Prediction Model of HBsAg Seroclearance in Patients with Chronic HBV Infection

Background Prediction of HBsAg seroclearance, defined as the loss of circulating HBsAg with or without development of antibodies for HBsAg in patients with chronic hepatitis B (CHB), is highly difficult and challenging due to its low incidence. This study is aimed at developing and validating a nomogram for prediction of HBsAg loss in CHB patients. Methods We analyzed a total of 1398 patients with CHB. Two-thirds of the patients were randomly assigned to the training set (n = 918), and one-third were assigned to the validation set (n = 480). Univariate and multivariate analysis by Cox regression analysis was performed using the training set, and the nomogram was constructed. Discrimination and calibration were performed using the training set and validation set. Results On multivariate analysis of the training set, independent factors for HBsAg loss including BMI, HBeAg status, HBsAg titer (quantitative HBsAg), and baseline hepatitis B virus (HBV) DNA level were incorporated into the nomogram. The HBsAg seroclearance calibration curve showed an optimal agreement between predictions by the nomogram and actual observation. The concordance index (C-index) of nomogram was 0.913, with confirmation in the validation set where the C-index was 0.886. Conclusions We established and validated a novel nomogram that can individually predict HBsAg seroclearance and non-seroclearance for CHB patients, which is clinically unprecedented. This practical prognostic model may help clinicians in decision-making and design of clinical studies.


Introduction
HBV infection continues to be a global health problem. Worldwide, around 2 billion people have evidence of past or present infection with HBV and an estimated 257 million are chronically infected [1]. Almost half of the world's population resides in areas of high HBV endemicity, with the highest prevalence in Africa and East Asia. In addition, in China, approximately 300 million people suffer from hepatopathy, having a major impact on the global burden of liver diseases [2]. Patients with chronic HBV infection have an increased risk of developing sequelae such as cirrhosis and hepatocellular carcinoma (HCC). Chronic infection is characterized by the persistence of hepatitis B virus surface antigen (HBsAg) for at least 6 months (with or without concurrent hepatitis B virus e-antigen (HBeAg)). Persistence of HBsAg is a surrogate marker for the risk of developing chronic liver disease and HCC. Recent studies have focused Hindawi BioMed Research International Volume 2020, Article ID 6820179, 7 pages https://doi.org/10.1155/2020/6820179 on the role of HBsAg quantification in seroclearance of HBsAg which usually indicates that HBV infection has been cured [3].
Nomograms are widely used as prognostic devices in medicine, especially for individualized estimation of cancer survival. With the ability to generate the probability of a clinical event by integrating diverse prognostic and determinant variables, nomograms can meet our need for biologically and clinically integrated models and fulfill our drive towards personalized medicine [4]. Current evidence suggests that the occurrence of HBsAg seroclearance in patients with chronic HBV infection is a rare event that occurs at 1% to 2% per year, usually after a long duration of sustained biochemical remission [5], and its probability to be forecasted is rarely known. In this study, we sought to develop a clinical nomogram for predicting the rate of HBsAg loss of patients with CHB.

Study Design and Patients.
From 2009.1 to 2018.6, a total of 3220 patients were diagnosed with CHB and were followed up every 3 to 6 months by the Infectious Diseases Department of The Third Affiliated Hospital of Sun Yat-sen University, Guangzhou, China. All patients had been HBsAg positive for more than 6 months and were excluded for at least one of the following conditions: lost to follow-up for over 12 months, presence of comorbidities (hepatitis A/C/E virus coinfection, autoimmune liver diseases, other malignant tumors, renal insufficiency, hepatolenticular degeneration, and alcoholic liver disease), having received immunosuppressive (transplantation) therapy, and cases of data loss. All enrolled patients signed informed consent.

Variables and Data Collection.
We recorded the following data for each patient: gender, age, body mass index (BMI), alcohol history, family history, diagnosis, treatment, and other laboratory indexes, such as alanine aminotransferase (ALT), aspartate aminotransferase (AST), total bilirubin, and albumin (ALB). The reference ranges of the biochemical index are as follows: ALT: 3-35 U/l; AST: 15-40 U/l; total bilirubin: 4.0-23.9 μmol/l; and ALB: 36.0-51.0 g/l. Furthermore, we also collected the following serological and virological markers: hepatitis B virus surface antigen (HBsAg), hepatitis B virus surface antibody (HBsAb), hepatitis B virus e-antigen (HBeAg), hepatitis B virus E antibody (HBeAb), and hepatitis B virus core antibody (HBcAb), measured by chemiluminescence immunoassay technique. HBsAg loss was defined as two consecutive HBsAg titers < 0:05 IU/ml, measured with the Elecsys HBsAg II Quant kits (Germany). The baseline HBV load was measured by nucleic acid fluorescent quantitative polymerase chain reaction (PCR), and the chemicals were purchased from DAAN GENE (Guangzhou, China). The lower limit of HBV DNA detection was 100 IU/ml. Logarithmic transformation of quantitative HBsAg values and baseline HBV DNA values were finally obtained.

Definitions.
Chronic hepatitis B was defined as follows: (1) HBsAg present for ≥6 months, (2) serum HBV DNA varies from undetectable to several billion IU/ml, and (3) normal or elevated ALT and/or AST levels [6]. HCC was diagnosed by at least two imaging studies (i.e., hepatic ultrasound together with CT, MRI, or both), and most cases were histopathologically confirmed according to the AASLD (2018) guidelines [6].

Statistical Analyses.
Frequency and percentage (%) were used to describe categorical variables, median and interquartile range were used for non-parametric continuous variables. Mann-Whitney U test and χ 2 test were used for intergroup differences. COX regression analysis was used to analyze both the univariate-and multivariate-adjusted rate ratios (with 95% confidence intervals) of HBsAg loss. Variables significant in univariate analyses were included in multivariate analyses. The statistical analysis was carried out using IBM SPSS 25.0, the p value was taken bilaterally, and p < 0:05 indicated statistically significant difference.
The nomogram was built based on the results of multivariate analyses of BMI, quantitative HBsAg and HBeAg status in the primary cohort, and baseline HBV DNA values. SAS 9.4 was used to randomly divide both the HBsAg seroclearance and nonseroclearance groups into a 1 : 2 ratio: two-thirds of cases used as training sets and the remaining one-third as validation sets. R 3.5.1 (http://www.r-project .org/) was used for constructing the nomograms with the survival using rms, grid, and ggplot2 graphics package. Nomogram validation consisted of discrimination and calibration by using the validation set. Discrimination and predictive performance of the nomogram were evaluated using a concordance index (C-index). C-index values range from 0.5 to 1.0, with 0.5 indicating random chance and 1.0 indicating a perfect ability to correctly discriminate the outcome using the nomogram [7]. p value < 0.05 was considered statistically significant. The calibration curve was derived based on regression analysis.

Clinical Data Characteristics.
Finally, 1398 CHB patients were enrolled in this study (Figure 1), and the cohort was divided into two sets: the training set (n = 918, HBsAg loss cases: 35) and validation set (n = 480, HBsAg loss cases: 22). After comparison between the two sets of data, there was no statistical difference regarding index variables (all the p values were >0.05). Long-term follow-up cohorts for the training and validation sets showed male predominance (72.77%; 77.08%), with differences found in median age (34 years; 35 years), median BMI (21.80; 22.04), report of familial history of HBV (60.57%; 61.25%), and median baseline HBV DNA (4.750 log 10 IU/ml; 4.197 log 10 IU/ml). AST was within the normal range (median value: 36 U/l) while ALT was mildly elevated (median value: 38 U/l) in both sets. Besides, the median follow-up time was 75 months in both sets (Table 1). In addition, the clinical features of patients with HBsAg loss are shown in Supporting Information Table S-1.

Prognostic Nomogram for HBsAg Loss of CHB.
A nomogram that incorporated the above-mentioned significant prognostic factors and baseline HBV DNA load, another important factor reported in the literature, was established ( Figure 2). The prognostic nomogram for HBsAg loss of CHB patients showed that variable "quantitative HBsAg" substantially contributed to prognosis, followed by BMI, HBeAg status, and baseline HBV DNA load. Harrell's Cindex for HBsAg loss prediction was 0.913 (95% CI, 0.868 to 0.958). Each subtype within the above-mentioned variables was assigned a score on the point scale. By summing up the total score and locating it on the total point scale, it was easy to draw a straight line down to determine the estimated probability of HBsAg non-seroclearance at each score point.

Discussion
Despite the introduction of an effective HBV vaccine decades ago, the burden of chronic HBV infection remains a public health concern, particularly in endemic regions of Asia and sub-Saharan Africa [8]. HBsAg loss is regarded as a positive achievement in the natural history, particularly if it occurs before the accrual of significant liver disease and is deemed as a functional "cure." However, HBsAg is infrequently cleared in CHB patients [9]. Chu and Liaw [5] reported that the predictive factors for HBsAg seroclearance can be divided into host factors including age, gender, normal alanine aminotransferase levels, and viral factors including HBeAg negativity at baseline, HBV DNA negativity by hybridization at baseline, genotype, and hepatitis C virus superinfection. A systematic review and meta-analysis of HBsAg clearance rates and predictors of clearance by Yeo et al. [10] showed that favorable factors for HBsAg loss included HBeAg seronegativity, low quantitative HBsAg values, and low HBV load at baseline. In our study, the independent risk factors of   [11,12]. Based on such findings, it can be inferred that BMI may be a contributing factor for HBsAg seroclearance. A nomogram is a simple graphical representation of a prediction model that generates a numerical probability of a clinical event and can be an important component of modern medical decision-making, if carefully constructed to answer a focused question and appropriately interpreted and applied [4]. It is a powerful unprecedented tool that can be harnessed to predict individual outcomes, in this case seroclearance. In the field of hepatopathies, nomograms are widely used in HCC [13] and acute on chronic liver failure (ACLF) [14]. Based on the factors mentioned above, the regression model can calculate the probability of target events in a certain period of time, such as 3, 5, and 10 years. At present, to our knowledge, there is no literature regarding any validated model that can reliably predict HBsAg loss. We envisaged the possibility of constructing a nomogram predicting HBsAg seroclearance as our clinical endpoint, incorporating factors which were proven to be independent by multivariate Cox regression analysis. Antiviral therapy for CHB patients is a long process that often leads to poor patient compliance. It has become a major conundrum for clinicians to estimate the probability of HBsAg loss and Nonetheless, our nomogram is limited by the retrospective nature of data collection and other shortcomings of our study design that affect its robustness and reliability. Firstly, HBsAg seroclearance is a low probability event, and the number of patients with HBsAg loss enrolled in our study was relatively small. Secondly, both the training and validation sets came from our follow-up cohort. Indeed, the presence of an external validation group would have improved the clinical value of our nomogram. Lastly, differences in antiviral therapy regimens and the HBV genotype, which  Figure 2: Nomogram predicting the probability of HBsAg non-seroclearance at 3, 5, and 10 years, using BMI, quantitative HBsAg (qHBsAg), HBeAg status, and baseline HBV DNA. To use the nomogram, an individual patient's value is located on each variable axis, and a line is drawn upward to determine the number of points received for each variable value. The sum of these numbers is located on the total points axis, and a line is drawn downward to the non-seroclearance axes to determine the likelihood of 3-year, 5-year, and 10-year non-seroclearance. BMI: body mass index; qHBsAg: quantitative hepatitis B virus surface antigen; HBeAg: hepatitis B virus e-antigen (0 and 1 represent negative and positive status, respectively); baseline HBV DNA: baseline hepatitis B viral deoxyribonucleic acid load.  Figure 3: The calibration curves for predicting HBsAg seroclearance for the training cohort (a) and the validation cohort (b). Nomogrampredicted probability of HBsAg seroclearance is plotted on the x-axis; actual HBsAg seroclearance is plotted on the y-axis.