INTRODUCTION
The COVID-19 pandemic caused by infection with SARS-CoV-2 was officially announced in March 2020 by the CDC and WHO [1, 2]. As of this publication, more than 100 million infections and over 2.6 million deaths have been reported worldwide. Majority of the subjects have asymptomatic infections. The rate of fatality is disproportionately high in the elderly and patients with comorbidities such as diabetes, cardiac disease, and kidney disease [3, 4]. The consequences of the pandemic are fraught with potential loss of lives, social and economic distress, and the uncertainty of disease progression because of variable individual pathogenesis.
A unique and dysregulated immune response has been shown to be a hallmark of COVID-19 [5-9]. Figure 1 schematically depicts the cascade of events that contribute to the progression of disease. Mathematical models have been utilized by several investigators to understand the mechanisms of disease pathogenesis, immune pathways involved and course of viral infections [10, 11]. In this article, we have proposed a predictive model that utilizes the levels of clinical and laboratory parameters to determine the severity of clinical outcomes ranging from asymptomatic to mild, moderate, severe, and critical disease states. The proposed model can be useful to predict clinical outcome at the individual-level and develop efficient and effective treatment strategies to manage public health challenges at the population-level.
The questions the model attempts to answer are: i) At an individual level, what is the probability of an individual infected with SARS-CoV-2, given the clinical signs and laboratory values on various days, likely to progress to severe disease, and ii) At a population level, what are the prioritized clinical and laboratory parameters that are most likely to contribute to progression to severe disease. We have used a multiple regression based model to predict severity of the outcome of COVID-19. To evaluate the combinatorics that are not observed in the sample, we have applied resampling methods based on Monte Carlo simulation.