Background
Coronavirus disease 2019 (COVID-19), a newly emerged respiratory disease caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has recently become the most important global public health emergency. Because of the coronavirus's novel nature, it remains difficult to come up with specific remedies that will allow us to prevail over COVID-19. It is now widely recognized that a large-scale epidemic of COVID-19 can cause many deaths and more emergency patients, which presents a severe challenge to regional healthcare systems [
1]. Rational medical resource allocation and efficiency of emergency rescue, which will be key measures to reduce the mortality of disease, depend on early prediction for length of hospital stay and disease progression.
Among the COVID-19 cases, about 81% are in mild or moderate condition, and 19% are severe or critical cases [
2]. Mild patients do not need hospitalization, while some moderate patients may need. The length of hospital stay means the amount of time patients spend on medical resources. Thus, identifying factors affecting hospitalization duration to assess the risk stratification of patients will help to shorten hospital stay to the briefest amount of time possible and alleviate the burden of medical resources. Compared with moderate patients, severe and critical patients are more likely to progress rapidly and have adverse outcomes [
3]. Predicting patients at high risk of progression, who often require more care and precise treatment, will improve the prognosis.
A recent systematic review critically appraised published and preprint reports of prediction models for prognosis of patients with COVID-19 [
4]. The most reported predictors of severe prognosis included age, sex, C-reactive protein (CRP), lactate dehydrogenase (LDH) and lymphocyte count. However, all included studies were rated at high risk of bias, mostly because of small sample sizes (ranging from 26 to 577 patients) and high risk of model overfitting. Reporting quality varied substantially between studies, and calibration of predictions was rarely assessed. In addition, findings from previous studies are inconsistent. For example, although several studies reported older patients had longer length of hospital stay, other studies have shown that demographic variables including age may not be good indicators for length of stay [
5‐
7]. Therefore, sharing data with large sample sizes and updating of COVID-19 prognosis related prediction models are urgently needed.
Here, we performed a retrospective cohort study in 2425 cases from one of the largest special hospital of COVID-19 in Wuhan, China. Our aims were to construct two risk stratification scoring systems for predicting length of hospital stay and disease progression in moderate and severe patients, respectively. We present the following article in accordance with the STROBE reporting checklist.
Discussion
In this cohort study, we identified risk factors for hospitalization duration and disease progression in patients from a large special hospital of COVID-19 in Wuhan, China. In particular, more clinical symptoms, abnormal platelet count, higher CRP, lower albumin, higher LDH on admission and higher body temperature during hospitalization were significantly associated with long-stay duration in moderate patients. History of respiratory disease, lower platelet count, higher NLR, higher creatinine, higher LDH, prolonged PT, and higher body temperature were associated with increased risk of disease progression in severely ill patient. Additionally, we built two easy-to-use risk stratification score systems that can be used by clinicians, named as STPCAL and TRPNCLP, to estimate the risk of hospital stay duration and disease progression, respectively.
CRP is an important biomarker to reflect cell injury, inflammation and tissue damage. The increase of CRP may indicate the state of inflammatory reaction and the degree of damage to the immune system caused by viral infection. Several studies have found that CRP levels were associated with COVID-19 severity [
13,
14]. Furthermore, a preprint study has also shown that CRP is one of the earliest biomarkers that changes to reflect physiological complications and could be used as an effective biomarker for predicting progression of COVID-19 infection [
15]. The deficiency of nutritional intake, consumption of albumin by the synthesis of acute inflammatory protein, and the abnormal distribution of albumin caused by pulmonary exudation are reflecting by the decrease of albumin [
16]. A recent meta-analysis has shown decreased serum albumin level has been associated with severe COVID-19 and mortality. Low albumin level can help to early recognition of severe COVID-19 [
17]. Our findings provided an important piece of evidence that both elevated CRP and decreased albumin were independently associated with hospital long-stay in patients with moderate COVID-19.
LDH represents the glucose metabolism of body tissue, high LDH levels are associated with cell damage occurring in various diseases, including inflammatory pulmonary disorders. Up to date, more and more convincing evidence links LDH as a biomarker to the development and severity of COVID-19 infection [
18]. Han et al. reported that LDH was an important indicator to reflect the disease severity of COVID-19 patients [
19]. They found that LDH was positively correlated to the indicators of inflammation, heart and liver function damage, but negatively correlated with lymphocyte count. Several studies suggest that LDH was a predictor of COVID-19 progression and mortality [
13,
20]. Our study further demonstrated the role of LDH in COVID-19, suggesting that LDH could be an auxiliary marker predicting a longer hospital stay and disease progression.
Platelet count is a simple and easy-to-use biomarker in clinical practice. Current studies have shown that a variety of cytokines, including IL-3, IL-6, IL-9, and IL-11, can promote the production of megakaryocytes and release of platelet. However, severe infections could cause secondary thrombocytopenia, such as disseminated intravascular coagulation (DIC), which is associated with significant bleeding manifestations and more common in fatal outcomes. Interestingly, we found that among moderate patients, normal range and increased platelets levels in patients predicted longer average hospital stay compared to patients with thrombocytopenia. Qu et al. reported that patients with platelet peaks have a longer average hospital stay, which is consistent with our finding [
21]. Increased platelets activated by excessive inflammation affect abnormal coagulation state and faster aggregation, leading to thrombotic disease [
22]. Moreover, pulmonary micro-thrombosis disturbs the blood oxygen transport to reduce lung function of the patients, which may be related to the longer course of the disease. Different with what was found in moderate patients, we found that lower platelet count could predict the progression in severely ill patients. Lower platelet count was associated with disease severity score and considered to be a risk factor for death in patients with severe acute respiratory syndrome (SARS) [
23]. Studies also reported that thrombocytopenia could increase the risk of severe, in-hospital mortality or bleeding complications during hospitalization of COVID-19 and, thus, should serve as an indicator of deterioration during hospitalization [
24‐
26]. Recently, data suggested that coagulation disorder caused by COVID-19 may be different from common infection-induced DIC. Increasing in circulating biomarkers may directly bind to platelet receptors, followed by platelet hyperactivation and aggregation, during such hyperactivation, platelet count is lower. Hyperresponsive platelets could contribute to the cytokine storm, while platelets were excessively consumed in severe COVID-19 patients due to the activation of coagulation pathway by cytokine storm, resulting in microcirculatory coagulation disorders and forming a vicious circle [
27,
28]. Therefore, we suspected that inflammation levels caused by infection leads to slightly activation of platelets during early-stage COVID-19, thrombocytopenia representing derangement of platelet function may be associated with hematopoietic inhibition, pulmonary damage, secondary infections and increased consumption of megakaryocytes and platelet during later-stage of the progression of the disease, reflecting conditions that are more prone to progression [
21,
27]. But for moderate patients, the clinical value of lower platelet count predicting shorter hospital stay needs to be explored by further studies.
Furthermore, prothrombin time could be used for early diagnoses of DIC. Compared with survivors, non-survivors had longer prothrombin time [
29,
30]. Increasing prothrombin time has been found to be significantly correlation with disease progression of COVID-19 [
31]. Prolonged prothrombin time may indicate excessive consumption of coagulating factors. In our study, prolonged prothrombin time has been identified as a risk factor to predict disease progression. Our findings confirmed that blood coagulation dysfunction may play a central role in the deterioration of the disease, and suggested that patients with the above coagulation-related indexes should be closely monitored [
32].
Several studies revealed the differences of baseline leucocyte count among patients with different clinical types of COVID-19. Compared with survivors, non-survivors had more significantly increased leucocyte count [
14,
30], which may be driven by elevated neutrophils. Li et al. found that higher neutrophil and lower lymphocyte count could predict in-hospital mortality for COVID-19 patients [
25]. NLR is an effective index reflecting the imbalance between neutrophil count and lymphocyte count, which is related to multiple organ injury. Elevated NLR may indicate the immunologic abnormality, and was related with severity of COVID-19 and in-hospital death [
33]. Furthermore, Yang et al. identified NLR as discriminator to improve prediction for poor clinical outcome in COVID-19 patients [
34]. Our findings were consistent with these lines of evidence, which suggested that NLR could be an important early prediction marker for disease progression in severely ill patients.
SARS-CoV-2 may also invade renal tubular epithelial cells though angiotensin converting enzyme II (ACE2) receptor, which expressed not only in respiratory organs but also in the kidney [
35]. Therefore, researchers began to be concerned about the renal function of COVID-19 patients. Creatinine, as a commonly used clinical index, reflects the state of renal function. One meta-analysis showed the significant association of elevated creatinine with severe or fatal patients [
14]. Cheng et al. reported that higher serum creatinine was a risk factor of in-hospital mortality of COVID-19 patients [
36]. Patients with elevated plasma creatinine are more likely to be admitted to ICU and develop acute renal injury, which was strongly related to increased mortality [
36,
37]. Our finding, consistent with the previous results, demonstrated that higher creatinine was involved in the disease progression.
Cardiovascular diseases may be a significant determinant of disease progression among COVID-19 patients. Some studies suggest that screening for acute coronary syndrome may be underestimated in the context of COVID-19 outbreak. Besides, unstable hemodynamic and pro-inflammatory state caused by acute respiratory failure of COVID-19 may promote the occurrence of acute coronary syndrome and lead to a poor prognosis of patients [
38]. Li et al. reported that there was a significant positive association between cardiovascular disease and in-hospital mortality of COVID-19. However, this result was obtained from an unadjusted meta-analysis [
39]. In our study, we found that severely ill patients with underlying cardiovascular comorbidities were more likely to suffer a poor prognosis. However, Cardiovascular disease did not as an independent risk factor for prognosis in multivariate analysis. Its role may have been offset by some other biomarkers. Besides, studies have shown the role of cardiac troponin in worsening clinical outcomes of patients [
40,
41]. However, our research failed to collect the results of cardiac troponin test, and more reliable data are needed to warrant the relationship between it and the prognosis of COVID-19. Several studies also suggested that features derived from CT scoring were predictors for prognosis of COVID-19 [
4]. However, radiological features were not included in our final prediction models. Considering the reason, it may be that patients with moderate, severe and critical type were separated into two groups, and the radiological characteristics of patients with the same clinical type may not be significantly different in our study. Besides, this study was based on a single center with limited population representation.
Currently, prediction models for COVID-19 related mortality have been published [
20,
42]. Several studies have focused on identifying risk factors related to the progression to severe or critical disease in patients with COVID-19, using a nomogram to predict the risk of disease progression visually [
13,
43,
44]. However, except for the small sample size, Wynants et al. pointed out that most of these models exclude patients who are still in hospital by the end of the study and had high risk of over-fitting [
4]. Furthermore, most of published studies have not concerned the differences in clinical endpoints between moderate and severely ill patients. Thus, considering the balance between practicality and accuracy, we constructed simplified prediction scores for risk stratification of the hospital stay and disease progression for patients with different clinical types, respectively. Scholars have constructed a score named MuLBSTA for the poor prognosis of viral pneumonia, which was in line with COVID-19 patients [
30]. But as shown by the results of comparison of AUC, sensitivity, specificity, NRI and IDI, MuLBSTA scores are not as good as our TRPNCLP score. The prediction scores we constructed performed well with good discrimination and calibration. In addition, according to the score, patients can be divides into low-, medium- and high-risk groups to guide the clinical decision. For example, a 67-year old severely ill patient with maximum body temperature of 37.4 ℃, respiratory disease, platelet count level of 164 x10
9/L, NLR of 20.6, creatinine of 56.40 umol/L, LDH of 361.90 U/L, and PT of 15.81 s. According to TRPNCLP score, this case receives a total of 12 points (4 points for temperature, 2 for respiratory disease, 0 for platelet count, 3 for NLR levels, 0 for creatinine, 1 for LDH, and 2 for PT), and would be predicted to have a high-risk of progression. The risk stratification scores constructed in this study might help clinicians classify patients accurately in the face of limited health resources and improve the survival rate of severe and critical patients.
To the best of our knowledge, this study is the largest cohort study on subgroup patients with moderate and severe COVID-19. However, the current study has several limitations. Firstly, it is a single-center study. Although the discrimination and calibration of prediction models were internally verified by a bootstrap method, the models are needed to be verified in independent external populations. Secondly, the roles of some biomarkers (such as IL-6, cTn and procalcitonin) may be ignored or underestimated in the predicting models because data were extracted from a real-world clinical patient cohort, and not all laboratory tests were done in all patients.