Our retrospective study was conducted at River Road Testing Lab [2], Louisiana State University (LSU), Pennington Biomedical Research Center, and Our Lady of the Lake Regional Medical Center. The study protocol was reviewed and approved by LSU Health Sciences Center [IRB#20-979 and exempt under 45CFR46.104 (d), category 4] on May 14, 2020. The protocol was conducted in accordance with relevant guidelines and institution policies. Because remnant nasopharyngeal swab samples received during routine care from SARS-CoV-2 infected patients were utilized in our study, informed consent was waived. All study sites worked under their approved biosafety protocols for handling SARS-CoV-2 specimens. Additionally, the authors vouch for the accuracy of the data reported.

Because age is a significant predictor for COVID-19 severity, a random stratified sampling scheme was applied whereby patients were classified according to age using the following categories: 18–59, 60–79, 80+. However, the youngest patient was 22 years of age. The SARS-CoV-2 infection was laboratory-confirmed by quantitative reverse transcription PCR (qRT-PCR) using FDA-approved CDC SARS-CoV-2 panel two [45]. Patients were randomly selected from within each group for a target of 95–97 per group. This resulted in 287 unique individuals during the period of March – May 2020. While demographic and clinical information were successfully collected from all 287 patient samples, leftover nasopharyngeal swab samples were available from 218 patients. The leftover nasopharyngeal swab samples for these 218 patients were used to evaluate immune and metabolic mediators at the site of infection–the nasal mucosa. Viral loads were determined in the samples from 174 patients.

Nasopharyngeal swab samples were not available from all patients in our cohorts. Our available data are detailed in Figure 1. Cell-free supernatants from nasopharyngeal swab samples were incubated at 56°C for 20 min to deactivate SARS-CoV-2 virions [48]. Then, heat-treated supernatants were subjected to electrochemiluminescence-based multiplex assays according to the manufacturer’s protocol (MSD). The following analytes were measured: C-peptide, GLP-1, Glucagon, Insulin, IP-10, Leptin, PYY, G-CSF, GM-CSF, IFN-α2a, IFN-γ, IFN-β, IL-10, IL-12p70, IL-13, IL-15, IL-17A, IL-18, IL-1β, IL-1α, IL-22, IL-23, IL-29, IL-33, TSLP, IL-4, IL-5, IL-6, MCP-1, MCP-2, MCP-3, MCP-4, MDC, MIP-1α, MIP-1β, MIP-3α, MIP-3β, and MIP-5. Samples with analytes concentrations below the limit of detection (LOD) of the assay are replaced by a value equal to the LOD divided by the square root of 2 [49]. The data summaries are shown in Supplementary Table S1.

Viral RNA nasopharyngeal swab samples were extracted, transcribed, and amplified as previously described [50]. The results were interpolated from an internal standard curve, produced by identical processing of serial dilution of a known copies-number of SARS-CoV-2 RNA stock (EDX, #COV019), hereafter referred to as the viral load.

To investigate risk factors, anonymized clinical, paraclinical, and demographic data from patients admitted to Our Lady of the Lake (OLOL), were extracted into a REDCap database. To anonymize information from patients, the remnant nasopharyngeal swab samples from OLOL were relabeled with lab sample’s ID before being sent to our lab. The lab’s sample ID was used to communicate between OLOL and our lab.

Patient disposition was defined as “severe” if the patient was admitted to the intensive care unit (ICU) or “deceased” if the patient died due to COVID-19 during the hospitalization and within 30 days of discharge. All other patients were coded as “less severe.” Obesity was determined by body mass index calculated from weight and height in the medical record, based on the CDC adult definition of “obese” [51]. Biological sex, age, and race were obtained from the medical record. Race was collected as African American (AA), Caucasian, Asian, and Other.

Patient cardiac, renal, pulmonary, hepatic, vascular, cancer, diabetes, and connective tissue comorbidities, as well as age at infection were also collected. To evaluate the combined impact of these comorbidities on COVID-19 severity, we used weighted Charlson Comorbidity Index (CCI) [52], a sum of weighted scores for each comorbidity.

The data are described using standard descriptive statistics. The relationship of demographic, clinical, virological, and immunological variables with the outcome of disease severity (non-severe, severe, and deceased) was examined using ordered logistic regression. The effect of variables on the development of severe and deceased COVID-19 was represented as an unadjusted odds ratio (ORs) with 95% confidence interval (CI). Severity was tested for association with age, viral load, immune mediators, demographical variables, and obesity status with ordinal logistic regression while multinomial and bivariate logistic regression were employed to test for association between viral load and race and sex, respectively. Correlation matrixes for analytes was conducted using non-parametric Spearman correlation with two-tailed p-values and 95% confidence interval (95% CI).

We use an exploratory approach to ensure a final model for predicting COVID-19 severity. The initial multivariable ordered logistic regression model included only variables with P ≤ 0.05 from the bivariate analyses. Because continuous variables were collected in different metrics, we used standard deviation units to standardize all variables [53]. Details regarding the selected variables are provided in Supplementary Table S1. A final ordered logistic regression model retained only variables with P ≤0.05.

Then, we used the final ordered logistic regression model to calculate the coefficients of a formula to predict a logit transformation of the probability of severe and critical COVID-19, respectively. Thus, the logarithm of the odds is log [P(Y ≤ j)/P(Y > j)] = logit [P(Y ≤ j)] = α_j - ∑β_iX_i [54]. With Y is an ordinal outcome with J the degree of disease severity (j = 1 for non-severe, j = 2 for severe, and j = 3 for deceased), P (Y ≤ j) is the cumulative probability of Y less than or equal to a specific degree of disease severity. X is the value of the particular predictors included in the final model. There are i = 3 predictors in our parsimonious model (i = 1, 2, and 3). α_j is the intercept for specific (j) degree of disease severity; β_i is the vector of regression coefficient or the effect of the individual (i) predictors on the specific outcome Y – degree of disease severity.

Statistical significance was assessed at the α = 0.05 level. The associations were evaluated using odds ratios (OR) with 95% confidence interval (95% CI) and *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.01, and ****p ≤ 0.0001 as statistically significant. All statistics were performed in R (version 4.0.4) in R Studio (1.4.17) and SPSS (IBM).

We included 287 COVID-19 patient samples in our study. These patients were admitted to local hospitals in Baton Rouge from March 2020 to May 2020 with a range of COVID-19 severity, including 72 deceased, 59 severe, and 156 non-severe patients (Table 1). Patients with severe COVID-19 were hospitalized for significantly more days compared to patients with less severe COVID-19 (Table 1). While race and ethnicity are not significantly different between the three groups, age and gender were not uniformly distributed across the three groups (Table 1). Although body mass index (BMI) as a continuous variable is not associated with COVID-19 severity, the BMI-based obesity classification is significantly different among the three groups (Table 1). We also evaluated the relationship of more than 20 significant comorbidities with COVID-19 severity (Supplementary Table S1). Among these comorbidities, the preexisting conditions related to hepatic, renal, cardiac, and pulmonary diseases were significantly associated with COVID-19 severity (Table 1). To evaluate the total burden of these comorbidities on COVID-19 severity, we used the Charlson Comorbidity Index (CCI) (53). Accordingly, the clinical severity significantly worsened as CCI increased (Figure 2B).

To explore early predictors of disease progression, we next examined the relationship of vital signs at triage with disease severity. At triage, patients with severe COVID-19 or deceased COVID-19 patients exhibited significantly higher body temperature and respiratory rate as compared to patients with non-severe COVID-19 (Figures 2B, C). Also, oxygen saturation to fraction of inspired oxygen ratios (SpO₂/FiO₂) at triage were significantly greater in non-severe COVID-19 patients compared to their severe or deceased peers (Figure 2D). There was no significant difference in either systolic (SBP) or diastolic blood pressure (DBP) among the three groups studied (Table 1).

To evaluate the local mucosal inflammatory responses, we subjected remnant nasopharyngeal swab samples to electrochemiluminescence -based multiplex assays examining 7 adipokines and 38 cytokines and chemokines. These analytes cover a wide range of inflammatory and metabolic pathways. We found increased levels of respiratory IL-6 and IL-10 in patients with more severe COVID-19 compared to less severe patients (Figures 3A, B). Similarly, patients with severe and critical COVID-19 exhibited significantly higher levels of monocyte chemoattractant protein −1 and −3 (MCP-1 and MCP-3) (Figures 3C, D). Although respiratory levels of insulin were significantly higher in severe compared to non-severe patients, there was no difference between severe and deceased patients regarding insulin levels (Figure 3E). We found that the distributions of viral load were significantly different between non-severe vs. deceased and severe vs. deceased (Figure 3F).

We next examined the relationship between viral load, local mucosal immune mediators, and early clinical indicators. To avoid the multiple comparisons problem, we only included variables significantly different between disease severity groups (Figure 4). We found that viral load was differentially correlated with immune response markers depending on whether data were aggregated over all patients or stratified based on disease severity (Figure 4). The pattern of correlation was not homogenous. In general, viral load was significantly and positively correlated with MCP-1, MCP-3, and IL-10. When we stratified for the degree of disease severity, several analytes were associated with viral load in one but not the other groupings. For example, IL-10 was positively and significantly correlated with viral load in deceased or non-severe patients but not in severe patients. Of note, there is no significant correlation between viral load and all other variables (Figure 4). Indeed, when we binned individuals into categories, ordered logistic regression revealed no association between increasing age group and viral load [OR: 0.86, 95% CI: (.73, 1.01)] (Supplementary Figure S1).

Next, we investigated whether viral load in this population was associated with being African American (“AA”) vs. not African American “(non-AA”). Logistic regression showed no association between viral load and identification as AA vs. not AA (Supplementary Figure S1). Similarly, we tested for an effect of biological sex (male vs. female) and found none (Supplementary Figure S1). On the other hand, viral load was significantly and positively correlated with age and CCI in deceased patients (Figure 4).

We observed a significant positive correlation among MCP-1, MCP-3, and IL-6 regardless of the degree of disease severity (Figure 4), suggesting common immune responses to SARS-CoV-2 infection. Although IL-10 and IL-6 were not correlated within the group of patients with severe COVID-19, they were directly proportional in deceased and non-severe groups.

Among early clinical and demographic indicators, we consistently observed a strong positive correlation between age at infection and CCI in all studied groups. In contrast, respiratory rate and SpO₂/FiO₂ ratio were not correlated with other variables regardless of disease severity. Intriguingly, in general, body temperature at triage was significantly correlated with BMI (Figure 4).

Having demonstrated the association between clinical and immunological factors with COVID-19 severity, we next sought to evaluate their potential predictability for disease severity. First, the relationship of individual factors with the severity of disease was assessed using ordered logistic regression (Figure 5A). To establish parsimonious models, we only tested the relationship of the severity of disease with factors that are not uniformly distributed among studied groups (non-severe, severe and deceased groups). We found that female (OR, 0.35; 95% CI, 0.19–0.64) individuals with high SpO₂/FiO₂ ratio at triage (OR, 0.37; 95% CI, 0.27–0.51) are at 2.86- and 2.7 fold lower risk of developing severe/critical COVID-19, respectively (Figure 5A). In contrast, higher CCI, age at infection, viral load, respiratory IL-10, and respiratory rate at triage were the drivers of worse clinical outcome. There was no significant association of other variables with COVID-19 severity by ordered logistic regression (Figure 5A). However, it is noteworthy that BMI-based obesity was significantly different among the three groups of patients (Non-severed vs. Severe vs. Deceased) with greater numbers of overweight and obese individuals classified as severe (33 obese, 17 overweight, 6 healthy weight) or deceased (33 obese, 24 overweight, 14 healthy weight) as compared to healthy weight and this is reflected in Table 1.

Among these 7 variables (gender, SpO₂/FiO₂ ratio at triage, CCI, respiratory IL-10, viral load, respiratory rate at triage, and age at infection), only CCI, SpO₂/FiO₂ ratio at triage, gender, and respiratory rate at triage were retained as potential risk factors in the final ordered logistic regression (Figure 5B). Because respiratory rate and SpO₂/FiO₂ ratio are clinically similar variables, only SpO₂/FiO₂ ratio, CCI, and gender were included in developing the parsimonious model. The combined influence of gender, SpO₂/FiO₂ ratio at triage, and CCI in the final ordered logistic regression model is illustrated in Figure 6. With the same levels of three risk factors, the probability of patients with non-severe COVID-19 developing severe disease (non-severe that escalates to severe) is higher than the probability of patients with severe COVID-19 progressing to more critical COVID-19 (severe that escalates to death). The effect of CCI and SpO₂/FiO₂ ratio at triage on the development of severe and deceased COVID-19 was greater in patients identifying as male. For example, a female patient with SPO₂/FiO₂ ratio of 224 at triage and CCI of 10 has a 14.9% probability of developing severe COVID-19 (if admitted without severe symptoms) and a 4.6% probability of progressing to critical/deceased COVID-19 (if admitted with severe symptoms). These probabilities will be significantly higher for male patients. For a male patient with the same SpO₂/FiO₂ ratio of 224 at triage and CCI of 10, the probabilities increase to 30.2% (non-severe progressing to severe) and 10.7% (severe progressing to critical/deceased). Therefore, males with reduced SpO₂/FiO₂ ratio and increased CCIs are at greater risk of developing severe and critical/deceased COVID-19.