|Ahead of print publication
Prognostic factors and clinical nomogram predicting survival in high-grade glioma
Thara Tunthanathip1, Sanguansin Ratanalert2, Sakchai Sae-Heng1, Thakul Oearsakul1, Ittichai Sakarunchai1, Anukoon Kaewborisutsakul1, Thirachit Chotsampancharoen3, Utcharee Intusoma4, Amnat Kitkhuandee5, Tanat Vaniyapong6
1 Department of Surgery, Division of Neurosurgery, Faculty of Medicine, Prince of Songkla University, Songkhla, Thailand
2 School of Medicine, Mae Fah Luang University, Chiang Rai, Thailand
3 Department of Pediatrics, Division of Hematology/Oncology, Faculty of Medicine, Prince of Songkla University, Songkhla, Thailand
4 Department of Pediatrics, Division of Pediatric Neurology, Faculty of Medicine, Prince of Songkla University, Songkhla, Thailand
5 Department of Surgery, Division of Neurosurgery, Faculty of Medicine, Khon Kaen University, Khon Kaen, Thailand
6 Department of Surgery, Division of Neurosurgery, Faculty of Medicine, Chiang Mai University, Chiang Mai, Thailand
|Date of Submission||06-Apr-2019|
|Date of Decision||13-Aug-2019|
|Date of Acceptance||18-Oct-2019|
|Date of Web Publication||03-Nov-2020|
Department of Surgery, Division of Neurosurgery, Faculty of Medicine, Prince of Songkla University, Hat Yai, Songkhla 90110
Source of Support: None, Conflict of Interest: None
Background: Genomic-based tools have been used to predict poor prognosis high-grade glioma (HGG). As genetic technologies are not generally available in countries with limited resources, clinical parameters may be still necessary to use in predicting the prognosis of the disease. This study aimed to identify prognostic factors associated with survival of patients with HGG. We also proposed a validated nomogram using clinical parameters to predict the survival of patients with HGG.
Methods: A multicenter retrospective study was conducted in patients who were diagnosed with anaplastic astrocytoma (WHO III) or glioblastoma (WHO IV). Collected data included clinical characteristics, neuroimaging findings, treatment, and outcomes. Prognostic factor analysis was conducted using Cox proportional hazard regression analysis. Then, we used the significant prognostic factors to develop a nomogram. A split validation of nomogram was performed. Twenty percent of the dataset was used to test the performance of the developed nomogram.
Results: Data from 171 patients with HGG were analyzed. Overall median survival was 12 months (interquartile range: 5). Significant independent predictors included frontal HGG (hazard ratio [HR]: 0.62; 95% confidence interval [CI]: 0.40–0.60), cerebellar HGG (HR: 4.67; 95% CI: 0.93–23.5), (HR: 1.55; 95% CI: 1.03–2.32; reference = total resection), and postoperative radiotherapy (HR: 0.18; 95% CI: 0.10–0.32). The proposed nomogram was validated using nomogram's predicted 1-year mortality rate. Sensitivity, specificity, positive predictive value, negative predictive value, accuracy, and area under the curve of our nomogram were 1.0, 0.50, 0.45, 1.0, 0.64, and 0.75, respectively.
Conclusion: We developed a nomogram for individually predicting the prognosis of HGG. This nomogram had acceptable performances with high sensitivity for predicting 1-year mortality.
Keywords: High-grade glioma, nomogram, validation
|How to cite this URL:|
Tunthanathip T, Ratanalert S, Sae-Heng S, Oearsakul T, Sakarunchai I, Kaewborisutsakul A, Chotsampancharoen T, Intusoma U, Kitkhuandee A, Vaniyapong T. Prognostic factors and clinical nomogram predicting survival in high-grade glioma. J Can Res Ther [Epub ahead of print] [cited 2020 Dec 3]. Available from: https://www.cancerjournal.net/preprintarticle.asp?id=299884
| > Introduction|| |
The term high-grade glioma (HGG) refers to tumors that are classified as anaplastic astrocytoma (AA) (WHO Grade III) and glioblastoma (GBM) (WHO Grade IV) according to their anaplastic features., The treatment strategy for HGG is the highest resection, followed by radiotherapy and temozolomide for GBM or recurrent AA. However, the median survival time of AA and GBM was 2–5 years and 12–18 months, respectively.,,, The prognostic factors of the HGG have been reported in the literature. Age of patients, Karnofsky Performance Status (KPS), the extent of resection, postoperative radiotherapy, tumor grade, and histology were associated with outcome.,,,,,
Currently, a nomogram has been used to predict individualized median survival time and survival probabilities each time point in various diseases, particularly GBM. Nomograms which based on O6-methylguanine-DNA methyltransferase (MGMT) promoter methylation status were proposed to predict the survival of patients with newly diagnosed GBM.,,, Form literature review, accuracy, and precision of nomogram's prediction have been debated. Gittleman et al. developed an MGMT promoter methylation-based nomogram to estimate individual predicted 6-, 12-, and 24-month survival probabilities from training dataset and validated with independent dataset. The results were that the nomogram provided an individualized estimate of survival time. However, Parks et al. studied in the validation of MGMT promoter methylation-based nomogram predicting survival time in patients with GBM. The calculator gives both inaccurate and imprecise predictions that 23% of predictions were within 25% of the actual survival.
Since the estimation of this tool informs as predicted survival time and predicted probability, the individualized validation of survival among patients' challenges for general practice. Interpretation of prediction with continuous results is not simple, whereas the interpretation of a test with binary results is straight forward., For example, predicted the 1-year probability of nomogram is 50%, what we should interpret these results for an individual in the real-world applications. Therefore, determining the most appropriate cutoff point of nomogram may be alternative methods for validating outcome as binary classifiers., Moreover, genomic technologies and services have been clustered in some centers in a real-world situation. In countries with limited resources, clinical factors have still used for predicting prognosis. Hence, we aimed to develop and validate the clinical-based nomogram to predict survival of patients with HGG for using in the general practice.
| > Methods|| |
Study designs and population
The study was a multicenter, retrospective cohort review of medical records of the three university hospitals (Project of Hospital-Based Central Nervous System Tumor Registry: Multicenter study). We enrolled consecutive patients who were newly diagnosed with AA or GBM. The inclusion criteria for the study were patients who had histologically confirmed by a pathologist between January 2009 and December 2017. The data comprised of the demographics, neuroimaging, treatment, and outcome.
The KPS score is a scale for evaluating functional impairment. These scores range from 0 to 100. Therefore, KPS scores which were dichotomized into two groups included KPS score <80 and >80 groups. Magnetic resonance images (MRIs) of the brain were reviewed to estimate tumor size, tumor location, and other characteristics of the tumor by neurosurgeons. The postoperative residual tumor was measured from postoperative MRI or contrast-enhanced computerized tomography of the brain.
According to Vecht et al., the extent of resection was postoperatively assessed. Gross total resection was defined as gross macroscopic tumor resection or when the surgeon felt that only a minimal amount of tumor (<5% of residual tumor) was detected on postoperative neuroimaging. Subtotal resection was defined as resection after which 5% to <25% of the residual tumor was visible on postoperative neuroimaging. Partial resection was defined as resection after which more than 25% of the residual tumor was evident on postoperative neuroimaging. Moreover, a biopsy was defined as an operation for tissue diagnosis only, and no attempt was made to remove the tumor.
For outcome assessment, the follow-up data were collected until December 2018 including update status (death or survival) and cause of death. Follow-up data were collected mainly when patients visited outpatient clinics and/or their relatives and death record from the local municipality.
Nomogram development and deployment
Using split methods for validating nomogram, the total data were spat into developing dataset (80%) and deploying dataset (20%) as in [Figure 1]. Cox proportional hazard regression analyses were used for each fold, and the predictive models were developed from the developing dataset. Using the significant parameters (P < 0.05), nomogram from each fold was developed by Zhang and Kattan method  for predicting 1-, 2-, and 5-year mortalities. The bootstrap method with 1000 replicates was applied for the internal validity of each model. The “rms” package was used to develop nomogram and analyze the bias-corrected concordance index that evaluated the predictive discrimination of the model. The concordance index is the probability of concordance between predicted probability and response.
|Figure 1: Workflow of nomogram development and deployment. Using split test, 80% of total data (white boxes) were used for developing nomogram from Cox regression analysis, and 20% of total data (Gray box) were used for testing performances of the nomogram|
Click here to view
From the deploying dataset, the nomogram's performances were evaluated, 1-, 2-, and 5-year mortalities, as binary classifiers (death or survival) with an optimal cutoff total point. Using the receiver operating characteristic (ROC) curve and the largest area under the ROC (AUC) chose the optimal cutoff point in each nomogram. Furthermore, the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy were determined in each nomogram. The statistical analysis was performed using the R version 3.4.0 software (R Foundation, Vienna, Austria). Moreover, ROC and AUC were created by “PlotROC” package.
The study was performed with permission from the research ethics committee (REC 61-203-10-1).
| > Results|| |
Clinical characteristics of the 171 patients with HGG are shown in [Table 1]. Two-third of HGG was dominant in males with the mean age of 50.2 (standard deviation [SD] 15.3) years. Common presentations were hemiparesis, progressive headache, and seizure. The seizure was observed as the first presentation in 30% and 25.5% of patients with AA and GBM, respectively. The common tumor location involved the frontal and temporal lobe in one-third of cases. The periventricular, basal ganglion, and pineal HGG were found in 5.3%, 0.6%, and 0.6%, respectively. In addition, the mean tumor volume was 5.2 (SD 1.7), and multiple HGG was observed in 20.5%. The most common of the extent of resection was partial resection, whereas the rates of total, subtotal, and biopsy were 28.7%, 3.5%, and 13.5%, respectively. Most of the patients (88.3%) underwent radiotherapy after resection. In addition to GBM treatment, temozolomide was used in one-third of cases for concomitant adjuvant therapy. From a mean follow-up of 19.4 months (SD 24.1), the overall median survival time was 26 months (95% confidence interval [CI]: 19.0–41.0), whereas the 1-, 2-, and 5-year survival probabilities were 45.3%, 16.0%, and 3.3%, respectively. The median survival time of AA was 12 months (95% CI: 7–26), whereas GBM was 11 months (95% CI: 9–13) that there was no significant difference by log-rank test (P = 0.4).
|Table 1: Demographic data of high-grade astrocytoma (WHO Grade III-IV) (n=171)|
Click here to view
In univariate analysis, significant independent predictors included frontal HGG (hazard ratio [HR]: 0.56; 95% CI: 0.37–0.85), cerebellar HGG (HR: 6.43; 95% CI: 1.52–27.16), biopsy (HR: 1.54; 95% CI: 1.03–2.30; reference = total resection), and postoperative radiotherapy (HR: 0.15; 95% CI: 0.08–0.26). Therefore, the significant model of the multivariable analysis consisted of frontal HGG (HR: 0.62; 95% CI: 0.40–0.60), cerebellar HGG (HR: 4.67; 95% CI: 0.93–23.5), biopsy (HR: 1.55; 95% CI: 1.03–2.32; reference = total resection), and postoperative radiotherapy (HR: 0.18; 95% CI: 0.10–0.32).
Using split validation, developing dataset was used for Cox proportional hazard regression analysis as shown in [Table 2] and nomogram development as shown in [Figure 2]. Therefore, deploying dataset was used for testing the performance of nomogram. The application of nomogram is simple in general practice. For example, in [Figure 3], a 48-year-old male with corpus callosum tumor (25 points from the nonfrontal group) underwent biopsy (15 points). The pathological diagnosis was GBM, and the patient received the concurrent radiotherapy (no point). Consequently, total points equaled 40 points which approximately corresponds to 50%–60% of 1-year survival probability, 10%–15% of 2-year survival probability, <2% of 5-year survival probability, and 15–20 months of median survival time.
|Table 2: Cox proportional hazard regression analysis for mortality of the highest performance nomogram (n=137)|
Click here to view
|Figure 2: The nomogram predicts 1-, 2-, and 5-year survival probabilities and median survival time (months). To use the nomogram, draw a straight line upward from the patient's characteristics of the frontal tumor, cerebellar tumor, the extent of resection, radiotherapy to the upper points scale, and the sums of the scores of all variables. Then, draw another straight line down from the scale of the total points through the 1-, 2-, 5-year, and median survival time. This is the probability of the presence of prognosis in an individual|
Click here to view
|Figure 3: Splenium tumor. A 48-year-old male with infiltrative tumor at the splenium of the corpus callosum. (a) Axial postcontrast T1-weighted image. (b) Coronal postcontrast T1-weighted image. (c) Sagittal postcontrast T1-weighted image. The patient underwent biopsy, and pathological diagnosis was glioblastoma. Therefore, the patient received the radiotherapy and died in 17 months after surgery|
Click here to view
[Figure 4] shows the nomogram's performance. For predicting 1-year mortality, the nomogram had good performances showing a sensitivity of 1.0, specificity of 0.50, PPV of 0.45, NPV of 1.0, accuracy of 0.64, and AUC of 0.75 while nomogram's performances dropped for predicting 2- and 5-year mortalities as shown in [Table 3].
|Figure 4: Receiver operating characteristic curve and area under the curve of a nomogram predicting mortality with cutoff 15 points. (a) Predicted 1-year mortality. (b) predicted 2-year mortality. (c) predicted 5-year mortality|
Click here to view
|Table 3: Performance of nomogram for predicting 1-, 2-, and 5-year mortalities from validation data at cutoff 15-point|
Click here to view
| > Discussion|| |
In the present cohort study, we observed poor prognosis of HGG. There was no significant difference between AA and GBM survivals. Similarly, Noiphithak and Veerasarn reported that the survivals of Thai patients with AA and GBM were not different between groups. Equally, prior western studies have been reported median survival time of HGG range 1–5 years. According to histology, AA had survival time significantly longer than GBM depended on each cohort.,, For the primary analyses, factors significantly associated with survival were some tumor location, the extent of resection, and postoperative adjuvant therapies.
Paldor et al. reported that frontal GBM had a better prognosis than nonfrontal GBM. Similarly, HGG localized to the frontal lobe had a significantly better prognosis than nonfrontal HGG. Because the frontal tumors are generally more amenable to complete surgical resection, these carry a better prognosis. Besides, cerebellar HGGs were reported that these groups had a poorer prognosis than supratentorial HGG. From previous studies, HGG localized in the cerebellum had an independent poor prognostic significance as shown by Cox proportional hazard regression analysis., In the present study, the median survival time of cerebellar HGG was 2 months, whereas noncerebellar HGG had median survival time 148 months. The results in our cohort showed concordance outcome with the prior study.
The potential effects of treatments on prognosis were the extent of resection, RT in AA, and radiotherapy with temozolomide in GBM.,, However, the present study had shorter survival compared with prior studies. Although the efficacy of chemotherapy significantly prolongs survival, temozolomide has been limited in certain health welfares in Thailand. Accessibility to temozolomide was about 32.6% of GBM. Therefore, survival in the present cohort had poorer than the literature. However, the use of postoperative radiotherapy for AA and GBM was seen to be independent favorable prognostic factors in the present study.
A nomogram is a simple tool which predicts the prognosis. However, biomarker-based nomograms may have limitations for real-world practice because genetic technologies have not still been worldwide available and have increased the cost of treatment. Furthermore, the lack of nomogram's validation has been observed from the literature review.
Parks et al. validated MGMT-based nomograms for predicting median survival time in a patient with GBM that there was only a weak-positive correlation between the predicted and actual survival among patients (R2 of 0.07). In addition, Gittleman et al. validated nomogram for individualized estimation of survival among patients with newly diagnosed GBM from independent validation datasets that discussed some limitations from switching the training and independent validation datasets for comparisons such as the Cox proportional hazard regression model estimates, the nomogram point assignments, or the concordance indices. Therefore, we proposed nomogram validation as binary classifiers each time point for testing tool's performance. The nomogram of our cohort had acceptable performances for predicting 1-year mortality that had a high level of sensitivity. For general practice, the high-sensitivity nomogram could be applied as a screening tool for decision-making treatment strategies. Because of the high cost in neurooncology treatment such as chemotherapy or genetic technologies, clinical parameters have still been necessary for predicting prognosis in the real-world setting, and cost-effective analysis of genetic technologies are needed to evaluate for the maximum health benefits in countries with limited resources. However, nomogram's performances dropped for predicting 2- and 5-year mortalities that need external validation in the future.
Finally, certain limitations of the present study should be acknowledged. As the retrospective design, the possibility of bias and confounding factors cannot be excluded. However, we presented to adjust the model with multivariable analysis for tackling this limitation. For the future work in this field, the external validation should be prospectively conducted to test this nomogram's performance in the future. In addition, the IDH1mutation of HGGs did not perform in the present study, because these genetic investigations have not routinely estimated in Thailand.
| > Conclusion|| |
Our study proposed nomogram using clinical predictors. This nomogram had acceptable performances and a high level of sensitivity for predicting 1-year mortality. For implication, the high-sensitivity nomogram could be useful to guide health-care workers for decision-making treatment strategies and advising patients.
The authors would like to offer special thanks to Assoc. Prof. Paramee Thongsuksai for advice about the manuscript preparation. In addition, the authors would like to thank Mrs. Supaporn Sainamsai for interdepartment and interinstitute coordination.
Financial support and sponsorship
This study was financially supported by the Faculty of Medicine, Prince of Songkla University, Thailand.
Conflicts of interest
There are no conflicts of interest.
| > References|| |
Komori T. The 2016 WHO classification of tumours of the central nervous system: The major points of revision. Neurol Med Chir (Tokyo) 2017;57:301-11.
Casartelli G, Dorcaratto A, Ravetti JL, Sola S, Vitali A, Merlo DF, et al.
Survival of high grade glioma patients depends on their age at diagnosis. Cancer Biol Ther 2009;8:1719-21.
Buckner JC. Factors influencing survival in high-grade gliomas. Semin Oncol 2003;30:10-4.
Stupp R, Mason WP, van den Bent MJ, Weller M, Fisher B, Taphoorn MJ, et al.
Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma. N Engl J Med 2005;352:987-96.
Wen PY, Kesari S. Malignant gliomas in adults. N Engl J Med 2008;359:492-507.
Noiphithak R, Veerasarn K. Clinical predictors for survival and treatment outcome of high-grade glioma in Prasat Neurological Institute. Asian J Neurosurg 2017;12:28-33.
] [Full text]
Ushio Y, Kochi M. Prognostic factors in malignant gliomas. Gan To Kagaku Ryoho 1996;23:643-8.
Gorlia T, van den Bent MJ, Hegi ME, Mirimanoff RO, Weller M, Cairncross JG, et al.
Nomograms for predicting survival of patients with newly diagnosed glioblastoma: Prognostic factor analysis of EORTC and NCIC trial 26981-22981/CE.3. Lancet Oncol 2008;9:29-38.
Gittleman H, Lim D, Kattan MW, Chakravarti A, Gilbert MR, Lassman AB, et al.
An independently validated nomogram for individualized estimation of survival among patients with newly diagnosed glioblastoma: NRG oncology RTOG 0525 and 0825. Neuro Oncol 2017;19:669-77.
Molitoris JK, Rao YJ, Patel RA, Kane LT, Badiyan SN, Gittleman H, et al.
Multi-institutional external validation of a novel glioblastoma prognostic nomogram incorporating MGMT methylation. J Neurooncol 2017;134:331-8.
Woo P, Ho J, Lam S, Ma E, Chan D, Wong WK, et al.
A comparative analysis of the usefulness of survival prediction models for patients with glioblastoma in the temozolomide era: The importance of methylguanine methyltransferase promoter methylation, extent of resection, and subventricular zone location. World Neurosurg 2018;115:e375-85.
Parks C, Heald J, Hall G, Kamaly-Asl I. Can the prognosis of individual patients with glioblastoma be predicted using an online calculator? Neuro Oncol 2013;15:1074-8.
Habibzadeh F, Habibzadeh P, Yadollahie M. On determining the most appropriate test cut-off value: The case of tests with continuous results. Biochem Med (Zagreb) 2016;26:297-307.
Hajian-Tilaki K. The choice of methods in determining the optimal cut-off value for quantitative diagnostic test evaluation. Stat Methods Med Res 2018;27:2374-83.
Akobeng AK. Understanding diagnostic tests 3: Receiver operating characteristic curves. Acta Paediatr 2007;96:644-7.
Tunthanathip T, Sae-Heng S, Oearsakul T, Sakarunchai I, Kaewborisutsakul A, Taweesomboonyat C. Machine learning applications for the prediction of surgical site infection in neurological operations. Neurosurg Focus 2019;47:E7.
Vecht CJ, Avezaat CJ, van Putten WL, Eijkenboom WM, Stefanko SZ. The influence of the extent of surgery on the neurological function and survival in malignant glioma. A retrospective analysis in 243 patients. J Neurol Neurosurg Psychiatry 1990;53:466-71.
Zhang Z, Kattan MW. Drawing nomograms with R: Applications to categorical outcome and survival data. Ann Transl Med 2017;5:211.
Swets JA. ROC analysis applied to the evaluation of medical imaging techniques. Invest Radiol 1979;14:109-21.
Paldor I, Pearce FC, Drummond KJ, Kaye AH. Frontal glioblastoma multiforme may be biologically distinct from non-frontal and multilobar tumors. J Clin Neurosci 2016;34:128-32.
Karremann M, Rausche U, Roth D, Kühn A, Pietsch T, Gielen GH, et al.
Cerebellar location may predict an unfavourable prognosis in paediatric high-grade glioma. Br J Cancer 2013;109:844-51.
Babu R, Sharma R, Karikari IO, Owens TR, Friedman AH, Adamson C. Outcome and prognostic factors in adult cerebellar glioblastoma. J Clin Neurosci 2013;20:1117-21.
Mann CJ. Observational research methods. Research design II: Cohort, cross sectional, and case-control studies. Emerg Med J 2003;20:54-60.
[Figure 1], [Figure 2], [Figure 3], [Figure 4]
[Table 1], [Table 2], [Table 3]