Early Prediction of COVID-19 Outcome: Contrasting Clinical Scores and Computational Intelligence Methods

Triaging incoming patients is critical for an optimal allocation of hospital resources, especially during a pandemic, when these tend to be quickly depleted. A typical approach for predicting patients’ outcomes relies on clinical scores such as the Charlson Comorbidity Index (CCI). CCI-based triaging is a reliable approach for estimating the mortality risk in the general patients’ population. However, this score is not optimized for predicting mortality in specific populations such as the one represented by COVID inpatients, often the most represented population in the emergency department cohorts during the current pandemic. Motivated by this, this chapter describes the development of a new COVID-19-specific clinical score: The General Assessment of SARS-CoV-2 patients Score (GASS). The score builds on the clinical experience gained during the first phase of the pandemic, and it is based on both clinical and laboratory data. It was aimed at predicting the 30-day mortality outcome of hospitalized COVID-19 patients and showed markedly better accuracy than the CCI. Furthermore, this chapter introduces an additional predictive model based on a classical Computational Intelligence method. Specifically, it describes the development and validation of a feedforward artificial Neural Network (NN) that automatically maps patients’ clinical and laboratory data to a 30-day mortality-risk score. Critically, the NN-based method was shown to be more accurate at predicting 30-day mortality of COVID-19 patients than both the CCI and GASS scores. However, the intrinsic black-box nature of the NN-based method makes it hard to reach an intuitive understanding of the internal computations underlying its decision process. This might affect its general acceptance among clinicians, and lead them to prefer using the GASS score.