Characterization of Conrmed and Suspected COVID-19 Pneumonia Patients in a Retrospective Cohort Study in Wuhan

Background: A methodical comparison of conrmed and suspected COVID-19 patients has not been previously reported. Therefore, we thoroughly analyzed the demographic and clinical characteristics between these groups to identify mortality risk factors. Methods: A retrospective cohort of 1,276 hospitalized COVID-19 pneumonia patients at Tongren Hospital (Wuhan, China; January 27 to March 3, 2020) was studied. Cox regression analyses were performed to evaluate multiple mortality risk factors. Results: Both cohorts of conrmed (n=797) and suspected (n=479) patients exhibited typical demographic, clinical, and radiological characteristics. Treatment methods were consistent and both groups shared similarities in many demographic and clinical characteristics: age ( ≥ 65, 45.9% vs 41.8%, P=0.378) and lung disease (12.5% vs 14.6%, P=0.293). However, conrmed patients exhibited more severe disease manifestations than those in suspected patients: a higher incidence of fever (65.4% vs 58.0%, P<0.01), lower lymphocyte count (1.12×10 9 /L vs 1.22×10 9 /L, P=0.022), higher C-reactive protein (CRP) (11.60 mg/L vs 7.61mg/L, P=0.021), and more severe radiographic manifestations (lung infection incidence, 3.8% vs 3.0%, P=0.014; ground-glass opacity lesion incidence, 2.3% vs 2.0%, P=0.033). The dynamic proles of lymphocytes, monocytes, D-dimer, and CRP, clearly delineated conrmed patients from suspected patients exhibiting critical illness. Cox regression analysis demonstrated that lung disease (adjusted hazard ratio 8.972, 95% CI: 3.782-21.283), cardiovascular disease (3.083, 1.347-7.059), neutrophil count (1.189, 1.081-1.307), age (1.068, 1.027-1.110), and ground-glass opacity lesions (1.039, 95%


Background
Severe acute respiratory syndrome conoravirus-2 (SARS-CoV-2) was rst reported in Wuhan in December 2019 [1] and the World Health Organization (WHO) named it severe coronavirus disease 2019 (COVID- 19) on February 11, 2020. As of September 24, 2020, over 30.6 million con rmed cases and 950000 deaths worldwide (WHO Coronavirus Disease (COVID-19) Dashboard) have been reported due to SARS-CoV-2, and the numbers are still rising every day. The experiences in combating COVID-19 and patient management in Wuhan are informative for medical intervention and the formulation of public policy against COVID-19 pneumonia worldwide.
While many reports focusing on con rmed COVID-19 patients have been published, a number of cases of COVID-19-suspected patients (SARS-CoV-2 RNA negative) during the outbreak have been overlooked. The WHO [2][3][4], Centers for Disease Control and Prevention (CDC) [5,6] and the National Institute for Health and Care Excellence (NICE) [7] suggest these patients seek medical attention if they have a fever, cough, and di culty breathing. Chinese hospital criteria are similar, except mildly ill patients have been asked to isolate in temporary hospitals reconstructed from gymnasiums and exhibition centers [2,8]. Moreover, many guidelines for the management of suspected COVID-19 patients suggest that for people with a high likelihood of infection, based on exposure history and/or clinical presentation, a single negative test does not completely exclude the possibility COVID-19 infection because false-negatives are possible of PCR test [3,7,8].
Some reports [9][10][11] reveal epidemiological, demographic, and clinical aspects of COVID-19 treatment, and provide valuable information for clinical intervention and the formulation of public policies. However, a signi cant number of patients, suspected to have COVID-19, who presented with typical symptoms including pneumonia and lung radiographic abnormalities, tested negative for SARS-CoV-2 more than once in the hospital. These suspected patients were treated in a hospital designated for COVID-19 cases in Wuhan to reduce mortality from February 4, 2020 to March 3, 2020. The suspected patients analyzed in this study tested negative for SARS-CoV-2 RNA, even after many tests. The lack of con rmation by a positive RNA test challenges the precision of diagnosis and the rational treatment for these patients, which may undermine the optimal distribution of limited medical resources during the outbreak. Further, the distinguishing clinical characteristics between con rmed and suspected COVID-19 patients are lacking. Currently, there is insu cient evidence to formulate a speci c management plan for these suspected patients.
The objective of this cohort study was to compare and contrast the clinical characteristics and outcomes between con rmed and suspected COVID-19 pneumonia patients. Here, we reveal that suspected COVID-19 pneumonia patients display technical differences from con rmed patients. More importantly, we con rm multiple risk factors for mortality in con rmed and suspected patients. Further, our study provides instructive information for combating and managing the COVID-19 pandemic.

Setting and sample
This cohort study was conducted in the endemic area of COVID-19 at the Tongren Hospital of Wuhan University (Wuhan, Hubei province, China). The hospital originally provided broad service to the local community but was reassigned by the government to exclusively hospitalize COVID-19 patients. All the suspected and con rmed COVID-19 pneumonia patients from Jan 27 to Mar 3, 2020 were included in our study cohort.

Study design
The diagnosis of con rmed and suspected COVID-19 pneumonia patients was performed in accordance with the tentative fth revised edition of "The diagnosis and treatment plan of Coronavirus Pneumonia 2019," issued by the National Health Commission of the People's Republic of China [8]. The patients were consecutively admitted to the Tongren Hospital of Wuhan University from Jan 27 to Mar 3, 2020.

Data collection
Electronic medical records were collected and reviewed by practicing physicians and a research team from Tongren Hospital of Wuhan University. Of 1,369 patients, 93 patients were excluded: 16 had normal computerized tomography (CT) scan results and were not administered an RNA SARS-CoV-2 test, 37 had no CT scan results and no SARS-CoV-2 RNA test results, 40 had no CT scan results and tested negative for SARS-CoV-2 RNA. In total, 1,276 cases were included in this study. 797 patients were con rmed as SARS-CoV-2 positive by quantitative reverse transcription polymerase chain reaction (qRT-PCR). 479 suspected cases tested negative, at least once, for SARS-CoV-2 RNA during their hospital stay. The demographic, clinical, radiographic, laboratory, and patient outcome information was obtained with data collection forms from electronic medical records. The radiographic features of chest CT scans were measured by arti cial intelligence technology [12,13]. The time of disease onset was de ned as the day when the rst symptoms were noticed. Patients were categorized as critical or non-critical based on their medical records.

Laboratory procedures
The diagnostic criteria and primer sets recommended by the National Institute for Viral Disease Control and Prevention, China (http://www.chinaivdc.cn/kyjz/202001/t20200121_211337.html) were followed.
Brie y, SARS-CoV-2 RNA was detected by a qRT-PCR assay following the manufacturer's instructions (BGI Biotech Ltd, Wuhan, China). A probe speci c for the highly conserved region of SARS-CoV-2 was designed and conjugated with a uorescein (FAM) tag. Brie y, throat swab samples were collected, and RNA was extracted from the samples within 2 hours. A one-step qRT-PCR assay was carried out using the following cycle parameters: 50 °C for 20 min and then 95 °C for 10 min, followed by 40 cycles of 95 °C for 15 sec, 60 °C for 30 sec. A cycle threshold (Ct) value of less than 37 cycles was de ned as positive test result, and a Ct value of 40 or more was de ned as a negative test result. A retest was required for con rmation of results with an intermediate Ct value (between 37 and 40). If the retest Ct value was less than 40 cycles and the ampli cation curve displayed an increasing slope,the sample was veri ed as positive; otherwise it was determined as negative.

De nitions
Fever was de ned as axillary temperature of at least 37.3 ℃. Moderate cases were de ned as patients displaying signs and symptoms which might include fever and respiratory or pneumonia manifestations. Patients were categorized as severe cases if they met one or more of three criteria: respiratory rate ≥ 30 breaths/min, oxygen saturation ≤ 93% on room air, or PaO2/FiO2 ≤ 300 mmHg (1 mmHg = 0.133 kPa). Critical cases were determined based on whether the patient met any three criteria: respiratory failure occurred and mechanical ventilation was required, shock occurred, or the case was complicated with other organ failure that required monitoring and treatment in the ICU.
The chest imaging features of COVID-19 patients was as follows: (1) In the early stage, the radiographs show multiple, small, patched shadows and interstitial changes, especially in the lung periphery. (2) As the disease progresses, the CT scans reveal further development of multiple ground glass shadows and in ltration shadows in both lungs. (3) In severe cases, lung consolidation may occur. Pleural effusion is seldom found in patients with COVID-19 [1].

Statistical analysis
Statistical analysis was performed with SPSS 22.0 (version 22.0, IBM, Armonk, NY, USA), unless otherwise stated. The normal distribution assumption was tested using the Kolmogorov-Smirnov test (P > 0.05). The continuous variables were compared using the Mann-Whitney U test and are presented as the median value (25th percentile − 75th percentile). Categorical variables were compared using the Chisquare test or Fisher's exact probability test and are presented as numbers and percentages. A Cox regression were performed to analyze the effect of several risk factors on survival. P values < 0.05 was considered statistically signi cant for two-tailed tests. Abundance levels of the clinical index at each individual time point were plotted using GraphPad Prism 8.0.

Clinical characteristics of con rmed and suspected COVID-19 patients
Both con rmed and suspected patients displayed typical clinical symptoms including fever, dry coughing, fatigue, chest tightness, and sputum production. Major comorbidities included hypertension, diabetes, cardiovascular disease, lung disease, liver disease, and chronic kidney disease. Of all the patients, 11.6% were treated under critical conditions and the overall fatality rate was 4.4% (Table 1). Although most laboratory parameters remained within the normal range, we noticed a high alveolar-arterial oxygen gradient (AaDpO 2 , 90 mmHg), suggesting a defect of lung oxygen exchange due to impaired oxygenation, typical of lung dysfunction in both groups of COVID-19 patients.

Vital characteristics demonstrate signi cant differences between con rmed and suspected patients
We further compared the clinical indices of con rmed and suspected patients. These two groups of patients displayed comparable lengths of time from the onset of symptoms to hospital admission, and both groups presented with a similar spectrum of disease (Table 1). However, these two groups of patients did show signi cant differences in many aspects. The suspected patients had a lower incidence of fever than con rmed patients (P < 0.01). Additionally, although there was no signi cant difference (P = 0.447) between the con rmed and suspected patient groups regarding disease classi cation (moderate, severe, and critical patients), the outcome of disease appeared signi cantly different (P = 0.002). Consistently, the con rmed patients also required a longer hospital stay (17 days vs 15 days, P < 0.001) ( Table 1). The differences in these two group were also re ected by several laboratory parameters, as evidenced by signi cantly increased C-reactive protein (CRP) (11.60 mg/L vs 7.61 mg/L, P = 0.021) and decreased lymphocytes (1.12 × 109/L vs 1.22 × 109/L, P = 0.022) in con rmed patients compared with that in suspected patients (Table 2).

Arti cial intelligence-aided analysis of CT scans in con rmed and suspected patients
The chest CT scans among both groups of patients, quanti ed by arti cial intelligence-aided technology, also exhibited signi cant differences in the incidence of lung infection. The radiographic data revealed that an overall 3.5% of the total COVID-19 patients suffered lung infection (Table 3). Notably, the con rmed patients displayed lung infection more frequently than suspected patients, as detected by CT scan (3.8% vs 3.0%, P = 0.014). Ground-glass opacity lesions in the lungs are a unique abnormality for COVID-19 patients [14].As expected, the con rmed patients more commonly exhibited ground-glass opacity than suspected patients (2.3% vs 2.0%, P = 0.033) ( Table 3). These observations suggest that con rmed patients tend to be more severe than suspected patients. Of note, the con rmed patients showed a higher incidence of fatality than suspected patients (5.8% vs 2.1%, P < 0.01) ( Table 1); however, the multiple factors analysis found that clinical type (con rmed vs suspected) was not signi cantly associated with mortality (Table 4).  In the Cox model, death is 1 and survival is 0. A patient with lung disease history or cardiovascular disease history is 1 and all others are 0. A patient been con rmed is 1 and else is 0.

Dynamic pro les of laboratory parameters in critical COVID-19 patients
We extracted the laboratory parameters from critical cases and compared con rmed patients with suspected patients. The dynamic pro les of several laboratory parameters were clearly distinguishable between con rmed and suspected patients (Fig. 1). The con rmed patients maintained low counts of lymphocytes and monocytes overtime, whereas suspected patients tended to recover at late stage ( Fig. 1A, B). The con rmed patients also displayed higher D-dimer and CRP levels (with a bell-shaped curve) than the suspected patients (with a atter curve) (Fig. 1C, D).

Risk factors associated with mortality in con rmed patients and suspected patients
We further assessed risk factors for the mortality of COVID-19. A Cox proportional hazard model was used to analyze the risk factors for mortality in all con rmed and suspected patients.

Discussion
The rapid transmission of SARS-CoV-2 has caused a global outbreak of COVID-19, which often overwhelms the capacities of local hospitals in the affected areas. This situation is complicated by the fact that a large number of patients presenting to these local hospitals were negative for SARS-CoV-2, and repeat testing was required to properly categorize them (the median number of RNA tests per patient is 2, range 1-9 in this study). These suspected patients displayed typical COVID-19 symptoms and lung abnormalities, which challenged the precise diagnosis and administration of appropriate treatment. Due to the exacerbating situation, the suspected patients were clinically diagnosed and treated as COVID-19 in Wuhan from February 4, 2020 to March 3, 2020. Nevertheless, these cases have been insu ciently characterized due to a lack of su cient clinical evidence at the time. In this retrospective cohort study, we have thoroughly analyzed the clinical data from these patients to improve the diagnosis and treatment of suspected and con rmed COVID-19 patients in the ongoing pandemic.
In fact, the suspected patients were largely comparable to the con rmed patients (Tables 1-2), except that the suspected patients tended toward a less severe presentation, for example, shorter hospital stays and less frequent ventilator usage. This observation is supported by another study [15]. It is worth noting that our serological test for SARS-CoV-2 antibodies revealed that most of the suspected patients were positive for SARS-CoV-2 IgM or IgG (Table 3). A previous study also found that the suspected patients who originally tested negative via qRT-PCR eventually turned out to be positive patients when repeat tests (up to 6) were performed over time [16]. These observations strongly argue that the majority of suspected patients meeting hospital criteria for moderate to critical illness are indeed real COVID-19 patients. They do have typical pneumonia symptoms and may progress to severe illness (10.2%, Table 1), or even death (2.1%, Table 1). Therefore, these patients must be properly treated and managed. Self-isolation and closely monitoring viral RNA levels and disease progression may be the optimal choice for managing these suspected patients during the COVID-19 outbreak, as practiced in affected areas.
Our observations on suspected patients may re ect special feature of coronavirus infection. A previous study reported that MERS-CoV-negative patients showed typical MERS symptoms, display less severe disease manifestation, and survived much better than MERS-CoV-positive patients [17]. Given that viral load is a very important factor in disease severity and outcome, suspected patients may simply have a viral load that is lower than the limit of detection. To support this, we found that the qRT-PCR Ct value was correlated with disease severity and death (Table 3). This study provides data for improving the current management guidelines for suspected patients in various countries; a negative RNA test alone should not exclude patients from the diagnosis of COVID-19. RNA test results should be considered in combination with radiologic and serological features. Negative RNA test results may also be due to errors in throat swab collection, or simply due to the fact that throat swabs may not be the best specimen for SARS-CoV-2 detection [14,18]. Further, we could not exclude that variant SARS-CoV-2 strains may exist in these patients that could not be recognized by the present primer sets. Neither could we exclude the possibility that other unidenti ed pathogens may have caused the pneumonia in these patients.
Risk factors for mortality have been implicated in several studies. Patients in these studies are from the very early stages of the COVID-19 outbreak in Wuhan, and the reported fatality is relatively high (54 deaths out of 191 patients in one study) [19]. These patients are likely highly selected and may not be representative of the global patient population. In this study, our patients were admitted from Jan 27 to Mar 3, 2020, a time period that spanned the major course of the epidemic in Wuhan.
Furthermore, we used multiple-variable Cox regression, instead of logistic regression, so that the potential effect of the length of the hospital stay on the clinical outcome was considered. It is worth pointing out that the effect of clinical type (con rmed vs suspected) on mortality was not signi cant in the Cox model (Table 4). This further supports the need to treat both types of patients if they present with moderate to serious illness. In addition to age (a well-known risk factor), lung disease was found to be a signi cant factor in our model. It is conceivable that underlying lung disease may worsen lung dysfunction and failure, representing a major cause of severe COVID-19 [20]. The cardiovascular system is known to be an important target for SARS-CoV-2, due to the expression of ACE2, and myoglobin and LDH levels re ect impairment in the heart [21]. Our results are consistent with the idea that COVID-19 mortality is most likely due to multiple organ failure. Additionally, we observed that CRP and B-type natriuretic peptide levels were associated with life-threatening infections and impaired heart function, leading to death in suspected patients. Ground-glass opacity lesions represent a unique feature of COVID-19, and the increased incidence of ground-glass opacity lesions was identi ed in our model as a signi cant risk factor. Our ndings demonstrate the valuable application of using AI-aided quanti cation of chest CT scan in precise diagnosis of COVID-19. Interestingly, the decreased adjusted hazard ratio of albumin was also identi ed, which supports the use of albumin in combating SARS-CoV-2 infection. Monitoring these factors on admission may help predict the prognosis of patients and aid in the formulation of rational patient management.

Limitations of this study
This study has several limitations. First, all of the data in this study were collected from a single hospital; a multiple-center study may help reveal more aspects of this disease. Second, only throat swabs were used, and the serum of patients was not obtained to evaluate viremia. Third, many patients still remained in hospital, or had been transmitted to other hospitals, by the last follow-up day. Therefore, the information of these patients is missing for the assessment of risk factors. However, the above limitations do not negate our main ndings.

Conclusion
Suspected patients with moderate to severe illness, who may have a fever and di culty breathing, should seek medical attention to reduce fatality. Multiple factors on admission need to be considered to assess the mortality risk of con rmed and suspected patients for the administration of appropriate clinical management.