Skip to main content
Advertisement

Main menu

  • Home
  • Articles
    • Newest Articles
    • Current Issue
    • Methods & Resources
    • Author Interviews
    • Archive
    • Subjects
  • Collections
  • Submit
    • Submit a Manuscript
    • Author Guidelines
    • License, Copyright, Fee
    • FAQ
    • Why submit
  • About
    • About Us
    • Editors & Staff
    • Board Members
    • Licensing and Reuse
    • Reviewer Guidelines
    • Privacy Policy
    • Advertise
    • Contact Us
    • LSA LLC
  • Alerts
  • Other Publications
    • EMBO Press
    • The EMBO Journal
    • EMBO reports
    • EMBO Molecular Medicine
    • Molecular Systems Biology
    • Rockefeller University Press
    • Journal of Cell Biology
    • Journal of Experimental Medicine
    • Journal of General Physiology
    • Journal of Human Immunity
    • Cold Spring Harbor Laboratory Press
    • Genes & Development
    • Genome Research

User menu

  • My alerts

Search

  • Advanced search
Life Science Alliance
  • Other Publications
    • EMBO Press
    • The EMBO Journal
    • EMBO reports
    • EMBO Molecular Medicine
    • Molecular Systems Biology
    • Rockefeller University Press
    • Journal of Cell Biology
    • Journal of Experimental Medicine
    • Journal of General Physiology
    • Journal of Human Immunity
    • Cold Spring Harbor Laboratory Press
    • Genes & Development
    • Genome Research
  • My alerts
Life Science Alliance

Advanced Search

  • Home
  • Articles
    • Newest Articles
    • Current Issue
    • Methods & Resources
    • Author Interviews
    • Archive
    • Subjects
  • Collections
  • Submit
    • Submit a Manuscript
    • Author Guidelines
    • License, Copyright, Fee
    • FAQ
    • Why submit
  • About
    • About Us
    • Editors & Staff
    • Board Members
    • Licensing and Reuse
    • Reviewer Guidelines
    • Privacy Policy
    • Advertise
    • Contact Us
    • LSA LLC
  • Alerts
  • Follow LSA on Bluesky
  • Follow lsa Template on Twitter
Research Article
Transparent Process
Open Access

A prediction model for COVID-19 liver dysfunction in patients with normal hepatic biochemical parameters

Jianfeng Bao, Shourong Liu, View ORCID ProfileXiao Liang, Congcong Wang, Lili Cao, Zhaoyi Li, Furong Wei, Ai Fu, Yingqiu Shi, Bo Shen, Xiaoli Zhu, Yuge Zhao, Hong Liu, Liangbin Miao, Yi Wang, Shuang Liang, Linyan Wu, View ORCID ProfileJinsong Huang  Correspondence email, View ORCID ProfileTiannan Guo  Correspondence email, View ORCID ProfileFang Liu  Correspondence email
Jianfeng Bao
1Department of Hepatology, Affiliated Hangzhou Xixi Hospital, Zhejiang University School of Medicine, Hangzhou, China
Roles: Conceptualization, Funding acquisition, Writing—original draft, Project administration, Writing—review and editing
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
Shourong Liu
1Department of Hepatology, Affiliated Hangzhou Xixi Hospital, Zhejiang University School of Medicine, Hangzhou, China
Roles: Supervision
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
Xiao Liang
2Westlake Laboratory of Life Sciences and Biomedicine, Key Laboratory of Structural Biology of Zhejiang Province, School of Life Sciences, Westlake University, Hangzhou, China
3Institute of Basic Medical Sciences, Westlake Institute for Advanced Study, Hangzhou, China
4Center for Infectious Disease Research, Westlake Laboratory of Life Sciences and Biomedicine, Hangzhou, China
Roles: Methodology, Writing—review and editing
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
  • ORCID record for Xiao Liang
Congcong Wang
5Insititute of Hepatology and Epidemiology, Affiliated Hangzhou Xixi Hospital, Zhejiang University School of Medicine, Hangzhou, China
Roles: Data curation
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
Lili Cao
6Department of Nursing, Affiliated Hangzhou Xixi Hospital, Zhejiang University School of Medicine, Hangzhou, China
Roles: Data curation
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
Zhaoyi Li
5Insititute of Hepatology and Epidemiology, Affiliated Hangzhou Xixi Hospital, Zhejiang University School of Medicine, Hangzhou, China
Roles: Data curation
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
Furong Wei
5Insititute of Hepatology and Epidemiology, Affiliated Hangzhou Xixi Hospital, Zhejiang University School of Medicine, Hangzhou, China
Roles: Data curation
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
Ai Fu
5Insititute of Hepatology and Epidemiology, Affiliated Hangzhou Xixi Hospital, Zhejiang University School of Medicine, Hangzhou, China
Roles: Data curation
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
Yingqiu Shi
2Westlake Laboratory of Life Sciences and Biomedicine, Key Laboratory of Structural Biology of Zhejiang Province, School of Life Sciences, Westlake University, Hangzhou, China
3Institute of Basic Medical Sciences, Westlake Institute for Advanced Study, Hangzhou, China
4Center for Infectious Disease Research, Westlake Laboratory of Life Sciences and Biomedicine, Hangzhou, China
Roles: Data curation
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
Bo Shen
7Department of Laboratory Medicine, Taizhou Hospital of Zhejiang Province Affiliated to Wenzhou Medical University, Linhai, China
Roles: Data curation
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
Xiaoli Zhu
7Department of Laboratory Medicine, Taizhou Hospital of Zhejiang Province Affiliated to Wenzhou Medical University, Linhai, China
Roles: Data curation
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
Yuge Zhao
8Department of Pathology, Affiliated Hangzhou Xixi Hospital, Zhejiang University School of Medicine, Hangzhou, China
Roles: Data curation
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
Hong Liu
8Department of Pathology, Affiliated Hangzhou Xixi Hospital, Zhejiang University School of Medicine, Hangzhou, China
Roles: Data curation
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
Liangbin Miao
5Insititute of Hepatology and Epidemiology, Affiliated Hangzhou Xixi Hospital, Zhejiang University School of Medicine, Hangzhou, China
Roles: Data curation
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
Yi Wang
5Insititute of Hepatology and Epidemiology, Affiliated Hangzhou Xixi Hospital, Zhejiang University School of Medicine, Hangzhou, China
Roles: Data curation
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
Shuang Liang
2Westlake Laboratory of Life Sciences and Biomedicine, Key Laboratory of Structural Biology of Zhejiang Province, School of Life Sciences, Westlake University, Hangzhou, China
3Institute of Basic Medical Sciences, Westlake Institute for Advanced Study, Hangzhou, China
4Center for Infectious Disease Research, Westlake Laboratory of Life Sciences and Biomedicine, Hangzhou, China
Roles: Data curation
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
Linyan Wu
6Department of Nursing, Affiliated Hangzhou Xixi Hospital, Zhejiang University School of Medicine, Hangzhou, China
Roles: Data curation
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
Jinsong Huang
1Department of Hepatology, Affiliated Hangzhou Xixi Hospital, Zhejiang University School of Medicine, Hangzhou, China
Roles: Supervision, Funding acquisition
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
  • ORCID record for Jinsong Huang
  • For correspondence: huangjinsongyz@126.com guotiannan@westlake.edu.cn hzxxlf@outlook.com
Tiannan Guo
2Westlake Laboratory of Life Sciences and Biomedicine, Key Laboratory of Structural Biology of Zhejiang Province, School of Life Sciences, Westlake University, Hangzhou, China
3Institute of Basic Medical Sciences, Westlake Institute for Advanced Study, Hangzhou, China
4Center for Infectious Disease Research, Westlake Laboratory of Life Sciences and Biomedicine, Hangzhou, China
Roles: Supervision, Funding acquisition, Methodology, Writing—review and editing
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
  • ORCID record for Tiannan Guo
  • For correspondence: huangjinsongyz@126.com guotiannan@westlake.edu.cn hzxxlf@outlook.com
Fang Liu
5Insititute of Hepatology and Epidemiology, Affiliated Hangzhou Xixi Hospital, Zhejiang University School of Medicine, Hangzhou, China
Roles: Conceptualization, Resources, Data curation, Formal analysis, Funding acquisition, Validation, Visualization, Methodology, Writing—original draft, Project administration, Writing—review and editing
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
  • ORCID record for Fang Liu
  • For correspondence: huangjinsongyz@126.com guotiannan@westlake.edu.cn hzxxlf@outlook.com
Published 19 October 2022. DOI: 10.26508/lsa.202201576
  • Article
  • Figures & Data
  • Info
  • Metrics
  • Reviewer Comments
  • PDF
Loading

Abstract

Coronavirus disease 2019 (COVID-19) patients with liver dysfunction (LD) have a higher chance of developing severe and critical disease. The routine hepatic biochemical parameters ALT, AST, GGT, and TBIL have limitations in reflecting COVID-19–related LD. In this study, we performed proteomic analysis on 397 serum samples from 98 COVID-19 patients to identify new biomarkers for LD. We then established 19 simple machine learning models using proteomic measurements and clinical variables to predict LD in a development cohort of 74 COVID-19 patients with normal hepatic biochemical parameters. The model based on the biomarker ANGL3 and sex (AS) exhibited the best discrimination (time-dependent AUCs: 0.60–0.80), calibration, and net benefit in the development cohort, and the accuracy of this model was 69.0–73.8% in an independent cohort. The AS model exhibits great potential in supporting optimization of therapeutic strategies for COVID-19 patients with a high risk of LD. This model is publicly available at https://xixihospital-liufang.shinyapps.io/DynNomapp/.

Introduction

Coronavirus disease 2019 (COVID-19), which is caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has become a pandemic. Different levels of liver disease have been reported in more than half of the infected population (Cai et al, 2020; Jothimani et al, 2020) and are recognized as an essential component of COVID-19 (Marjot et al, 2021).

In the early stage of COVID-19, some patients with normal hepatic biochemical parameters (alanine aminotransferase, aspartate aminotransferase, γ-glutamyl transferase, and total bilirubin) experience progression to liver dysfunction (LD) after the initiation of hepatotoxic-antiviral drugs, with even dramatic progression to liver failure without predictable circumstances (Carothers et al, 2020; Weber et al, 2020). With the reduction in the rate of COVID-19–related severity and death, a means of decreasing the incidence of non-SARS-CoV-2–related complications has become increasingly important for patients with mild disease.

Given the important role of the liver in drug metabolism, coagulation, albumin synthesis, and the generation of acute-phase reactants, abnormal liver function can influence the pathologic course of the systemic disease COVID-19. Accumulating evidence suggests that elevated expression of serum hepatic biochemical parameters is related to adverse events, including mechanical ventilation, shock, and increased risk of death (Cai et al, 2020; Ding et al, 2021; Mao et al, 2020; Weber et al, 2021; Yadav et al, 2021; Yip et al, 2021). However, hepatic abnormalities involve various manifestations, including liver congestion, inflammatory response, drug-induced liver damage, and hepatocyte infection (Marjot et al, 2021), with associated difficulty in identifying susceptible patients at an early stage. Considering the limitations of common hepatic biochemical parameters in representing the actual liver function status of COVID-19 patients, early identification of subgroups with a high risk of COVID-19–based LD through new biomarkers is needed for medical intervention. However, there is no reliable strategy thus far for effective prediction of subsequent LD among COVID-19 patients at admission.

Proteomics profiling has the ability to shed light on molecular changes reflected in sera from COVID-19 patients (Nie et al, 2021; Shen et al, 2020). In this study, we developed and validated an exceptionally parsimonious model through proteomic analysis to predict subsequent risk of liver damage among COVID-19 patients with normal hepatic biochemical parameters on admission, in accordance with Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) standards.

Currently, many clinical predictive models might not have wide adaptability in clinical practice, often because of modeling of complex and multiple variables. In addition, most such models have been classified as having a high risk of bias and might not be generalizable, which is often caused by nonadherence to reporting standards and best practice methods during model development (Wolff et al, 2019; Wynants et al, 2020). The model reported here includes only two variables, namely, sex and angiopoietin-like 3 (ANGL3), which is a liver-specific secreted protein (Gaudet et al, 2017) screened by random forest (RF) and CoxBoost algorithm analyses from 1,517 proteins of proteomic data. The generalizability concerning discrimination, calibration, and clinical utility of this prediction model was comprehensively evaluated according to TRIPOD guidelines. We call this the protein-based AS LD model. The prediction tool and model parameters are freely available online to facilitate risk stratification for therapeutic interventions.

Results

Baseline clinical characteristics of the discovery cohort

As shown in Fig 1, 109 participants were recruited between January 2020 and April 2020. During this period, 660 samples were collected from all of these patients. Ninety-eight (89.9%) individuals with available liver outcomes were included in the discovery cohort. A total of 397 serum samples from these individuals were designated as a source of material for identifying biomarkers. The baseline characteristics of the discovery cohort are shown as stratification for LD (Table S1); 52 (53.1%) of the patients experienced LD, with a median follow-up time of 19 (median value and interquartile range [IQR]: 10.2–31.0) days. Significant differences were observed in sex and body mass index (BMI) (P < 0.05).

Figure 1.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure 1. Study design and modeling flowchart.

RF, random forest; MDA, the mean decrease of accuracy; ROC, receiver operator characteristic curve.

Table S1 Baseline characteristics of the COVID-19 patients in the discovery cohort.

Protein selection

Proteomic analysis identified 1,517 proteins from the serum samples. 20 proteins were preliminarily screened by a RF algorithm with a mean decrease in accuracy (MDA) of higher than 3 (Fig S1). The expression patterns of these 20 proteins among different grades of LD samples were compared using principal coordinates analysis (PCoA) mapping (Fig S2). The first and second principal components (PCoA1 accounted for 43.16% of the variance and PCoA2 for 11.01%) were plotted using the proteomic data, revealing a certain degree of distinction among the four groups.

Figure S1.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure S1. Top 20 proteins prioritized by random forest analysis ranked by the mean decrease of accuracy.
Figure S2.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure S2. PCoA plot of 397 samples using the top 20 proteins quantified.

PCoA, principal coordinate analysis.

A heat map was plotted to illustrate the expression patterns of the above 20 proteins in different grades of LD samples (Fig 2A). AL1A1 was first removed owing to a negligible difference between the LD groups. Thirteen proteins were retained according to the following rules: AUC > 0.70 for any grade of LD (Fig 2B) and significant multivariable associations with LD (P < 0.05) (Fig 2C). To minimize potential collinearity and overfitting for variables in the final model, we removed six proteins (FAAA, PSA7, PSB1, PSA5, PSA3, and PSA1) that correlated strongly with ALDOB and PSA4 (Spearman correlation coefficient > 0.6) (Fig 2D).

Figure 2.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure 2. Liver dysfunction (LD) associates with a set of dysregulated proteins in the discovery cohort.

(A) Heat map of 20 proteins identified by random forest algorithm in the four grades of LD. (B) The AUC assesses the discriminating accuracy of each of the 20 proteins in differentiating LD from COVID-19 patients. (C) The association analysis between each of the 20 proteins and LD using the Cox model. (D) Correlation analysis between the 20 proteins. LD, liver dysfunction.

Levels of the retained proteins (ANGL3, ALDOB, ADH4, ACY1, ADH1B, ARLY, and PSA4) were significantly different between the normal and LD groups (Fig 3A). Patients in the LD group exhibited higher levels of ANGL3 (median IQR: 1.06 [1.27–0.95] versus 1.03 [1.11–0.89], P < 0.01), ALDOB (median IQR: 1.52 [1.91–1.07] versus 0.84 [1.02–0.68], P < 0.001), ADH4 (median IQR: 0.59 [1.11–0.59] versus 0.59 [0.59–0.41], P < 0.001), ACY1 (median IQR: 0.56 [0.79–0.56] versus 0.56 [0.56–0.43], P < 0.001), ADH1B (median IQR: 0.99 [1.50–0.82] versus 0.79 [0.82–0.51], P < 0.001), ARLY (median IQR: 0.75 [1.02–0.75] versus 0.75 [0.75–0.61], P < 0.001), and PSA4 (median IQR: 0.96 [1.32–0.85] versus 0.81 [0.90–0.68], P < 0.001).

Figure 3.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure 3. Differentially expressed proteins between groups in the discovery cohort and the AS model for prediction of COVID-19–related liver dysfunction (LD) in the development cohort.

(A) Violin plot showing the expression values of the seven retained proteins between the normal and LD group. The black dot represents the median value. The error bar represents the interquartile range. (B) The discrimination and calibration of the AS model in the development cohort. (C) Decision curve analyses for the 18 candidate models in the development cohort. Net benefit is shown with piecewise linear function for each candidate model compared with the treat all and treat none approaches. LD, liver dysfunction; ANGL3.SE, ANGL3.Sex; ANGL3.HB, ANGL3.HBV; ANGL3.HY, ANGL3.Hypertension; ANGL3.SM, ANGL3.Smoking; ANGL3.CT, ANGL3.Chest CT, ANGL3.FEV, ANGL3.Fever; ANGL3.CO, ANGL3.Cough; ANGL3.FEE, ANGL3.Feebleness; ADH4.SE, ADH4.Sex; ADH4.HB, ADH4.HBV; ADH4.HY, ADH4.Hypertension; ADH4.SM, ADH4.Smoking; ADH4.CT, ADH4.Chest CT; ADH4.FEV, ADH4.Fever; ADH4.CO, ADH4.Cough; ADH4.FEE, ADH4.Feebleness.

Baseline characteristics of the development cohort

In the development cohort, 30 patients (40.5%) experienced LD within a median follow-up of 22.0 (11.0–32.0) days (Table 1). No significant differences were observed between the patients with normal liver function and those who experienced LD during the follow-up period.

View this table:
  • View inline
  • View popup
Table 1.

Characteristic of the candidate predictors considered during CoxBoost variable selection among COVID-19 patients with normal hepatic biochemical parameters at baseline.

Predictor selection

Baseline measurements included 31 clinical variables at admission and peak values of seven prioritized proteins within 7 d of hospitalization, which were simultaneously introduced into the CoxBoost model for predictor selection. After CoxBoost processing, 10 variables remained as significant predictors of LD, including ADH4, hypertension, sex, fever, ANGL3, cough, feebleness, smoking, Hepatis B virus (HBV), and chest computed tomography (CT) (Fig S3). The step number and the penalty of the CoxBoost classifier were 101 and 198, respectively.

Figure S3.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure S3. The significant predictors of LD after CoxBoost processing.

PII*, double pneumonia. CT, computed tomography.

Model selection

To maximize the clinical convenience and applicability of our prediction model, we limited the predictors to two in the final model and constructed 19 models as follows: ANGL3.Sex (ANGL3.SE), ANGL3.HBV (ANGL3.HB), ANGL3.Hypertension (ANGL3.HY), ANGL3.Smoking (ANGL3.SM), ANGL3.Chest CT (ANGL3.CT), ANGL3.Fever (ANGL3.FEV), ANGL3.Cough (ANGL3.CO), ANGL3.Feebleness (ANGL3.FEE), ANGL3.ADH4, ANGL3, ADH4.Sex (ADH4.SE), ADH4.HBV (ADH4.HB), ADH4.Hypertension (ADH4.HY), ADH4.Smoking (ADH4.SM), ADH4.Chest CT (ADH4.CT), ADH4.Fever (ADH4.FEV), ADH4.Cough (ADH4.CO), ADH4.Feebleness (ADH4.FEE), and ADH4 (Fig 3C).

To assess the model performance over the entire period, we calculated the time-dependent areas under the curves (AUCs) and presented them by graphical visualization (Figs 3B and S4). The ANGL3-based and ADH4-based models yielded AUC values from 0.48–0.83 to 0.53–0.81 over time, respectively. The calibration curves of ANGL3-based models showed better performance than those of ADH4-based models (Figs 3B and S5). The discrimination and calibration of the model ADH4 could not be calculated.

Figure S4.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure S4. The time-dependent AUCs of the 17 models trained by Cox regression in the development cohort.

ANGL3.SE, ANGL3.Sex; ANGL3.HB, ANGL3.HBV; ANGL3.HY, ANGL3.Hypertension; ANGL3.SM, ANGL3.Smoking; ANGL3.CT, ANGL3.Chest CT, ANGL3.FEV, ANGL3.Fever; ANGL3.CO, ANGL3.Cough; ANGL3.FEE, ANGL3.Feebleness; ADH4.SE, ADH4.Sex; ADH4.HB, ADH4.HBV; ADH4.HY, ADH4.Hypertension; ADH4.SM, ADH4.Smoking; ADH4.CT, ADH4.Chest CT; ADH4.FEV, ADH4.Fever; ADH4.CO, ADH4.Cough; ADH4.FEE, ADH4.Feebleness.

Figure S5.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure S5. Calibration curves for the 17 models in the development cohort.

ANGL3.SE, ANGL3.Sex; ANGL3.HB, ANGL3.HBV; ANGL3.HY, ANGL3.Hypertension; ANGL3.SM, ANGL3.Smoking; ANGL3.CT, ANGL3.Chest CT, ANGL3.FEV, ANGL3.Fever; ANGL3.CO, ANGL3.Cough; ANGL3.FEE, ANGL3.Feebleness; ADH4.SE, ADH4.Sex; ADH4.HB, ADH4.HBV; ADH4.HY, ADH4.Hypertension; ADH4.SM, ADH4.Smoking; ADH4.CT, ADH4.Chest CT; ADH4.FEV, ADH4.Fever; ADH4.CO, ADH4.Cough; ADH4.FEE, ADH4.Feebleness.

Decision curve analyses of the 19 models at a point estimate of 28 d are shown in Fig 3C. Threshold probabilities for the net benefit associated with the application of the ANGL3-based and ADH4-based models in detecting LD ranged from 0.00–0.78 and 0.00–0.46, respectively. Notably, the ANGL3.SE (AS) model showed a higher net benefit than any other ANGL3-based model and the treat all or treat none strategies.

After performing an omni-ensemble analysis with discrimination, calibration, and clinical utility assessment, we prioritized the best-of-state AS model as the final LD prediction model for COVID-19 patients. As shown in Fig 3B, the time-dependent AUCs were 0.60–0.80 for predicting LD, indicating favorable discrimination by the AS model. The calibration curves of the AS model showed adequate agreement between the predicted and observed probability in the development cohort. Decision curves revealed that the AS model achieved more net benefits than others for a broad range of threshold probabilities in the development cohort (Fig 3C).

Model validation

Internal validation

We validated the discrimination of the AS model internally using bootstrapping of 1,000 resamples and the multiple fractional polynomial regression model (Fig S6). The time-dependent AUCs showed results similar to those of the primary analyses.

Figure S6.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure S6. The internal discrimination validation of the AS model.

(A) The time-dependent AUCs of the AS model in the bootstrap model. (B) The time-dependent AUCs of the AS model in the multiple fractional polynomial regression model.

External validation

The external validation–independent cohort included 13 patients with a median follow-up time of 22.0 (15.2–32.8) days and a median number of 2.0 (1.0–3.0) measurements; 6 (46.2%) were male (Table S2). The outcome of LD eventually occurred in 7 (53.8%) of these patients, and significant differences in sex were observed. Applying the AS model to the data for 42 serum samples led to the correct assignment 29 (69.0%), 29 (69.0%), and 31 (73.8%) times when the cutoff was defined as 0.3, 0.5, and 0.7, respectively (Fig 4A). The accuracies of the AS model in the external cohort were approximately 70%, similar to the AUCs for the development cohort (Fig 3B).

Table S2 Characteristics of the COVID-19 patients in the external cohort.

Figure 4.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure 4. External validation, nomogram, and online web-based calculator of AS model.

(A) The external discrimination validation of the AS model using another dataset from Taizhou hospital. Predicted probability comes from the AS model. Observed events represent the actual occurrences of LD in the external cohort. (B) Characteristics in the nomogram to predict the risk of liver dysfunction (LD) in COVID-19 patients with normal liver biochemical parameters on admission. Patient predictive values are located on the axis of each variable; a line is then drawn upwards at a 90° angle to determine the number of points for that particular variable. The sum of these numbers is located on the total score axis, and a line is drawn at a 90° angle downward to the LD risk axes. (C) The free online tool for identifying the patients with low risk of liver dysfunction (LD) in COVID-19 patients. Male was represented at an arbitrary value of 1 (female = 2). ANGL3 represents the relative expression during proteomics analysis. LD, liver dysfunction.

Construction of the nomogram and web-based calculator

Based on the final AS model, a nomogram at 14, 21, 28, 35, and 42 d after hospitalization was constructed using the Cox model (Fig 4B). Fig S7 describes the use of the nomogram to determine the total score and risk of LD in a patient from the development cohort.

Figure S7.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure S7. Practice example of determining the risk of liver dysfunction (LD) by nomogram.

The blue line represents the values of a randomly selected patient from the development dataset who is male (28.34 points) with ANGL3 of 1.31 (70.48 points). These values, when plotted, correspond vertically to the points scale (top), and points are then summed to give a total point score (bottom). The presented patient has a total point score of 98.82 (red line) with an associated probability of LD of 0.06, 0.20, 0.44, 0.58, and 0.66 at 14, 21, 28, 35, and 42 d, respectively. LD, liver dysfunction.

A free calculator was established online to enable clinicians to automatically acquire a survival plot and survival probabilities (with 95% CI) of normal liver function in COVID-19 patients by entering values for ANGL3 and sex (https://xixihospital-liufang.shinyapps.io/DynNomapp/) (Fig 4C).

Discussion

In this study, we developed and validated an AS model and built a free online calculator for the prediction of LD in COVID-19 patients with normal hepatic biochemical parameters on admission. Notably, the AS model only integrates two predictors, namely, a novel biomarker ANGL3 and sex, with maximized generalization and simplicity. The prediction model can be freely accessed at an online web server (https://xixihospital-liufang.shinyapps.io/DynNomapp/).

The predictor ANGL3 obtained among 1,517 proteins is a specific secreted protein originally expressed in the liver, enhancing expression of plasma cholesterol (Gaudet et al, 2017). Interestingly, it has been speculated that cholesterol facilitates SARS-CoV-2 infection by promoting viral spike protein–mediated entry via ACE2 and furin processing (Shoemark et al, 2021). Previous reports have also highlighted the pivotal role of host cholesterol metabolism in the viral life cycle during HCV infection (Foka et al, 2014). Taken together, ANGL3 might indirectly participate in SARS-CoV-2 viral replication through cholesterol metabolism, leading to aggravated LD. The positive association of ANGL3 with LD incidence observed in our study confirms this hypothesis.

The protein ADH4, one of the markers screened by the CoxBoost algorithm, is an enzyme involved in reducing retinaldehyde, which also regulates cholesterol metabolism (Pares et al, 2008). In this study, ADH4-based model calibration did not perform well, potentially because ADH4 is not a liver-specific enzyme and is expressed in many epithelial tissues of the body, such as the blood vessels, skin, cornea, and gastrointestinal mucosa (Pares et al, 2008).

Another predictor of the AS model was sex. Previous studies have reported that males are more prone to LD (Hundt et al, 2020; Wang et al, 2020), consistent with our results (Fig S3), probably resulting from smoking and alcohol use. Moreover, the net benefit comparison between the AS and the ANGL3.SM model showed comparable performance above a threshold of 0.35, suggesting that smoking might be a critical factor in LD.

In the present study, the AS model development followed the principle of at least 10 events per variable (Mallett et al, 2010; Peduzzi et al, 1996) to reduce the risk of overfitting caused by the small sample size of our cohort. With few exceptions, most published COVID-19 prediction models lack comprehensive evaluation of calibration, clinical utility, and external applicability (Wynants et al, 2020). Our study followed the TRIPOD guidelines for model development and evaluated all three aspects mentioned above. We also retained ANGL3 as a continuous variable without arbitrary categorization to avoid loss of information (Gupta et al, 2021). Furthermore, time-dependent discrimination of the AS model provides a dynamic reference for clinicians to predict the appropriate incidence of LD for clinical decision-making.

Our study has several limitations. First, as a single-center and small sample size study, more extensive multicenter clinical research works are required to further validate the potential of the AS model for predicting COVID-19–related LD. Second, although the AS model–based nomogram and online calculator can identify individuals at high risk of LD among COVID-19 patients with normal hepatic biochemical parameters, our study cannot ascertain which treatment strategy might improve the outcomes of these patients. Third, to conclude, the simple AS model presented in this study can predict the incidence of LD earlier than the routine biochemical parameters alanine aminotransferase, aspartate aminotransferase, γ-glutamyl transferase, and total bilirubin. The nomogram and free web-based calculator make the prediction model easily accessible. COVID-19 patients with normal hepatic biochemistry might benefit from early prediction of the probability of subsequent LD to inform clinical decision-making.

Materials and Methods

Participants and outcome definitions

The retrospective study cohort was recruited from the Affiliated Hangzhou Xixi Hospital, Zhejiang University School of Medicine (Zhejiang, Southeast China; simplified as Xixi Hospital) from January 2020 to April 2020. Proteomic experiments were conducted among 660 serum samples from 109 COVID-19 patients at the Key Laboratory of Structural Biology of Zhejiang Province, School of Life Sciences, Westlake University (Zhejiang, Southeast China). All serum samples were collected concurrently with clinical blood examination by nurses and subsequently stored at −80°C by two laboratory technicians. Demographic, clinical, and outcome data were collected from the Data Center Laboratory of Xixi Hospital.

Patients with PCR-confirmed positive SARS-CoV-2 RNA (according to the test methods described previously) (Liu et al, 2020) were screened for eligibility. This study is reported following TRIPOD guidance (Collins et al, 2015); it was approved by the Institutional Review Board of Xixi Hospital. Informed consent was waived because of the retrospective nature of this study.

The primary outcome was the subsequent LD after 7 d of admission. Definitions and diagnostic criteria of normal liver function, LD, and its grade (mild, moderate, severe) are summarized in Table S3 (Cai et al, 2020; Liu et al, 2021).

Table S3 Diagnostic criteria of liver dysfunction.

Proteome analysis

Proteome experiment

Serum samples were inactivated and sterilized at 56°C for 30 min and processed as in the previous study with some modifications (Shen et al, 2020). Briefly, 4 μl serum from each sample was depleted using High Select Top-14 Abundant Protein Depletion MiniSpin Columns (Thermo Fisher Scientific). Then elutes were denatured, reduced, and alkylated to derive protein lysates. The solutions were diluted with 200 μl 100 mM triethylammonium bicarbonate followed by a double-step trypsinization (enzyme-to-substrate ratio kept at 1:40 in each step). 10% TFA was added to each sample to quench the enzymatic reaction. Digested peptides were cleaned up with SOLAμ (Thermo Fisher Scientific) according to the manufacturer’s instructions and labeled by TMTpro 16plex reagents (Thermo Fisher Scientific). The reagents were all MS grade. 660 samples in total (633 serum samples for experiment and 27 randomly selected serum samples for quality control) were divided into 44 batches for the labeling experiment, each batch containing 15 samples and one pool. The labeled samples in each batch were combined and fractionated. Each sample was separated into 30 fractions and then consolidated into 10 fractions; dried and re-dissolved fractions with 2% acetonitrile (ACN)/0.1% formic acid of MS grade. The re-dissolved peptides were analyzed with a U3000 HPLC system coupled to an Orbitrap Exploris 480 (Thermo Fisher Scientific) in data-dependent acquisition mode combined to a front-end high-field asymmetric waveform ion mobility spectrometry (FAIMS). Peptides of each fraction were loaded onto a pre-column (3 μm, 100 Å, 20 mm × 75 mm i.d.) and separated with a 75 min LC gradient at a flow rate of 300 nl/min (analytical column: 1.9 mm, 120 Å, 150 mm × 75 mm i.d.; buffer A: 2% ACN, 98% H2O containing 0.1% FA; buffer B: 98% ACN, 2% H2O containing 0.1% FA). FAIMS was operated at two different CVs, −48 and −68 V, respectively. For MS acquisition, the RF level was set at 50% and ion transfer tube temperature at 320°C. The turbo-TMT mode was enabled. The scan range of MS1 was 350–1,800 m/z. Resolutions were set to 60,000 for MS1 and 30,000 for MS2. Normalized AGC target was set at 300% for MS1 and 200% for MS2. The maximum injection time was set as 50 ms for MS1 and 86 ms for MS2. Dynamic exclusion was on, and mass tolerance was set to ±10 ppm. Intensity threshold was set at 20,000 for MS1, and HCD collision energy was fixed to 38%. MS/MS data were recorded in centroid mode. Isolation window was set to 0.7 m/z.

Proteomic data analysis

MS data for proteomics were processed with Proteome Discoverer (version 2.4.0.305; Thermo Fisher Scientific) against a manually annotated and reviewed Homo sapiens protein FASTA database (Swiss-Prot, 27 April 2020). For parameter settings, enzyme digestion was set to full-specific trypsin with two missed cleavages. Static modifications were TMTpro of lysine residues and N-terminus peptides and carbamidomethylation of cysteine. Dynamic modifications were set to oxidation of methionine and acetylation of N-terminus peptides. Mass tolerance for precursors and product ions was fixed at 10 ppm and 0.02 Da, respectively. The peptide–spectrum match allowed 1% target false discovery rate (strict) and 5% target false discovery rate (relaxed). 1,517 proteins in total were quantified. The protein abundance ratio of samples to the pooled sample within each batch was considered as the relative protein abundance ratio for the subsequent analysis.

Quality control of proteome data

Multiple quality control of proteome data was performed as previously stated (Shen et al, 2020). Briefly, mouse liver digest was used as a standard sample for instrument performance evaluation. To avoid carry-over, a blank sample (buffer A) was run between every four sample injections. A pooled peptide sample labeled by TMT pro-126 was contained in each batch for aligning data from different batches and evaluation of quantitative accuracy. We first evaluated the quantitative ratio distribution of 633 samples. The outliers (above upper quartile) were imputed as two times the interquartile. Then we checked the correlation of 27 technical replicates by Pearson correlation coefficient (Fig S8A). R package limma was used to remove batch effects from experiments and instruments. There were no significant differences between 44 batches using principal component analysis (Fig S8B).

Figure S8.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure S8. Quality control of proteomic data.

(A) Violin plot showing the Pearson correlation coefficient of the technical replicates. (B) Principal component analysis of samples from different batches.

Candidate protein selection

Ninety-eight of the 109 patients with LD outcomes were enrolled as the discovery cohort for identifying biomarkers. We estimated the MDA for each of the 1,517 quantified proteins by the RF algorithm in the discovery cohort and prioritized 20 proteins with an MDA larger than 3 for subsequent analysis (Fig S1). In RF analysis, 500 trees were built with 10-fold cross-validation, and this was repeated 100 times.

We performed univariable and multivariate Cox models, including age, sex, and BMI, and AUCs (equal to the concordance statistics) to estimate the ability of each protein to discriminate LD versus normal (Fig 2B and C). A protein–protein correlation plot was constructed for the 20 proteins (Fig 2D). Some strong-correlation proteins were eliminated from candidate variates to minimize potential collinearity and overfitting bias. ANGL3, ALDOB, ADH4, ACY1, ADH1B, ARLY, and PSA4 were eventually retained (Fig 3A).

Selection of potential clinical predictors and model development

We included multiple clinical characteristics in the model development, including demographic variables, medical history, symptoms, and chest CT results on hospital admission. The demographic variables examined included age, sex, BMI, smoking status, highest temperature, systolic blood pressure, and diastolic blood pressure. Medical history included medications, HBV, hypertension, cardiovascular disease, tumor, HIV, HCV, chronic obstructive pulmonary disease, and diabetes. We also considered symptoms, such as cough, rhinorrhea, fever, pidiarrhea, rigor, nausea, dyspnea, muscular soreness, feebleness, headache, chest congestion, sore throat, and sputum. Chest CT results included normal, single pneumonia, and double pneumonia.

Of 98 individuals in the discovery cohort, 74 patients hospitalized with normal liver function for 7 d were selected as the development cohort. Peak values of ANGL3, ALDOB, ADH4, ACY1, ADH1B, ARLY, and PSA4 within 7 d of admission to the hospital were subject to next-stage selection together with the other 31 clinical variables, as described above.

Given that time may affect the discrimination of the prediction model, the CoxBoost algorithm (an XGBoost-based survival estimator model) (Unterhuber et al, 2021) was applied to select the most potent predictors that could be retained in the final model. To maximize the generalizability and simplicity of the prediction model, we restricted the number of variables to 2, and the combination was set to contain at least one protein marker. Each combination was introduced into a Cox regression model in view of the long-term horizon for predictions (during inpatient and non-inpatient periods) to quantify AUC values for given times.

Model selection

During selection, the discrimination (ability to distinguish individuals who will develop LD from those who will not, as quantified as the AUC), calibration (consistency between the predicted and observed probability, as evaluated by a calibration plot), and clinical utility (as quantified by the net benefit of a decision curve) were assessed among the above-combined prediction models (Steyerberg & Vergouwe 2014). An ideal calibration slope is 1, which suggests that the risk of observed outcomes matches the risk predicted. Decision curve analysis permits evaluation of clinical utility by quantifying the trade-off between correct discrimination of true positives and incorrect discrimination of false positives, as weighted based on the threshold probability (Vickers et al, 2019). The threshold probability represents the benefit ratio for the intervention. In this article, decision curve analysis was conducted at 28 d after hospital admission to quantify the net benefit of implementing the model in clinical practice. For all of the decision curves, data were fitted with a piecewise linear function among all replicates with the following algorithm: double yi1, yi2; yi1=a1+k1xi1;yi2=yi1+k2(xi2−xi1);if (x<xi1)y=a1+k1xelse if (x<xi2)y=yi1+k2(x−xi1);elsey=yi2+k3(x−xi2);Finally, the model that performed comprehensively best across discrimination, calibration, and clinical utility was used to construct a dynamic nomogram plot, which was then used to build a free online calculator.

Model validation

We utilized bootstrapping of 1,000 resamples (with replacement) and multiple fractional polynomial regression modeling to internally validate the stability of the final model using complete development case data.

To further validate the generalizability of the AS model, we collected 42 serum samples from 13 COVID-19 patients with normal hepatic biochemical parameters at Taizhou Hospital of Zhejiang province, affiliated to Wenzhou Medical University between January 2020 and February 2020 at the time of hospitalization for the external validation cohort. The sex information and the maximum values of ANGL3 within 7 d of hospital admission were selected for validation.

A result was considered statistically significant when the two-tailed P-value was below 0.05. R software versions 3.6.3 and 4.0.5 (www.r-project.org) were used for the statistical analyses.

Data Availability Statement

The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding authors.

Acknowledgements

We thank Westlake University Supercomputer Center for assistance in data generation and storage and the Mass Spectrometry & Metabolomics Core Facility at the Center for Biomedical Research Core Facilities of Westlake University for sample analysis. The authors disclosed receipt of the following financial support for the research, authorship, and publication of this article: Natural Science Foundation of Zhejiang Province (LQ22H100001), Medical and Health Research Project of Zhejiang Province (2022508831), the Hepatology (Traditional Chinese and Western Medicine) of Hangzhou Medical Peak Subject, Science and Technology Development Program of Hangzhou (202004A20), Hangzhou Pharmaceutical Health and Technology Project (Z20220098), the National Key R&D Program of China (2020YFE0202200), the National Natural Science Foundation of China (81972492, 21904107), Zhejiang Provincial Natural Science Foundation for Distinguished Young Scholars (LR19C050001), and Westlake Education Foundation, Tencent Foundation (2020).

Author Contributions

  • J Bao: conceptualization, funding acquisition, project administration, and writing—original draft, review, and editing.

  • S Liu: supervision.

  • X Liang: methodology and writing—review and editing.

  • C Wang: data curation.

  • L Cao: data curation.

  • Z Li: data curation.

  • F Wei: data curation.

  • A Fu: data curation.

  • Y Shi: data curation.

  • B Shen: data curation.

  • X Zhu: data curation.

  • Y Zhao: data curation.

  • H Liu: data curation.

  • L Miao: data curation.

  • Y Wang: data curation.

  • S Liang: data curation.

  • L Wu: data curation.

  • J Huang: supervision and funding acquisition.

  • T Guo: supervision, funding acquisition, methodology, and writing—review and editing.

  • F Liu: conceptualization, resources, data curation, formal analysis, funding acquisition, validation, visualization, methodology, project administration, and writing—original draft, review, and editing.

Conflict of Interest Statement

The authors declare that they have no conflicts of interest.

  • Received June 23, 2022.
  • Revision received September 30, 2022.
  • Accepted September 30, 2022.
  • © 2022 Bao et al.
Creative Commons logoCreative Commons logohttps://creativecommons.org/licenses/by/4.0/

This article is available under a Creative Commons License (Attribution 4.0 International, as described at https://creativecommons.org/licenses/by/4.0/).

References

  1. ↵
    1. Cai Q,
    2. Huang D,
    3. Yu H,
    4. Zhu Z,
    5. Xia Z,
    6. Su Y,
    7. Li Z,
    8. Zhou G,
    9. Gou J,
    10. Qu J, et al.
    (2020) Covid-19: Abnormal liver function tests. J Hepatol 73: 566–574. doi:10.1016/j.jhep.2020.04.006
    OpenUrlCrossRefPubMed
  2. ↵
    1. Carothers C,
    2. Birrer K,
    3. Vo M
    (2020) Acetylcysteine for the treatment of suspected remdesivir-associated acute liver failure in covid-19: A case series. Pharmacotherapy 40: 1166–1171. doi:10.1002/phar.2464
    OpenUrlCrossRef
  3. ↵
    1. Collins GS,
    2. Reitsma JB,
    3. Altman DG,
    4. Moons KGM
    (2015) Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (tripod): The tripod statement. BMJ 350: g7594. doi:10.1136/bmj.g7594
    OpenUrlCrossRefPubMed
  4. ↵
    1. Ding ZY,
    2. Li GX,
    3. Chen L,
    4. Shu C,
    5. Song J,
    6. Wang W,
    7. Wang YW,
    8. Chen Q,
    9. Jin GN,
    10. Liu TT, et al.
    (2021) Association of liver abnormalities with in-hospital mortality in patients with COVID-19. J Hepatol 74: 1295–1302. doi:10.1016/j.jhep.2020.12.012
    OpenUrlCrossRef
  5. ↵
    1. Foka P,
    2. Karamichali E,
    3. Dalagiorgou G,
    4. Serti E,
    5. Doumba PP,
    6. Pissas G,
    7. Kakkanas A,
    8. Kazazi D,
    9. Kochlios E,
    10. Gaitanou M, et al.
    (2014) Hepatitis c virus modulates lipid regulatory factor angiopoietin-like 3 gene expression by repressing hnf-1α activity. J Hepatol 60: 30–38. doi:10.1016/j.jhep.2013.08.016
    OpenUrlCrossRef
  6. ↵
    1. Gaudet D,
    2. Gipe DA,
    3. Pordy R,
    4. Ahmad Z,
    5. Cuchel M,
    6. Shah PK,
    7. Chyu KY,
    8. Sasiela WJ,
    9. Chan KC,
    10. Brisson D, et al.
    (2017) Angptl3 inhibition in homozygous familial hypercholesterolemia. N Engl J Med 377: 296–297. doi:10.1056/NEJMc1705994
    OpenUrlCrossRefPubMed
  7. ↵
    1. Gupta RK,
    2. Harrison EM,
    3. Ho A,
    4. Docherty AB,
    5. Knight SR,
    6. van Smeden M,
    7. Abubakar I,
    8. Lipman M,
    9. Quartagno M,
    10. Pius R, et al.
    (2021) Development and validation of the ISARIC 4c deterioration model for adults hospitalised with COVID-19: A prospective cohort study. Lancet Respir Med 9: 349–359. doi:10.1016/s2213-2600(20)30559-2
    OpenUrlCrossRef
  8. ↵
    1. Hundt MA,
    2. Deng Y,
    3. Ciarleglio MM,
    4. Nathanson MH,
    5. Lim JK
    (2020) Abnormal liver tests in covid-19: A retrospective observational cohort study of 1,827 patients in a major U.S. Hospital network. Hepatology 72: 1169–1176. doi:10.1002/hep.31487
    OpenUrlCrossRefPubMed
  9. ↵
    1. Jothimani D,
    2. Venugopal R,
    3. Abedin MF,
    4. Kaliamoorthy I,
    5. Rela M
    (2020) Covid-19 and the liver. J Hepatol 73: 1231–1240. doi:10.1016/j.jhep.2020.06.006
    OpenUrlCrossRefPubMed
  10. ↵
    1. Liu F,
    2. Liu H,
    3. Yu WY,
    4. Liu Z,
    5. Zhang X,
    6. Wang Y,
    7. Miao LB,
    8. Li ZY,
    9. Huang JS,
    10. Bao JF
    (2021) The associations of lymphocyte ratio and neutrophil ratio on liver dysfunction in covid-19 patients. Front Immunol 12: 717461. doi:10.3389/fimmu.2021.717461
    OpenUrlCrossRef
  11. ↵
    1. Liu F,
    2. Xu A,
    3. Zhang Y,
    4. Xuan W,
    5. Yan T,
    6. Pan K,
    7. Yu W,
    8. Zhang J
    (2020) Patients of covid-19 may benefit from sustained lopinavir-combined regimen and the increase of eosinophil may predict the outcome of covid-19 progression. Int J Infect Dis 95: 183–191. doi:10.1016/j.ijid.2020.03.013
    OpenUrlCrossRef
  12. ↵
    1. Mallett S,
    2. Royston P,
    3. Dutton S,
    4. Waters R,
    5. Altman DG
    (2010) Reporting methods in studies developing prognostic models in cancer: A review. BMC Med 8: 20. doi:10.1186/1741-7015-8-20
    OpenUrlCrossRefPubMed
  13. ↵
    1. Mao R,
    2. Qiu Y,
    3. He J-S,
    4. Tan J-Y,
    5. Li X-H,
    6. Liang J,
    7. Shen J,
    8. Zhu L-R,
    9. Chen Y,
    10. Iacucci M, et al.
    (2020) Manifestations and prognosis of gastrointestinal and liver involvement in patients with covid-19: A systematic review and meta-analysis. Lancet Gastroenterol Hepatol 5: 667–678. doi:10.1016/s2468-1253(20)30126-6
    OpenUrlCrossRef
  14. ↵
    1. Marjot T,
    2. Webb GJ,
    3. Barritt AS IV.,
    4. Moon AM,
    5. Stamataki Z,
    6. Wong VW,
    7. Barnes E
    (2021) Covid-19 and liver disease: Mechanistic and clinical perspectives. Nat Rev Gastroenterol Hepatol 18: 348–364. doi:10.1038/s41575-021-00426-4
    OpenUrlCrossRefPubMed
  15. ↵
    1. Nie X,
    2. Qian L,
    3. Sun R,
    4. Huang B,
    5. Dong X,
    6. Xiao Q,
    7. Zhang Q,
    8. Lu T,
    9. Yue L,
    10. Chen S, et al.
    (2021) Multi-organ proteomic landscape of covid-19 autopsies. Cell 184: 775–791.e14. doi:10.1016/j.cell.2021.01.004
    OpenUrlCrossRef
  16. ↵
    1. Pares X,
    2. Farres J,
    3. Kedishvili N,
    4. Duester G
    (2008) Medium- and short-chain dehydrogenase/reductase gene and protein families: Medium-chain and short-chain dehydrogenases/reductases in retinoid metabolism. Cell Mol Life Sci 65: 3936–3949. doi:10.1007/s00018-008-8591-3
    OpenUrlCrossRefPubMed
  17. ↵
    1. Peduzzi P,
    2. Concato J,
    3. Kemper E,
    4. Holford TR,
    5. Feinstein AR
    (1996) A simulation study of the number of events per variable in logistic regression analysis. J Clin Epidemiol 49: 1373–1379. doi:10.1016/s0895-4356(96)00236-3
    OpenUrlCrossRefPubMed
  18. ↵
    1. Shen B,
    2. Yi X,
    3. Sun Y,
    4. Bi X,
    5. Du J,
    6. Zhang C,
    7. Quan S,
    8. Zhang F,
    9. Sun R,
    10. Qian L, et al.
    (2020) Proteomic and metabolomic characterization of covid-19 patient sera. Cell 182: 59–72.e15. doi:10.1016/j.cell.2020.05.032
    OpenUrlCrossRefPubMed
  19. ↵
    1. Shoemark DK,
    2. Colenso CK,
    3. Toelzer C,
    4. Gupta K,
    5. Sessions RB,
    6. Davidson AD,
    7. Berger I,
    8. Schaffitzel C,
    9. Spencer J,
    10. Mulholland AJ
    (2021) Molecular simulations suggest vitamins, retinoids and steroids as ligands of the free fatty acid pocket of the sars-cov-2 spike protein. Angew Chem Int Ed Engl 60: 7098–7110. doi:10.1002/anie.202015639
    OpenUrlCrossRefPubMed
  20. ↵
    1. Steyerberg EW,
    2. Vergouwe Y
    (2014) Towards better clinical prediction models: Seven steps for development and an abcd for validation. Eur Heart J 35: 1925–1931. doi:10.1093/eurheartj/ehu207
    OpenUrlCrossRefPubMed
  21. ↵
    1. Unterhuber M,
    2. Kresoja KP,
    3. Rommel KP,
    4. Besler C,
    5. Baragetti A,
    6. Kloting N,
    7. Ceglarek U,
    8. Bluher M,
    9. Scholz M,
    10. Catapano AL, et al.
    (2021) Proteomics-enabled deep learning machine algorithms can enhance prediction of mortality. J Am Coll Cardiol 78: 1621–1631. doi:10.1016/j.jacc.2021.08.018
    OpenUrlCrossRef
  22. ↵
    1. Vickers AJ,
    2. van Calster B,
    3. Steyerberg EW
    (2019) A simple, step-by-step guide to interpreting decision curve analysis. Diagn Progn Res 3: 18. doi:10.1186/s41512-019-0064-7
    OpenUrlCrossRefPubMed
  23. ↵
    1. Wang M,
    2. Yan W,
    3. Qi W,
    4. Wu D,
    5. Zhu L,
    6. Li W,
    7. Wang X,
    8. Ma K,
    9. Ni M,
    10. Xu D, et al.
    (2020) Clinical characteristics and risk factors of liver injury in covid-19: A retrospective cohort study from wuhan, China. Hepatol Int 14: 723–732. doi:10.1007/s12072-020-10075-5
    OpenUrlCrossRefPubMed
  24. ↵
    1. Weber S,
    2. Hellmuth JC,
    3. Scherer C,
    4. Muenchhoff M,
    5. Mayerle J,
    6. Gerbes AL
    (2021) Liver function test abnormalities at hospital admission are associated with severe course of sars-cov-2 infection: A prospective cohort study. Gut 70: 1925–1932. doi:10.1136/gutjnl-2020-323800
    OpenUrlAbstract/FREE Full Text
  25. ↵
    1. Weber S,
    2. Mayerle J,
    3. Irlbeck M,
    4. Gerbes AL
    (2020) Severe liver failure during sars-cov-2 infection. Gut 69: 1365–1367. doi:10.1136/gutjnl-2020-321350
    OpenUrlFREE Full Text
  26. ↵
    1. Wolff RF,
    2. Moons KGM,
    3. Riley RD,
    4. Whiting PF,
    5. Westwood M,
    6. Collins GS,
    7. Reitsma JB,
    8. Kleijnen J,
    9. Mallett S,
    10. Groupdagger P
    (2019) Probast: A tool to assess the risk of bias and applicability of prediction model studies. Ann Intern Med 170: 51–58. doi:10.7326/M18-1376
    OpenUrlCrossRefPubMed
  27. ↵
    1. Wynants L,
    2. Van Calster B,
    3. Collins GS,
    4. Riley RD,
    5. Heinze G,
    6. Schuit E,
    7. Bonten MMJ,
    8. Dahly DL,
    9. Damen JAA,
    10. Debray TPA, et al.
    (2020) Prediction models for diagnosis and prognosis of covid-19: Systematic review and critical appraisal. BMJ 369: m1328. doi:10.1136/bmj.m1328
    OpenUrlAbstract/FREE Full Text
  28. ↵
    1. Yadav DK,
    2. Singh A,
    3. Zhang Q,
    4. Bai X,
    5. Zhang W,
    6. Yadav RK,
    7. Singh A,
    8. Zhiwei L,
    9. Adhikari VP,
    10. Liang T
    (2021) Involvement of liver in covid-19: Systematic review and meta-analysis. Gut 70: 807–809. doi:10.1136/gutjnl-2020-322072
    OpenUrlFREE Full Text
  29. ↵
    1. Yip TCF,
    2. Lui GCY,
    3. Wong VWS,
    4. Chow VCY,
    5. Ho THY,
    6. Li TCM,
    7. Tse YK,
    8. Hui DSC,
    9. Chan HLY,
    10. Wong GLH
    (2021) Liver injury is independently associated with adverse clinical outcomes in patients with covid-19. Gut 70: 733–742. doi:10.1136/gutjnl-2020-321726
    OpenUrlAbstract/FREE Full Text
PreviousNext
Back to top
Download PDF
Email Article

Thank you for your interest in spreading the word on Life Science Alliance.

NOTE: We only request your email address so that the person you are recommending the page to knows that you wanted them to see it, and that it is not junk mail. We do not capture any email address.

Enter multiple addresses on separate lines or separate them with commas.
A prediction model for COVID-19 liver dysfunction in patients with normal hepatic biochemical parameters
(Your Name) has sent you a message from Life Science Alliance
(Your Name) thought you would like to see the Life Science Alliance web site.
CAPTCHA
This question is for testing whether or not you are a human visitor and to prevent automated spam submissions.
Citation Tools
Nomogram for COVID-19 liver dysfunction
Jianfeng Bao, Shourong Liu, Xiao Liang, Congcong Wang, Lili Cao, Zhaoyi Li, Furong Wei, Ai Fu, Yingqiu Shi, Bo Shen, Xiaoli Zhu, Yuge Zhao, Hong Liu, Liangbin Miao, Yi Wang, Shuang Liang, Linyan Wu, Jinsong Huang, Tiannan Guo, Fang Liu
Life Science Alliance Oct 2022, 6 (1) e202201576; DOI: 10.26508/lsa.202201576

Citation Manager Formats

  • BibTeX
  • Bookends
  • EasyBib
  • EndNote (tagged)
  • EndNote 8 (xml)
  • Medlars
  • Mendeley
  • Papers
  • RefWorks Tagged
  • Ref Manager
  • RIS
  • Zotero
Share
Nomogram for COVID-19 liver dysfunction
Jianfeng Bao, Shourong Liu, Xiao Liang, Congcong Wang, Lili Cao, Zhaoyi Li, Furong Wei, Ai Fu, Yingqiu Shi, Bo Shen, Xiaoli Zhu, Yuge Zhao, Hong Liu, Liangbin Miao, Yi Wang, Shuang Liang, Linyan Wu, Jinsong Huang, Tiannan Guo, Fang Liu
Life Science Alliance Oct 2022, 6 (1) e202201576; DOI: 10.26508/lsa.202201576
Twitter logo Facebook logo Mendeley logo
  • Tweet Widget
Issue Cover

In this Issue

Volume 6, No. 1
January 2023
  • Table of Contents
  • Cover (PDF)
  • About the Cover
  • Masthead (PDF)
Advertisement

Jump to section

  • Article
    • Abstract
    • Introduction
    • Results
    • Discussion
    • Materials and Methods
    • Data Availability Statement
    • Acknowledgements
    • References
  • Figures & Data
  • Info
  • Metrics
  • Reviewer Comments
  • PDF

Subjects

  • Medical Research

Related Articles

  • No related articles found.

Cited By...

  • No citing articles found.
  • Google Scholar

More in this TOC Section

  • HU modulates thiol–disulfide homeostasis
  • GSK3A, a proviral host factor for HAdV-B7 replication
  • PSME3 regulates myogenesis
Show more Research Article

Similar Articles

EMBO Press LogoRockefeller University Press LogoCold Spring Harbor Logo

Content

  • Home
  • Newest Articles
  • Current Issue
  • Archive
  • Subject Collections

For Authors

  • Submit a Manuscript
  • Author Guidelines
  • License, copyright, Fee

Other Services

  • Alerts
  • Bluesky
  • X/Twitter
  • RSS Feeds

More Information

  • Editors & Staff
  • Reviewer Guidelines
  • Feedback
  • Licensing and Reuse
  • Privacy Policy

ISSN: 2575-1077
© 2025 Life Science Alliance LLC

Life Science Alliance is registered as a trademark in the U.S. Patent and Trade Mark Office and in the European Union Intellectual Property Office.