gms | German Medical Science

As the amount of genetic, genomic, proteomic, and phenomic data is increasing at an unprecedented rate, the complexity of data management is also increasing. In addition, comprehensive analyses of data ranging from diplotypes of SNPs to gene expression and side effect profiles, require flexible statistical tools.

The traditional role of IT (informatics) support to clinical investigators has involved assisting them in the design of their databases after the study had been designed with assistance from a biostatistician. Once the data for an interim or final analysis has been collected, informatics support is provided for statistical analyses and the biostatistician assists in the analysis. At the RUH GCRC, we are developing a Web-based Interactive Study Design, Operation and Management module (WISDOM) to improve the overall quality of clinical trials by making knowledge about the study design available during all stages of development and progress of a clinical study.

In a typical research environment, various software tools are used to design studies, write protocols, manage and audit data entry and storage, monitor safety, analyze data statistically. Maintaining consistency across such different, often idiosyncratic programs and documents is cumbersome and error-prone, because of redundancy. To match flexible tools precisely to a particular aspect of an analysis, both in terms of the experimental design and the biological background, a precise declaration of the medical and experimental context in which the data was observed (meta data) is necessitated. As statistical methods play an increasingly important role, the metadata to be represented depends to a large extend on the needs of the statistical methods. Formally distinguishing declarative and procedural knowledge between medical background, experimental design, and scientific objectives led to a representation of metadata in five layers corresponding to different levels of abstraction [Ref. 1].

The ACCESS layer contains knowledge of how to store and retrieve data (e.g., SQL statements) from the different data bases and how to send it to different software packages. The SEMANTICS level contains a description of the experimental design, e.g., how the variables relate to each other and the design matrix underlying each of the statistical methods. The STRATEGY layer contains a description of the variables’ characteristics, such as causality (strata / interventions / observations), discrete / continuous, nominal < ordinal < interval < absolute, measurement units (MODEL), and the rules on how to find corresponding methods (STATISTICS). The DOMAIN layer describes knowledge about the variables, such as ranges/interactions or expected adverse experiences (MEDICINE), and specifications of the output, such as level of detail (PRESENTATION). At the INTERACTION layer, an audit of previous dialog actions is kept and the rights and preferences of each user are being described.

Having meta data entered into the data base management system during the design stage of a study and made available during all subsequent steps allows for providing new informatics services, where the users are freed from the need of repeatedly entering such information [Ref. 2]. Figure 1 [Fig. 1] shows how DESIGN and MODEL knowledge is entered into WISDOM. After the first year of development, WISDOM already provides several novel biomedical informatics services that utilize this meta data.

• Preparing structured information to be included with study protocols, e.g., study time line, inclusion / exclusion criteria, anticipated AEs.
• Automatically creating a data base (currently: Oracle) on a secure server without the need for any additional user interaction.
• Providing Web-based access to the data base through browser windows, where users can exchange rectangular ranges of data between commercially available spreadsheet programs and the database using standard cut-and-paste functionality.
• Generating CRFs for data entry, either on paper, or as Web forms.

A previous knowledge-based approach [Ref. 3] suffered from technological shortcomings. In particular, it lacked sufficiently user-friendly tools for knowledge acquisition. Thus, when it was commercialized by SAS, Inc. in 1989 as JMP, only a small portion of its potential was realized. As new technologies for developing graphical user interfaces become available we have now developed a knowledge acquisition paradigm (Figure 1 [Fig. 1]), where clinical trials, genetic studies, or gene expression experiments can be described interactively in sufficient detail to extract DESIGN and MODEL knowledge sufficient to support a wide range of study related activities.

The current version of WISDOM already provides tangible benefits to our investigators. Data is being stored on a secure server with routine back up and Web access independent of operating systems. We are now working on incorporating these data into study management and other components of the overall GCRC IT. In particular, we will draw on the PANOS experience to facilitate graphical and statistical analyses. As WISDOM will contain sufficient information to perform the statistical analyses addressing the primary objectives, these analyses will be carried out automatically after the data has been entered or at any predefined interim time point.

The modular design also allows for applications in other areas or research than clinical studies. Currently, we are working on developing a MODEL knowledge base for microarray experiments [Ref. 4], so that WISDOM will be able to function as a knowledge based interface to https://mustat.rockefeller.edu, our multi-processor Web server for multivariate ordinal data, in general, and for data mining based on genetic and genomic profiles [Ref. 5] in stratified designs [Ref. 6].

This work was partially supported by National Center for Research Resources at the National Institutes of Health; contract/grant number: M01-RR00102.