Trauma Outcome Prediction in the Era of Big Data: From Data Collection to Analytics
Physiological data of unprecedented volume is generated daily in healthcare. The Big Data approach in medical data analysis could assist clinicians in doing fast and accurate diagnosis and planning early therapeutic interventions, hence improving patients’ outcomes and reducing healthcare cost. However, it is still challenging to reliably collect a large amount of data from bedside monitors and to discover knowledge from the physiological data.
In the area of trauma patient care, early recognition and mitigation of secondary injury and hemorrhage could prevent death caused by massive bleeding or improve patients’ life quality after treatment. We describe methods that are used in a level I regional trauma center for massive physiological data collection and analysis. An almost 100% collection rate of continuous, automated, high-fidelity data provides the basis for clinical data mining. Within the machine learning framework, we introduce feature design, feature selection, model performance evaluation, and computing issues, which are critical components in large-scale medical data analysis. Validated knowledge from massive data can be used by clinicians in a simple way for decision making or prioritizing care in the busy hospital environment.