Predictive Parameters for the Worsening Clinical Course of Mild COVID-19 Pneumonia

Background: This study aimed to determine parameters for worsening oxygenation in mild COVID-19 pneumonia. Methods: This retrospective cohort study included conrmed COVID-19 pneumonia in a single public hospital in South Korea from January to April 2020. Parameters were compared between the two groups on the basis of clinical course: the desaturation group was dened as those with oxygen saturation ≤ 94% on ambient air, or received oxygen or mechanical ventilation (MV) throughout the clinical course versus the nonevent group who were without any respiratory event up to 28 days. The severity and extent of viral pneumonia from an initial single chest CT were calculated using articial intelligence (AI) algorithms and measured visually by a radiologist. Results: We included 136 patients with 32 (23.5%) in the desaturation group, of whom two needed MV and one died. Initial vital signs and duration of symptoms showed no difference between the two groups, however, univariate logistic regression analysis revealed that a variety of parameters at admission were associated with an increased risk of a desaturation event. In a sex-, age-, and comorbid illness-matched case-control study, ferritin ≥ 280 μg/L (OR 3.600, 95% CI 1.142-11.346; p=0.029), LDH ≥ 240 U/L (OR 3.600, 95% CI 1.142-11.346; p=0.029), pneumonia burden (OR 1.010, 95% CI 1.002-1.019; p=0.021), and extent (OR 1.194, 95% CI 1.017-1.401; p=0.030) by AI, and visual severity scores (OR 1.146, 95% CI 1.005-1.307; p=0.042) were the predictive parameters for worsening clinical course with desaturation. Conclusion: Our study presents initial CT parameters measured by AI or visual severity scoring as well as serum markers of inammation at admission as the best parameters for predicting worsening oxygenation in the COVID-19 pneumonia cohort. Initial chest CT scans may help clinicians diagnose viral pneumonia and evaluate the prognosis in mild COVID-19.


Introduction
Coronavirus disease (COVID-19) has now become a pandemic since the rst case of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) in Wuhan, China [1]. The symptoms vary widely, from asymptomatic disease to pneumonia, to life-threatening complications and ultimately death. Sometimes it shows rapid progression of respiratory failure soon after the onset of dyspnoea and hypoxemia [2]. It is important to unravel the risk factors associated with hypoxemia and identify at an early stage patients who are susceptible to severe or critical illness upon infection.
Older patients and those with pre-existing respiratory or cardiovascular conditions, diabetes, hypertension, and cancer appear to be at the greatest risk for poor outcomes or severe complications [1,[3][4][5]. Leucocytosis, low lymphocyte count, serum levels of D-dimer, lactate dehydrogenase (LDH), ferritin, cardiac troponin, and higher concentrations of proin ammatory cytokines were reported to be risk factors associated with severe and fatal COVID-19 in hospitalised patients [1,6,7]. Typical imaging features of COVID-19 were established to help early screening and evaluation of the severity and extent of disease.
CT involvement scores can assist in evaluation of the severity and extent of the disease [8]. Previous studies estimating risk factors included mostly people with severe and fatal diseases at presentation, and data are lacking for subclinical or mild COVID-19 patients. We explored potential risk factors associated with worsening oxygenation in a retrospective cohort of mild patients with COVID-19 pneumonia.

Study design and participants
This retrospective, single-centre study, recruited patients from January 30 th to April 15 th , 2020 in a public teaching hospital that also served as a COVID-19 dedicated hospital in Seoul, South Korea. We reviewed patients with COVID-19 pneumonia who were previously discharged or had been hospitalised for at least 28 days or more. Ethical approval was provided by the institutional ethics review board of our hospital (2020-04-024). According to hospital protocol, all patients with con rmed COVID-19 were admitted to the hospital and routinely underwent non-contrast CTs. The inclusion criteria were patients aged ≥18 years who (a) were positive for nucleic acid detection of SARS-CoV-2, (b) underwent a chest CT within 24 hours of admission, and (c) showed evidence of viral pneumonia in the initial chest CT scan. The presence of ground-glass opacity (GGO) or consolidation in the bilateral peripheral posterior lungs on chest CT scans without any other speci ed causes or pathogen in a COVID-19 con rmed patient was categorised as CT-documented COVID-19 pneumonia [9]. The exclusion criteria were (a) patients with symptom onset more than 14 days; (b) normal ndings on initial chest CT; and (c) lung abnormalities other than viral pneumonia.
The clinician's decision dictated treatment. Multidisciplinary teams screened for signs of acute deterioration and the need for intensive care unit (ICU) admission daily using an in-hospital electronicmedical-record monitoring system.

Data collection and de nitions
Demographic characteristics of the patients, laboratory and radiological data were reviewed. The need for oxygen devices such as nasal oxygen supply, high-ow oxygen therapy, or respiratory support such as invasive MV were assessed.
For classi cation of the clinical course with poor oxygenation throughout the initial 28 day hospitalisation, we grouped patients into the desaturation group and the nonevent group. The desaturation group was de ned as having any of the following criteria for admission or transfer to the ICU: (1) desaturation event with peripheral oxygen saturation of haemoglobin (SpO 2 ) on room air with rest ≤ 94% in at least two consecutive measurements or (2) respiratory failure requiring high-ow oxygen therapy or MV.
According to the "Diagnosis and Treatment Program of Pneumonia of COVID-19", severe cases were de ned as having any of the following criteria: (1) respiratory rate ≥ 30 breaths/min; (2) resting blood oxygen saturation ≤ 93%; or (3) partial pressure of arterial blood oxygen/fraction of inspired oxygen ≤ 300 mmHg. Critical patients needed to meet one of the following conditions: (1) respiratory failure requiring MV, (2) shock, or (3) other organ failure requiring ICU monitoring or treatment [10]. We de ned "mild COVID-19 patients" as those who did not meet any of the severe or critical criteria on admission.
Chest CT acquisition Chest CTs were performed using a 64-detector row CT scanner (SOMATOM Sensation 64, Siemens Healthcare, Erlangen, Germany) on admission (low-dose (reference mAs, 40) with automatic exposure control; slice thickness, 1.0 mm; and reconstruction interval, 3.0 mm). Axial images were reconstructed with a sharp reconstruction kernel at a 3 mm slice thickness. The CTs were performed with the patient in the supine position at full inspiration without contrast medium. Axial images were reconstructed with a sharp reconstruction kernel at a 3 mm slice thickness.
Quantitative CT analysis CT images of each patient were uploaded in a free software for arti cial intelligence (AI)-driven automatic CT analysis of COVID-19 pneumonia (MEDIP v1.2.0.0, MEDICALIP Co. Ltd., Korea), released on a website (http://medicalip.com/mobile/shop/covid19.php) on March 18, 2020. The deep learning software automatically generated a volumetric mask of the lung and pneumonia. The CT parameters from the automatic quantitative software were: the total lung volume (cm 3 ), the pneumonia volume (cm 3 ), pneumonia extent, and the mean lung densities of the total lung and the pneumonia.
An experienced thoracic radiologist (Y. K.L), blinded to the clinical data and laboratory indicators, evaluated the visual severity scores of the CT images. The CT severity score is adapted from a method previously used in patients after SARS [11]. The extent of lung lesions was scored from 0 to 5 in each lobe and summed up (0, normal; 1, 1-5% involvement; 2, 6-25% involvement; 3, 26-50% involvement; 4, 51-75%; 5, 76-100% involvement). The CT severity score was de ned as the sum of the individual scores in the ve lobes, which ranged from 0 to 25 points.

Statistical analysis
Continuous variables were represented as mean and standard deviation (SD) or median and interquartile ranges (IQR). A one-sample Kolmogorov-Smirnov test was performed to check the normality of all variables. To compare the two groups, the Mann-Whitney U test was used for continuous variables, and the chi-square or Fisher's exact test was used for categorical variables. Correlation analysis was performed to identify associations between continuous variables. Univariate logistic regression analysis was performed to identify the clinical variables associated with desaturation events. Finally, a propensity score matching analysis was performed to minimise the effect of potential confounding factors that could affect the worsening clinical course of COVID-19 pneumonia patients. We selected 26 sex-, age-, and underlying-disease-matched patients from each group to perform a case-control study at a ratio of 1:1, and a univariate logistic regression analysis was also performed with matched data.
All statistical analyses were performed using SPSS statistical software version 23.0 (IBM Corp., Armonk, NY, USA) and R software version 3.6.2 0 (R Foundation for Statistical Computing, Vienna, Austria). Pvalues were based on a two-sided signi cance level of 0.05.

Demographics and clinical courses of enrolled patients
A total of 244 patients were admitted with laboratory-con rmed COVID-19 ( Figure 1). After excluding 97 patients with normal baseline CT scans, four patients whose symptom onset was >14 days, and seven patients with bacterial coinfection or mycobacterial disease, 136 patients were included in this study (56 men, 80 women; mean age, 44.8±17.4 years; range, 18-95 years). No patient received MV or high-ow oxygen therapy on admission, only one patient had SpO 2 on room air ≤ 93%, and two patients showed respiratory rate ≥ 30 breaths/min, there was no shock or organ failure at presentation. Most patients, except for three patients, were classi ed as having mild COVID-19 on admission [10].
Among the 136 patients, 32 patients (23.5%) were classi ed into the desaturation group, and 104 remained in the nonevent group (76.5%) during 28 days of hospitalisation. All patients in the desaturation group had oxygen desaturation ≤ 94% in at least two consecutive measurements, and 10 patients required oxygen supplementation. Only two patients in the desaturation group needed MV, while a total of ve patients needed high-ow oxygen therapy before or without MV. No patient needed extracorporeal membrane oxygenation (ECMO). During hospitalisation, 10 patients were admitted to the ICU, including two patients who developed acute respiratory distress syndrome (ARDS), and one patient died declining to apply the MV in accordance with a previously signed advance directive. Among the patients in the desaturation group, time intervals for oxygen desaturation ≤ 94% were a median 2 (IQR, 1-5) days from admission and 8 (IQR 6-11) days from symptom or disease onset, respectively. The duration of oxygen support was a median 3 (IQR 1-7) days after admission and 12 (IQR 9-13) days from symptom or disease onset.

Baseline characteristics and laboratory indices between the two groups
The baseline characteristics of COVID-19 patients according to the clinical course are summarised in Table 1. The average interval for admission from symptom or disease onset was 6.8 days. Only 71 of 136 patients (52.2%) complained of respiratory symptoms, and only 23 of 136 patients (16.9%) had a body temperature (BT) ≥37.5°C on admission. Intervals between symptom or disease onset and admission, and initial symptoms showed no difference between the two groups. The rst recorded vital signs were similar in the two groups.
Patients in the desaturation group were older (P <0.001), and had more underlying diseases than those in the nonevent group (40.6% vs. 22.1%, P = 0.038). Hypertension was the most common underlying disease in both groups, and dementia was more frequent (15.6% vs. 1.0%, P = 0.003) in the desaturation group than in the nonevent group. The desaturation group showed higher body mass index (BMI) and 53.1% had obesity based on a BMI ≥25 kg/m 2 (vs. 26.9%, P = 0.006).
There were numerous differences in laboratory ndings between the two groups (Table 1), including higher serum levels of C-reactive protein (CRP), ferritin, and LDH as well as lower lymphocyte counts in the desaturation group than in the nonevent group (all P <0.05). The desaturation group had higher pneumonia severity index (PSI); (56.5 vs. 37.1, P <0.001) and MuLBSTA score (6.0 vs. 3.9, p = 0.001) than the nonevent group, the modi ed early warning score (MEWS) showed no difference (1.4 vs. 1.3, P = 0.737).

Relationships between CT features and worsening oxygenation
The pneumonia volume quanti ed using AI was signi cantly larger in the desaturation group (median 45.2 cm 3 vs. 10.7 cm 3 , P = 0.001) as well as the extent of opacities (median 2.4% vs. 0.5%, P = 0.001).
Visual severity scores of the CT images were also signi cantly higher in the desaturation group (8.0±5.2 vs. 4.6±3.2, P <0.001). In particular, 37.5% of the desaturation group showed a visual severity score ≥11/25 (vs. 4.8% of nonevent group); (Table 2). Figure 2 shows examples of COVID-19 pneumonia lesions from each group detected by free software for AI-driven automatic CT analysis.

Predictors of worsening clinical course
We initially evaluated the results using univariate analysis (Table 3). Our analysis revealed that elderly patients, dementia, obesity, higher pneumonia severity scores, such as the PSI and MuLBSTA score, higher levels of initial CRP, ferritin, and LDH, and lower lymphocyte levels were associated with the desaturation group compared to the nonevent group. In addition, pneumonia volume and the extent of the CT by AI, and visual severity scores, were higher in the desaturation group when compared with nonevent group (p < 0.05).
After matching on the propensity score, 52 patients were included in the analysis, and between-group differences for all covariates were eliminated ( 1.005-1.307; p = 0.042) were the parameters for predicting worsening clinical course with oxygen desaturation in this cohort (P < 0.05); (Table 4).

Discussion
An estimated 2.3% of COVID-19 patients require tracheal intubation currently [12]. A prominent complication of advanced COVID-19 is acute hypoxemic respiratory failure requiring oxygen and MV. It is estimated that 15-20% of infected people develop severe pneumonia and 5%-10% require critical care [12]. A recent report according to the Chinese classi cation showed that 14% of patients were of the severe type and 5% were of the critical type [13]. Severe illness of COVID-19 usually begins approximately 1 week after the onset of symptoms [2], while patients have variable phases of symptoms and disease onset at admission [12]. In the present study, most patients were classi ed as having mild COVID-19 at presentation [10], and no patient received MV or high-ow oxygen therapy on admission. In the beginning, only three patients were classi ed into the desaturation group. Patients in this cohort were admitted an average of 6.8 days from symptom onset, and 12.5% were asymptomatic on admission. Nevertheless, 23.5% of these mild patients proceeded to oxygen desaturation during a median 2 days from hospitalisation. This study was undertaken to determine the predictive parameters for a worsening clinical course in patients with mild COVID-19 pneumonia admitted at a relatively early phase from symptom or disease onset.
Early case series from the Korean Cohort Study showed that 21.4% of patients received oxygen therapy, with no requirement for MV [14]. In the Daegu and Gyeongbuk area, Republic of Korea, substantial community outbreaks were linked to a large religious meeting, and a retrospective study from that area (mean age of 55.4±17.1 years) showed that 13.3% of patients required ICU admission with MV use in 11.2%, ECMO in 4.1%, and a mortality rate of 5.1% [15]. The fatality rate of South Korea, which is 2.32% (to date) nationwide, is much lower than the rates reported in China or other countries [4,6,[16][17][18][19]. Although recent data (median age of 51 years) outside Wuhan, China showed an 0.8% mortality rate [20]. Studies on COVID-19 have been limited to patients with severe or critical disease admitted to the ICU. In contrast, a recent report showed that more than 80% of COVID-19 cases were mild in China [4,6,13,[16][17][18][19]. In our study, most cases were mild, as evidenced by relatively low rates of ICU admission and mortality rates that would be related to younger ages (mean 44.8 ± 17.4) and lower comorbidity rates (26.5%) than that of previous reports. In metropolitan Seoul, all con rmed cases were either hospitalised at dedicated hospitals or monitored at community isolation centres. Many cases of COVID-19 were diagnosed by surveillance testing in South Korea, so that a large proportion of the patients in our cohort had mild or subclinical symptoms at the beginning of hospitalisation. The comprehensive hospital management with close monitoring led to mostly favourable outcomes, in addition to early case detection and immediate admission strategy of South Korea.
To our knowledge, markers for the prediction of worsening oxygenation among patients with initially mild COVID-19 pneumonia are not de ned. In this cohort, only 16.9% showed fever based on a BT ≥37.5°C.
Most of the rst recorded vital signs were within normal ranges and showed no difference between the two groups, suggesting that the worsening clinical course cannot be distinguished by initial vital signs. Apart from that, we noted that more patients in the desaturation group had higher pneumonia severity scores, and several laboratory parameters such as lymphopenia, higher ferritin and LDH, as well as older age and comorbidities such as dementia and obesity were associated with a worsening clinical course and oxygen desaturation. Feng  features combined with traditional clinical biomarkers in predicting progression to worsening course in the early stages of COVID-19. We used free software and did not analyse the parameters for each lobe. We found that the AI-driven parameters of pneumonia volume and extent of the whole lung in the initial chest CT predicted the worsening oxygen saturation in mild COVID-19 pneumonia. Our study con rms the additional bene t of the visual severity scores summing the estimated extent of each lobe and the integration of clinical and laboratory parameters.
We experienced that hypoxemia was the rst step for quick deterioration in some patients and observed that some patients are relatively asymptomatic although they have a good degree of hypoxemia for inexplicable reasons, referred to as "silent hypoxemia" in Wuhan [22]. We designed this study to identify parameters at an early stage to predict whether oxygen will decrease. This study has limitations. First, it is a single-centre retrospective study. In our hospital, the standard cutoff value for admission or transfer to the ICU and change or add-on of antiviral agents was 94% of the SpO 2 on ambient air. These criteria should be regarded as empirical, as there is no robust supporting evidence. Second, fewer patients were severe or critical compared to other studies. The majority of the desaturation group recovered without oxygen or MV, or ECMO. These favourable outcomes may not re ect the full prognosis of patients with COVID-19 in other areas. We did not analyse the factors that could affect outcomes such as regional differences, medical resources, and treatment options we applied as in a clinical trial. Nevertheless, these data, including patients with mild symptoms in a relatively early stage, are thought to be useful in predicting parameters for clinical course and triaging patients in real practice. Third, we did not analyse the CT scans at the time of deterioration or follow-up, and the features or patterns with GGO or consolidation, though the extent and burden of the total GGO/consolidation of the initial CT proved to be one of the predictive parameters for the clinical course of mild COVID-19 pneumonia.
Our study presents initial CT parameters measured by AI or visual extent scoring as well as serum markers of in ammation such as ferritin and LDH at admission as the best parameters for predicting worsening oxygenation in mild COVID-19 pneumonia patients. Funding:

None
Availability of data and materials: The datasets analyzed during the current study available from the corresponding author on reasonable request.

Page 10/19
Ethics approval and consent to participate: The study protocol was approved by the ethics institutional review boards of participating centers (Institutional Review Board of Seoul Medical Center (2020-04-024)). Due to the retrospective design of the study and there was minimal to no risk or bene t to enrolled patients, an informed consent was waived.

Consent for publication:
Not applicable (The manuscript does not contain any individual persons' data).     Figure 1 Study ow diagram.