We retrospectively extracted the data from the electronic charts of consecutive patients with confirmed COVID-19 at Huoshenshan hospital (Wuhan, China) from 5 February to 23 March 2020. The Huoshenshan hospital, a makeshift hospital with 1,000 beds, was opened by the government on 5 February 2020, and assigned to treat exclusively COVID-19 patients. This study was approved by the National Health Commission of China and the institutional review board at Huoshenshan hospital. Written informed consent was waived by the ethics committee of the Huoshenshan hospital for patients with emerging infectious diseases.

Inclusion criteria for the study were (i) hospitalized patients with confirmed COVID-19 infection; (ii) age >18 years old. Patients with no data on LFTs were excluded from the study. COVID-19 was diagnosed by clinical manifestations, chest computed tomography (CT), and confirmed by real-time polymerase chain reaction (RT-PCR) according to World Health Organization (WHO) interim guidance (30), and the New Coronavirus Pneumonia Prevention and Control Program (7th edition) published by the National Health Commission of China (31). The severity of COVID-19 was categorized as mild, severe, or critical (31, 32). Mild type was defined as having slight clinical symptoms without signs of pneumonia or with mild pneumonia (multiple small patchy shadows and interstitial changes, mainly in the outer zone of the lung and under the pleura) by radiography (31, 32). Severe cases were characterized by dyspnoea, respiratory frequency ≥30/min, blood oxygen saturation ≤ 93%, PaO₂/FiO₂ ratio <300 mmHg, and/or lung infiltrates >50% within 24–48 h (31, 32). Such patients were considered as critical case if they developed respiratory failure requiring mechanic ventilation, septic shock, and/or multiple organ dysfunction/failure (31, 32).

Baseline data collected within 24 h after admission include patient demographics, clinical features at inclusion, clinical history, comorbidities, initial blood pressure, and heart rate, laboratory values (peripheral white blood cell, neutrophil, lymphocyte, hemoglobin, platelet count, creatinine, blood urea nitrogen, potassium, sodium, D-dimer, prothrombin time, activated partial thromboplastin time, international normalized ratio, creatine kinase, lactate dehydrogenase, procalcitonin, and c-reactive protein), and radiological reports. Data regarding the specific drug therapy provided during the hospitalization also were collected. Liver function tests [alanine aminotransferase (ALT), aspartate aminotransferase (AST), albumin, total bilirubin, (TBIL), alkaline phosphatase (ALP), and gamma-glutamyltransferase (GGT)] from the time of hospital admission until discharge or death were obtained. The performing of LFT was determined by the attending physicians based on the demand of clinical decision. LFTs were considered as abnormal when at least one among AST, ALT, albumin, TBIL, ALP, and GGT were above the upper limit of normal (ULN) of laboratory reference range standards (i.e., AST >40 U/L, ALT >45 U/L, albumin <35 g/L, TBIL >26 μmol/L, ALP >125 U/L, GGT >60 U/L). All data were reviewed and confirmed by two certified investigators (Yong Lv and Huahong Xie) to ensure accuracy.

The primary endpoint was all-cause mortality during hospitalization. Secondary endpoints included ICU admission and use of mechanical ventilation. All clinical outcomes were obtained from clinical charts and assessed on April 15, 2020, when all survived patients were discharged and the Huoshenshan Hospital was shut down. The criteria for discharge are: (i) throat swab specimens collected 24 h apart were negative for tests of SARS-CoV-2; (ii) body temperature was normal for three consecutive days; (iii) symptoms of COVID-19 were resolved; (iv) the radiographic findings of COVID-19 significantly improved (31).

For all analyses, missing data of the covariates were imputed with multiple imputations methods (detailed in Supplementary Materials and Methods). Data are presented as frequencies (percentage), mean ± standard deviation (SD), or medians with interquartile range (IQR) as appropriate. Comparisons of variables between groups were performed using Student t-test, non-parametric Mann-Whitney U-test, chi-squared test, or Fisher's exact test as appropriate. Dynamic changes in liver function were presented using locally weighted scatterplot smoothing (LOESS). The cumulative probability model [an ordinal regression model for continuous outcomes (33)] was used to evaluate the association of baseline characteristics and treatment before peaking of FLTs with the peak levels of LFTs in hospital, where the liver function markers were treated as continuous response variables. The non-linear relationships between liver function markers and the risk of the evaluated outcomes were visualized using restricted cubic splines by entering the liver function markers as a continuous variable into the logistic regression analysis. Cumulative risks of death was assessed with Kaplan-Meier curves and compared using the log-rank test. Cumulative incidences of ICU admission or mechanical ventilation were estimated in a competing risks setting, where the death competed with the event of interest. The contribution of each variable to the risk of developing the endpoint was reported as a hazard ratio (HR) with 95% confidence interval (CI). We assessed the unadjusted and confounder-adjusted effects of LFTs on the evaluated outcomes using Cox regression models. Age, gender, severity of COVID-19 (severe/critical vs. mild), comorbidities (include hypertension, cardiovascular disease, diabetes, chronic pulmonary diseases, cerebrovascular disease, malignancy, and autoimmune disease) and chronic liver diseases (include hepatitis B virus infection, hepatitis C virus infection, and autoimmune liver disease) were considered as potential confounders. We assessed the heterogeneity in the effect of LFTs across the severity of COVID-19 by including a LFTs-by-COVID-19-severity interaction term in the Cox regression models. A significant interaction would indicate that the effect of LFTs was different across the severity of COVID-19. Significance was established at p < 0.05. All statistical calculations were performed using R 3.6.1(http://www.R-project.org/) with the add-on packages Hmisc, rms, riskRegression, pec, prodlim, and cmprsk.

During the study period, 2,922 patients with confirmed COVID-19 were admitted to the Huoshenshan hospital, and 10 patients were excluded because of incomplete relevant data. Ultimately, 2,912 patients with COVID-19 were included in the study. In the entire cohort, the mean age was 58.4 ± 14.4 years, and 1,512 (51.9%) were female. On admission, the severity of COVID-19 was mild in 2,160 (74.2%) patients, severe in 714 (24.5%) and critical in 38 (1.3%). Among the 752 serious and critically ill patients, 54 (7.2%) patients had multiple organ dysfunction syndromes. A total of 1236 (42.4%) patients had comorbidities, with hypertension (910 patients [31.2%]) being the most common one, followed by diabetes (392 patients [13.5%]). Sixty-eight patients (2.3%) had chronic liver disease, among which 58 had hepatitis B virus infection, 8 had hepatitis C virus infection, and 2 autoimmune liver disease. No patients had cirrhosis. The most common symptoms of COVID-19 were fever (2,057 patients [70.6%]), followed by cough (2,001 patients [68.7%]), fatigue (1,461 patients [50.2%]), dyspnea (1,394 patients [47.9%]), myalgia (774 patients [26.6%]), anorexia (523 patients [18.0%]), and expectoration (420 patients [14.4%]). Nausea, vomiting, abdominal pain, diarrhea, headache, dizziness disorders of consciousness were rare.

On admission 1,414 patients (48.6%) had abnormal LFTs, with ALT, AST, TBIL, ALP, and GGT above ULN in 662 (22.7%), 221 (7.6%), 52 (1.8%), 135 (4.6%), and 536 (18.5%) patients, respectively, and hypoalbuminemia (<35g/L) in 737 (25.3%) patients. The baseline characteristics of the study population according to normal and abnormal LFTs on admission are summarized in Table 1. Compared with patients with normal LFTs, patients with abnormal LFTs were older, with more severe COVID-19 disease and more likely to have symptoms of fever, cough, expectoration, dyspnea, fatigue, myalgia, anorexia, and nausea. The mean values of white blood cell count, neutrophil count, lymphocyte count, hemoglobin, platelet count, creatinine, D-dime, activated partial thromboplastin time, creatine kinase, lactate dehydrogenase, procalcitonin, and C-reactive protein were also higher in patients with abnormal LFTs. Mover, patients with abnormal LFTs had a higher likelihood of receiving antiviral therapy, antibiotics, immunoglobin, glucocorticoid therapy, high flow nasal cannula, and continuous renal replacement therapy during hospitalization (Table 2).

When stratified according to the severity of COVID-19 infection, patients with severe or critical COVID-19 had higher median values of AST, TBIL, ALP, and GGT and lower median value of albumin. The proportion of patients with abnormal AST, albumin, TBIL, ALP, and GGT were higher in severe or critical cases (Figure 1 and Supplementary Tables 1, 2). Nevertheless, the median value of ALT and the proportion of patients with abnormal ALT were not significantly across the severity of COVID-19.

Figure 2 depicts the dynamic trajectories of ALT, AST, albumin, TBIL, ALB, and GGT according to normal or abnormal LFTs on admission. The ALT, AST, TBIL, ALP, and GGT values in the abnormal LFTs group increased slightly within the first 5 days after admission and trended downwards thereafter, while those values in the normal LFTs group tended upwards for the entire in-hospital duration. The albumin values trended downwards in both groups within the first 5 days of hospitalization and then fluctuated slightly for the entire duration of follow-up.

When stratified according to the severity of COVID-19 (mild vs. severe/critical), the dynamic curves of LFTs showed downward trends of ALT, AST, TBIL, ALP, and GGT and a upward trend of albumin in both mild and severe/critical groups (Supplementary Figure 2). Furthermore, the values of ALT, AST, TBIL, ALP, and GGT were higher and albumin was lower in patents with the outcomes of death, ICU admission, and mechanical ventilation compared those without in most time-points (Supplementary Figures 3–5).

The cumulative probability model revealed the association between baseline characteristics and hospital treatment on peak ALT, AST, TBIL, ALP, GGT levels, and nadir albumin levels in the entire cohort (Supplementary Figure 6 and Figures 3, 4). Younger age, male gender, use of antibiotics, increased hemoglobin, increased C-reactive protein, and increased lactate dehydrogenase were factors positively associated with elevated ALT levels. Male gender, diabetes, higher C-reactive protein, and increased lactate dehydrogenase were the leading factors positively associated with elevated AST levels. Total bilirubin levels rise were tightly associated with male gender, decreased creatinine, decreased platelet count, increased C-reactive protein, and increased lactate dehydrogenase. Older age, male gender, antiviral, antibiotics, systemic corticosteroids use, hemoglobin reduction, C-reactive protein, and lactate dehydrogenase elevation were main factors positively correlated with decreased albumin levels. Alkaline phosphatase levels were closely linked with older age, male gender, platelet count, C-reactive protein, and lactate dehydrogenase elevation. Male gender, white blood cell, platelet count, hemoglobin, C-reactive protein, and lactate dehydrogenase increase were identified as factors positively associated with elevated GGT levels. C-reactive protein, lactate dehydrogenase platelet count, hemoglobin, and male gender were common factors positively associated with ALT, AST, TBIL, ALP, GGT elevation, and albumin reduction during hospitalization. To predict the peak (nadir) value of these LFTs, nomograms that incorporated the significant risk factors were established.

During a median 13 (IQR: 8–19) days of hospitalization, 61 patients (2.1%) died, 106 patients (3.6%) admitted or transfer to ICU, and 75 patients (2.6%) required mechanical ventilation (Table 2 and Supplementary Figure 7). The 30-day cumulative incidences of death was significantly higher in patients with abnormal LFTs on admission compared with those with normal LFTs (abnormal vs. normal: 3.3 vs. 0.47%; HR 8.32, [95%CI 3.79 −18.26]; p < 0.001, Figure 5A). Similarly, patients with abnormal LFTs on admission had a higher 30-day cumulative incidences of ICU admission (5.9 vs. 1.2%; HR 5.18 [95%CI 3.12–8.60]; p < 0.001; Figure 5C) and mechanical ventilation requirement (4.2 vs. 0.8%; HR 5.14 [95%CI 2.82–9.34]; p < 0.001, Figure 5E). This pattern persisted after adjusting for potential confounders, with the adjusted HRs of abnormal LFTs were 3.66 (95%CI 1.64–8.19, p = 0.002, Figure 5B) for death, 3.12 (95%CI 1.86–5.23, p < 0.001, Figure 5D) for ICU admission, and 3.00 (95%CI 1.63–5.52, p < 0.001, Figure 5F) for mechanical ventilation requirement. Furthermore, these effects were homogeneous across the severity of COVID-19 (P_interaction > 0.1 for all comparisons, Figure 6 and Supplementary Figures 8, 9). Notably, chronic liver disease was not associated with an increased risk of either of these adverse outcomes (Figures 5, 6 and Supplementary Figures 8, 9).

The relationship between the baseline ALT, AST, albumin TBIL, ALP as well as GGT and death rate during hospitalization was depicted in Figure 7. The increased ALT, AST, TBIL, ALP, GGT, and decreased albumin on admission had a non-linear positive association with the risk of death, which was homogeneous across the severity of COVID-19. Similar results were observed for the secondary endpoint of ICU admission (Supplementary Figure 10) as well as mechanical ventilation requirement (Supplementary Figure 11).

When stratified according to different levels of LFTs, abnormal levels of baseline AST, albumin, TBIL, ALP, and GGT were significantly associated with the risk of death, ICU admission, and mechanical ventilation (Figure 8 and Supplementary Figures 12, 13). Among them, AST over three-fold ULN and albumin <30 g/L had the highest risks of death, ICU admission, and mechanical ventilation. The elevation of ALT tended to be associated with increased risks of those outcomes. Nevertheless, the difference did not reach significance.

Among the 1,498 patients with normal LFTs upon admission, 368 patients (24.6%) developed de novo abnormalities of LFTs (Supplementary Table 3). Univariable and multivariable logistic regression analysis showed that lymphocyte count (OR 1.13, 95%CI: 1.03–1.25, p = 0.007), use of quinolones (OR 1.48, 95%CI: 1.10–1.98, p = 0.010), and cephalosporins (OR 4.80, 95%CI: 1.58–14.59, p = 0.006) were independently were associated with de novo abnormalities of LFTs (Supplementary Figure 14 and Figure 9). Compared with those without de novo abnormal LFTs, patients with de novo abnormal LFTs had higher risk of death, ICU admission as well as the mechanical ventilation requirement. The trends persisted after adjusting for potential confounders, but the differences were not significant (Supplementary Table 4 and Figure 10).

Overall, 1,782 patients (61.2%) had abnormal LFTs during hospitalization in the entire cohort. Peak ALT, AST, TBIL, ALP, and GGT above ULN was observed in 855 (29.4%), 321 (11.0%), 103 (3.5%), 166 (5.7%), and 642 (22.0%) patients, respectively. Nadir albumin <35 g/L was observed in 1,114 (38.3%) patients during hospitalization. The baseline characteristics of patients grouped according to LFTs abnormalities during hospitalization are shown in Supplementary Table 5. Compared with those with normal liver function impairment during hospitalization, patients with abnormal LFTs had severer COVID-19 disease and more common of respiratory and digestive symptoms. Similar to baseline LFTs abnormality, abnormal LFTs during hospitalization, peak AST, TBIL, ALP, GGT, and nadir albumin but not peak ALT were significantly (or a trend toward) associated with adverse outcomes of COVID-19 (Figures 11–13, Supplementary Tables 6, 7, and Supplementary Figures 15–19).