NCOG-39. DISCOVERY OF CLINICAL AND DEMOGRAPHIC DETERMINANTS OF SYMPTOM BURDEN IN PRIMARY BRAIN TUMOR PATIENTS USING NETWORK ANALYSIS AND UNSUPERVISED CLUSTERING

BACKGROUND

Precision health approaches to managing symptom burden in primary brain tumor (PBT) patients are imperative to improving patient outcomes and quality of life. Network Analysis (NA) identifies complex symptom co-severity patterns across large patient cohorts. Unsupervised clustering unbiasedly stratifies patients into clinically relevant subgroups based on symptom patterns. This is the first study to use NA and unsupervised clustering to explore PBT patients’ clinical and demographic determinants of symptom burden.

METHODS

Symptom severity data reported using the MDASI-BT from a two-institutional cohort of 1,128 PBT patients was analyzed. Gaussian Graphical Model networks were constructed for the entire cohort and for subgroups identified by unsupervised clustering. Network characteristics were analyzed and compared using permutation-based statistical tests.

RESULTS

NA on the entire cohort revealed that the majority of PBT patients experience symptoms on four core dimensions that drive overall symptom burden: cognitive, physical, focal neurologic, and affective. These dimensions substantially overlap with factor groupings defined during initial construct validation of the MDASI-BT. Fatigue/drowsiness scored the highest in all network centrality measures, indicating a pivotal role in the symptom experience. Unsupervised clustering identified four patient subgroups: PC1 (n = 683), PC2 (n = 244), PC3 (n = 92), and PC4 (n = 109). Moderately accurate networks could be constructed for PC1 and PC2, but not for PC3 and PC4 due to their small size. The PC1 network closely resembles the all-patient network, and these patients have the highest interference scores among the subgroups with fatigue/drowsiness as the primary driver. PC2 represents an older subgroup in which cognitive symptoms drive symptom burden.

CONCLUSIONS

This novel study identified clinically relevant subgroups of patients with unique symptom burdens. With further validation, our approach may inform more personalized and effective symptom management by identifying symptoms to prioritize for targeting in individual patients.

Analysis of SARS-CoV-2 infection associated cell entry proteins ACE2, CD147, PPIA, and PPIB in datasets from non SARS-CoV-2 infected neuroblastoma patients, as potential prognostic and infection biomarkers in neuroblastoma

SARS-CoV-2 viral contagion has given rise to a worldwide pandemic. Although most children experience minor symptoms from SARS-CoV-2 infection, some have severe complications including Multisystem Inflammatory Syndrome in Children. Neuroblastoma patients may be at higher risk of severe infection as treatment requires immunocompromising chemotherapy and SARS-CoV-2 has demonstrated tropism for nervous cells. To date, there is no sufficient epidemiological data on neuroblastoma patients with SARS-CoV-2. Therefore, we evaluated datasets of non-SARS-CoV-2 infected neuroblastoma patients to assess for key genes involved with SARS-CoV-2 infection as possible neuroblastoma prognostic and infection biomarkers. We hypothesized that ACE2, CD147, PPIA and PPIB, which are associated with viral-cell entry, are potential biomarkers for poor prognosis neuroblastoma and SARS-CoV-2 infection. We have analysed three publicly available neuroblastoma gene expression datasets to understand the specific molecular susceptibilities that high-risk neuroblastoma patients have to the virus. Gene Expression Omnibus (GEO) GSE49711 and GEO GSE62564 are the microarray and RNA-Seq data, respectively, from 498 neuroblastoma samples published as part of the Sequencing Quality Control initiative. TARGET, contains microarray data from 249 samples and is part of the Therapeutically Applicable Research to Generate Effective Treatments (TARGET) initiative. ACE2, CD147, PPIA and PPIB were identified through their involvement in both SARS-CoV-2 infection and cancer pathogenesis. In-depth statistical analysis using Kaplan-Meier, differential gene expression, and Cox multivariate regression analysis, demonstrated that overexpression of ACE2, CD147, PPIA and PPIB is significantly associated with poor-prognosis neuroblastoma samples. These results were seen in the presence of amplified MYCN, unfavourable tumour histology and in patients older than 18 months of age. Previously, we have shown that high levels of the nerve growth factor receptor NTRK1 together with low levels of the phosphatase PTPN6 and TP53 are associated with increased relapse-free survival of neuroblastoma patients. Interestingly, low levels of expression of ACE2, CD147, PPIA and PPIB are associated with this NTRK1-PTPN6-TP53 module, suggesting that low expression levels of these genes are associated with good prognosis. These findings have implications for clinical care and therapeutic treatment. The upregulation of ACE2, CD147, PPIA and PPIB in poor-prognosis neuroblastoma samples suggests that these patients may be at higher risk of severe SARS-CoV-2 infection. Importantly, our findings reveal ACE2, CD147, PPIA and PPIB as potential biomarkers and therapeutic targets for neuroblastoma.

Abstract T MP52: Probabilistic Atlasing of Acute Ischemic Stroke Topology

Background: Knowing which areas of the brain are most vulnerable to acute ischemic stroke can focus care on more effective, targeted therapies. By displaying lesion incidence, these susceptible regions are clearly illustrated and can be correlated with characteristics common to that population. While previous studies created DWI lesion atlases, we aimed to map the spatio-temporal topology of multi-modal MRI sequences from a large cohort of stroke patients onto a standard atlas coordinate system.

Methods: Pre-treatment ADC and perfusion-weighted imaging (PWI) and post-treatment FLAIR MRI sequences for 241 acute ischemic stroke patients which were retrospectively processed. Lesion locations were circumscribed semi-automatically on the ADC and FLAIR images. Perfusion parameters were extracted from the PWI images using Bayesian hemodynamic parameter estimation. To account for anatomical variation, custom software was created to co-register the MRIs onto an atlas. Decomposing by NIHSS score items, the software then created probabilistic maps by overlaying and averaging the subgroup’s lesions and parameters within the atlas.

Results: An multi-dimensional set of annotated, co-registered stroke images for ADC, PWI and FLAIR sequences was established. Incidence maps based on NIHSS score items displayed anatomic localization of presenting symptoms in various imaging modes, before and after reperfusion or other therapies (Figure). A software suite was produced that co-registers, atlases, and calculates incidence maps for any set of images.

Conclusions: A framework to obtain spatial incidence maps was applied to ADC, PWI and FLAIR MRI of ischemic stroke. This instrument can co-register any annotated imaging parameters and compute lesion maps for any metric of interest. This novel atlasing method may be used to elucidate stroke etiology, predict lesion progression and identify optimal treatments based on individual imaging features at presentation.