Clinical Data Science

点击翻译按钮获取中文摘要

The following themes are researched: 1.Generating insights from data repositories: Using the repositories available to us, such as the CMS Virtual Research Data Center (VRDC) or AllofUs dataset, our objective is to answer concrete clinical questions, taking into account not only the features of the repository (e.g., size and data elements available), but also its limiting characteristics (i.e., data granularity and dataset population). We intend to explore both hypothesis- and data-driven approaches to investigating clinical questions. Additionally, by collaborating with external institutions with access to rich EHR data (e.g., Observational Health Data Science and Informatics collaborative; OHDSI), as we already have, we will be able to access a larger set of repositories and investigate a broader set of clinical questions. 2.Expertise with available repositories: There has been significant growth in number of institution or network centric IDRs and similar growth in number of available clinical trial repositories. Researchers are facing a difficult task of choosing the most appropriate repository for a given research question. We plan to acquire practical expertise with advantages and limitations of both clinical and research datasets whenever possible either through their active use or though published reports otherwise. 3. Characterizing data repositories: To facilitate the choice of appropriate repository or to facilitate improvement of a repository over time, we plan to develop methods to best characterize the repository size, population characteristics, clinical breadth and depth of data, and data quality. We also expect to contribute to the development of best practices for repository creation and maintenance through dataset characterization. 4. Integrating data repositories: While it is valuable to analyze individual repositories, more benefits may come from integrating individual repositories into larger repositories, for example to support large-scale analyses, meta-analyses, and comparisons across repositories (e.g., for reproducibility testing). Integrating repositories rests, in a large part, on the transformation of local repositories using a homegrown data model into repositories based on a common analytical model, supporting federated queries across repositories. The emergence of common data models (CDMs) for an analytic purpose reflects a vision for analytical interoperability. Integrated data repositories not only share a harmonized information model, but also commit to target terminologies for coding biomedical entities (e.g., RxNorm for drugs, SNOMED CT for diagnoses, and LOINC for clinical observations). In terms of data integration, under a project within NIH Office of AIDS research, we analyzed Common Data Elements (CDEs) in HIV domain. In FY20, our research focused on: - HIV Common Data Elements (theme 4). Common Data Elements (CDE) allow harmonization of data across studies. - Descriptive research within HIV domain (theme 1,2). We used Center for Medicare and Medicaid Services (CMS) Data Warehouse. - Dental informatics (theme 1) - Optimal representation of drug exposures within databases that use Observational Medical Outcomes Partnership (OMOP) model (theme 4) - Comparative safety and effectiveness (osteoporosis focused research) (theme 1) - Optimal data sharing and general CDEs (across clinical domains; general medicine) - Clinical characterization of asthma and COPD using FDA Sentinel database (collaboration with FDA; two 2020 American Thoracic Society Annual Conference abstracts not listed) Publications Generated during the 2020 Reporting Period