Anna Bakidou, Electrical engineering

Anna Bakidou is a PhD student in the research group Biomedical signals and systems

Discussion leader is Associate professor Carl Nettelblad, Uppsala University

Examiner is Professor Tomas McKelvey, Division of Signal processing and Biomedical engineering

Main supervisor is Dr. Stefan Candefjord, Division of Signal processing and Biomedical engineering

Abstract

Trauma is the leading cause of death for young adults and in the case of severe injury, time to correct treatment is important to increase the chance of survival. Emergency medical service clinicians are often first on site in the case of a trauma incident, and guidelines are used to assess the patient’s condition, needed care level and appropriate transport destination. Although current guidelines are easy to use, there are reports of a large proportion of patients being transported to inappropriate care levels. This indicates a need to improve the assessment and prioritization guidelines, and Artificial Intelligence (AI) has been suggested as a possible solution with promising results in related domains, e.g., international studies and earlier publications by the research group concerning On-Scene Injury Severity Prediction (OSISP) for traffic accidents.
In this seminar, we introduce the project Acute Support Assessment and Prioritizing Point-of-Care Trauma aiming at developing AI based algorithms for a decision support for patients affected by trauma. Two studies conducted within the project will be presented. In study 1, data from the Swedish Trauma Registry was used to evaluate if an AI based OSISP model can complement clinical practice in predicting injury severity. Predictors were selected based on statistical tests, models were assessed with cross-validation and target outcome was severe injury or not. Sensitivity analyzes included four different imputation methods to manage missing data, and four different definitions of severe injury. Results showed an increased performance for all models compared to the clinical outcome, with areas under the receiver operating characteristic curve between 0,80-0,89. After promising results from study 1, study 2 continued the work by optimizing the OSISP models and externally validating them with data from the Norwegian Trauma Registry. The seminar will end with reflections about the conducted studies and discuss plans to continue the work.