The Korea Disease Control and Prevention Agency provided the data for this study. These data included basic information, laboratory results at the time of initial examination, the hospitalization history, hospital conditions, clinical severity, general blood test results, and history of chronic disease.

We included all patients who were confirmed to have COVID-19, hospitalized, and discharged in Korea between January 21, 2020 and April 30, 2020, excluding 14 cases involving pregnancy, 27 cases with no clinical severity reported, and 4 cases with incomplete data. The final sample included 5,571 patients.

This study was focused on the presence of comorbidities. We included the following common comorbid conditions reported by patients at the time of hospitalization: DM, HTN, heart failure (HF), cardiac conduction disease (CCD), asthma, COPD, chronic kidney disease (CKD), malignancy, chronic liver disease (CLD), rheumatic disease/autoimmune disorder (RDAD), and dementia. These underlying diseases are the most commonly observed comorbidities among COVID-19 patients [19,22].

The outcome variable was the clinical severity score (CSS). The CSS was initially classified into 8 categories: 1 (no activity limitation), 2 (limited activity but no oxygen [O₂] required), 3 (O₂ with nasal prongs), 4 (O₂ with facial mask), 5 (non-invasive ventilation), 6 (invasive ventilation), 7 (multiorgan failure/extracorporeal membrane oxygenation), and 8 (death). Few cases (n = 106, 1.9%) were categorized between 4 and 7 (4, n= 43; 5, n= 33; 6, n= 19; 7, n= 11), making statistical analyses using all the 8 levels of the variable invalid. To increase the statistical power with a balanced sample distribution, we converted this raw scale into new CSS categories. Risk factor analyses commonly involve dichotomous variables such as alive/dead or non-severe/severe. However, this could result in a loss of information regarding the original variable. To accurately examine the impacts of comorbidities on severity, we converted the raw scale into 3 new categories: 1, mild disease (O₂ treatment not needed, original scale= 1-2); 2, severe disease (O₂ treatment needed, original scale= 3-7); and 3, death (raw scale= 8).

First, patients’ demographic characteristics were summarized in relation to the means and interquartile ranges of the CSS groups. We performed statistical tests to identify any significant inter-group differences with respect to symptoms and comorbidities.

Second, the relationships between COVID-19 severity and comorbidities were examined using 2-mode network analysis. Comorbidity prevalence was also summarized via a similar network analysis [30,31]. A single comorbidity was expected to yield a different CSS from 2 or more concurrent comorbidities. In the network, nodes represent other diseases present among the COVID-19 patients. Lines (links) indicate the coappearance of 2 or more underlying diseases in a patient. A comorbidity network, which is essentially a constellation of diseases, can effectively indicate the relative prevalence rates of underlying diseases [17]. The comorbidity networks for severe disease and death were compared with those for mild disease. Since the COVID-19 disease course unfolds differently according to the pathological history of each individual, simply understanding the virus is not enough to understand individual variation in response to the virus. Applying a network perspective helps elucidate the complex pathodynamics of COVID-19. A symptom or chronic disease network should assist healthcare workers in more easily identifying groups at highrisk for poor outcomes of COVID-19. We compared the relative connectivity of each node with other underlying diseases using eigenvector and weighted degree centralities. Eigenvector centrality is a measure of the influence of a node in a network. Relative centrality scores are assigned to each node based on its connectivity to other nodes. Weighted degree centrality refers to the sum of the weights assigned to capture the strengths of the links to other directly connected diseases. While eigenvector centrality can be used to identify how diseases are commonly connected to all other diseases, weighted degree centrality can capture the strength of these connections. We used both measures to depict the COVID-19 comorbidity network structure.

Next, we used the correlation structure between each pair of comorbidity and symptom variables to identify any associations. A p-value was derived for each of the estimated correlation coefficients based on the Pearson correlation test [32]. The statistical significance of the association was described using edge bundling for graph visualization. This analysis was used to investigate the association between each underlying disease and the risk of developing COVID-19 symptoms.

Finally, we conducted multinomial logistic regression analysis [33-35] to identify risk factors associated with illness severity and mortality among COVID-19 patients by comparing the severe illness and death CSS groups with the mild group. This enabled an in-depth examination that may have been obscured if we had used only 2 categories, such as survival and death. After all, some comorbidities may be associated with severe disease but not lead to death, or vice versa. Our outcomes of interest were the risk factors statistically associated with the presence of severe shortness of breath (SOB); therefore, ventilation treatment was included as a risk factor, as were the risk factors associated with death. Two levels of model adjustment were used: age, sex, and body mass index (BMI) (the most common control variables in the existing comorbidity studies [7,22,36]) only (model 1) and additional adjustment of model 1 for all of the comorbidity covariates (model 2). All statistical analyses were performed using R version 4.0.0 (R Foundation for Statistical Computing, Vienna, Austria), and graphical visualization was performed with Gephi software. The multinomial function from the nnet package was used to construct multinomial logistic regression models. We considered a pvalue < 0.05 to indicate statistical significance.

This study was approved by the Korea University Institutional Review Board (IRB-2020-0177).

Figure 1 presents the comorbidity networks for the hospitalized COVID-19 patients. HTN and DM were the most common underlying diseases. The overall comorbidity network is shown in Figure 1A. These comorbidities were not only most common among COVID-19 patients, but they also often appeared concomitantly with other chronic diseases. The comorbidity networks for mild disease (n= 4,762), severe disease (n= 575), and death (n= 234) are shown in Figure 1B-1D. As severity increased, centralities of disease severity emerged more prominently as key nodes in the network. The main components of the severe and fatal networks were HTN, DM, CCD, and dementia. Thus, the severe and fatal cases had distinct underlying comorbidity patterns compared to the mild cases. The eigenvector centrality and weighted degree centrality of each comorbidity are summarized in Table 3. Very little variation in degree centrality was present among the 11 underlying diseases because they were commonly co-observed among the confirmed COVID-19 patients. The clinical importance of a disease is indicated by a significant difference in its centrality across severity groups. Although high prevalence alone does not mean that a disease is a critical contributor to increased disease severity, its relative connectivity to other underlying diseases is an important factor for identifying patients at greater risk. For example, although HTN was the most common disease present among the hospitalized COVID-19 patients, the relative centrality of HTN was particularly highly pronounced among the fatal cases. HF was also found to be one of the most prevalent comorbidities, yet its relative centrality among patients with mild and severe cases was low. DM was the most central underlying disease among the mild and severe cases.

Figure 2A shows the associations between comorbidities and COVID-19 symptoms. The comorbidity most commonly and strongly associated with multiple symptoms was HTN, followed by DM, HF, asthma, and CCD. Among the reported symptoms, SOB consistently showed a strong correlation with comorbidities, including DM, HTN, HF, CCD, asthma, and COPD. Although SOB was exhibited in only 11.9% of the reported cases in our study (a relatively small proportion compared to cough, fever, and sputum), it is noteworthy that it was the most prevalent symptom strongly associated with comorbidities.

Regarding the impact of comorbidities on symptoms, COVID-19 patients with HTN were at a greater risk of having 1 or more of these symptoms: ACC, VN, SOB, fatigue, sputum production, and fever, listed in the order of the strongest to the weakest association (p < 0.05). Patients with DM were more likely to have ACC, SOB, fatigue, and VN than patients without DM. HF was associated with ACC, SOB, and VN. Patients with asthma showed an increased risk of experiencing SOB, VN, and sputum production. Patients with CCD were more likely to experience SOB, ACC, VN, and fever. COPD was associated with a higher risk of SOB, sputum production, cough, and fever. Patients with CKD were more likely to experience fatigue and fever. CLD was only associated with fatigue, and dementia was associated with ACC and SOB.

In model 1 (adjusted for age, sex, and BMI only), CKD and HF were significant risk factors for both severe disease and mortality. DM, HTN, malignancy, and dementia showed strong associations with increased mortality alone, whereas COPD was associated with severe disease alone.

Model 2 included additional adjustment for all of the comorbidity covariates (Table 4 and Figure 2B). In this model, age and sex (female) were associated with both severe disease and death, as was CKD, as is consistent with the results from model 1. Patients with COPD had a greater risk of developing breathing difficulties, but no significantly higher risk of death. DM, HF, malignancy, and dementia were strongly associated with the risk of death.

More specifically, the risk of severe disease increased with the presence of COPD (odds ratio [OR], 3.19; 95% confidence interval [CI], 1.35 to 7.52) and CKD (OR, 2.58; 95% CI, 1.19 to 5.61). HF, which was associated with severe breathing difficulties in model 1, showed a weaker association in model 2. The risk of death was most strongly associated with CKD (OR, 5.35; 95% CI, 2.00 to 14.31), followed by malignancy (OR, 3.38; 95% CI, 1.59 to 7.17), HF (OR, 3.15; 95% CI, 1.22 to 8.15), dementia (OR, 2.62; 95% CI, 1.45 to 4.72), and DM (OR, 2.26; 95% CI, 1.46 to 3.49). Other comorbidities rather weakly associated (p< 0.1) with increased mortality were asthma and HTN, although the associations were not statistically significant at a level of significance of 0.05 (asthma: OR, 2.20; 95% CI, 0.86 to 5.59; p= 0.10 and HTN: OR, 1.49; 95% CI, 0.95 to 2.31; p= 0.08).