Using Medbook Workbench To Create Evidence Streams To Guide Medical Decisions

Background: Clinical datasets typically have small sample size limiting the ability to generate inference with strong statistical significance. By combining: 1) existing datasets; 2)integrated analysis of different data types 3) curated pathways and 4) state of the art machine learning classifiers into one analysis pipeline, clinicians and researchers may be able to combine multiple observations to spot a pattern in an individual patient sample to base a critical decision about therapy. However, few tools exist to enable a diverse set of analytical results to be coherently pooled and disseminated to a research team. New web-based modalities are urgently needed to meet the demands of new genomics-based oncology use cases. Methods: Building on the success of MSKCC cBioPortal and UCSC Cancer Genomics Browser, an easy to use workbench allows users to generate new datasets by integrating these types of data: Import public datasets from NCBI, Wrangle data into standard format, Import clinical from OnCore, RNA Seq analysis and mutation analysis, Unsupervised clustering analysis, Sharing Data with collaborators in a secure manner, Differential gene expression analysis, Pathway enrichment analysis, Training classifiers to recognize events on existing datasets, Applying classifiers to new datasets to infer molecular events. We have created a new Medical Information System called MedBook to give a context for integrating multiple observations about patients and their biopsies into a unified social network. Data can be shared with clinicians and collaborators in a secure manner. Borrowing terminology from systems such as FaceBook and Google Plus, we describe an organization of collaborative analyses as evidence “streams” that allow sharing, annotation, and nucleation points for further analyses. Streams are composed of “evidence cards” that encapsulate figures and/or tables. Discussion threads allow interpretation and commenting on findings associated with each evidence card. Results: We demonstrate the utility of the stream concept using a gene expression based signature that predicts small cell disease in castration resistant prostate cancer. Importantly, the stream concept allows 1) redefinition of the signature to incorporate additional patient samples and clinical definitions on-the-fly, 2) the application of the signature to query samples, 3) viewing the predictions in cBioPortal to assist clinicians judgment of samples in the context of other relevant genomics events, 3) extend the results by leveraging additional bioinformatics apps like the Medbook Workbench and observation Deck 4) the recording of the provenance, and 5) creates a focal point for deliberation about patients, signatures, genes, mutations, pathways, and clinical inferences. Conclusions: Prior to MedBook, there was a cognitive gap between individual doctor observations and large research paper driven cohort meta-analysis. MedBook is able to fill the gap by facilitating the online integration necessary for data analysis in a dynamic clinical environment. Citation Format: Robert Baertsch, Chris Wong, Jack Youngren, Josh Stuart, Eric Small, Ted Goldstein. Using Medbook Workbench to create evidence streams to guide medical decisions. [abstract]. In: Proceedings of the AACR Special Conference on Translation of the Cancer Genome; Feb 7-9, 2015; San Francisco, CA. Philadelphia (PA): AACR; Cancer Res 2015;75(22 Suppl 1):Abstract nr A1-46.