Exploitation of Bayesian Networks for Clinical Decision Support on the Battlefield
Abstract
Purpose: The purpose of this research is to optimize the care of battlefield trauma patients through the development of Bayesian-network (BN) machine learning-powered clinical decision-support (CDS) tools. Scope: The scope of the research encompasses the refinement of existing BNs, and development and prototyping of new BNs designed for pre-hospital, en-route, and deployed healthcare facility stages of care, such that CDS prototypes are available for piloting and assessment in future, real-world clinical studies. Major Findings - Year 4 Data. Transfer of US DoD data to Barts Health/QMUL Data Safe Haven Environment achieved with Data Sharing Agreement in place. Model generation. 1. Transfusion prediction model novel structure developed, reflecting non-compressible versus compressible hemorrhage. Novel outcome variables (predicted outcome of between 1 and 21+ units at incremental time segments of 0-2,2-6,6-12 and 12-24hrs) featured. Parameter learning completed. Trained on 995 patients. Probabilistic confusion matrix constructed to assess model performance as traditional performance metrics (AUROC) unsuited. Model performance assessed as Excellent or good in 81.4 percent of predictions. External validation awaited. 2. Survival-MIL model developed with TBI fragment AUROC at 0.93 (0.92-0.95). Final model structure (TAKI,TBI, TIC) developed. Overall performance of initial model AUROC 0.81(0.77-0.85) model refinement in progress. Software + system engineering. Back-end, front-end and manager API test suites are completed using relevant test packages for python and JavaScript languages.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 2023
- Accession Number
- AD1219069
Entities
People
- Nigel R. Tai
Organizations
- Queen Mary University of London