Translating the vast data generated by genomic platforms into accurate predictions of clinical outcomes is a fundamental challenge in genomic medicine. Many prediction methods face limitations in learning from the high-dimensional profiles generated by these platforms, and rely on experts to hand-select a small number of features for training prediction models. In this paper, we demonstrate how deep learning and Bayesian optimization methods that have been remarkably successful in general high-dimensional prediction tasks can be adapted to the problem of predicting cancer outcomes. We perform an extensive comparison of Bayesian optimized deep survival models and other state of the art machine learning methods for survival analysis, and describe a framework for interpreting deep survival models using a risk backpropagation technique. Finally, we illustrate that deep survival models can successfully transfer information across diseases to improve prognostic accuracy. We provide an open-source software implementation of this framework called SurvivalNet that enables automatic training, evaluation and interpretation of deep survival models.

Oligodendrogliomas are diffusely infiltrative gliomas defined by IDH-mutation and co-deletion of 1p/19q. They have highly variable clinical courses, with survivals ranging from 6 months to over 20 years, but little is known regarding the pathways involved with their progression or optimal markers for stratifying risk. We utilized machine-learning approaches with genomic data from The Cancer Genome Atlas to objectively identify molecular factors associated with clinical outcomes of oligodendroglioma and extended these findings to study signaling pathways implicated in oncogenesis and clinical endpoints associated with glioma progression. Our multi-faceted computational approach uncovered key genetic alterations associated with disease progression and shorter survival in oligodendroglioma and specifically identified Notch pathway inactivation and PI3K pathway activation as the most strongly associated with MRI and pathology findings of advanced disease and poor clinical outcome. Our findings that Notch pathway inactivation and PI3K pathway activation are associated with advanced disease and survival risk will pave the way for clinically relevant markers of disease progression and therapeutic targets to improve clinical outcomes. Furthermore, our approach demonstrates the strength of machine learning and computational methods for identifying genetic events critical to disease progression in the era of big data and precision medicine.

Translating the vast data generated by genomic platforms into accurate predictions of clinical outcomes is a fundamental challenge in genomic medicine. Many prediction methods face limitations in learning from the high-dimensional profiles generated by these platforms, and rely on experts to hand-select a small number of features for training prediction models. In this paper, we demonstrate how deep learning and Bayesian optimization methods that have been remarkably successful in general highdimensional prediction tasks can be adapted to the problem of predicting cancer outcomes. We perform an extensive comparison of Bayesian optimized deep survival models and other state of the art machine learning methods for survival analysis, and describe a framework for interpreting deep survival models using a risk backpropagation technique. Finally, we illustrate that deep survival models can successfully transfer information across diseases to improve prognostic accuracy. We provide an opensource software implementation of this framework called SurvivalNet that enables automatic training, evaluation and interpretation of deep survival models.Advanced molecular platforms can generate rich descriptions of the genetic, transcriptional, epigenetic and proteomic profiles of cancer specimens, and data from these platforms are increasingly utilized to guide clinical decision-making. Although contemporary platforms like sequencing can provide thousands to millions of features describing the molecular states of neoplastic cells, only a small number of these features have established clinical significance and are used in prognostication [1][2][3][4] . Making reliable and accurate predictions of clinical outcomes from high-dimensional molecular data remains a major challenge in realizing the potential of precision genomic medicine.Traditional Cox proportional hazards models require enormous cohorts for training models on high-dimensional datasets containing large numbers of features. Consequently, a small set of features is selected in a subjective process that is prone to bias and limited by imperfect understanding of disease biology. High-dimensional learning problems are common in the machine-learning community, and many machine-learning approaches have been adapted to predicting survival or time to progression 5 . Prior knowledge has been used to reduce dimensionality by learning gene signatures of cancer hallmarks to generate intermediate features that successfully predict outcomes 6,7 . Regularization methods for Cox models like elastic net have been developed to perform objective and data-driven feature selection with time-to-event data 8 . Random forests are reputed to resist overfitting in high-dimensional prediction problems, and have been adapted to survival modeling 9 . Neural network based approaches have been used in low-dimensional survival prediction problems 10 , but subsequent evaluation of these methods found no performance improvement over ordinary Cox regression 11. The difficulty of deconstructing these black...

Sepsis, a life-threatening organ dysfunction, is a clinical syndrome triggered by acute infection and affects over 1 million Americans every year. Untreated sepsis can progress to septic shock and organ failure, making sepsis one of the leading causes of morbidity and mortality in hospitals. Early detection of sepsis and timely antibiotics administration is known to save lives. In this work, we design a sepsis prediction algorithm based on data from electronic health records (EHR) using a deep learning approach. While most existing EHR-based sepsis prediction models utilize structured data including vitals, labs, and clinical information, we show that incorporation of features based on clinical texts, using a pre-trained neural language representation model, allows for incorporation of unstructured data without an explicit need for ontology-based named-entity recognition and classification. The proposed model is trained on a large critical care database of over 40,000 patients, including 2805 septic patients, and is compared against competing baseline models. In comparison to a baseline model based on structured data alone, incorporation of clinical texts improved AUC from 0.81 to 0.84. Our findings indicate that incorporation of clinical text features via a pre-trained language representation model can improve early prediction of sepsis and reduce false alarms.

Objective Sepsis has a high rate of 30-day unplanned readmissions. Predictive modeling has been suggested as a tool to identify high-risk patients. However, existing sepsis readmission models have low predictive value and most predictive factors in such models are not actionable. Materials and Methods Data from patients enrolled in the AllofUs Research Program cohort from 35 hospitals were used to develop a multicenter validated sepsis-related unplanned readmission model that incorporates clinical and social determinants of health (SDH) to predict 30-day unplanned readmissions. Sepsis cases were identified using concepts represented in the Observational Medical Outcomes Partnership. The dataset included over 60 clinical/laboratory features and over 100 SDH features. Results Incorporation of SDH factors into our model of clinical and demographic features improves model area under the receiver operating characteristic curve (AUC) significantly (from 0.75 to 0.80; P < .001). Model-agnostic interpretability techniques revealed demographics, economic stability, and delay in getting medical care as important SDH predictive features of unplanned hospital readmissions. Discussion This work represents one of the largest studies of sepsis readmissions using objective clinical data to date (8935 septic index encounters). SDH are important to determine which sepsis patients are more likely to have an unplanned 30-day readmission. The AllofUS dataset provides granular data from a diverse set of individuals, making this model potentially more generalizable than prior models. Conclusion Use of SDH improves predictive performance of a model to identify which sepsis patients are at high risk of an unplanned 30-day readmission.