Diffuse low-grade and intermediate-grade gliomas (together known as lower-grade gliomas, WHO grade II and III) develop in the supporting glial cells of brain and are the most common types of primary brain tumor. Despite a better prognosis for lower-grade gliomas, 70% of patients undergo high-grade transformation within 10 years, stressing the importance of better prognosis. Long noncoding RNAs (lncRNAs) are gaining attention as potential biomarkers for cancer diagnosis and prognosis. We have developed a computational model, UVA8, for prognosis of lower-grade gliomas by combining lncRNA expression, Cox regression and L1-LASSO penalization. The model was trained on a subset of patients in TCGA. Patients in TCGA, as well as a completely independent validation set (CGGA) could be dichotomized based on their risk score, a linear combination of the level of each prognostic lncRNA weighted by its multivariable cox regression coefficient. UVA8 is an independent predictor of survival and outperforms standard epidemiological approaches and previous published lncRNA-based predictors as a survival model. Guilt-by-association studies of the lncRNAs in UVA8, all of which predict good outcome, suggest they have a role in suppressing interferon stimulated response and epithelial to mesenchymal transition. The expression levels of 8 lncRNAs can be combined to produce a prognostic tool applicable to diverse populations of glioma patients. The 8 lncRNA (UVA8) based score can identify grade II and grade III glioma patients with poor outcome and thus identify patients who should receive more aggressive therapy at the outset.

Background: While clinical factors such as age, grade, stage, and histological subtype provide physicians with information about patient prognosis, genomic data can further improve these predictions. Previous studies have shown that germline variants in known cancer driver genes are predictive of patient outcome, but no study has systematically analyzed multiple cancers in an unbiased way to identify genetic loci that can improve patient outcome predictions made using clinical factors. Methods: We analyzed sequencing data from the over 10,000 cancer patients available through The Cancer Genome Atlas to identify germline variants associated with patient outcome using multivariate Cox regression models. Results: We identified 79 prognostic germline variants in individual cancers and 112 prognostic germline variants in groups of cancers. The germline variants identified in individual cancers provide additional predictive power about patient outcomes beyond clinical information currently in use and may therefore augment clinical decisions based on expected tumor aggressiveness. Molecularly, at least 12 of the germline variants are likely associated with patient outcome through perturbation of protein structure and at least five through association with gene expression differences. Almost half of these germline variants are in previously reported tumor suppressors, oncogenes or cancer driver genes with the other half pointing to genomic loci that should be further investigated for their roles in cancers. Conclusions: Germline variants are predictive of outcome in cancer patients and specific germline variants can improve patient outcome predictions beyond predictions made using clinical factors alone. The germline variants also implicate new means by which known oncogenes, tumor suppressor genes, and driver genes are perturbed in cancer and suggest roles in cancer for other genes that have not been extensively studied in oncology. Further studies in other cancer cohorts are necessary to confirm that germline variation is associated with outcome in cancer patients as this is a proof-of-principle study.

In this paper we propose a heterogeneous modeling framework which achieves individualwise feature selection and individualized covariates' effects subgrouping simultaneously. In contrast to conventional model selection approaches, the new approach constructs a separation penalty with multi-directional shrinkages, which facilitates individualized modeling to distinguish strong signals from noisy ones and selects different relevant variables for different individuals. Meanwhile, the proposed model identifies subgroups among which individuals share similar covariates' effects, and thus improves individualized estimation efficiency and feature selection accuracy. Moreover, the proposed model also incorporates within-individual correlation for longitudinal data to gain extra efficiency. We provide a general theoretical foundation under a double-divergence modeling framework where the number of individuals and the number of individual-wise measurements can both diverge, which enables inference on both an individual level and a population level. In particular, we establish strong oracle property for the individualized estimator to ensure its optimal large sample property under various conditions. An efficient ADMM algorithm is developed for computational scalability. Simulation studies and applications to post-trauma mental disorder analysis with genetic variation and an HIV longitudinal treatment study are illustrated to compare the new approach to existing methods.

This work is motivated by multimodality breast cancer imaging data, which is quite challenging in that the signals of discrete tumor-associated microvesicles (TMVs) are randomly distributed with heterogeneous patterns. This imposes a significant challenge for conventional imaging regression and dimension reduction models assuming a homogeneous feature structure. We develop an innovative multilayer tensor learning method to incorporate heterogeneity to a higher-order tensor decomposition and predict disease status effectively through utilizing subject-wise imaging features and multimodality information. Specifically, we construct a multilayer decomposition which leverages an individualized imaging layer in addition to a modality-specific tensor structure. One major advantage of our approach is that we are able to efficiently capture the heterogeneous spatial features of signals that are not characterized by a population structure as well as integrating multimodality information simultaneously. To achieve scalable computing, we develop a new bi-level block improvement algorithm. In theory, we investigate both the algorithm convergence property, tensor signal recovery error bound and asymptotic consistency for prediction model estimation. We also apply the proposed method for simulated and human breast cancer imaging data. Numerical results demonstrate that the proposed method outperforms other existing competing methods.

The use of RSL is a viable option in the community setting, with several benefits over WL. While operative times were slightly longer with RSL, there was no difference in specimen size or re-excision rate for malignant lesions.