Objectives: Our goal was to produce a decision support platform with the following capabilities: 1. Build a typical dashboard in hours, with more complex dashboards taking a few days. 2. Build most dashboards without any programming, using configuration by science officers or business process analysts. 3. Use standard components and web technologies to simplify maintenance and improve interoperability. 4. Provide flexible access to data for new software modules. 5. Support sophisticated and intuitive searching, with support for querying by terms meaningful to users.
Methods: We designed a data warehouse with an extensible data model for storing a wide variety of data about scientific research, and defined lightweight methods for extracting and loading data from internal and external sources. Next, we added a robust, web-native data-access layer (HTSQL), a rapid web-application development framework (HTRAF), and, finally, a visual application builder (the HTRAF VAB). The first pass at semantic search was handled via AlchemyAPI web service.
Results:
We developed the platform using an agile methodology over 10 weekly iterations. We then took two weeks to configure the first prototype of a decision-support tool suite: a set of interconnected dashboards for reviewing grants and grant applications and making decisions about their status. The prototype included the following screens: application dashboard, grant dashboard, publication dashboard, scientist profile, collaboration dashboard and home/search page. We were able to deliver three complete designs of the suite (a total of 21 different screens) for a usability review. Data provided by the prototype was judged useful for helping science officers make decisions about grant applications, and one design was selected as the most intuitive. Preliminary semantic search functionality showed some promise, but was too immature to evaluate.
Conclusions:
A well-designed generic platform can facilitate inexpensive and rapid delivery of tools to support a wide range of decision-making tasks. The largest challenge was successfully integrating data sources with various degrees of cleanliness and completeness. Data cleaning and normalization are likely to remain challenging, as the number of data sources pulled into the platform continues to increase. We expect to make the results of this project available to the research community and believe it will help inform decisions about scientific directions in autism research.