Sepsis is a dysregulated host response to an infection, which is associated with organ failure and can lead to death of the patient. The global burden of sepsis is high, as it affects 30 million people per year, leading to the death of 20% of these people. Recognition of early sepsis is critical to allow timely initiation of adequate treatment: antibiotics and supportive care. Clinical sepsis criteria to facilitate its diagnosis using a combination of vital parameters have a very limited sensitivity in the early phase and most physicians diagnose sepsis based on the clinical impression, also known as “gut feeling”. Importantly, the clinical impression of the physician is stronger associated with and better in predicting severity-of-illness than clinical sepsis criteria. The estimation by the physician is not only based on vital parameters such as body temperature, heart rate or blood pressure, but also takes the patient’s physical appearance into account and the pattern of parameters. Since rapid recognition of patients in need of medical care is critical among patients admitted to the emergency department (ED), we use big data to develop novel algorithms to improve early recognition of sepsis and identify which patients benefits the most from which therapy (personalized medicine) using deep learning. To facilitate this kind of research, we have set up the Acutelines data-biobank at the ED of the UMCG. The purpose of the Acutelines data-biobank is to improve prevention, recognition and treatment of acute conditions. A trained team of researchers screens all patients entering the ED, followed by data/biomaterial collection depending on broad selection criteria. In addition to demographic and medical data from the electronic patient file, we collect and store biomaterials (blood, urine, stool) for biomarker discovery, take a photograph of the face to predict deterioration using computer vision techniques and record continuous electrophysiological waveforms (i.e. ECG, PPG, EMG) to identify features predictive of deterioration. In the current project, we will identify novel risk markers predictive of deterioration in patients with sepsis at the ED and its interaction with therapy. Herefore, we make use of electrophysiological waveform analysis, photographs and the clinical impression of the health care professional that will be pre-processed prior to intergration into machine-learning models combined with demographic data and vital parameters.
To develop novel algorithms to identify patients who will benefit the most from specific therapy in sepsis (personalized medicine), we aim to:
- Continuously improve the data warehouse structure to allow collection of high quality big data.
- Develop algorithms to pre-process complex data (i.e. photographs, electrophysiological waveforms).
- Integrate data in ML-models to identify risk markers predictive of deterioration in sepsis.
- Identify risk markers predictive of efficacy of given therapy by integrating treatment data in the models.