Predictors of mortality in patients with coronavirus disease 2019: a systematic review and meta-analysis

Background Coronavirus disease 2019 (COVID-19) is associated with a high mortality rate, especially in patients with severe illness. We conducted a systematic review and meta-analysis to assess the potential predictors of mortality in patients with COVID-19. Methods PubMed, EMBASE, the Cochrane Library, and three electronic Chinese databases were searched from December 1, 2019 to April 29, 2020. Eligible studies reporting potential predictors of mortality in patients with COVID-19 were identified. Unadjusted prognostic effect estimates were pooled using the random-effects model if data from at least two studies were available. Adjusted prognostic effect estimates were presented by qualitative analysis. Results Thirty-six observational studies were identified, of which 27 were included in the meta-analysis. A total of 106 potential risk factors were tested, and the following important predictors were associated with mortality: advanced age, male sex, current smoking status, preexisting comorbidities (especially chronic kidney, respiratory, and cardio-cerebrovascular diseases), symptoms of dyspnea, complications during hospitalization, corticosteroid therapy and a severe condition. Additionally, a series of abnormal laboratory biomarkers of hematologic parameters, hepatorenal function, inflammation, coagulation, and cardiovascular injury were also associated with fatal outcome. Conclusion We identified predictors of mortality in patients with COVID-19. These findings could help healthcare providers take appropriate measures and improve clinical outcomes in such patients. Supplementary Information The online version contains supplementary material available at 10.1186/s12879-021-06369-0.

items: study participation, study attrition, prognostic factor measurement, outcome measurement, study confounding, and statistical analysis and reporting [41].

Statistical analysis
The pooled unadjusted estimates for each predictor were performed if data from at least two studies were available. The results as risk ratios (RRs) for dichotomous data and weighted mean differences (WMDs) for continuous outcomes were presented, both with 95% con dence intervals (CIs). For studies that presented data as medians and interquartile ranges, we calculated the means and standard deviations based on the formulas by Wan et al. [42]. To provide conservative pooling estimates, we applied a random-effects model in all analyses.
Heterogeneity was assessed using the I 2 statistic, and a value of >50% was considered signi cant heterogeneity. In the presence of signi cant heterogeneity, sensitivity analyses were performed to assess the stability of the results by omitting the largest (or smallest) study. We also calculated the xed-effects model for additional sensitivity analyses. Publication bias was examined by Egger's test. All statistical analyses were performed using RevMan version 5.3 (the Cochrane Collaboration) and Stata 15.0 software (StataCorp, College Station, Texas, USA).
Pooled adjusted estimates for predictors were not available because only few original studies reported adjusted data, different types of effect measures were used (such as odds ratios and hazard ratios), and potential overlap of the patients existed between the included studies.
Therefore, we presented adjusted data using qualitative analysis.
The characteristics of the studies included in the systematic review are presented in Table 1. The sample size in each study ranged from 54 to 5688, and the mortality rate ranged from 3.1% to 61.5%. All the studies were from China, except for three studies conducted in the USA [22,24] and Italy [11]. The mean (or median) age of the patients ranged from 46 to 69 years, and the proportion of female patients ranged from 18.0% to 60.2%. The quality assessment of the included studies is presented in Table S2.

Meta-analyses of unadjusted estimates
We conducted meta-analyses of unadjusted estimates for 106 potential predictors of mortality in patients with COVID-19 (Table 2 and Figure  2). All individual forest plots and more details are presented in the Supplementary Table S2-S107 and Figure S1-S106. c Values are expressed as medians (interquartile ranges) or means ± standard deviations.
Compared with survivors, the mean age of the deceased patients was signi cantly higher (MD, 13 Symptoms of dyspnea (RR, 1.98; 95% CI, 1.70 to 2.30) were more common in the nonsurvivor group than in survivor group. No associations between the remaining clinical symptoms and mortality were observed. Antiviral agent and immunoglobulin therapies were not associated with mortality. However, patients who received glucocorticoids (RR, 1.79; 95% CI, 1.25 to 2.55) or antibiotics (RR, 1.20; 95% CI, 1.02 to 1.40) were more likely to die than those who did not.
The values of the following laboratory parameters were signi cantly higher in the deceased patients than in survivors: white blood cells  Oxygen treatment was applied more often in nonsurvivors than in survivors (high ow nasal cannula, RR, 10.6; 95% CI, 5.97 to 18.8; noninvasive ventilation, RR, 5.12; 95% CI, 3.98 to 6.57; invasive ventilation, RR, 29.3; 95% CI, 21.5 to 39.9). Renal replacement therapy was also observed to be more prevalent in the deceased patients than in the surviving patients (RR, 53.5; 95% CI, 22.4 to 127. 3). No correlation between the use of extracorporeal membrane oxygenation (ECMO) and mortality was observed.
A signi cant difference between nonsurvivors and survivors was also observed for the following variables: respiratory rate, heart rate, respiratory rate >30 (or 24) breaths per min, partial pressure of oxygen (PaO 2 ), partial pressure of carbon dioxide (PaCO 2 ), peripheral oxygen saturation (SpO 2 ), ratio of partial pressure of oxygen to fraction of inspired oxygen (PaO 2 /FiO 2 ), bilateral pneumonia, Acute Physiology and Chronic Health Evaluation II (APACHEII) score, and Sequential Organ Failure Assessment (SOFA) score.
Signi cant heterogeneity was observed in the analyses of 54 tested predictors. Sensitivity analyses did not change the conclusions about the most tested variables while the results were inconsistent in the following 14 tested predictors: preexisting malignancy, symptoms of anorexia, antiviral therapy, immunoglobulin therapy, CRP ≥10 mg/L, shock, acute liver injury, MON and PLT count, APTT, γ-glutamyl transpeptidase, PaO 2 , SpO 2 , and IL-6 (Supplementary Table S108-S109). No signi cant publication bias was observed for any risk factors except for two predictors, current smoking status and antibiotic therapy (Table 2 and Figure 2).

Discussion
To the best of our knowledge, this is the most comprehensive systematic review and meta-analysis evaluating predictors of mortality in patients with COVID-19. Important risk factors associated with an increased fatality rate included older age, male sex, current smoking, baseline comorbidities (especially chronic kidney, respiratory, and cardio-cerebrovascular diseases), symptoms of dyspnea, complications during hospitalization, corticosteroid use and a severe condition. Additionally, a series of abnormal biomarkers of hematologic parameters (especially WBC, NEU, and LYM counts), hepatorenal function (especially Cr, BUN, and AST), in ammation (especially PCT, CRP, ferritin, and the ESR), coagulation (especially D-dimer and PT), and cardiovascular injury (especially hs-cTnI and NT-proBNP) were also associated with fatal outcomes.
Similar to the two previous emergences of coronavirus diseases, severe acute respiratory syndrome (SARS) and Middle East respiratory syndrome (MERS), the outbreak of COVID-19 has posed great challenges for public health. Although most COVID-19 cases are mild, patients with severe conditions may quickly progress to ARDS, multiple organ failure and even death. The present study identi ed predictors of mortality that clinicians and other healthcare providers can consider when discussing the expected prognosis of patients with COVID-19 and thus take appropriate measures.
Advanced age has been identi ed as an independent risk factor for mortality in SARS [43,44] and MERS [45,46]. Our meta-analysis con rmed that older age was also correlated with an increased mortality rate in patients with COVID-19. Several factors might contribute to this mortality risk, including age-related physiological changes, impaired immune function, and preexisting illnesses. The present meta-analysis found that patients with current smoking were 2.95 times more likely to die than nonsmoker. Upregulation of angiotensin-converting enzyme 2 (ACE2) expression in airways might explain the increased risk of death rate in current smokers with COVID-19 infection [47]. The pooled analysis indicated that male sex was associated with a 30% increased risk of mortality among patients with COVID-19. Although the factors accounting for the sex difference in the incidence of death remain unknown, we suggest that smoking might be one of the contributing factors. Of note, the included studies were primarily from China, where the proportion of adult men who smoke (>50%) is much higher than that the proportion of adult women (<3%) [48]. It is possible that the sex differences in the survival rate are due to the different proportions of smoking between sexes.
Accumulated evidence has shown that COVID-19 infection is more likely to occur in patients with preexisting conditions than in those without preexisting conditions [49]. Similar to SARS [44,50] and MERS patients [51], preexisting conditions were also found to have an important effect on the prognosis in COVID-19 patients. Our pooled analysis showed that the risk of mortality in COVID-19 patients with any comorbidity was 2.85 times higher than that in those without preexisting conditions. Patients with chronic kidney disease, cerebrovascular disease, respiratory disease, cardiovascular disease, diabetes mellitus and hypertension had approximately 8-fold, 8-fold, 4-fold, 3-fold, 2-fold, and 2-fold higher risks of mortality than individuals without these conditions, respectively. Considering that these comorbidities are predictors of poor outcomes, optimum control of these conditions may be bene cial for the management of COVID-19. A recent large-sample study investigated the correlation of blood glucose control and outcomes in COVID-19 patients with diabetes [52]. The results indicated that patients with wellcontrolled blood glucose, maintaining glycemic variability within 3.9 to 10.0 mmol/L, had signi cantly improved survival compared to patients with poorly controlled blood glucose [52].
The symptoms of COVID-19 infection are nonspeci c. Patients with COVID-19 can present with fever, cough, muscle aches, fatigue, headache, gastrointestinal symptoms, and dyspnea [53]. Of them, the symptom of dyspnea was signi cantly associated with an increased mortality rate in the pooled analysis, corroborating the ndings of two studies [4,31] that demonstrated dyspnea was correlated with a higher mortality rate even after adjustment for age, sex, and other confounding factors. As dyspnea can be easily observed in clinical practice, it may be a valuable predictor to help identify individuals at high risk for fatal outcomes that may need additional attention. Additionally, blood gas analysis may be a useful tool to determine the severity of dyspnea. Decreased SpO 2 may re ect severe dyspnea, indicating an increased mortality risk, as shown in the present meta-analysis.
Dramatically reduced LYM levels as well as CD3, CD4, and CD8 cell counts in the deceased patients suggests that SARS-CoV-2 may act on T lymphocytes, and viral replication contributes to the destruction of T lymphocytes, decreasing immune function. Not surprisingly, patients with poor immune function were more likely to suffer from acute infection than those with normal immune function, and patients suffering from acute infection were more likely to die than those without this complication. In the present meta-analysis, a positive correlation between the PCT level or WBC count and mortality was observed, indicating that an increased WBC count (≥10×10 9 /L) may be a useful predictor and that PCT-guided antibiotic therapy might be bene cial in COVID-19 infection.
Compared with that in survivors, serum concentrations of Cr were higher in patients in the deceased group, indicating worse kidney function, although the mean values remained within the normal range. A single-cell transcriptome analysis indicated that the cytopathic effects of SARS-CoV-2 on podocytes and proximal straight tubule cells may contribute to the development of AKI in patients with COVID-19 [54].
Increased baseline Cr or BUN, peak Cr >133 μmol/L, and the presence of hematuria, hematuria or AKI were identi ed as independent predictors of in-hospital mortality in a large-sample prospective study evaluating 701 COVID-19 cases [7]. In the present meta-analysis, the development of AKI and increased SCr (>133 μmol/L) were associated with an approximately 9.6-fold and 3.6-fold increased risk of mortality among patients with COVID-19, respectively. Considering the great impact of renal damage on prognosis, close monitoring of renal function-related parameters is required.
Regarding markers of liver injury, the pooled analysis demonstrated statistically higher levels of AST and TBIL in nonsurvivors than in survivors, and patients with increased AST (>40 U/L) were approximately twice as likely to die compared to patients with normal values. Although increased ALT (>40 U/L) was also correlated with an increased risk of death, no signi cant difference in the mean levels of ALT between the nonsurvivors and survivors was observed in our meta-analysis. In a large-sample longitudinal study evaluating 5771 patients with COVID-19 infection, elevation of AST was correlated with the highest mortality risk compared to other markers of liver injury, such as ALT, TBIL and alkaline phosphatas. In addition, the elevation of AST occurred before the elevation of ALT [55]. These ndings indicate that AST may be a better liver injury marker for predicting clinical outcomes than the other markers, and frequent monitoring of AST and early detection of liver injury are suggested. Decreased albumin (<35 g/L) has been identi ed as an independent predictor of severe infection requiring intensive care unit (ICU) admission in MERS infection [56]. In our study, serum albumin was also signi cantly lower in the deceased patients than in the surviving patients, indicating that malnutrition might contribute to the adverse outcome of COVID-19 infection and that nutritional support may be bene cial in the management of this disease.
Cytokine storm, also known as hypercytokinemia, refers to the excessive and uncontrolled release of pro-inflammatory cytokines. Huang et al.
[57] found markedly higher plasma levels of cytokines in COVID-19 patients requiring ICU admission than in those not treated in the ICU, indicating that cytokines are correlated with disease severity. In the present meta-analysis, numerous in ammatory biomarkers (ESR, CRP, PCT, ferritin, and IL-6) were higher in deceased patients than in survivors, providing further evidence for the presence of a cytokine storm that can contribute to the fatal outcome of COVID-19 patients. Mehta et al. [58] suggested that each COVID-19 patient with a severe condition should be screened for hyperin ammation considering laboratory trends (increasing ferritin, decreasing PLTs or ESR) and HScore [59].
Cardiovascular complications of COVID-19, such as cardiac injury, heart failure, and arrhythmia, were more prevalent in patients who died than in patients who survived. Among them, cardiac injury was correlated with the highest mortality risk and has been widely studied. The possible mechanisms of cardiac injury caused by COVID-19 may involve cardiac stress due to respiratory failure and hypoxemia, direct myocardial infection by the virus, and indirect damage from the systemic inflammatory response [60]. In the present meta-analysis, several indicators of cardiovascular injury, such as hs-cTnI, NT-proBNP, CK-MB and myoglobin, were signi cantly higher in patients in nonsurvivor group than in those in the survivor group. Increased hs-cTnI and NT-proBNP among COVID-19 patients have been identi ed as independent risk factors for mortality even after adjustment for age and other confounding factors [9,10]. Therefore, we suggest frequent measurement of hs-cTnI and NT-proBNP should be required in the management of COVID-19, especially for patients with preexisting cardio-cerebrovascular disease.
Coagulation dysfunction is common in patients with COVID-19. We identi ed signi cantly lower PLTs in patients with a fatal outcome, and thrombocytopenia (PLT <125 × 10 9 /L) was correlated with a 4.65-fold increased risk of mortality. Increased D-dimer and prolonged PT were also observed more frequently in nonsurvivors than in survivors. These ndings indicated that excessive activation of the coagulation cascade and PLTs existed in the progression of COVID-19 infection. The underlying mechanisms of activated coagulation remain unclear but may be due to in ammatory responses induced by SARS-CoV-2 [61]. Coagulation screening, especially the determination of D-dimer and PLT levels, has been suggested.
To date, no effective pharmacological interventions have been identi ed to treat COVID-19. Antiviral agents have been widely applied in the management of COVID-19 infection. However, there was no signi cant difference between nonsurvivors and survivors regarding antiviral therapy e cacy. Corticosteroid treatment for COVID-19 infection is a hot topic. A recently published meta-analysis investigated the impact of corticosteroids on the outcomes of patients with coronavirus infections, including SARS, MERS, and COVID-19 [62]. The results demonstrated that the use of corticosteroids signi cantly delayed the clearance of the virus and did not improve the survival rate, shorten the hospitalization time, or reduce the ICU admission and mechanical ventilation rate. In the present meta-analysis, we found that corticosteroid treatment was correlated with an elevated risk of mortality in COVID-19 patients, with a pooled RR of 1.79 (95% CI 1.25 to 2.55). Current evidence and our ndings further support the recommendations by the Infectious Diseases Society of America (IDSA) denouncing the use of corticosteroids in patients with COVID-19 pneumonia [63]. Of note, a subset analysis of 84 patients with ARDS secondary to COVID-19 infection found an improved survival rate in patients who received methylprednisolone, indicating corticosteroid therapy might be bene cial for COVID-19 patients who develop ARDS. The timing, dose, and duration of corticosteroid therapy in patients with COVID-19 remain to be identi ed in future studies.
Patients who have indications for organ supportive care, such as the need for invasive mechanical ventilation and renal replacement therapy, tend to be sicker than other patients, which could explain the increased death rate among those patients. Other indicators of severe conditions, such as the APACHEII and SOFA scores, can also be used to predict the prognosis of COVID-19 infection.
The present study has several strengths. First, the current study was performed based on recent guidelines for the systematic review and metaanalysis of prognostic factors. Second, we tried our best to avoid potential patient overlap by checking the settings and patient recruitment periods. Third, many potential predictors of mortality in COVID-19 patients were tested.
The present study also had several limitations and should be interpreted cautiously. First, only unadjusted prognostic effect estimates were pooled because only few original studies reported adjusted data, different types of effect measures used (such as odds ratios and hazard ratios), and potential overlap of the patients between the included studies existed. It is possible that the unadjusted estimates of several factors may become nonsigni cant after adjustment. Second, the present meta-analysis contained substantial heterogeneity in the analysis of some tested factors, which could be explained by differences in patient populations and severity of the disease. Third, although a total of 27 studies were included in the meta-analysis, the number of studies for analysis of each predictor was insu cient to allow for meaningful subgroup analyses.

Conclusion
In conclusion, the present meta-analysis provides evidence of correlations between important prognostic factors and survival in patients with COVID-19. Clinicians and other healthcare providers should consider these factors when discussing the expected prognosis of COVID-19 patients and take appropriate measures accordingly. Further studies are required to provide a better understanding of the pathophysiological mechanisms of the association between these predictors and COVID-19 infection.  Pooled analyses of predictors for mortality in COVID-19 patients strati ed by risk factor type.