Currently, no clinical prediction model exists that would allow clinicians to prospectively identify patients with high risk of returning with worsened symptoms after ED discharge and to discriminate such patients from subjects with low risk of re-admission. As ED clinicians need to make high-quality decisions in a timely fashion. Evidence-based clinical decision rules can support ED clinicians to make empirically informed decisions and to meet stringent time constraints. Clinicians support rules can also be used to perform a risk-based approach to ED care, where potential COVID-19 cases with low risk of “severe” disease course are identified early on. This approach can also provide important opportunities to take up measure to avoid that low- risk COVID-19 cases compete for resources that must remain reserved for COVID-19 patients with a high risk for a “severe” disease course. Such a risk-based approach can improve the overall efficiency but also the quality of care for individual patients. To develop such prediction model is of mounting clinical importance since the global pandemic will lead to situation where the available resources become scares and must therefore be distributed as efficiently as possible. Empirically-informed guidelines may ease the burden of ED clinicians and improve the standard of care for patients with COVID-19. At present, however, it remains poorly understood exactly how and why clinical decisions are made – and, therefore, how to improve upon current practice. The present study aims to build a prediction model to discriminate between predicted “severe” and “non-severe” disease course that is informed by the best, currently available evidence based on a systematic scoping review of the literature (Tricco et al. 2016). We propose drawing data from electronic medical records (EMR) of COVID-19 patients who have presented to the CUIMC adult ED, with the goal of identifying which variables predict key outcomes such as patients’ discharge disposition (e.g., hospitalization, ICU admission, re-admission to the ED or ICU within the next 72 hours) and the overall length of stay. Traditional statistical approaches may not be well suited to analyze the large numbers of variables that contribute to optimal care of COVID-19 patients in the ED. Therefore, we will apply more complex prediction models(including natural language processing of clinician notes), capable of accounting for substantially more variables, may be better equipped to analyze the patient population and corresponding clinical decisions.
