Bioinformatic Generated and Learned Acute Assessment Models (BioGLAAM)

Hemorrhagic shock remains a leading cause of death for soldiers on the battlefield. The first 60 minutes following a traumatic injury is vital to saving lives; therefore, it is critical to provide medical personnel with real time monitoring of soldiers with traumatic injuries. Recent research indicates, however, that no single measure alone is sufficient to determine the severity of hemorrhage in trauma patients. A new approach is needed to identify a reliable early indicator of blood loss from easily acquired physiological signals. To address this issue, we propose to design and demonstrate the feasibility of Bioinformatic Generated and Learned Acute Assessment Models (BioGLAAM), a machine learning approach for producing an early detector of hemorrhagic shock using physiologic signals. BioGLAAM uses Dynamic Bayesian Networks (DBNs), a model-based approach, to discover patterns within existing data sets that serve as reliable indicators of blood loss. This model is augmented by incorporating subject matter expertise directly, leveraging both automation and domain knowledge. Furthermore, DBNs are structured hierarchically to promote both off-line and on-line adaptation to individual physiologies. The resulting models can then be embedded in resource-constrained portable devices, providing medical personnel with an increased capability for early hemorrhagic shock detection.