This meta-analysis is reported following with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement and has been registered at the International Prospective Register of Systematic Reviews (numberCRD42020196659) (12).

We search PubMed/MEDLINE, Scopus, Web of Science, Cochrane Central Register of Controlled Trials, Embase, clinical trials, China National Knowledge Infrastructure (CNKI) and Google Scholar databases for articles published until 1 August 2022, using the keywords. “coronavirus,” “2019-nCoV,” “COVID-19,” “SARS-CoV-2,” “2019 novel coronavirus disease,” “coronavirus disease 2019,” “laboratory,” “clinical characteristics.” In addition, the WHO publication database, lancet, New England Journal of Medicine, JAMA, BMJ were screened to find potentially relevant publications. To ensure the comprehensiveness and accuracy of the research, we also consulted the references of the attached literature.

Study designs eligible for inclusion were cross sectional, case-control, and cohort studies. To be eligible, the studies must clearly indicate that the patients have been diagnosed as COVID-19 and were positive for SARS-CoV-2RNA. Studies were excluded if patients were asymptomatic carriers and did not fulfill the inclusion criteria. Studies were considered eligible for inclusion in this meta-analysis if they compared laboratory parameters between patients with severe or non-serious diseases or between survivors and non-survivors. Since the pre-prints have not been peer-reviewed, we did not include these papers in our analysis to avoid the spread of any potential misinformation. The following studies were excluded: duplicate publications, reviews, editorials, single case reports, small case series (< 10 cases), studies did not include the biomarkers required for the meta-analysis or missing research data. Two authors independently screened the title and abstract based on these selection criteria, discussed the differences with another author and subsequently resolved through negotiation.

Two authors (HYZ, HJW) independently extracted the data, differences were resolved in a discussion, or, if a consensus could not be reached, the third author resolved (WFS). We extracted the following variables: first author; publication year; study design; country; the number of participants in severe and non-severe disease groups; levels of laboratory indexes (d-dimer levels, lymphocyte subsets, cytokines) in different groups. If further information is needed, we will email the corresponding author, if there is no response, we will exclude the study. Stratified data or interquartile range (IQR) were converted to mean (±SD) using mathematical formulas for meta-analysis (13, 14). We planned to use the Cochrane Risk of Bias tool to assess the risk of bias in randomized trials, but our search did not find any eligible randomized trials. We used the Newcastle-Ottawa scale to rate the risk of bias in the non-randomized study. There are 8 criteria for NOS, with scores ranging from 0 (high-risk deviation) to 9 (low-risk deviation). Studies with a NOS score> 7 were considered high- quality.

We used Stata (version 12.0) for all statistical analysis. We used the I² statistic and Cochran's Q-test to assess statistical heterogeneity. We believe that an I² statistics of 0–25% shows a low heterogeneity, the medium heterogeneity is 26–75%, and the high heterogeneity is 76–100%, the heterogeneity p-value is < 0.05. If there is heterogeneity, the random effect model was used, otherwise, the fixed-effect model was used. Weighted mean difference (WMD) with 95% confidence intervals (95% CI) was calculated for D-dimer, lymphocyte subsets, and cytokines. We studied the influence of a single study on the overall risk estimate and eliminated each study in each round to test the robustness of the main results. If a meta- analysis included more than ten studies, publication bias was assessed by Begg and Egger test. We defined significant publication bias as a p-value < 0.05.The trim-and-fill computation was used to estimate the effect of publication bias on the interpretation of the results.

We identified 5561 studies, of which 32 (data from 3337 participants) were included in our analysis (10, 11, 15–44) (Figure 1). The basic characteristics of the articles included in the study are shown in Table 1. Five of the 32 trials were published in 2022, four trials in 2021, and 23 trials were published in 2020. Two studies were from Italy, two studies were from Turkey, one study was from Germany, one study was from the United States, one study was from Spain, 1 study was from India, one study was from the Republic of North Macedonia, and the rest were from China. Twenty-eight studies were grouped by severity and non-severity, and four studies were grouped by survivors and non-survivors of COVID-19. Among these articles, 21 articles described the D-dimer levels of COVID-19 patients, 28 articles described the lymphocyte subsets levels of COVID-19 patients, 17 articles described the cytokine levels of COVID-19 patients. All the included studies were rated high quality according to the NOS scores and details are shown in Table 2.

Twenty-one articles showed that compared with patients in non-severe group, patients in severe group had higher levels of D-dimer (WMD = 1.217 mg/L, 95%CI = [0.788, 1.646], P < 0.001) (Figure 2). A sensitivity analysis was carried out by excluding one study at a time and reanalyzing the entire data set. We found that the results were not influenced by excluding any one specific study (Supplementary material 1A). A funnel plot based on the d-dimer levels of patients showed p-values of 0.001 in Egger's test (Supplementary material 1B). There was publication bias in the studies included in the meta-analysis. After filling one trial, the revised result was still consistent using random model (WMD = 0.306 mg/L, 95%CI = [−0. 167, 0.778], P = 0.205) or fix model (WMD = 0.127 mg/L, 95%CI = [0.058, 0.195], P < 0.001) (Supplementary material 1C). Besides, using standard mean difference (SMD) for the meta-analysis still did not change the conclusion (Table 3).

A random-effects model was used to analyze the correlation between the level of patient lymphocyte subsets and the severity of COVID-19. Compared with the non-severe group, patients in the severe group had lower levels for lymphocytes (WMD = −0.468( × 10⁹/L), 95%CI = [−0.543, −0.394], P < 0.001) (Figure 3), T cells (WMD = −446.746 (/μL), 95%CI = [−619.607, −273.885], P < 0.001), B cells(WMD = −60.616(/μL), 95%CI = [−96.452, −24.780], P < 0.001), NK cells (WMD = −68.297(/μL), 95%CI = [−90.600, −45.994], P < 0.001), CD3+T cells (WMD = −487.870(/μL), 95%CI = [−627.248, −348.492], P < 0.001), CD4+T cells (WMD = −290.134(/μL), 95%CI = [−370.834, −209.435], P < 0.001), CD8+T cells (WMD = −188.781(/μL), 95%CI = [−227.806, −149.757], P < 0.001) (Supplementary materials 2A–F). Furthermore, the neutrophil-to-lymphocyte ratio (NLR) (WMD = 6.939, 95%CI = [4.581, 9.297], P < 0.001) (Figure 4) of severe patients was higher than that of non-severe patients. Sensitivity analyses indicated that the results were robustness by removing any one specific study in all the lymphocyte subsets and NLR between non-severe and severe groups (Supplementary materials 2G–N). No significant publication bias was detected in most of the studies except for NK cells (p = 0.034) and neutrophil-to-lymphocyte ratio (NLR) (p = 0.016) (Supplementary materials 2O–T). When applying the trim-and-fill method, there were no any trials trimmed and filled in NK cells and neutrophil-to-lymphocyte ratio(NLR) (Supplementary materials 2U,V). Besides, using standard mean difference (SMD) for the meta-analysis still did not change the conclusion (Table 3).

The random-effect results demonstrated that compared with patients in the non-severe group, patients in the severe group had higher levels for interleukin (IL)-2 (WMD = 0.371 pg/ml, 95%CI = [−0.190,0.932], P = 0.004), IL-4 (WMD = 0.139 pg/ml, 95%CI = [0.060, 0.219], P = 0.717) (Supplementary materials 3A,B), IL-6 (WMD = 44.251 pg/ml, 95%CI = [27.010, 61.493], P < 0.001) (Figure 5), IL-10 (WMD = 3.718 pg/ml, 95%CI = [2.648, 4.788], P < 0.001) (Supplementary material 3C). Sensitivity analysis by removing one study in each turn, the result indicated that the main result was robustness (Supplementary materials 3D–G). A funnel plot based on the IL-6 levels of patients showed p-values of 0.031 in Egger's test (Supplementary material 3H). No significant publication bias was detected in IL-10 (Supplementary material 4I). When applying the trim-and-fill method, there were no any trials trimmed and filled in IL-6 (Supplementary material 3J). Besides, using standard mean difference (SMD) for the meta-analysis still did not change the conclusion (Table 3).