Emergency department (ED) clinicians need to make fast and accurate risk estimates in a stressful environment. At the same time, the amount of available information (in electronic health records etc) is exploding, as is our knowledge of risk factors. The ED physician can no longer grasp and use all accessible information. To counter this, risk prediction algorithms have been created to support the clinician. The clinical decision support systems (CDSS) used today are generally simple rule-based schemes based on manual data input, and the patient’s risk is classified as high to low in categories. However, the schemes often do not include important risk factors, let alone the vast prognostic information available in medical records and population registers. Further, the algorithms do not take data interactions or synergistic risk effects into account, and their accuracy have been questioned. Studies suggest that many of these algorithms are not better than the unstructured assessment by inexperienced clinicians.
There is thus a strong need for improved methods risk assessment at the ED. Our lack of reliable risk prediction at the ED probably leads to suboptimal (or even erroneous) prioritization, monitoring, diagnostic tests and discharge or admission to in-hospital care, and to increased cost. Indeed, studies suggest that a significant number of patients die soon after discharge from the ED and that over 20% of all admissions to in-hospital care may be unnecessary.
In this project, we aim to develop CDSS based on extensive register data and machine learning as well as advanced statistical methods. The goal is to improve the rapid identification of low-risk patients for safe, early discharge from the ED, and high-risk patients for immediate treatment, monitoring and hospital admission.
