Clinical laboratory parameters associated with severe or critical novel coronavirus disease 2019 (COVID-19): A systematic review and meta-analysis

Background To date, several clinical laboratory parameters associated with Coronavirus disease 2019 (COVID-19) severity have been reported. However, these parameters have not been observed consistently across studies. The aim of this review was to assess clinical laboratory parameters which may serve as markers or predictors of severe or critical COVID-19. Methods and findings We conducted a systematic search of MEDLINE, Embase, Web of Science, CINAHL and Google Scholar databases from 2019 through April 18, 2020, and reviewed bibliographies of eligible studies, relevant systematic reviews, and the medRxiv pre-print server. We included hospital-based observational studies reporting clinical laboratory parameters of confirmed cases of COVID-19 and excluded studies having large proportions (>10%) of children and pregnant women. Two authors independently carried out screening of articles, data extraction and quality assessment. Meta-analyses were done using random effects model. Meta-median difference (MMD) and 95% confidence interval (CI) was calculated for each laboratory parameter. Forty-five studies in 6 countries were included. Compared to non-severe COVID-19 cases, severe or critical COVID-19 was characterised by higher neutrophil count (MMD: 1.23 [95% CI: 0.58 to 1.88] ×109 cells/L), and lower lymphocyte, CD4 and CD8 T cell counts with MMD (95% CI) of -0.39 (-0.47, -0.31) ×109 cells/L, -204.9 (-302.6, -107.1) cells/μl and -123.6 (-170.6, -76.6) cells/μl, respectively. Other notable results were observed for C-reactive protein (MMD: 36.97 [95% CI: 27.58, 46.35] mg/L), interleukin-6 (MMD: 17.37 [95% CI: 4.74, 30.00] pg/ml), Troponin I (MMD: 0.01 [0.00, 0.02] ng/ml), and D-dimer (MMD: 0.65 [0.45, 0.85] mg/ml). Conclusions Relative to non-severe COVID-19, severe or critical COVID-19 is characterised by increased markers of innate immune response, decreased markers of adaptive immune response, and increased markers of tissue damage and major organ failure. These markers could be used to recognise severe or critical disease and to monitor clinical course of COVID-19.

a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 manual search of the medRxiv pre-print server to identify latest relevant studies that might still be undergoing peer-review. The search was limited to the years 2019-2020 and there was no limitation regarding language of publication.

Study selection
Following deduplication of records retrieved during the systematic search, we exported retained articles into Covidence review manager to facilitate the screening of titles and abstracts, which was followed by a full text review to determine eligibility [23].
Two authors (JM and PP) independently carried out title and abstract screening and full text evaluation of all articles using the eligibility criteria listed in the previous section. The discrepancies in study selection were resolved through adjudication by a third author (KM). To avoid including data on the same patient populations more than once in the meta-analysis, we matched studies based on the location of the study (hospital, town) and the period over which data was collected. For two or more studies conducted at the same location over the same or overlapping periods, we included only the largest study, unless one of the smaller studies presented relevant information not included in the larger study.

Data extraction and data items
Two authors (JM and PP) independently extracted, verified and summarized data from each study included in the meta-analysis. The information extracted from the selected studies included: study author(s), study sponsors, date of publication, study period, study location, study design, sample size, sample characteristics (age, gender, comorbidities), exposure characteristics (study definition of severity of COVID-19, timing of classification of disease severity [on admission or otherwise], number of cases with non-severe COVID-19, number of cases with severe or critical COVID- 19), timing of blood sample collection (on admission or otherwise), clinical laboratory parameters stratified by COVID-19 severity, mean (standard deviation [SD]) and/or median (interquartile range [IQR] or minimum-maximum [total] range) of clinical laboratory parameters when reported on continuous scales, and numbers (percentages) of cases above and below cut-off values when reported on categorical scales. Discrepancies in collected data were resolved by re-checking the primary studies until consensus was reached. For the studies which had unclear severity classification, the authors were contacted to seek additional clarification. Studies in the Chinese language were translated into English language by a Chinese native speaker. The extracted data were exported into R programming software.

Summary measures and data synthesis
Where clinical laboratory parameters were measured on a continuous scale, we pooled median differences from each study using the quantile estimation method [26]. The result of this analysis was expressed as a meta-median difference (MMD) accompanied by a corresponding 95% confidence interval (CI). We preferred median differences over mean differences because clinical laboratory indices are usually skewed, and mean values could be influenced by outlier values, particularly in small samples. We performed a sensitivity analysis by pooling mean differences from each study using inverse variance weighting. Where the studies reported only median (IQR or total range) values, we computed mean (SD) using methods previously described [27,28].
Where clinical laboratory parameters were measured on a categorical scale, we computed prevalence ratios for each study using counts of events in the exposure and comparator group and calculated meta-prevalence ratios (MPR) and the 95% CIs using the Mantel-Haenszel method.
Meta-analysis was conducted using random effect models. We assessed clinical heterogeneity (age distribution, comorbidities criteria of severity) and study methodological heterogeneity (timing of blood sample collection) and considered the potential impact of these factors on the meta-analysis results. We assessed statistical heterogeneity using Cochran's Q test and calculated the I 2 statistic, which was interpreted using cut-offs of 25%, 50%, and 75% for low, moderate, and substantial heterogeneity, respectively. We performed influence analysis using the Leave-One-Out-method to identify studies that have a high influence on our results [25]. Additional sensitivity analyses were performed by excluding 'outlier' studies. A study-specific estimate was considered an outlier if its confidence interval did not overlap with the confidence interval of the meta-estimate [25].
To detect possible publication bias, funnel plots were constructed for the 4 laboratory parameters with the highest number of individual studies. Egger's test was carried out to assess statistical symmetry of the plots.

Study selection
We identified 3,779 studies through database searching and from other sources (Fig 1). After removing duplicates, 1722 unique records were screened, and of those, 1398 were removed after title and abstract review. Additional 257 records were excluded due to lack of COVID-19 severity classification, lack of laboratory parameter records or ineligible study design. Of the 67 remaining studies, another 22 were excluded because they used data from the same locations or covered overlapping periods (see S2 Text in S1 File). A total of 45 studies were retained for meta-analyses.

Study characteristics
The characteristics of included studies are detailed in Table 1 and S3 Table in S1 File. All studies included in the meta-analyses were observational and hospital based. The majority of 45 studies (87%) were from China; and of those, 14 were from Wuhan and 25 from other locations in China. Two studies were from the USA, and the remaining 4 studies were from France, Germany, Japan and Singapore. All studies were published in 2020 and the data collection covered the period from December 25 th , 2019 to April 2 nd , 2020. The median population size of the included studies was 97 (IQR: 49-221). Data were collected retrospectively in all but one study [30]. COVID-19 severity was classified using China National Health Commission guidelines (20 studies), WHO guidelines (4 studies), American Thoracic Society guidelines (2 studies), Berlin criteria (1 study), Complementary and Natural Healthcare council (1 study), or unspecified guidelines (17 studies). Studies classified severity on admission (22 studies, 49%), on the ward (11 studies, 24%), or during unspecified periods (12 studies, 27%). Clinical laboratory tests were done on admission (33 studies, 73%), post-admission (5 studies, 11%) and at unspecified periods (7 studies, 16%). The highest number of laboratory parameters  Qu et al. [64] Hospital (  reported in a single study was 30 [31] and the lowest number of laboratory parameters reported in a single study was 2 [32]. The median (or mean) age of patients in the included studies ranged from 35 years to 67 years, and the proportion of male patients ranged from 30% to 81%. Patients in the included studies had varying proportions of comorbidities such as hypertension, diabetes and cancer as detailed in Table 1 (and S3 Table in S1 File).

Synthesis of results
Results of meta-analyses are reported in Table 2, and Forest plots and Leave-One-Out analysis are displayed in S4 Fig in S1 File.
As pertains to haematological parameters, the majority of studies reported higher white cell count and higher neutrophil count in severe or critical COVID-19 patients relative to nonsevere COVID-19 patients. Median difference in individual studies ranged from -1.6 to 7.3 (×10 9 cells/L) for white cell count and from -1.0 to 5.2 (×10 9 cells/L) for neutrophil count. The MMD estimates (10 9 cells/L) were 0.87 (95% CI: 0.35 to 1.40; I 2 : 80.5%) for white cell count and 1.23 (95% CI 0.58 to 1.88; I 2 : 90%) for neutrophil count. When the results were expressed in terms of ratio measures, patients with severe or critical COVID-19 had significantly higher

PLOS ONE
Clinical laboratory parameters associated with severe or critical novel coronavirus disease 2019 (COVID-19)

Assessment of threats to validity
The threats to validity in this meta-analysis fall into two broad categories: risk of bias in individual studies, and publication bias across the body of literature. Assessments of these two categories of threat to validity are presented below.
Using the NIH study quality assessment tools, 28 studies (62.2%) were rated as having a low risk of bias, 14 studies (31.1%) were rated as having a medium risk of bias, and 3 studies (6.7%) were rated as having a high risk of bias. The majority of studies had a clearly defined study objective (97.8%), a well-defined study population (100%), and had comparable subjects (100%). In contrast, no study provided a sample size calculation or power description. All the studies were rated as having a high risk of bias for the element assessing a temporal sequence between the laboratory measure and disease severity, and none of the reported results was adjusted for potential confounding (Fig 2 and S5 Table in S1 File). Our results did not markedly differ in sensitivity analyses after excluding studies with a high risk of bias.
The symmetry of funnel plots obtained from the 4 laboratory parameters with the highest number of individual studies was assessed using Egger's test. The symmetrical funnel plots for C-reactive protein (p: 0.155) and creatinine (p: 0.415) suggested no evidence of publication bias whereas asymmetrical funnel plot for white cell count (p: 0.004) and lymphocyte count (p: 0.005) indicated significant influence of smaller studies, which may be indicative of publication bias (Fig 3). Important to note that Egger's test may not be robust for C-reactive protein, white cell count and lymphocyte parameters due to substantial heterogeneity (I 2 >75%).

Additional analysis
In sensitivity analyses excluding outlier studies, statistical heterogeneity was reduced, and the meta-estimate of most laboratory parameters were not markedly altered. In sensitivity analysis using mean differences (S6 Table in S1 File), there was substantial heterogeneity for most laboratory parameters, and the associations observed from using median differences persisted.

Discussion
COVID-19 is a rapidly evolving pandemic with significant global morbidity and mortality. The aim of this meta-analysis was to investigate which clinical laboratory parameters may be associated with severe or critical COVID-19 disease. Out of the 39 clinical laboratory parameters evaluated, we found that derangements in 36 clinical laboratory parameters were

PLOS ONE
Clinical laboratory parameters associated with severe or critical novel coronavirus disease 2019  significantly associated with severe or critical COVID-19. Whilst some of the observed associations may not be clinically relevant, certain, more pronounced laboratory abnormalities may have important clinical implications. Markers of an overactive innate immune system such as markedly elevated neutrophil-to-lymphocyte ratio (NLR), IL-6, serum ferritin and C-reactive protein, and markers of a deficient adaptive immune system such as lymphocytes and CD4 count could help recognise potential severe infections during triage, while markers of organ failure could be helpful in monitoring evolution of hospitalised COVID-19 patients.
Following infection with a virus, the innate immune system in activated. This early response is nonspecific and serves to limit virus multiplication during the acute phase [33]. The adaptive immune system is activated a few days later and is responsible for a more specific response, which is immunomodulatory (via engagement of helper T cells and regulatory T cells) and produces 'immunological memory' [33]. Elevated lymphocyte count is commonly found in most viral infections, and the magnitude and quality of T cell responses may determine the fate of these infections [34,35]. Failure to mount an appropriate adaptive immune response means the innate immune response remains continuously stimulated with deleterious effects on the lungs and other organs. We found that severe or critical COVID-19 patients had increased markers of innate immune system activity compared to patients with non-severe disease. This is evidenced in the significantly higher levels of neutrophils, IL-6, and acute phase reaction markers such as CRP, ESR and serum ferritin, as well as decreased concentrations of albumin and prealbumin. Severe or critical COVID-19 patients also exhibited defective adaptive immune response evidenced by significantly lower levels of lymphocytes and their subsets (CD3, CD4, CD8). CD4 count is currently being used to define severe cases of HIV infection [36]. In the case of HIV, the virus directly infects CD4 cells using the envelope

PLOS ONE
Clinical laboratory parameters associated with severe or critical novel coronavirus disease 2019 (COVID- 19) glycoprotein gp120. Various authors have suggested that SARS-CoV-2 could deplete lymphocytes directly by infecting T lymphocytes, or indirectly through lymphocyte apoptosis induced by persistent elevated inflammatory cytokines [15,37,38]. Since severe COVID-19 patients display reduced lymphocyte count, it is likely that the cytokine release syndrome observed in some patients with severe or critical COVID-19 is mediated by interferons, TNFs, and cytokines secreted by non-T cell leucocytes such as macrophages, neutrophils and NK cells which are all key elements of innate immunity to viruses [39].
These findings could be applied clinically to identify severe or critical COVID-19 patients. For example, routine monitoring of NLR may provide insight into the functioning of both the innate and adaptive immune responses and help predict the clinical course of COVID-19. Despite the 45 studies included in this review, only five reported results for NLR; all these five studies found a significant association between increased NLR and severe or critical COVID-19 disease.
We also found that patients with severe or critical COVID-19 had significantly higher biomarkers of tissue and organ damage such as LDH, liver enzymes, kidney function parameters and markers of myocardial function. These observed associations could be explained by 3 mechanisms. First, the virus may cause direct organ damage by attaching to the ACE2 receptors, which are commonly expressed in the lungs, heart, arteries, kidneys and intestines [40]. The second, more indirect mechanism is systemic hyperinflammation caused by the cytokine release syndrome mediated by the innate immune system [40]. Systemic hyperinflammation affects all organs and could also explain the significantly increased expression of markers of disseminated intravascular coagulation (high D-dimer and depleted platelet count) in severely or critically ill patients [41,42]. The third, also indirect, mechanism by which severe or critical COVID-19 causes multiple organ damage is hypoxia resulting from respiratory failure.
Once the mechanisms of COVID-19 induced organ damage are better understood, markers reflecting the pathophysiological changes caused by these mechanisms may find their way into clinical practice. Based on the results of our meta-analysis especially promising may be markers of immune function such as NLR, IL-6, C-reactive protein, serum ferritin, lymphocytes, CD4 count, and markers of coagulation and organ damage such as D-dimer, LDH, troponin I and liver enzymes.
The strengths and limitations of this review and meta-analysis need to be considered in the context of rapidly evolving literature. On the one hand, our study identified some associations that deserve further consideration and may lead to improvements in the risk stratification, monitoring and management of COVID-19 patients. On the other hand, it is important to emphasize that our analyses need to be viewed as hypothesis-generating rather than hypothesis-testing. Due to the large number of associations examined simultaneously there is a considerable likelihood of false-positive findings. This limitation can be addressed in future, more focused, studies that will take into consideration prior knowledge and reduce the likelihood of false-positive results through application of Bayesian and empirical-Bayes methods [43]. Our review is also affected by the limitations of the underlying literature. Of those, perhaps the most important is the cross-sectional nature of the analyses used in most publications. Although it is plausible that markers of immune function can be used to predict disease severity, the evidence would have been stronger if the studies were able to perform laboratory testing of COVID-19 patients before their disease severity was known. In addition, many studies from China reported on overlapping patient populations. While we tried to exclude studies that relied on the same data, it is possible that some of the associations examined in this metaanalysis were based on non-independent observations.
In conclusion, compared to non-severe COVID-19, severe or critical COVID-19 is associated with increased markers of innate immune response such as neutrophil count, NLR, IL-6, CRP and serum ferritin; decreased markers of adaptive immune response such as lymphocyte, CD4 and CD8 counts; and increased markers of tissue damage and major organ failure including D-dimer LDH, Troponin I, CK-MB, AST, ALT, urea, and creatinine. Based on the results of our meta-analysis, especially promising markers are NLR, IL-6, serum ferritin, lymphocyte and CD4 counts, D-dimer and troponin I. The clinical value of these markers should be explored further to assess the risk of severe or critical disease and to monitor the clinical course of COVID-19.