DISCUSSION
We have evaluated multiple regression analysis for mathematically modeling the course of COVID-19 to predict clinical outcome. The premise of this model is that quantitatively measured clinical and laboratory parameters involved in the pathogenesis of disease progression can be mathematically mapped to a multiple-regression model. COVID-19 is initiated by infection of the subject with SARS-CoV-2 with subsequent replication in the epithelial cells of the lung. The factors that contribute to the viral load include number of cells that express the ACE2 and other receptors, and inflammatory cytokines. Comorbidities contribute towards a more serious disease progression. Virus infection of antigen presenting cells, such as dendritic cells, macrophages, and other cell types including endothelial cells, results in activation of biochemical signals, which lead to secretion of a battery of cytokines that include IL1β and IL-6. The viral infection as well as inflammatory cytokines cause fever and an increase in serum inflammatory factors such as D-Dimer and Ferritin. Induction of an inflammatory response contributes to reduction of the total numbers of lymphocytes from circulation. The inflammation results in a loss of lung function (e.g., reduction in blood-oxygen levels), cardiac function (blood pressure) and can culminate in multi-organ failure.
Subjects with a normal immune response can generally mount an adequate innate and adaptive response to the virus. These individuals clear the virus by generating adaptive T cell responses and neutralizing antibodies. Subjects with comorbid conditions can have compromised immune function which could result in dysfunctional activation of inflammatory responses, leading to worse clinical outcomes.
Selection of the parameters that were included in the model building process was influenced by their perceived significance from current research reports. This list of factors is by no means complete and it is expected that in due course a more comprehensive list will emerge. This report provides a basis for creating a tool, independent of the number and type of parameters, that could find utility in predicting the disease outcome using those parameters.
Viral Load. Association of viral load and progression of diseases has been reported for several viral infections [42-44]. Viral load in COVID-19 is measured by qRT-PCR of SARS-CoV-2 using primers for the spike gene [43]. The correlation of high viral load with severity of disease progression has been extensively demonstrated. The systemic dissemination of the virus has been associated with expression of the ACE2 receptor on endothelial cells [21]]. Comorbid conditions could enhance the expression of receptors and enable distribution of virus, thereby enhancing the viral load, which can result in progression of disease.
IFNα. The critical role of Type I interferons in innate and adaptive immunity, leading to both protective and pathogenic responses, has been reported in the case of several viral and bacterial infections [45]. SARS-CoV-2 infection has been shown to result in a diverse range of effects on Type I immune responses. Most patients elicit a strong IFNα response along with a battery of inflammatory cytokines, some of which progress to a cytokine storm [46, 47]. Specific blocking of the type I mediated signal transduction by various proteins of SARS-CoV-2 has been demonstrated [48]. A remarkably high proportion of male subjects experiencing severe or critical COVID-19 disease expressed an inability to produce sufficient levels of IFNα due to various types of errors in the IFN genes. Curiously, majority of the male subjects possessed circulating IFNα autoantibodies that had the ability to neutralize the endogenously produced cytokine, thereby effectively reducing the available IFNα. The discovery of these two mechanisms for lowering IFNα levels underscores its relevance in controlling the progression of disease in individuals infected with the SARS-CoV-2 [49].
D-Dimer. D-Dimer is routinely measured in clinical situations because its levels correlate with serious underlying conditions including venous thromboembolism, cancer and sepsis [48]. In the case of COVID-19 patients, introduction of the virus brings about infection-induced inflammatory alterations leading to coagulopathy. Lungs being the target of SARS-CoV-2, acute injury to the lung as well as multi organ failure have been caused by the virus-induced cascade of the inflammatory pathway. In an early study on 41 COVID-19 patients, those with severe disease had higher levels of D-Dimer along with high levels of IL-8, TNFα and IL-2R [31]. Male patients were found to have higher levels of IL-6, IL-2R, Ferritin and other markers of inflammation compared to female. High levels of IL-6 showed a statistically significant correlation with severe disease in a retrospective study as well [27]. One can hypothesize that such patients would likely benefit from anticoagulation therapy.
Ferritin. A high level of ferritin, measure of stored iron, was found to be associated with severe disease in COVID-19 patients and was linked to high fatality rates in a 72 patient prospective study [33, 50, 51]. In another study on 39 patients, those with mild COVID-19 symptoms had lower levels of ferritin while those with moderate or severe symptoms expressed higher levels of ferritin [50].
Lymphopenia. Loss of lymphocytes after viral infections has been associated with severe disease. The mechanisms involved in lymphodepletion can been implicated to be due to cell death, cytokine storm and/or redistribution of lymphocyte populations [3, 33, 37]. In this model, we have utilized lymphopenia as a measure of severity of disease progression. Loss of immune function could result in several potential mechanisms of pathogenesis including autoimmunity, hyperactivation, increased susceptibility to infections and organ dysfunction.
Neutralizing Antibodies. Induction of neutralizing antibodies directed to the receptor-binding domain of the spike protein is critical for restricting entry of the virus into the cells and has been one of the central tenets of a protective immune response. In this model, we have used a range of IgG titers to spike protein for the simulated data set [52]. However, the role of neutralizing antibodies induced in a large proportion of subjects following natural infection is still being studied [53]. Some subjects do not elicit strong antibody responses. Sub-optimal levels of antibodies may catalyze generation of virus mutants [54]. Neutralizing antibodies to the virus have generally not correlated with reduced severity of disease in the primary infection. In addition, it will be interesting to decipher the role of pre-existing antibodies reported recently in the modulation of disease and its impact on vaccination regimens. Thus, the mechanisms involved in the induction of antibodies, the repertoire and diversity of responses, and effects on protection versus progression, remains to be clearly established.
The predictive model can have multiple applications, such as forecasting the percentage of the population that will progress to severe disease in each geography, enabling logistics planning for hospital beds, health care providers and personal-protective safety equipment. Analysis of the coefficient of correlations of parameters with outcome of disease may provide clues to a better understanding of the mechanism of action of disease pathogenesis. The model can predict the probability of disease progression at an individual level, based on parameter data, and can be used to understand the effect and impact of therapeutic interventions. The predictive model can be utilized to analyze large amounts of data to develop algorithms for personalized treatment regimens.
In summary, we have developed a probabilistic model that can be utilized to predict progression of disease following infection with SARS-CoV-2. This model was developed using simulated data based on published levels of COVID-19 related clinical and laboratory parameters and provides an approach to predicting the outcome of disease. Validation of the model will require existing data and the clinical outcomes of patients. Prediction of disease progression can be highly valuable at an individual as well as population level.