Comparative Analysis of Survival Outcomes and Prognostic Factors of Supratentorial versus Cerebellar Glioblastoma in the Elderly: Does Location Really Matter?

Advancing precision prognostication in neuro-oncology: Machine learning models for data-driven personalized survival predictions in IDH-wildtype glioblastoma

Background

Glioblastoma (GBM) remains associated with a dismal prognoses despite standard therapies. While population-level survival statistics are established, generating individualized prognosis remains challenging. We aim to develop machine learning (ML) models that generate personalized survival predictions for GBM patients to enhance prognostication.

Methods

Adult patients with histologically confirmed IDH-wildtype GBM from the National Cancer Database (NCDB) were analyzed. ML models were developed with TabPFN, TabNet, XGBoost, LightGBM, and Random Forest algorithms to predict mortality at 6, 12, 18, and 24 months postdiagnosis. SHapley Additive exPlanations (SHAP) were employed to enhance the interpretability of the models. Models were primarily evaluated using the area under the receiver operating characteristic (AUROC) values, and the top-performing models indicated by the highest AUROCs for each outcome were deployed in a web application that was created for individualized predictions.

Results

A total of 7537 patients were retrieved from the NCDB. Performance evaluation revealed the top-performing models for each outcome were built using the TabPFN algorithm. The TabPFN models yielded mean AUROCs of 0.836, 0.78, 0.732, and 0.724 in predicting 6, 12, 18, and 24 month mortality, respectively.

Conclusions

This study establishes ML models tailored to individual patients to enhance GBM prognostication. Future work should focus on external validation and dynamic updating as new data emerge.

Exploring the relationship between age and prognosis in glioma: rethinking current age stratification

Background

The age of glioma plays a unique role in prognosis. We hypothesized that age is not positively correlated with survival prognosis and explored its exact relationship.

Methods

Glioma was identified from the SEER database (between 2000 and 2018). A multivariate Cox proportional regression model and restricted cubic spline (RCS) plot were used to assess the relationship between age and prognosis.

Results

A total of 66465 patients with glioma were included. Hazard ratios (HR) for ten-year by age: 0–9 years, HR 1.06 (0.93–1.20); 10–19 years: reference; 20–29 years, HR 0.90 (0.82–1.00); 30–39 years, HR 1.14 (1.04–1.25); 40–49 years, HR 2.09 (1.91–2.28); 50–59 years, HR 3.48 (3.19–3.79); 60–69 years, HR 4.91 (4.51–5.35);70–79 years, HR 7.95 (7.29–8.66); 80–84 years, HR 12.85 (11.74–14.06). After adjusting for covariates, the prognosis was not positively correlated with age. The smooth curve of RCS revealed this non-linear relationship: HR increased to 10 years first, decreased to 23 years, reached its lowest point, and became J-shaped.

Conclusion

The relationship between age and glioma prognosis is non-linear. These results challenge the applicability of current age groupings for gliomas and advocate the consideration of individualized treatment guided by precise age.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12883-022-02879-9.

Optimal managements of elderly patients with glioblastoma

Optimizing the management of elderly patients with glioblastoma is an ongoing task in neuro-oncology. The number of patients with this tumor type is gradually increasing with the aging of the population. Although available data and practice recommendations remain limited, the current strategy is maximal safe surgical resection followed by radiotherapy in combination with temozolomide. However, survival is significantly worse than that in the younger population. Surgical resection provides survival benefit in patients with good performance status. Hypofractionated radiotherapy decreases toxicities while maintaining therapeutic efficacy, thus improving treatment adherence and subsequently leading to better quality of life. The intensity of these treatments should be balanced with patient-specific factors and consideration of quality of life. This review discusses the current optimal management in terms of efficacy and safety, as well as future perspectives.

Novel Predictive Models for High-Value Care Outcomes Following Glioblastoma Resection

BACKGROUND

Treating patients with glioblastoma (GBM) requires extensive medical infrastructure. Individualized risk assessment for extended length of stay (LOS), nonroutine discharge disposition, and increased total hospital charges is critical to optimize delivery of care. Our study sought to develop predictive models identifying independent risk factors for these outcomes.

METHODS

We retrospectively reviewed patients undergoing GBM resection at our institution between January 2017 and September 2020. Extended LOS and elevated hospital charges were defined as values in the upper quartile of the cohort. Nonroutine discharge was defined as any disposition other than to home. Multivariate models for each outcome included covariates demonstrating P ≤ 0.10 on bivariate analysis.

RESULTS

We identified 265 patients undergoing GBM resection, with an average age of 58.2 years. 24.5% of patients experienced extended LOS, 22.6% underwent nonroutine discharge, and 24.9% incurred elevated total hospital charges. Decreasing Karnofsky Performance Status (KPS) (P = 0.004), increasing modified 5-factor frailty (mFI-5) index (P = 0.012), lower surgeon experience (P = 0.005), emergent surgery (P < 0.0001), and larger tumor volume (P < 0.0001) predicted extended LOS. Independent predictors of nonroutine discharge included older age (P = 0.02), decreasing KPS (P < 0.0001), and emergent surgery (P = 0.048). Nonprivate insurance (P = 0.011), decreasing KPS (P = 0.029), emergent surgery (P < 0.0001), and larger tumor volume (P = 0.004) predicted elevated hospital charges. These models were incorporated into an open-access online calculator (https://neurooncsurgery3.shinyapps.io/gbm_calculator/).

CONCLUSIONS

Several factors were independent predictors for at least 1 high-value care outcome, with lower KPS and emergent admission associated with each outcome. These models and our calculator may help clinicians provide individualized postoperative risk assessment to glioblastoma patients.

Rapid Brainstem Infiltration of a Cerebellar Glioblastoma

A 79-year-old female complained of a one-month history of imbalance and headache. Brain MRI showed an irregular rim enhancing solid and cystic mass centered in the superomedial left cerebellar hemisphere. Resection of the lesion was recommended; however, the patient opted to undergo the procedure the following month because of the nearby Christmas holidays. When the patient returned 30 days later, a new brain MRI showed an enlargement of the cerebellar mass, extending to the brainstem and infiltrating the left brachium pontis, left posterior aspect of the tegmentum of the pons, and posterolateral medulla oblongata. Subtotal resection was performed without complications, and pathology was compatible with a primary cerebellar glioblastoma negative for IDH1/2 gene mutation. Because of the poor prognosis, the patient and her family members opted for hospice treatment, with the patient dying three weeks later. This case illustrates that cerebellar glioblastoma can rapidly infiltrate the brainstem.

Nomograms for predicting the overall survival of patients with cerebellar glioma: an analysis of the surveillance epidemiology and end results (SEER) database

At present, our understanding of cerebellar glioma is still insufficient. This study collected information on patients in the SEER database to identify the predictive factors for patients with cerebellar glioma. Data from patients with cerebellar glioma diagnosed from 1975 to 2018 were retrieved from the Surveillance Epidemiology and End Results Database. We randomly divided the patients into a training group and a validation group, established a nomogram based on the training group, and used the validation group data to verify the clinical value of the model. A total of 508 patients were included in this study. Multivariate analysis was performed based on the data before randomization, and the results showed that the patient's age, WHO grade, histological type, and extent were significantly correlated with the survival rate. The C-index of the OS nomograms of the training cohort was 0.909 (95% CI, (0.880–0.938)) and 0.932 (95% CI, (0.889–0.975)) in the validation group. The calibration curve of OS for 3 and 5 years showed that there was good consistency between the actual survival probability and the predicted survival probability. For patients with cerebellar glioma, the age at diagnosis, WHO grade of the glioma, histological type, and extension are the four factors that most strongly affect the overall survival outcomes. Furthermore, our model may be a useful tool for predicting OS in these patients.