gms | German Medical Science

Introduction: For assessing data quality (DQ; i.e., the degree to which data fits requirements), data dictionaries are a valuable source of information at the variable level. This is the case because metadata in data dictionaries specifies data requirements (e.g., measurement limits, expected probability distributions, or the assignments of process variables, e.g. from examiners to outcome variables) that enable automated DQ reporting [1], [2].

However, different machine-readable formats are used for data dictionaries, including spreadsheet formats with pre-specified columns [1], [2], [3], XML formats [4], [5], generic databases (PostgreSQL, MySQL [2], or study specific databases such as Opal [3]). This work shows how a modular architecture of an R analysis pipeline facilitates using metadata and data from different sources to enable DQ assessments (DQA).

Methods: R packages have been implemented as part of the Square2 DQA web application [2], [6] to create configurable DQ reports from various data sources, such as Opal servers, spreadsheet files, and database management systems. One module has been developed for each supported data source and output format. Each module implements interface methods for all the applicable steps, i.e., reading, evaluating, and writing, of each input (data, metadata, report configuration) and output (analysis results).

Results and discussion: The DQ pipeline can be controlled from the graphical user interface in Square2, where users can further configure data and metadata sources and the output (e.g., variable selection, comments, and layout). Afterwards, a template-based report is generated using RMarkdown [7], flexdashboard [8], and dataquieR [1].

Based on experiences with DQA in SHIP [9] and NAKO [10], and by reviewing similar pipelines from other groups, the presented software targets typical steps, data-sources and output formats in DQA workflows.

The modular approach represents a major advantage in enabling comparable DQ reports when using data from different sources. No additional preprocessing steps by the user are necessary. Furthermore, the modular architecture facilitates extensions. New data sources and data formats for input/output can be installed as additional modules. The architecture also allows reusing the modules in other DQA pipelines, for example when manually creating RMarkdown reports.

Die Autoren geben an, dass kein Interessenkonflikt besteht.

Die Autoren geben an, dass kein Ethikvotum erforderlich ist.