Study identification and selection
3,601 unique citations were identified, of which 648 were selected for full-text evaluation. Of those, under full-text evaluation, 189 observational studies comprising 57,563 COVID-19 patients were included for the final analysis. References of the included studies can be found in the Supplemental Material, and the flow chart of study selection results can be found in Fig. 1. In brief, the majority of the studies were conducted in the Asia Pacific region (n = 134, 72%), with 21, 20 and 14 studies coming from Europe, the US, and Middle-East, respectively. No study was found in South America. The majority of the studies were conducted in adults (n = 159, 84.13%), 22 studies (11.64%) included both pediatric and adult patients, and 8 studies (4.23%) included pediatric patients only. The mean age of study patients ranged from 2.5 to 75.25 years. The proportion of males ranged from 0 to 100%. Detailed characteristics of the included studies can be found in Supplemental Table S1.
Serum hemoglobin levels
Due to heterogeneity among studies, we were not able to provide a pooled estimate on prevalence of anemia among Covid-19 patients. Based on findings from 139 observational studies comprising 40,450 individuals, pooled mean hemoglobin level was 129.7 g/L (95% Confidence Interval (CI) 128.51; 130.88; I2 = 98.2%, P-value for heterogeneity < 0.001, Table 1). In the subgroup analyses, mean hemoglobin levels were lower in studies including subjects above median age (55.25 years) and having a higher proportion of male participants, prevalent cardiovascular disease, diabetes and hypertension, and a higher proportion of ICU admission (Supplementary Table S2). Similarly, when fitting regression lines, significant linear trends were observed for mean hemoglobin levels decreasing with advancing mean age of study participants, increasing proportion of subjects with diabetes, hypertension and overall comorbidities in each study (Supplemental Figure S7). Compared to moderate COVID-19 cases (using data from 63 studies with 21,605 individuals), severe cases had lower hemoglobin levels [weighted mean difference (WMD), − 4.08 g/L (95% CI − 5.12; − 3.05); I2 = 59%, P-value for heterogeneity < 0.001] (Table 2, Supplemental Figure S1). The mean difference was larger in studies with a higher proportion of individuals with hypertension (Supplemental Table S3). In addition, we observed a lower mean difference in hemoglobin levels between severe and moderate COVID-19 cases with an increasing number of proportion of study participants with diabetes (Supplemental figures S8). Based on pooled estimates from 27 studies, we did not find a significant difference in hemoglobin levels between COVID-19 survivors and non-survivors [WMD, − 0.26 g/L (95% CI − 2.37; 1.85); I2 = 74%, P-value for heterogeneity < 0.001] (Table 2, Supplemental Figure S2). In subgroup analyses, none of the patient characteristics were identified as a source of heterogeneity (Supplemental Table S4). However, when fitting a regression line, the WMD of mean hemoglobin levels between survivors and non-survivors decreased with a higher percentage of diabetic patients in each study (Supplemental figures S9).
Table 1 Characteristics of studies included in meta-analysis of mean hemoglobin, ferritin, and other biomarkers levels and the main meta-analysis results Table 2 Meta-analysis of differences in mean hemoglobin, ferritin and other biomarkers levels based on disease severity and vital status Among studies included in the narrative synthesis, Huang et al. [21] reported reduction in hemoglobin levels in 38.2% of hospitalized COVID-19 patients, but did not specify the definition of decreased hemoglobin. While Wang et al. [1] reported reduced hemoglobin levels (< 110 g/L) in 19.23% of the study population admitted to hospital. In contrast, Xu et al. [22] studied asymptomatic patients and reported none of the cases had decreased hemoglobin levels, albeit, not defining the cut-off of decreased levels. Based on retrospective data from 245 individuals with COVID-19, Liu et al. [23] found that the unadjusted association between baseline hemoglobin levels and all-cause mortality during hospitalization was non-significant, and the odds ratio of death with increasing serum hemoglobin level was 0.98 (95% CI 0.96, 1.00, p = 0.05). Cai et al. [24] studied factors associated with ICU admission and found no link between hemoglobin levels and odds of being admitted at ICU. In a study by Cen et al. [25] hemoglobin levels below 110 g/L were linked with disease progression in patients with COVID-19; univariable hazard ratio was 3.91 (95% CI 2.99–5.10). In addition, Giacomelli et al. [26] reported anemia (defined as hemoglobin levels below 125 g/L) was more prevalent in Covid-19 non-survivors (66.7%) compared to survivors (42.7%).
Serum ferritin levels
Based on findings from 54 observational studies, including 24,262 COVID-19 patients, pooled mean ferritin level in COVID-19 patients was 777.33 ng/mL (95% CI 701.33; 852.77), I2 = 95.5%, P-value for heterogeneity < 0.001 (Table 1). In subgroup analyses, mean pooled ferritin levels were higher in studies including older population, higher proportion of hypertensive patients and of patients who did not survive (Supplementary Table S5). In addition, when regressing the most important study participants characteristic against mean serum ferritin levels, we observed significant linear trends with mean serum ferritin levels increasing with advancing age, increasing proportion of male study participants, number of hypertensive patients and proportion of patients admitted to ICU (Supplemental figures S10).
Based on pooled estimates from 29 studies and 13,620 individuals, the mean difference in serum ferritin was higher in severe COVID-19 individuals compared to moderate cases; [WMD, 473.25 ng/mL (95% CI 382.52; 563.98); I2 = 91.8%, P-value for heterogeneity < 0.001] (Table 2, Supplemental figure S3). Similarly, when pooling the estimates from 18 observational studies and 7,190 individuals we found higher mean ferritin levels in non-survivors as compared to survivors [ WMD, 606.37 ng/mL (95% CI 461.86; 750.88); I2 = 90.9%, P-value for heterogeneity < 0.001] (Table 2, Supplemental figure S4). In subgroup analyses concerning disease severity and survival, high heterogeneity between the studies was not explained by any of studied patient's characteristics. We observed a linear trend with WMD of ferritin levels between severe and moderate COVID-19 cases increasing with advancing age, while no linear trends were observed between patients characteristics and WMD in ferritin levels among survivors and non-survivors (Supplemental Table S6&7, Supplemental figure S11 & 12).
In line with our findings, Zhou et al. found in a univariable analysis that odds of in hospital death were higher among patients with ferritin levels above 300 ng/mL compared to those with serum ferritin ≤ 399 ng/mL( odds ratio was 9.10 (95% CI 2.04; 40.58, p = 0.0038). Indeed, levels of serum ferritin were elevated in non-survivors compared with survivors (562 ng/mL ± 492 ng/mL) throughout the clinical course, and were increased with disease deterioration [3]. Similarly, in a study by Li et al. [27] ferritin was significantly higher in severe cases. Shah et al. reported no significant differences in serum ferritin levels and transferrin saturation between patients with non-severe versus severe hypoxemia. Yet, in patients with severe hypoxemia they reported significantly lower levels of serum iron [median 2.3 μmol/L (IQR, 2.2–2.5)] in comparison to non-severe hypoxemia (median 4.3 μmol/L. (IQR, 3.3–5.2)] [28].
Other biomarkers
We were able to pool the estimates from 13 and 4 studies on mean RBC and RDW comprising 1,382 and 1,538 subjects and the pooled mean RBC level and RDW were 4.09 × 1012/L (95% CI 3.9; 4,28) and 12.99% (95% CI 12.62;13.36) respectively. In addition, based on seven, five and three observational studies, pooled MCV, MCH and MCHC were 89.88 fL (95% CI 88.05; 91.75), 38.68 pg (95% CI 30.17;31.18) and 338.05 g/dL (95% CI 332.08; 344.03) respectively (Table 1, Supplemental Figure S5). Due to limited number of studies, we were able to perform meta-regression only for analyses concerning RBC. In subgroup analyses, mean RBC count was lower in studies including higher proportion of patients with comorbidities (73.2%); in line with this, with increasing proportion of subjects with comorbidities, RBC count was decreasing (Supplemental table S8 and Supplemental figure S13). When pooling the estimates from seven observational studies and 717 individuals with COVID-19, the mean difference in RBC count was lower in individuals with severe as compared to moderate disease [WMD, − 0.16 × 1012/ (95% CI − 0.31; − 0.014)], while RDW was higher [WMD, 1.82% (95% CI 0.10; 3.55] (Table 2, Supplemental Figure S6). We were not able to meta-analyze the other biomarkers of iron metabolism and anemia due to limited number of studies included in current review and/or high heterogeneity in reporting across the studies.
Quality of studies and publication bias
The majority of studies included in the current systematic review were evaluated as low risk of bias (n = 114, 60.3%), while the rest of the studies were evaluated as medium risk of bias (n = 75) (Supplemental table S1 and S9). The funnel plot on ferritin difference between survivors and non-survivors was asymmetrical and Egger's P value was 0.002 suggesting the presence of publication bias, while for the rest of the meta-analyses including 5 or more studies we found no evidence of publication bias (Supplemental figures S14-16).