Name
#109 The use of AI Natural Language Processing Tools in Clinical Decision Support Systems: an MTBI2 Presentation
Content Presented on Behalf of
Uniformed Services University
Services/Agencies represented
Uniformed Services University (USU)
Session Type
Posters
Room#/Location
Prince Georges Exhibit Hall A/B
Focus Areas/Topics
Medical Technology
Learning Outcomes
1. Understand current challenges in the field with big data research and data management.
2. Describe challenges across military research with resources and uniformity of data collection and management tools.
3. Understand how AI Natural Language Processing can assist military health research in the area of big data - for management and hypothesis generation
4. Appreciate practical application of this technology for use
Session Currently Live
Description
Research and clinical trials are often centered on addressing gaps within the Military Health System in the treatment of service members. The recent availability of large data lakes such as the Military Health Systems (MHS) Data Repository (MDR) and the Federal Interagency Traumatic Brain Injury Repository (FITBIR) have resulted in researchers obtaining access to a wide variety of data, some of which includes lifetime data sets for service members and their dependents. This has afforded the opportunity for complex modeling and information extraction, but in conjunction have generated large amounts of data, and associated regulatory burden for security, access and storage. Manual curation and extraction is impractical and extremely resource limiting. To improve efficiency and effectiveness, we present (1) our full featured unified platform for data collection and management and (2) the addition of natural language processing (NLP) and Artificial Intelligence (AI) tools for the data extraction centered on key medical terms. These topics are critical to address an outstanding, and important area of need within the DoD and fills a unique void. We have developed a standardized informatics platform through the MTBI2 Collection Access Sharing and Analytics (CASA) platform. At the core, the CASA platform is built on the NIH BRICS system however, there are several important integrated tools for clinical trials: full audit capabilities, samples tracking and management, offline capture, remote data collection, integrated patient management, data dictionary and real-time data access to support user friendly informatics capabilities. The CASA platform is able to manage research protocol setup, ingest data collected externally, manage subjects, facilitate regulatory compliance and end-to-end support of the clinical trials lifecycle (planning to execution to close-out). In addition, inbuilt API end-points for analysis using R and Jupyter Notebook are available in the CASA platform. We further add the ability to perform NLP and AI to ingest large amount of data, filter the data, identify and extract key medical terms and furthermore compile knowledge summaries and present these in the form of clinical decision support tools. The crux of the NLP are autoencoders that are trained using collected data in the CASA data repository. The key strength of these additional AI tools is the ability to quickly parse through large amounts of data to help users find critical information and supporting artifacts. The AI tools also have chat bot type features to answer follow on prompts. We have successfully deployed our CASA platform to support a variety of active DoD funded clinical trials ranging from sleep disorders, chronic migraine, and PTSD involving service members; across multiple military treatment facilities. The addition of the NLP and AI features for text feature extraction is being tested. The NLP and AI are being trained on terminology and characteristics of common data elements as curated by the National Library of Medicine to ensure consistency and quality checks are in place. Our goal is to connect to the MHS Genesis system and MRS to obtain large amounts of data for training and fine tuning these models. The practical development and implementation of a standardized informatics platform is valuable as it provides explicit process for data regulatory compliance, monitoring, curation and sharing, with a well-documented data dictionary; features that promote data sharing and serve to accelerate translational therapeutics and cross validation of research findings. The implementation of NLP and AI to extract pertinent information based on specific clinical or scientific questions, and automatically generate meaningful supporting information from the available data creates a strong platform that could serve to help with clinical decision making. Nonetheless the strength of the AI tools are limited by the amount of data available and more work is being performed to increase data availability and training efficiency of the underlying models.