Data were analyzed using SPSS (version 25.0; SPSS, Inc., Chicago, IL), R software (4.0.3; R Foundation for Statistical Computing, Vienna, Austria), and SAS (9.4; SAS Institute Inc., Cary, NC). Continuous variables were expressed as mean ± SD or median (interquartile range [IQR]), as appropriate, while categorical variables were presented as frequency (percentage). Qualitative and quantitative differences between groups were analyzed by chi‐squared test or Fisher’s exact tests for categorical parameters and Student t test, Mann‐Whitney U test, one‐way ANOVA, or Kruskal‐Wallis test for continuous parameters, as appropriate. Median ALT level before and after HBV antiviral treatment and corticosteroid therapy between groups was compared by Quade’s analysis of covariance (ANCOVA).

Propensity score (PS), the conditional probability of having current HBV infection, was estimated among three groups of patients with current, past, and no HBV infections, to control for confounders and reduce selection bias.( 21 , 22 ) Twenty‐one clinical characteristics were included in the PS. We developed PS by generalized boosted models (GBMs) to capture nonlinear effects and interaction terms. GBM has been shown to provide less prediction error and more stable weights than logistic regression.( 23 , 24 , 25 ) The four stopping rules, namely the mean and maximum of the absolute standardized mean difference (ASMD) and of the Kolmogorov‐Smirnov statistic, were adopted to determine the optimal iteration of GBM. The stopping rule with overall the best balance of clinical characteristics and effective sample size was selected.( 26 ) In the inverse probability of treatment weighting (IPTW) analysis, we applied average treatment effect on the treated weighting, so the baseline clinical characteristics of patients with past or no HBV infection had nearly identical distributions after IPTW to those with current HBV infection.( 25 ) The balance of baseline clinical characteristics between patients was assessed by ASMD; an ASMD < 0.2 indicated a good balance.( 23 , 27 )

Before PS estimation, missing data were assumed missing at random and replaced by multiple imputation by chained equations to create 20 complete data sets after 10 initial burn‐in iterations.( 28 , 29 ) The imputed baseline variables (missing percentage) were serum creatinine (0.1%), albumin (0.1%), ALT (0.1%), total bilirubin (0.1%), ALP (0.1%), lactate dehydrogenase (1.1%), C‐reactive protein (1.0%), hemoglobin (0.04%), white cell counts (0.04%), lymphocyte (0.4%), neutrophil (0.4%), and platelet (0.04%). The variables included in the imputation models were all covariates included in PS estimation, occurrence of mortality, and the corresponding Nelson‐Aalen estimator of the cumulative hazard at the time of event or censoring.( 30 ) All imputed values were constrained within plausible ranges.

HRs and adjusted HRs (aHRs) with 95% CIs of current or past HBV infection referenced to no HBV infection on the primary endpoint were estimated by Cox proportional hazards regression. We adjusted for patients’ demographic, presence of acute liver injury, liver cirrhosis, comorbidities, and other relevant laboratory parameters, as shown in Supporting Table S4; backward stepwise elimination was performed to select statistically significant covariates. Weighted Cox proportional hazards regression was used in PS weighting and matching analysis. Clinical characteristics with ASMD ≥ 0.2 after PS balancing were adjusted in the weighted Cox model for double robustness. Robust (empirical) variance estimates were obtained to calculate 95% CIs.( 31 ) The overall coefficient estimates and standard errors were computed by combining the estimates obtained on each individual multiple imputation data set using Rubin’s rules.( 32 ) Schoenfeld residual plots were used to assess the proportional hazards assumption, which did not detect any significant violations.

ORs and adjusted ORs (aORs) with 95% CIs for acute liver injury were estimated by logistic regression. We included the following covariates: HBV exposure (current, past, or no HBV infection), age, gender, presence of cirrhosis and DM, and use of corticosteroids, remdesivir, interferon‐beta, ribavirin, lopinavir‐ritonavir, antibiotics, or antifungals during hospitalization; none were excluded in this analysis due to missing data. Significant covariates were selected by backward stepwise elimination. Goodness of fit was assessed by the Hosmer‐Lemeshow goodness‐of‐fit test. All statistical tests were two‐sided. Statistical significance was taken as P < 0.05.

Note: The content above has been extracted from a research article, so it may not display correctly.

Please log in to submit your questions online.
Your question will be posted on the Bio-101 website. We will send your questions to the authors of this protocol and Bio-protocol community members who are experienced with this method. you will be informed using the email address associated with your Bio-protocol account.

We use cookies on this site to enhance your user experience. By using our website, you are agreeing to allow the storage of cookies on your computer.