Laboratory Markers and Mortality in Novel Coronavirus Disease 2019 (COVID-19): A Systematic Review and Meta-Analysis

Background: Novel coronavirus disease 2019 (COVID-19) is associated with high mortality. Many laboratory values have been suggested to predict patients at risk for a poor outcome in COVID-19.The objective of this project was to systematically review and meta-analyze all laboratory markers associated with prognosis of mortality in patients with COVID-19. Methods: We searched OVID Medline, SCOPUS, MedRxIv, preprints.org, and Centers for Disease Control databases from November 2019 to April 10, 2020 for articles on laboratory values and mortality in COVID-19 and updated the search July 20, 2020. Teams of 2 independent reviewers reviewed titles and abstracts for studies that reported mortality and laboratory values and subsequently abstracted relevant data. Results: Our initial search identified 6,973 articles and a total of 96 articles (30 articles from first search and 66 from updated search) on 72 laboratory values were included. Many laboratory values were associated with mortality, but those most associated with mortality included lymphopenia (OR 0.30, 95% CI 0.24-0.36), thrombocytopenia (OR 0.46, 95% CI 0.35-0.60), elevated lactate dehydrogenase (OR 7.32, 95% CI 5.19-10.33), and ferritinemia (OR 5.19, 95% CI 3.07-8.62). All cardiac markers were associated with mortality, with troponin being the least associated. A low PaO2:FiO2 ratio was also associated with mortality (OR 0.13, 95% CI -0.06-0.28). Heterogeneity was high and risk of bias was moderate. Conclusions: This meta-analysis identified many laboratory abnormalities associated with mortality in COVID-19, though was limited by heterogeneity. Laboratory markers previously identified as associated with a poor prognosis in COVID-19 were confirmed to be those most associated with mortality in this large meta-analysis.


Introduction
Novel coronavirus disease 2019 (COVID-19) has caused over a million international deaths and overwhelmed healthcare systems internationally: Wuhan, China; Lombardy, Italy; and New York City, New York. This disease has led to overall mortality rates from 1-6%, [1][2][3] with the subset of patients who require critical care or intubation having mortality as high as 50-80% [4,5].
The vast majority of patients will be asymptomatic or minimally symptomatic [6] and can be treated at home. However, a small percentage of patients will require hospitalization for aggressive supportive care -supplemental oxygen, intubation, and sometimes cardiac support [7]. Those at most risk 10 participants, or studies with not enough data to address the research question. We did include preprint articles that are not yet peer reviewed. We planned to contact authors for updated data if data was incomplete. After full text review for inclusion, all authors independently identified articles for exclusion and any discrepancies were resolved with a consensus meeting among all 4 authors (JD, CT, SG, GH).

Data extraction and data synthesis
After title and abstract selection, a standardized abstraction form was developed by all of the authors. Two of three authors (CT, SG, GH) abstracted data from the studies and compared their results for discrepancies. These were resolved by consensus among all authors. Data included authors, publication year, population, setting, intervention, mortality, and other outcomes.

Risk of bias
Risk of bias within a study was assessed using the Quality in Prognostic Studies (QUIPS) tool. This was assessed by one author (JD) and verified by one other un-blinded author. Funnel plot analyses for each laboratory value was conducted for publication bias and heterogeneity was assessed using I 2 .

Statistical analysis
Statistical analysis was conducted in RevMansoftware, version 5 (Review Manager (RevMan) [Computer program]. Version 5.4, The Cochrane Collaboration, 2020). A random effects model and inverse variance weighting was used in all analyses. All laboratory values with more than 2 studies were included in the meta-analysis. All of these were included in a standardized mean difference analysis. Studies who reported averages (either mean or median) were also included in an absolute mean difference analysis for laboratory values that were conducted on the same scale. Laboratory values were transformed to International System units when able, but laboratory values on different scales (e.g., C-reactive protein), where different assays may affect results (e.g., D-Dimer), or similar lab tests with different implications and measures (troponin I and troponin T) were not combined for absolute mean difference analysis and only included in a standardized mean difference analysis. Authors of studies that did not report averages were contacted to request this data. Studies that reported median data were transformed using the method recommended by Cochrane [18] and modified for low sample sizes when applicable using the method described by Wan, et al. [19] or Hozo, et al. [20] as applicable based on available data. Data for odds ratios (ORs), hazard ratios, and relative risk were transformed to standardized mean differences and associated standard errors using the method described by Chinn, et al. [21]. Unadjusted/unweighted ratios were preferred when available. In studies with a group with no patients, a dummy variable of 0.5 was used in calculation of odds ratio/standardized mean difference, and in studies with a group with no variance, the average of the standard deviations from all other studies on that laboratory test or the average value itself was used, whichever was lower. If there was no difference and no variance reported, a standard error of 0.01 was used. Studies that reported multiple terioration or mortality would be helpful to clinicians when making disposition or treatment decisions.
Several laboratory values have been proposed to predict severe illness. These include nonspecific inflammatory markers (C-reactive protein and ferritin), [10] liver enzymes (AST, ALT, LDH), [11] and blood cell counts (platelets, white blood cells, and lymphocytes) [10]. D-Dimer has been highly associated with mortality, with the newest theories recognizing that this may be due to risk of thrombotic complications, and not just inflammation alone [12,13]. As with many other diseases, troponin elevations, too, have been associated with mortality [14,15].
Given the reports of associations of these laboratory values and prognosis, we sought to do a systematic review and meta-analysis of laboratory values associated with mortality in adult patients with COVID-19.

Methods
This study was conducted in accordance with Preferred Reporting in Systematic Review and Meta-Analysis Guidelines (PRISMA) [16]. The PRISMA Checklist is available in Supplement 1 and the research protocol is available from the authors upon request. This study had no human subjects and was, therefore, exempt from Institutional Review Board Review.

Search strategy and study selection
We searched OVID Medline, SCOPUS, MedRxIV, preprints. org, and Centers for Disease Control (CDC) databases from November 2019 to April 10, 2020. The search used keyword terms for "COVID-19", "SARS", or "coronavirus". The CDC database was already compiled and was only filtered for English articles. Per PRISMA guidelines, 1 sample search strategy, in detail, is available in the Supplement 2. A complete search strategy is available from the authors upon reasonable request. The search was updated on July 20, 2020 with a focused search in PubMed and MedRxIV, using the laboratory values found in the initial search. These 2 sources were chosen because they could conduct a focused search that was also date limited to the month level and included one published and one unpublished source. An example of the updated search strategy is also available in Supplement 2. We reviewed citations from included articles, review articles, and suggestions from select content experts to find relevant articles that may have been missed. We also contacted authors of studies that appeared to have underlying data that would answer the research question to ask for additional data. Titles and abstracts from these search methods were screened independently by 3 trained reviewers (CT, SG, GH). If any reviewer thought an article was potentially relevant, a full text copy of the article was ordered and again reviewed for inclusion by these 3 authors.
Inclusion criteria were any research study type that evaluated human adult patients with COVID-19 and any laboratory value associated with mortality. Exclusion criteria included opinions (i.e. editorials, letters without data, etc.), abstracts, non-English papers, studies only on children, studies with <   different patient cohorts were included in separate analyses, but studies that had one cohort but separated "control" outcomes (i.e. severe vs. non-severe illness) were combined into one control group (survived). Studies that reported different analyses of the same outcome were only included once, with the preference for average (mean/median) data. Preprint articles that were subsequently published were only included once. Statistical significance was set at an alpha of 0.001 to account for multiple analyses.

Results
Our initial search identified 6,973 unique articles. 147 of these were identified for full text review, and 30 were deemed to meet all inclusion and no exclusion criteria ( Figure  1). In the updated search, 61 new articles were identified (896 from medRxIV and 570 from PubMed), and 52 were included (18 from medRxIV and 34 from PubMed). Two studies in the initial search had since been published from a preprint form and these published versions were used. A total of 105 authors were contacted (90 for primary data and 15 for clarification of printed data). Thirteen of these authors responded (12.9%) and 10 (9.9%) of those provided new data. A total of 96 unique studies [11,12,14,15, with 30,985 patients and 72 different laboratory values were included. A summary of the characteristics of the included studies is in Table 1 and details of each study are available in Supplement 3.
The summary of the standardized mean difference analysis is available in Table 2. A summary of the results for the absolute mean difference analysis is available in Table 3. A forest plot and funnel plot for each laboratory value is available in Supplement 4, Supplement 5 and Supplement 6.

Hematologic
The most predictive complete blood cell count test was neutrophil percent (OR 26.47, 95% CI 5.68-121.09), though this was limited by a small number of studies and a wide confidence interval. Nonetheless, absolute neutrophil count (OR 4.66, 95% CI 3.30-6.45) and neutrophil to lymphocyte ratio (OR 3.36, 95% CI 1.60-6.57) were also the next most significant positive associations with mortality, and a low absolute lymphocyte count (OR 0.30, 95% CI 0.24-0.36) and lymphocyte percent (OR 0.05, 95% CI 0.02-0.11) were the most significant negative correlations, as was thrombocytopenia (OR 0.46, 95% CI 0.35-0.60). There was a moderate association with leukocytosis and mortality, as well (OR 3.68, 95% CI 2.91-4.74).

Risk of bias
The risk of bias within studies was low in 27% (n = 26), moderate in 63% (n = 60), and high in 11% (n = 10) of studies. The most common reason for high risk of bias was selective patient population. The most common reason for moderate risk of bias was either not accounting for patients still hospitalized or unclear reporting of the timing of either the laboratory value or the outcome (mortality). Overall heterogeneity was moderate to high with 58% (38/65) of laboratory values in the absolute mean difference analysis and 60% (43/72) in the standardized mean difference with I 2 values > 80%. Funnel plot analyses showed most values with a low risk of publication bias.

Discussion
This systematic review and meta-analysis identified many laboratory abnormalities associated with mortality in COVID-19, but was limited by heterogeneity and risk of bias. Laboratory tests that have been previously identified to have prognostic and diagnostic accuracy seem to be the most associated with mortality. This includes lymphopenia, thrombocytopenia, leukocytosis, elevated liver enzymes, elevated LDH, elevated cardiac enzymes, and elevated ferritin [10- 15,114].
In light of the pandemic nature of COVID-19, the ability to accurately identify patients at risk for poor outcome is paramount. Most patients with COVID-19 do not die from it. Given that this disease has placed a strain on healthcare resources in many locations and will continue to potentially do so, identification of patients who need some of these precious resources, like hospitalization or ICU care, will help in fair al-with mortality, except there was no significant association between CD19+/B cells and CD4+ to CD8+ ratio.

Coagulation
D-Dimer is the most significant coagulation parameter associated with mortality (OR 3.30, 95% CI 2.76-3.96), followed by prothrombin time/International Normalized Ratio (PT/ INR, OR 2.86, 95% CI 2.10-3.89). Because these parameters are laboratory specific, an absolute mean difference was not calculated for them.
An absolute mean difference of CRP was not calculated due to its dependence on lab variability. The absolute mean difference in interleukin 2 receptor was most apparent (506.43 U/mL, 95% CI 392.25-620.61).

Limitations
Our meta-analysis has some limitations. There is always a risk of missing potentially germane articles, but we attempted to combat this by using a pre-specified protocol, searching multiple databases, using two independent reviewers, and searching other systematic reviews for references. There was also a very high rate of no response to email inquiries of authors for data, which could introduce bias. All of the included studies were observational, and there was, in general, a moderate risk of bias within the studies. Our process of assessing quality with two authors was not blinded, so there, again, could be bias in the assessment of quality. There also was a significant amount of heterogeneity, which can be expected when such a large number of studies is analyzed but also raises questions as to whether there are subsets of patients in whom specific laboratory values might be more or less accurate.

Conclusion
Our meta-analysis identified multiple laboratory values associated with mortality in COVID-19, but was limited by heterogeneity. Many of these have been described previously: Lymphopenia, thrombocytopenia, leukocytosis, elevated D-Dimer, ferritinemia, elevated LDH, elevated liver enzymes, and cardiac injury. Clinicians can use this information to help predict which patients are most at risk for adverse outcomes with COVID-19 and how best to allocate limited healthcare resources.

Financial Support
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Conflict of Interest
To the best of our knowledge, the authors have no conflicts of interest, financial or other, to declare regarding this manuscript.
location. When effective targeted treatments are identified, these prognostic values may also help identify patients who are most likely to benefit from these therapies.
In regards to specific laboratory values, the combination of neutrophilia and lymphopenia has been combined into a relatively new value called the neutrophil to lymphocyte ratio (NLR). This has been studied in several infectious and inflammatory processes, like bacteremia, [115] pancreatitis, [116,117] and pulmonary embolism [118], and has shown some diagnostic and prognostic value. It appears to be significantly associated with mortality in our meta-analysis.
All cardiac biomarkers studied were significantly associated with mortality. This is not surprising, as the risk of cardiac involvement in COVID-19 has been well studied and has been previously associated with mortality [119]. COVID-19, too, has been reported to be associated with fatal myocarditis [120]. Interestingly, troponin, which is one of the more common cardiac biomarkers evaluated, is one of the least significant markers compared to other values like myoglobin and alpha hydroxybutyrate dehydrogenase.
Only the P/F ratio was significant when examining blood gas levels. One of the hallmarks of COVID-19 is a ARDS-like illness. P/F has been well studied in ARDS and has been shown to be associated with mortality, so its association with mortality in COVID-19 should be no surprise. More interesting, though, pH and pO 2 have no significance and lactic acid, interestingly, has only minimal significance.
When examining liver enzymes, AST, ALT, ALP, and GGT all have significant association with mortality. However, AST appears to be much more associated than ALT. This, in addition to the GGT elevation, has been described in the prior SARS outbreak [121], but is not typical of other viral hepatitis entities [122], like hepatitis B or C. It is more commonly seen in ischemic or congestive hepatopathy [123]. Therefore, direct viral injury may not be the only pathophysiology of liver injury in COVID-19 [124,125].
Nearly all markers of inflammation appear to be significantly associated with mortality in COVID-19, including tumor necrosis factor alpha and several interleukin levels. CRP appears to be the most significant inflammatory marker that is commonly available in clinical practice. Several other markers that have significant association with inflammation are also elevated, like D-Dimer, ferritin, and LDH. The pathophysiology of these values is poorly understood. D-Dimer has been hypothesized to be elevated not only due to inflammation but also due to a hypercoagulable state. However, recent research has shown that COVID-19 may not, in fact, be a hypercoagulable state [126] despite initial anecdotal reports and observations of this [127].
Our meta-analysis involved a very large number of studies so even small differences, which may not be clinically significant, were rendered statistically significant. For example, the absolute difference in white blood cell count associated with mortality was 2.38, which would not cause alarm to most clinicians. Also, several laboratory values, although significant, are not routinely available in clinical practice (e.g. lymphocyte subsets or interleukin values).