Smart Study Builder Challenge

Use artificial intelligence to read and understand study protocols, Statistical Analysis Plans (SAP), and metadata standards to create a knowledge graph by linking relevant elements together. The ideal solution would leverage smart machines to think, read, and act like a human when reviewing various types of clinical documentation. Utilizing Natural Language Processing (NLP), the ideal solution should be capable of extracting and linking, for example, end points in a protocol, to visit schedule, to statistical techniques, in a SAP.

The pharmaceutical industry creates multiple documents when conducting a clinical trial. These range from a clinical study protocol document to a statistical analysis plan (SAP). These documents can live in various formats, platforms, and file types. The documents can contain unstructured data. Pfizer is looking to accelerate the translation of data to knowledge graphs such that it can lead to actionable insights, precise therapeutic interventions, and health care strategies. 

Multiple selected companies participated in this targeted challenge. Each solution showed promise in its ability to build and train a “smart machine” to generate a document customized for a specific clinical trial protocol. The participating companies provided novel solutions using Natural Language Processing (NLP) technique to convert the protocol into useable “knowledge.” The accuracy varied among the participating companies--however a mechanism to “train the smart machine” would be necessary to improve accuracy. The results of this challenge showed that a knowledge graph is viable when built using NLP with pretrained models that are mature enough to perform named entity recognition. This technique can be leveraged to automate study build specifications and improve data lineage.