Journal of Infectious Diseases Advance Access published December 29, 2014

MAJOR ARTICLE

Predicting Hepatitis B Virus (HBV) Surface Antigen Seroclearance in HBV e Antigen– Negative Patients With Chronic Hepatitis B: External Validation of a Scoring System Downloaded from http://jid.oxfordjournals.org/ at D H Hill Library - Acquis Dept S on January 1, 2015

Jessica Liu,1,a Tai-Chung Tseng,2,9,a Hwai-I Yang,1,10,11 Mei-Hsuan Lee,3 Richard Batrla-Utermann,13 Chin-Lan Jen,1 Sheng-Nan Lu,12 Li-Yu Wang,4 San-Lin You,1 Pei-Jer Chen,5,7 Chien-Jen Chen,1,8 and Jia-Horng Kao5,6,7 1

Genomics Research Center, Academia Sinica, 2Division of Gastroenterology, Department of Internal Medicine, Taipei Tzuchi Hospital, The Buddhist Tzuchi Medical Foundation, 3Institute of Clinical Medicine, National Yang Ming University, 4MacKay College of Medicine, 5Division of Gastroenterology and Hepatology, Department of Internal Medicine, 6Hepatitis Research Center, 7Graduate Institute of Clinical Medicine, National Taiwan University College of Medicine, 8Graduate Institute of Epidemiology and Preventative Medicine, College of Public Health, National Taiwan University, Taipei, 9School of Medicine, Tzu Chi University, Hualien, 10Molecular and Genomic Epidemiology Center, China Medical University Hospital Taichung, 11Graduate Institute of Clinical Medical Science, China Medical University, Taichung, and 12Department of Gastroenterology, Chang-Gung Memorial Hospital, Kaohsiung, Taiwan; and 13Roche Diagnostics, Basel, Switzerland

Background. Hepatitis B virus (HBV) surface antigen (HBsAg) seroclearance is the ultimate serological end point in chronic hepatitis B. This study aimed to develop and validate a prediction score for spontaneous HBsAg seroclearance in HBV e antigen (HBeAg)-negative patients with chronic hepatitis B due to HBV genotype B or C. Methods. The development cohort included 2491 untreated participants from the community-based REVEALHBV study, who were HBeAg negative, anti–hepatitis C virus negative, and cirrhosis free. The independent validation cohort consisted of 1934 hospital-based individuals from the National Taiwan University Hospital. Clinical markers included in the model were age and serum HBV DNA and HBsAg levels. Cox proportional hazards regression models were used to create the prediction model. Results. A prediction score ranging from 0 to 27 was developed. Predicted probabilities of 5- and 10-year HBsAg seroclearance ranged from 0.95% to 30.49% and from 2.58% to 62.52%, respectively. When applied to the independent validation cohort, the areas under the receiver operating characteristic curves for the 5- and 10-year prediction of HBsAg seroclearance in the validation cohort were 0.82 (95% confidence interval [CI], .76–.88) and 0.74 (95% CI, .70–.78). Model fit was still adequate, according to Hosmer–Lemeshow goodness of fit tests. Conclusions. A clinically applicable prediction score for HBsAg seroclearance was developed and externally validated. This model can assist clinicians in further stratifying risk groups. Keywords.

epidemiology; viral hepatitis; prediction model.

Chronic hepatitis B is a global public health threat because of its widespread distribution and potential to cause adverse clinical outcomes such as cirrhosis, hepatic decompensation, hepatocellular carcinoma

Received 15 September 2014; accepted 12 November 2014. a J. L. and T.-C. T. contributed equally to this report. Correspondence: Chien-Jen Chen, ScD, Genomics Research Center, Academia Sinica, 128 Academia Rd, Section 2, Taipei 115, Taiwan ([email protected]). The Journal of Infectious Diseases® © The Author 2014. Published by Oxford University Press on behalf of the Infectious Diseases Society of America. All rights reserved. For Permissions, please e-mail: [email protected]. DOI: 10.1093/infdis/jiu659

(HCC), and even death [1–3]. The natural history of chronic hepatitis B is typically defined by several chronological phases, which are characterized by changes and interactions among several key seromarkers, including alanine aminotransferase (ALT), hepatitis B virus (HBV) surface antigen (HBsAg), HBV e antigen (HBeAg), anti-HBe, and HBV DNA [4, 5]. For patients with chronic hepatitis B, particularly those who are negative for HBeAg, HBsAg seroclearance has been well established as the ideal and ultimate treatment and clinical end point, as it confers an excellent prognosis and leads to extremely low rates of HCC, as long as there are no confounding factors at the time

A Prediction System for HBsAg Seroclearance

•

JID

•

1

METHODS Study Cohorts

The cohort of patients used to establish the prediction model (the development cohort) was a subset of the well-documented community-based REVEAL-HBV study, which has been previously described [26, 27]. A total of 2491 individuals aged 30–65 years at baseline who were HBsAg-positive, HBeAg-negative, anti-HCV negative, and free of liver cirrhosis at study entry were recruited in this study during 1991–1992. All individuals had measurements of HBV DNA and quantitative HBsAg at baseline. Participants did not receive antiviral treatment during the study period, as antiviral therapy was not reimbursed by the national health insurance for the majority of the study period and was only available under stringent criteria 2

•

JID

•

Liu et al

to higher-risk patients at the end of the study period. Participants were followed for a median duration of 10.1 years (interquartile range [IQR], 8.2 years). The independent cohort of patients used to validate the prediction model (the validation cohort) was from the hospitalbased ERADICATE-B cohort at the National Taiwan University Hospital in Taipei, Taiwan [22, 28]. This cohort consists of individuals who were older than 16 years at recruitment, in 1985, and who were followed up regularly for >3 years. To be consistent with the development cohort, a subset total of 1934 individuals aged 30–65 years at baseline who were also HBsAg positive, HBeAg negative, anti-HCV negative, and free of liver cirrhosis at study entry were included in the validation cohort. These patients also did not receive antiviral treatment during the study period and were followed for a median duration of 12.3 years (IQR, 5.1 years). Any patients who received antiviral therapy were excluded. If any antiviral therapy was indicated during the study period, patients were censored at the date when therapy began. Patients from both cohorts were either infected with HBV perinatally through vertical transmission or were infected during early childhood, which is often the case in Taiwan. All participants for this study provided written informed consent. This study was approved by the Institutional Review Board of the College of Public Health, National Taiwan University, Taipei, Taiwan. Definition of HBsAg Seroclearance

Follow-up in the development cohort was performed every 6– 12 months, during which participants were tested for HBV DNA, HBsAg and HBeAg serostatus, ALT and aspartate aminotransferase (AST) levels, and α-fetoprotein. HBsAg seroclearance in the development cohort was defined as the first instance when a participant tested seronegative for HBsAg and stayed persistently seronegative at each test afterward. For the validation cohort, liver biochemical tests, α-fetoprotein testing, and abdominal echography were performed every 3–6 months. Serum HBsAg levels were determined retrospectively. HBsAg seroclearance was defined as at least 2 consecutive tests, at least 1 year apart, that found an HBsAg level of 50 years [6–8]. Clearance of HBsAg, both spontaneous and as a result of antiviral treatment, however, is rare [7, 9–13]. Therefore, fully determining the determinants of HBsAg seroclearance and developing a useful prediction model is clinically important. Previously, low HBV DNA level was seen as the best predictor of HBsAg seroclearance [9]. However, technological advances allowing for the rapid and inexpensive quantification of HBsAg levels spurred many studies examining the role of quantitative HBsAg, which was then suggested as a promising marker of immunological response to antiviral therapy [14–19]. In addition, serum HBsAg levels also complement HBV DNA in predicting liver disease progression to liver cirrhosis and HCC and determining inactive carrier status, and they are strongly associated with HBsAg seroclearance, particularly in individuals with low viral loads [19–24]. Recently, we published a simple score-based system for predicting spontaneous HBsAg seroclearance that incorporated well-known predictors of HBsAg seroclearance, including age, smoking habit, body mass index (BMI), serum HBV DNA level, and HBsAg level [25]. In this study, the addition of quantitative HBsAg levels into previous viral load–based models was able to significantly improve the 5- and 10-year predictive accuracy of spontaneous HBsAg seroclearance in HBeAg-negative individuals with HBV genotype B or C infection [25]. Such a model creates a probability distribution for HBsAg seroclearance that takes into account the differing clinical profile of each patient, and it may potentially be useful to further stratify risk groups and refine clinical treatment of patients with chronic hepatitis B. However, although useful, this model has not been externally validated. Therefore, this study aimed to redevelop and validate a clinically applicable tool for predicting spontaneous HBsAg seroclearance in HBeAg-seronegative patients with chronic hepatitis B due to HBV genotype B or C infection.

and precore mutations were also not included, as they were not previously shown to influence HBsAg seroclearance rates and were also not available in the validation cohort. Measurements of and data on all included predictive factors were collected at study baseline in both development and validation cohorts.

Table 1. Baseline Characteristics of the Development and Validation Cohorts

Statistical Analysis

HBsAg seroclearance Overall, cases, no.

exp ðbage  SCORE 

^p ¼ 1  S 0

Pp

bMÞ i¼1 i i

;

where S0 is the baseline 5- or 10-year disease-free probability, βage is the regression coefficient for a 10-year increase in age, SCORE is the total sum of individual prediction scores, βi is the regression coefficient for the ith variable, and Mi is the mean level for the ith variable. Model validation was performed by applying the prediction scores and calculating a cumulative score for each patient in the validation cohort. Model discrimination was assessed using time-dependent receiver operating characteristic (ROC) curves, with each possible score as a cutoff point, and by examining sensitivity, specificity, and the area under the ROC (AUROC) for the 5- and 10-year prediction of spontaneous HBsAg seroclearance [30]. Calibration of the prediction model was assessed by comparing the mean predicted 5- and 10-year probability of seroclearance with the actual observed probability of HBsAg seroclearance among individuals with the same total prediction score in the validation cohort, as calculated with the Kaplan– Meier method, using a bootstrap method. However, if the number of HBsAg cases among a group with a particular score was smaller, groups were combined to form larger groups with equal distributions of cases. Model fit was assessed using the Hosmer–

Characteristic

523

Incidence, cases/ 261.76 10 000 personyears Follow-up duration, person-years Overall

Validation Cohort (n = 1934)

P Valuea

314 132.34