key: cord-0705227-d6gfs5qp
authors: Wang, Jing; Zheng, Yufen; Chen, Yijun; Hu, Xingzhong; Peng, Minfei; Fang, Yicheng; Shen, Bo; Lu, Guoguang
title: Laboratory indicators in COVID-19 and other pneumonias: Analysis for differential diagnosis and comparison of dynamic changes during 400-day follow-up
date: 2021-04-27
journal: Comput Struct Biotechnol J
DOI: 10.1016/j.csbj.2021.04.063
sha: 54b88e6eb73208213c1fcbf54da309a4d9359ca7
doc_id: 705227
cord_uid: d6gfs5qp

BACKGROUND: COVID-19 is spreading rapidly all over the world, the patients' symptoms can be easily confused with other pneumonia types. Therefore, it is valuable to seek a laboratory differential diagnostic protocol of COVID-19 and other pneumonia types on admission, and to compare the dynamic changes in laboratory indicators during follow-up. METHODS: A total of 143 COVID-19, 143 bacterial pneumonia and 145 conventional viral pneumonia patients were included. The model group consisted of 140 COVID-19, 80 bacterial pneumonia and 60 conventional viral pneumonia patients, who were age and sex matched. We established a differential diagnostic model based on the laboratory results of the model group on admission via a nomogram, which was validated in an external validation group. We also compared the 400-day dynamic changes of the laboratory indicators among groups. RESULTS: LASSO regression and multivariate logistic regression showed that eosinophils (Eos), total protein (TP), prealbumin (PA), potassium(K), high-density lipoprotein cholesterol (HDLC), and low-density lipoprotein cholesterol (LDLC) could differentiate COVID-19 from other pneumonia types. The C-index of the nomogram model was 0.922. Applying the nomogram to the external validation group showed an area under the curve (AUC) of 0.902. The 400-day change trends of the laboratory indexes varied among subgroups divided by sex, age, oxygenation index (OI), and pathogen. CONCLUSION: The laboratory model was highly accurate at providing a new method to identify COVID-19 in pneumonia patients. The 400-day dynamic changes in laboratory indicators revealed that the recovery time of COVID-19 patients was not longer than that of other pneumonia types.

The main manifestations of COVID-19 are fever, dry cough, and fatigue, with approximately 11.4% of patients having at least one gastrointestinal symptom [1] . Most severe cases of COVID-19 manifest with dyspnea after one week and rapidly progress to acute respiratory distress syndrome, septic shock, difficult-to-correct metabolic acidosis, coagulation dysfunction, and multiple-organ failure [2] . COVID-19 is highly infectious, and there were nearly 120 million confirmed cases by March 8, 2021 , with a mortality of 2.2%. Among COVID-19 patients in intensive care units (ICUs), the mortality is up to 48.7% [3] .

Pathogens of bacterial pneumonia mainly include Streptococcus pneumoniae, Staphylococcus, and Klebsiella pneumoniae [4] , while for conventional viral pneumonia, they mainly include influenza A, B, C virus, and adenovirus [5, 6] . Through a study of 836 COVID-19 patients, a low frequency of bacterial coinfection was found in the early COVID-19 hospital presentation, with no evidence of concomitant fungal infection, at least not in the early phase of COVID-19 [7] . Therefore, it can be inferred that coinfection is rare in the early stage of COVID-19. Because the clinical symptoms of COVID-19 are similar to those of bacterial pneumonia and conventional viral pneumonia, distinguishing COVID-19 patients from other pneumonia patients is of vital importance.

At present, research on the differential diagnosis of COVID-19 and other pneumonia has mainly focused on imaging tools such as X-ray and CT [8] . A deep-learning convolutional neural network with the feature of transfer learning was built that could accurately differentiate COVID-19 on portable chest X-ray against normal ones. This approach could help radiologists and frontline physiologists provide timelier and more accurate diagnoses [9] . Li et al. found that a peripheral distribution, a lesion range > 10 cm, the involvement of 5 lobes, the presence of hilar and mediastinal lymph node enlargement, and no pleural effusion were significantly associated with COVID-19 [10] . An AI system built by CT images can display parameters of the background, lung field, consolidation, ground-glass opacity (GGO), pulmonary fibrosis, interstitial thickening, and pleural effusion with an accuracy of approximately 91% [11] .

However, imaging tools involve radiation and so are not suitable for pregnant women, and the image definition is related to the patient's breathing motion during the scan and the accuracy of the instrument [12] . It will be of great clinical significance to find other methods, such as laboratory indicators, to identify COVID-19 patients in the early stage of the disease.

In the laboratory, COVID-19 is mainly diagnosed by collecting nasal and pharyngeal swabs for viral nucleic acid detection by real-time reverse transcription-polymerase chain reaction (RT-PCR) [13] . The turnaround time (TAT) of the RT-PCR test is approximately [3] [4] hours, it has a false-negative rate of 15%-20% [14] , and there may be a particular risk of contamination in the process of sample processing.

Research has shown that the differential diagnosis of pneumonia can also be assisted by common laboratory indicators. For example, patients with conventional viral pneumonia may have a lower white blood cell count (WBC) and lymphocyte count. In contrast, most bacterial pneumonia patients have a higher WBC count, and the neutrophil percentage increases. In addition, some infection indicators, such as CRP, SAA, and PCT, are helpful in the differential diagnosis of bacterial pneumonia and conventional viral pneumonia [15] . Previous studies have demonstrated that blood samples of COVID-19 patients are not infectious [16] , and the TATs of conventional laboratory tests, such as routine blood tests, biochemistry, and blood coagulation, are 2 hours or less, which is shorter than RT-PCR tests, and these tests are also available in African countries and primary hospitals [17, 18] . Therefore, it is of great epidemiological value to distinguish COVID-19 from bacterial pneumonia and common viral pneumonia through common laboratory indicators with economical methods.

So far, some studies have explored the differences in some laboratory indicators between COVID-19 and non-COVID-19 patients. A study comparing COVID-19 and influenza pneumonia indicated that both cohorts showed reduced lymphocyte numbers, but the influenza cohort displayed higher white blood cell counts and PCT values [19] . Another study revealed that the white blood cell subset counts most closely correlated with COVID-19 risk were lower neutrophil and eosinophil counts [20] . Hu et al. found that laboratory indexes were different between nucleic acid-positive and nucleic acid-negative patients, and laboratory differences were also observed in COVID-19 patients and influenza patients [21] . In addition, a review showed that lymphopenia and an increased neutrophil/lymphocyte ratio (NLR) were the most consistent abnormal routine blood results and were associated with the disease course, especially in severe patients [22] . Although they compared the differences in laboratory indexes between COVID-19 and non-COVID-19 patients, they did not use laboratory indexes for modeling. Wang et al. established a differential diagnostic model for COVID-19 patients and influenza patients by using routine blood parameters through a nomogram. Although the model had a high area under the curve (AUC = 0.913), it lacked external verification, and the model could not distinguish COVID-19 patients and bacterial pneumonia patients [23] .

Our study collected the results of standard laboratory indicators of bacterial pneumonia, conventional viral pneumonia, and COVID-19 patients on admission and during the 400-day follow-up, aiming to establish an early differential diagnostic model of COVID-19 and non-COVID-19 pneumonia patients through common laboratory indicators on admission, and to compare the recovery time and long-term dynamics of these laboratory indicators in different types of pneumonia patients.

From January 16, 2020, to April 1, 2020, a total of 144 patients with confirmed COVID-19 in Taizhou city were enrolled in the study. One patient missing more than 60% of the laboratory indicators was excluded, leaving 143 COVID-19 patients in the study. We also 

Diagnostic criteria included clinical manifestations, chest CT, and etiological examination.

RT-PCR of nasopharyngeal swabs was used to diagnose patients with COVID-19 and conventional viral pneumonia. Patients with bacterial pneumonia were identified by sputum culture.

EDTA-K2 anticoagulant samples were measured by a Sysmex 2100D routine hematology analyzer (Sysmex, Japan), and ESR was performed by using the Italian Ali-fax Test 1 automatic ESR analyzer. Serum samples were centrifuged at 3500 rpm for 5 minutes, and both CRP and SAA were performed by using an AU5800 (OLYMPUS, Japan). PCT was carried out by adopting the Roche Cobas e411 electrochemiluminescence analyzer (Roche Diagnostics, Germany). Sodium citrate plasma samples were centrifuged at 3500 rpm for 5 minutes, and coagulation parameters were gathered using a Sysmex CS 5100i (Sysmex, Japan) automatic hemagglutination analyzer. Arterial blood gas analysis was performed by using the GEM Premier 3500.

First, we collected all the patients' laboratory indicators on admission, including blood routine, biochemical, hemagglutination, infectious, and blood gas indexes. We established a laboratory differential model of COVID-19 with bacterial pneumonia and conventional viral pneumonia by LASSO regression, multivariate logistic regression, and a nomogram. We also collected the laboratory indexes of all patients during up to 400 days of follow-up to compare the recovery time and long-term dynamics of the laboratory indicators in patients with different types of pneumonia.

All the graphs were drawn and the corresponding statistical analyses done in R (Version: 4.0.2). Continuous variables are expressed as median (P25-P75). The Mann-Whitney U test was used for comparisons between two groups. The Kruskal-Wallis H test was used to compare three groups. Categorical variables are expressed as number (percentage) and were compared between groups using chi-square test. LASSO regression and multivariate logistic regression analyses were used to screen laboratory indicators. Box plots and heat maps were used to compare laboratory indexes between groups. A nomogram was built to determine the role of each laboratory parameter in the differential diagnosis of COVID-19, bacterial pneumonia, and conventional viral pneumonia. A locally weighted scatterplot smoothing (LOWESS) plot was used to compare the dynamics of the laboratory indicators in different groups. P ≤0.05 was considered to indicate statistical significance. Table   1 and Table S1. The first blood sampling times after admission following symptom onset in the model group and the validation group are displayed in Figure S1 .

In the model group, among the routine blood and infection indicators, the WBC counts [especially eosinophils (Eos)] and C-reactive protein (CRP) were lower in COVID-19 patients than other patients. Among liver function indicators, alanine aminotransferase (ALT), total protein (TP), globulin (Glb) and total bile acid (TBA) were the highest in COVID-19 patients.

For the renal function indicators, COVID-19 patients had the highest creatinine (Cr) value and the lowest retinol binding protein (RBP), blood urea nitrogen (BUN) and estimated glomerular filtration rate (eGFR) levels. For serum electrolytes, the lowest potassium (K), sodium (Na) and chlorine (Cl) levels were found in COVID-19 patients. Among the blood lipid indexes, the highest triglyceride (TG) and low-density lipoprotein cholesterol (LDLC) and the lowest highdensity lipoprotein cholesterol (HDLC) were found in COVID-19 patients. For coagulationrelated indexes, the fibrin (Fib) and D-dimer (DD) levels in COVID-19 patients were the lowest.

In addition, the highest levels of arterial oxygen saturation (SaO2) and oxygenation index (OI) were found in the COVID-19 group. Laboratory indicators of all cohorts and the model group on admission are displayed in Table 2 .

The laboratory indexes with significant differences among groups were included in the LASSO regression. The screened indicators, including Eos, HDLC, K, LDLC, TP, Na, WBC, DD, PA, Glu, Hb, ALP, Cl, OI, and CRP, were included in multivariate logistic regression. The indicators with a P value < 0.01 were selected as modeling indicators: Eos, HDLC, K, LDLC, TP, PA and ALP. Heatmaps and box plots showed the differences in the above indexes among the groups.

We further established a nomogram model consisting of Eos, HDLC, K, LDLC, TP, PA and ALP. The model's C-index was 0.922. The calibration curve, clinical decision curve and clinical impact curve showed that the model had high differential diagnostic ability ( Figure 3A -D). Applying the nomogram to the validation group showed that the area under the curve (AUC) of the model was 0.902, with a sensitivity and specificity of 82.5% and 80.0%, respectively ( Figure 3E ). The calibration curve of the validation group is displayed in Figure 3F .

The level of Eos in COVID-19 patients stayed the lowest out of the 3 groups within 200 days after onset, and its dynamic changes were similar to those in conventional viral pneumonia patients, especially influenza B pneumonia patients. In COVID-19 patients, Eos stopped rising 25 days after onset, and patients with an oxygenation index <300 mmHg had a very low level of Eos for a long time after the onset of the disease. It recovered by 350 days after the onset of the disease. Eos in bacterial pneumonia patients showed a bimodal trend, with the second peak appearing 100 days after onset. In particular, among patients under 50 years of age with an oxygenation index > 300 mmHg and gram-positive bacterial pneumonia, Eos exceeded the upper limit of the reference range at its second peak. In all 3 groups of patients, eosinophils peaked later in male patients, patients younger than 50 years of age and those with an oxygenation index > 300 mmHg.

The ALP levels of COVID-19 patients stayed within the reference range throughout the 400-day follow-up, while obvious fluctuations were observed in bacterial pneumonia and conventional viral pneumonia patients during the follow-up period. The dynamic change trend of ALP was opposite between males and females. Different change trends were found between patients aged over 50 and under 50 and between patients with oxygenation indexes higher than 300 mmHg and lower than 300 mmHg. The change trends of influenza A pneumonia and influenza B pneumonia patients were similar, but those of gram-positive bacteria and gramnegative bacteria pneumonia patients were opposite after 50 days.

The levels of total protein (TP) and prealbumin (PA) in the 3 groups were significantly different; COVID-19 patients had the highest TP level, and conventional viral pneumonia patients had the lowest TP level. TP and PA in the COVID-19 patients were low up to 20 days after onset and then quickly recovered and stayed in the normal range. TP and PA in female patients, patients aged > 50 years and patients with an oxygenation index < 300 mmHg were lower. In particular, in patients with an oxygenation index < 300 mmHg, although the TP level was normal, the PA level was below the lower limit of the reference range, and the change trend of PA in influenza B patients was similar to that in COVID-19 patients whose oxygenation index was lower than 300 mmHg. The trends of serum potassium level in COVID-19 patients and conventional viral pneumonia patients were basically the same: low at onset and then recovering and stabilizing in the normal range. In bacterial pneumonia patients, there was a downward trend 100 days after the onset, and it was even below the lower limit of the reference range at 180 days. A rise in the serum potassium levels of patients under 50 years old and with an oxygenation index below 300 mmHg was observed after an obvious downward trend.

High-density lipoprotein cholesterol (HDLC) and low-density lipoprotein cholesterol (LDLC) levels in bacterial pneumonia patients were relatively stable during the follow-up period, but the trends of change between gram-positive and gram-negative bacterial pneumonia were different. HDLC and LDLC levels were different between COVID-19 and conventional viral pneumonia patients, but the trends were similar. The LDLC levels of COVID-19 patients were the highest throughout the follow-up period. The HDLC and LDLC of patients of different sexes, ages and oxygenation index groups were similar during the follow-up period, showing a slight decrease at first, followed by an increase, before finally stabilizing within the normal range. HDLC and LDLC in patients with oxygenation index below 300 mmHg had a peak approximately 120 days after onset, and HDLC significantly exceeded the upper limit of the reference range, then returned to the reference range 400 days after onset. The change trends of HDLC and LDLC in conventional viral pneumonia patients in different age and oxygenation index groups were different: They peaked in patients under 50 years of age and patients with an oxygenation index lower than 300 mmHg at 120-150 days after onset at levels far above the upper limits of the reference ranges.

In this study, a differential diagnostic model was established in the form of a nomogram We collected all patients' laboratory results on admission for modeling. Because there were more laboratory indicators than members of each cohort, we first used LASSO regression to screen the indicators. LASSO regression was characterized by variable selection and regulation while fitting the generalized linear model. Therefore, whether the response variable is continuous, binary or multivariate discrete can be modeled and predicted by LASSO regression. In clinical applications, if the independent variables have multicollinearity or the number of variables is much larger than the sample size, LASSO regression should be done [26] .

According to the LASSO regression results, we further included the selected indicators in multivariate logistic regression, and we selected 7 indicators with P<0.01, including Eos, HDLC, K, LDLC, TP, PA and ALP, to establish the differential diagnostic model. Boxplots and heatmaps showed that LDLC and TP were higher in the COVID-19 group, and Eos, HDLC, K, PA and ALP were lower. During the follow-up of 400 days, it was found that there were significant differences in the dynamic changes between COVID-19 patients and other pneumonia patients.

Previous studies have found that eosinophils in COVID-19 patients are significantly low on admission, especially in critically ill patients, and the recovery of eosinophils could be used as a predictor of recovery [27] . In our study, eosinophil counts were always the lowest in In this study, we found that the total protein level of COVID-19 patients was higher than that of bacterial pneumonia and conventional viral pneumonia patients, but prealbumin was lower. Prealbumin is a specific indicator reflecting the synthetic function of the liver that is more sensitive and accurate than albumin [29] , and a previous study revealed lower prealbumin levels in COVID-19 patients than in non-COVID-19 patients [30], which was consistent with our study. Therefore, we can infer that acute liver injury may arise in the early stage of COVID-

A study showed that the ALP in COVID-19 patients was significantly lower than that in patients with influenza virus infection [31] and community-acquired pneumonia (CAP) [32] .

In this study, we also found that the ALP levels of COVID-19 patients were lower than those of bacterial pneumonia and conventional viral pneumonia patients on admission. Our study also found that sex and age influenced the changes in ALP in bacterial pneumonia and conventional viral pneumonia patients during follow-up, but no significant difference was observed in COVID-19 patients between subgroups, suggesting that ALP was not a sensitive indicator of the disease course of COVID-19 patients, which might be related to the mild degree of liver injury in COVID-19 patients.

It has been revealed that hypokalemia is the second most common complication in emergency patients with community-acquired pneumonia, which was related to the prolonged hospital stay, but it has nothing to do with pneumonia recurrence [33] . Our data showed that the serum potassium level of COVID-19 patients was the lowest at the onset of the disease, but 

In conclusion, the differential diagnostic model established by laboratory indicators on admission in this study is highly accurate when it is used to distinguish COVID-19 from bacterial pneumonia and conventional viral pneumonia. It can provide a new method for clinicians to identify COVID-19 patients, although a larger sample and prospective studies are still needed for further validation. More importantly, we compared the 400-day dynamic changes in laboratory parameters between groups and revealed that the recovery time of COVID-19 patients was not longer than that of bacterial pneumonia and conventional viral pneumonia patients. 

Epidemiological, clinical and virological characteristics of 74 cases of coronavirus-infected disease 2019 (COVID-19) with gastrointestinal symptoms

COVID-19: Learning from Lessons To Guide Treatment and Prevention Interventions

Risk Factors Associated With Mortality Among Patients With COVID-19 in Intensive Care Units in Lombardy, Italy

Infectious Diseases Society of, Guidelines for the management of adults with hospital-acquired, ventilator-associated, and healthcare-associated pneumonia

Viral pneumonia

Update in adult community-acquired pneumonia: key points from the new American Thoracic Society/Infectious Diseases Society of America 2019 guideline

Liver enzyme patterns in maternal deaths due to eclampsia: A South African cohort

Radiographic and CT Features of Viral Pneumonia

Deep-learning convolutional neural networks with transfer learning accurately classify COVID-19 lung infection on portable chest radiographs

Comparison of chest CT findings between COVID-19 pneumonia and other types of viral pneumonia: a two-center retrospective study

Applicable AI System for Accurate Diagnosis, Quantitative Measurements, and Prognosis of COVID-19 Pneumonia Using Computed Tomography

The influence of respiratory motion on CT image volume definition

COVID-19: Progress in diagnostics, therapy and vaccination

Routine blood tests as a potential diagnostic tool for COVID-19

Review of Non-Bacterial Infections in Respiratory Medicine: Viral Pneumonia

Viral load dynamics and disease severity in patients infected with SARS-CoV-2 in Zhejiang province, China

Resolution of anaemia in a cohort of HIV-infected patients with a high prevalence and incidence of tuberculosis receiving antiretroviral therapy in South Africa

Full blood count and white cell differential count reference ranges obtained from a healthy urban South African population residing in the Western Cape of South Africa

Clinical characteristics of COVID-19 and its comparison with influenza pneumonia

Epidemiological and Clinical Predictors of COVID-19

Laboratory data analysis of novel coronavirus (COVID-19) screening in 2510 patients

A summary of the diagnostic and prognostic value of hemocytometry markers in COVID-19 patients

Differentiating Between 2019 Novel Coronavirus Pneumonia and Influenza Using a Nonspecific Laboratory Marker-Based Dynamic Nomogram

Nomograms in oncology: more than meets the eye

Scientific research progress of COVID-19/SARS-CoV-2 in the first five months

Logistic LASSO Regression for Dietary Intakes and Breast Cancer

concentration; RDW: Red blood cell distribution width; PLT: Platelet; MPV: Mean platelet volume; Pct: Platelet crit ;ESR: Erythrocyte sedimentation rate; CRP:C-reactive protein; ALT: Alanine aminotransferase; AST: Aspartate aminotransferase; ALP: Alkaline phosphatase; GGT: γglutamine transferase

Albumin globulin ratio; TBA: Total bile acid; PA: Prealbumin; RBP: Retinol binding protein

BUN: Blood urea nitrogen; UA: Uric acid; eGFR: estimated Glomerular Filtration Rate; Cl: Chlorine; Ca: Calcium

TG: Triglyceride; CH: Total cholesterol; HDLC: High density lipoprotein cholesterol; LDLC: Low density lipoprotein cholesterol; PT: Prothrombin time; INR: International normalized ratio; APTT: Activated partial thromboplastin time

pH: Hydrogen ion concentration; PO2:Arterial partial pressure of oxygen; SaO2:Arterial oxygen saturation