3.1 Baseline data
All participants were divided into two groups by normal and elevation ferritin levels (ferritin ≤ 200 ng/mL and ferritin > 200 ng/mL) according to clinical reference range. As shown in Table 1, severe cases were more frequently found in patients with higher ferritin levels, compared to those who with normal ferritin levels (50.0% vs 2.9%, p < 0.001). There were significant differences of sex between the two groups, with male patients accounting for the 68.2% in elevation ferritin group but only 17.1% in normal ferritin group (p < 0.001). The average age of elevation ferritin group was also significantly older than the normal ferritin group (49.9 vs 36.8 years old). The other characteristics including systolic blood pressure (SBP), diastolic blood pressure (DBP), respire, pulse, smoking, comorbidities and symptoms of both groups were recorded but no statistical difference was found.
3.2 Laboratory examination
As displayed in Supplementary Table 1, the levels of triglyceride (TG), serum amyloid A protein (SAA), alanine aminotransferase (ALT), aspartate aminotransferase (AST), γ-glutamyl transferase (γ-GT), interleukin-6 (IL-6), creatinine (Cre), lactate dehydrogenase (LDH), fibrinogen (FIB), fibrinogen degradation product (FDP), C-reactive protein (CRP) and erythrocyte sedimentation rate (ESR) in elevation ferritin group were extremely higher than those of normal ferritin group (p < 0.001). The levels of fasting blood glucose (FBG), alkaline phosphatase (ALP), natural kill-cell (NK-cell), creatine kinase (CK), troponin (TNI), thrombin time (TT), D-dimer, haemoglobin, leucocytes, neutrophils and neutrophil-to-lymphocyte ratio (NLR) in elevation ferritin group were also higher than those of normal group (p < 0.05). However, the levels of high-density lipoprotein cholesterol (HDL-c), albumin, CD3 + and estimated glomerular filtration rate (eGFR) in elevation ferritin group were significantly lower than those of normal ferritin group (p < 0.05). The other indicators showed no insignificance between two groups.
Furthermore, a spearman’s correlation analysis was employed to demonstrate the relationship between ferritin and other covariates (Fig. 1). Ferritin was highly positively correlated to FIB, LDH, SAA, γ-GT, NK-cell, FDP, ESR, AST, ALT, CRP, CK, IL-6, Cre and TT (r > 0.4 and p < 0.001), and negatively associated with eGFR and CD3 + with a statistical significance (r<-0.45 and p < 0.0005).
3.3 Determination of ferritin and corresponding CT manifestation
As mentioned above, there existed significantly more severe cases in elevation ferritin group than the normal group. In addition, we compared the ferritin levels among severe and non-severe patients, finding that the ferritin level of severe group is significantly higher than that of non-severe group (median 921.3 (IQR 440.0-1609.8) vs median 130.7 (IQR 58.8-320.4), p < 0.001) (Fig. 2A). Furthermore, we recorded the ferritin levels which were calculated from the onset of symptom of each patient. Likewise, at every stage the ferritin level of sever group was markedly higher than their counterparts (Fig. 2B). On the other side, we observed the CT images that were corresponding to the ferritin values in three representative patients (Fig. 2C). The results demonstrated that the severity of pulmonary imaging was consistent with the ferritin level. Taken together, the ferritin level is able to reveal the severity of COVID-19.
3.4 Diagnostic ability of ferritin
The AUC of ROC plot for ferritin was 0.873 (95% CI: 0.798–0.954), larger than that of age (0.697, 95% CI: 0.567–0.828), CRP (0.730, 95% CI: 0.602–0.859) and lymphocytes% (LYM%, 0.717, 95% CI: 0.586–0.847), indicating ferritin possesses a considerable diagnostic value for severity of patients on admission. The best cutoff value of ferritin was 272.5 ng/mL, with a sensitivity of 96% and specificity of 70% (Fig. 3A). Furthermore, we established a multivariable diagnostic model incorporating sex, age, ferritin, LYM% and CRP (n = 69, 0.862 (95% CI: 0.761–0.962), sensitivity 77%, specificity 87%) (Fig. 3B). It is interesting to find that the diagnostic ability of integrated model showed the similar value as ferritin alone with the AUC showing (n = 69, 0.861 (95% CI: 0.773–0.949), sensitivity 95%, specificity 68%) (p = 0.981). These results suggest that ferritin may act as an independent risk factor on COVID-19 patients.
In addition, we employed logistic regression analysis to assess the risk factors related with severity illness (Fig. 3C). The results showed that a 100 ng/mL increase in serum ferritin, the odds ratio (OR) of severity illness was 1.17 (95% CI: 1.03–1.33, p = 0.014). After adjusted for sex, age, CRP or LYM%, the OR of severity illness was 1.20 (95% CI: 1.02–1.41, p = 0.014). Moreover, Fig. 3D showed that the OR of severity illness for serum ferritin was 10.78 (95% CI: 1.13-102.73, p = 0.039) when serum ferritin was calculated as categorical variable (> 200 ng/mL vs ≤ 200 ng/mL) in crude model. After adjusted for age, sex, CRP and LYM%, the OR of severity illness for patients with elevated ferritin level (> 200 ng/mL) was 18.75 (95% CI: 1.28, 275.41, p = 0.032), as compared with patients with normal ferritin level (≤ 200 ng/mL). Therefore, we established a nomogram based on the above risk factors to improve the availability in clinical practice (Fig. 4).
3.5 Prognostic ability of ferritin
Median duration of viral clearance was 6 days (IQR 2–25) in normal ferritin group (≤ 200 ng/mL, but elevation ferritin group (> 200 ng/mL) was 16 days (IQR 2–47). The viral clearance events in in normal group and elevation ferritin group by using Kaplan-Meier survival analysis were displayed in Fig. 5A (p < 0.001).
Besides, we recorded the length of hospitalization of all patients except 2 deceased ones and 8 patients that transferred to other hospitals. Our result showed that patients with higher ferritin levels on admission stayed in hospital for a longer time (18 days, IQR 3–47), when compared with the ones with normal ferritin levels (10 days, IQR 5–29). The discharge events in in normal group and elevation ferritin group by using Kaplan-Meier survival analysis are displayed in Fig. 5B (p < 0.001).