A prognostic model of newly diagnosed glioblastoma based on clinical characteristics

Using isocitrate dehydrogenase (IDH) mutations to classify survival outcomes of patients with glioblastoma multiforme was recommended based on novel histopathological classication of brain tumors. Considering this novel classication, it is unclear whether the extent of tumor resection (EOR) is still important. The aim of this study was to investigate prognostic value of clinical factors (age, sex, EOR, status of IDH mutations, and adjuvant therapy) in patients with newly diagnosed glioblastoma.


Abstract Background
Using isocitrate dehydrogenase (IDH) mutations to classify survival outcomes of patients with glioblastoma multiforme was recommended based on novel histopathological classi cation of brain tumors. Considering this novel classi cation, it is unclear whether the extent of tumor resection (EOR) is still important. The aim of this study was to investigate prognostic value of clinical factors (age, sex, EOR, status of IDH mutations, and adjuvant therapy) in patients with newly diagnosed glioblastoma.

Methods
In total, 269 patients were retrospectively enrolled and randomly divided into training (n = 179) and validation (n = 90) cohorts. Clinical information and survival outcomes were acquired from inpatient records and follow-ups. After adjusting for risk coe cients, the independent prognostic factors were selected in a multivariable analysis to generate a model to evaluate survival outcomes. Additionally, a receiver operating characteristic curve was used to assess accuracy for predicting survival outcomes at 12, 15, 18, and 24 months.

Results
Total resection of the contrast-enhanced region, age ≤ 60 years, received chemotherapy, and IDH mutations were favorable independent factors for overall survival. Area under the curve (AUC) for prediction of survival in the training cohort was 0.815, 0.851, 0.849, and 0.836 at 12, 15, 18, and 24 months, respectively. In the validation cohort, the AUC for prediction of survival was 0.780, 0.807, 0.836, and 0.849 at 12, 15, 18, and 24 months, respectively.

Conclusion
Total resection of the contrast-enhanced region is still crucial and recommended for patients with glioblastoma. Our prognostic model was able to predict survival outcomes, especially for long-term survival prediction.

Background
Glioblastoma multiforme (GBM) is the most common primary malignant brain tumor [1] and constitutes 80% of all primary malignancies in the central nervous system. [2] The median overall survival (OS) is only 14.6 months for primary GBMs after surgery, radiation therapy, and chemotherapy. [3] The generation of prognostic models to identify favorable and harmful factors was crucial to improve prognosis prediction of patients with GBM. Based on the ndings from previous studies, factors such as the extent of tumor resection (EOR), [4] Karnofsky Performance Scale (KPS) at diagnosis, [5] molecular biomarkers, [6,7] adjuvant treatment, [5,[8][9][10] and age [11] were associated with the prognosis of patients with GBM.
The accepted optimal treatment for GBM was surgical resection with radiation therapy and chemotherapy. [12,13] As the initial treatment for GBM, tumor resection has played a crucial role in improving survival outcomes among patients with GBMs using a multivariable Cox regression analysis. [14,15] A previous study recommended a predictive system with a continuous variable for EOR that enabled the prediction of the survival time for patients individually. This system was a large contribution to the prediction of survival outcomes. However, the previous system did not include information regarding IDH mutations. Based on the recommendations outlined in the 2016 World Health Organization classi cation of tumors of the central nervous system, it was recommend that the isocitrate dehydrogenase (IDH) mutation status be used to predict survival outcomes of patients. [16] Moreover, in high-grade glioma, gross total resection only affected the OS of patients with IDH1/2 mutations without 1p/19q co-deletion. [17] It remains unclear whether totally removing the region of T1 contrast-enhanced of GBM improves the OS of patients with wild type IDH. [18] Therefore, a novel evaluative system including information regarding EOR and the IDH mutation status is necessary to assist in predicting prognosis and guiding individualized adjuvant therapy.
The aims of this study were: 1) to identify the independent clinical variables that were signi cant predictors of OS in patients with newly diagnosed GBM; 2) to generate a scoring system to quickly evaluate the prognostic risks of patients with GBM; 3) to investigate whether total resection of the enhanced region on T1-images was bene cial for patients with wild type IDH.

Materials And Methods
This study was approved by the local institutional review board.

Study design and patients
We conducted a retrospective study of patients who underwent initial surgical resection from March 29, 2005 to April 26, 2018 at ** Hospital and had histologically con rmed GBM. All patients aged 18 to 75 years were eligible for inclusion (n = 444). Patients were excluded for the following reasons: IDH mutation status not available (n = 109); had GBM with an oligo component (n = 43); incomplete clinical variables, postoperative treatment, and follow-up information during the study period (n = 23). Thus, in total, 269 patients were enrolled and randomly classi ed into training (n = 179) and validation (n = 90) cohorts ( Fig. sup-1).

Data variables and outcomes
The clinical variables derived from inpatient records, included age, sex, IDH mutation status, and records regarding radiation therapy and chemotherapy. Age was dichotomized as > 60 and ≤ 60 years. Additionally, the IDH mutation status which was tested at the Molecular Pathology Testing Center of ** Neurosurgical Institute was classi ed as mutation and wild type. Moreover, total resection was de ned as the gadolinium-enhanced region totally removed based on the preoperative and postoperative T1weighted contrast-enhanced MRI. If the contrast-enhanced region were not totally removed, the EOR would be de ned as non-total resection. The preoperative KPS score was dichotomized as > 80 and ≤ 80, corresponding to those with and without neurological impairment.
Time to death was derived based on the information in the follow-up records. The OS was calculated as the time interval between the date of tumor resection and the date of death or date that the patient was last investigated; the nal follow-up date was February 23, 2020.

Statistical analysis
Statistical analysis was performed using SPSS (19.0 version, IBM) and GraphPad Prism 7 (GraphPad Software Inc, San Diego, USA). Student t-tests were used to assess the difference in means for continuous variables while chi-squared tests were used to assess the differences in categorical variables.
A Cox proportional hazards model was used to investigate the univariate and multivariate predictors of survival. After identifying the signi cant (p < 0.05) independent predictors of OS in the multivariate analysis, these predictors were selected to generate a model to derive a risk score for each patient based on the beta value. In this risk system, the score of each item was calculated using the following formula: Item score = variable value × beta value Subsequently, the nal score for each patient was equal to the sum of the score for each item.
The receiver operating characteristic curve (ROC) was used to analyze the survival outcomes at 12, 15, 18, and 24 months and to determine the cut off value for the risk score at each survival time point. Based on the cut-off value, all patients were classi ed into high or low risk groups.
The Kaplan-Meier method was used to illustrate the survival curves. Survival curves for the various subgroups were compared between the low-risk and high-risk groups in both the training and validation cohorts.

ROC analysis of the training and validation cohorts at each time point
Based on the results of multivariate Cox analysis, age, EOR, chemotherapy, and the IDH mutation status were selected to build a model (Table 3). In the training cohort, the results of the ROC analysis are shown in Fig. 1a and

Survival curve for the high and low risk groups
Based on the cutoff values for the risk score, the patients were divided into low (risk score ≤ -0.942) and high (risk score > -0.942) risk groups. The median OS for patients in the low risk group was signi cantly longer than that for the patients in the high risk group in the both the training and validation cohorts, and among all patients, respectively (training cohort, low-risk vs high-risk = 938 vs 333 days, p < 0.0001; validation cohort, low-risk vs high-risk = 1122 vs 378 days, p < 0.0001; all patients, low-risk vs high-risk = 1105 vs 363 days, p < 0.0001, Fig. 2).

ROC analysis of the risk score system and other category methods
We compared the predictive effort by using the risk score system and independently using EOR, IDH mutation status, and age with the ROC analysis at 12, 15, 18, and 24 months after tumor resection. We found that the AUCs of the risk score system were larger than any other independent factors. (Fig. 3 Furthermore, the prognostic factors for OS in patients with IDH mutation were shown in supplemental part 1.

Discussion
The current recommended treatment for patients with newly diagnosed GBM is resection followed by adjuvant chemoradiotherapy. Based on the new histopathological criteria for glioma, the presence of IDH mutations was highlighted as important to the distinct prognosis of patients with GBM. However, it remains unknown whether total resection is bene cial for patients with wild type IDH GBM. The current study generated a risk scoring system that included information regarding clinical characteristics and IDH mutation status to evaluate the risk factors associated with the prognosis of GBM patients and to predict the survival outcomes at 12, 15, 18, and 24 months. It was determined that total resection, age < 60 years, received chemotherapy, and IDH mutation were independent factors associated with improved OS. Moreover, this study con rmed that totally removing the region of contrast-enhancement of GBM was bene cial for prolonging the OS of those with wild type IDH.
Total resection of the contrast-enhancement region of tumor has been proven the most bene cial factor for improving OS in patients newly diagnosed GBM patients. [1,4,19,20] Removing more than 78% of contrast-enhancement of GBM has also been shown to be valuable for OS. [4] Moreover, Sawaya et al. recommended that more than 53.21% of the surrounding region of the T2-FLAIR abnormality beyond the 100% contrast-enhancement resection should be removed since that was able to bring more bene t for prolonging OS than only removing the region of contrast-enhancement. [15] Our ndings were consistent with previous studies. In our risk scoring system, the impact coe cient of total resection was nearly twice that of other variables. Our ndings re-veri ed that total resection was essential for improving OS of patients with GBM. Furthermore, referring to the literature review recently published, whether patients with GBM and wild type IDH have bene ts associated with totally resection still unknown. [18] Our results con rmed that total resection would not only signi cantly prolong OS of patients with GBM and IDH mutations but also would prolong OS in patients with IDH wild type (supplemental part 2).
Radiation therapy plus chemotherapy is helpful for prolonging OS of patients with newly diagnosed GBM. [10,11] Although the period of enrolled patients was too long to implement uniform treatment strategies for chemotherapy, having received chemotherapy was an independent favorable factor for OS. Hence, our ndings suggest that chemotherapy was necessary for patients with newly diagnosed GBM. Additionally, our results showed that having received radiation therapy was not a signi cant bene cial prognostic factor for OS. However, our ndings did not suggest that radiation therapy was unnecessary for the treatment of newly diagnosed GBM. As various previous studies have shown, [21][22][23] radiation therapy was undoubtedly bene cial for the OS of patients with GBM. In our study, almost all patients (nearly 90%) received radiation therapy. Hence, there was no signi cant difference in OS between the patients with and without radiation therapy. Hence, the variable of receiving radiation therapy was not selected to generate our risk system because all patients were recommended to receive radiation therapy after GBM resection at present.
Although the mutation rate in primary GBM was much lower than that in secondary GBM, IDH mutations were reported to have a strong positive correlation with OS in GBM. [24,25] Our previous study showed that the IDH mutation status was able to predict the OS of patients with secondary GBM by using microRNA signature. In the current study, the IDH mutation status could be applied to classify the survival outcomes of newly diagnosed GBM patients. Using the criteria regarding histological tumors in neural central system (WHO 2016), it was recommended that molecular characteristics (IDH mutation, 1p/19q co-deletion) be used to diagnose different kinds of glioma. [3,16] However, to our knowledge, [1,6,12,26] our risk system was the rst evaluative system, generated based on clinical characteristics, information regarding adjuvant treatment, and the IDH mutation status, to classify the survival outcomes of patients with GBM. Our ndings remedied the lack of inclusion of molecular information in the evaluative systems that were previously recommended.
Our ndings revealed that age > 60 years was signi cantly associated with decreased survival; similar ndings have been shown in previous studies. [27][28][29][30] As has been previously demonstrated, elderly people may have a decreased ability to withstand neurological insults caused by the tumor, surgery, and/or adjuvant therapy. [27] In our study, there was no signi cant difference in the number of younger (≤ 60 years) and elderly (> 60 years) patients with total resection and chemotherapy. Consequently, we inferred that the reason that elderly age was still a negative prognostic factor because elderly people have fewer chances to undergo secondary surgical treatment or adjuvant therapy. [30] Moreover, we found a signi cant decline in the preoperative KPS score among elderly patients when compared with younger patients (p < 0.0001). This nding was consistent with those from previous studies, in which lower KPS scores were negatively associated with OS. [27,31] These ndings are likely explained by the fact that elderly patients may harbor tumors with different molecular pro les and resistance genes that confer more aggressive behaviors. [32] The main advantage of our evaluative system was its stability; both the training and validation cohorts used the same cut-off values to classify survival outcomes. Moreover, the evaluative system was generated based on clinical characteristics and the IDH mutation status; these variables were processed in a binary manner. Our ndings indicated that patients would possibly be able to live longer if they were younger than 60 years, received total resection and chemotherapy, and had an IDH mutation. Hence, our system will be e cient and accurate in assisting clinical neurosurgeons or oncologists to predict survival outcomes of patients. Additionally, we found that the AUC for survival prediction at 18 and 24 months was higher than that at 12 and 15 months in both the training and validation groups. This indicates that our system will be more advantageous for a long-term versus short-term survival prediction.
Although our system had some advantages for survival prediction, it was not externally validated to assess its robustness. In the future, multicenter trials will be performed to validate our risk system.

Conclusions
For patients with newly diagnosed GBM, being younger than 60 years of age, total resection of the contrast-enhanced region, having received chemotherapy, and IDH mutations were independent factors associated with a favorable OS. Our prognostic system based on these factors enables to help to guide treatment strategy of GBM because this system was stable and able to predict the survival outcomes, in particular the long-term survival.  The results of Kaplan-Meier analysis for the overall survival times in the training and validation cohorts and in all patients. The cut-off value to distinct low-risk and high-risk was equal to -0.942.