy and expensive, preventing widespread application.

Based on the foregoing, this study gathered easily accessible and low-cost clinical data, hematological data, magnetic resonance imaging data, pathological and immunohistochemical data, and built a random forest prognosis model of GBM after surgical resection and the STUPP regimen. In the testing set and validation set, our prediction model’s sensitivity (0.870, 0.833), specificity (0.944, 0.967), and AUC (0.907, 0.900) respectively. Because of the small sample size, there is some over-fitting in the training set, resulting in a 100% accuracy rate. The reason for selecting this model is that it has the following advantages over other machine learning methods and traditional mathematical models: 1. it can be used for a wide range of data types. 2. It can handle numerous input variables. 3. It can assess the importance of variables. 4. It can maintain accuracy even when data is lost, and 5. It is faster to learn.

Without a doubt, this study has some limitations: 1. this is a retrospective study, the number of patients included is limited, and the follow-up time is inadequate. 2. Patients may receive different treatments after surgery, such as bevacizumab, and the duration of treatment may vary. This may cause research findings to deviate. More prospective clinical trials are needed to assess the prognostic value of hematological indicators in GBM patients. 3. The GBM random forest prognosis model developed in this study requires additional external validation.

High levels of NLR, MLR, PLR, FPR, and SII before surgery are prognostic risk factors for GBM patients. A high preoperative SII level is an independent risk factor for GBM prognosis. The random forest model that includes preoperative hematological markers has the potential to predict the individual GBM patient’s 3-year survival status after treatment, and assist the clinicians for making a good clinical decision.