ACRT Journal

Introduction

Chronic hepatitis B (CHB) is a major global health threat induced by hepatitis B virus (HBV) infection, and it results in significant worldwide socio-economic burdens [1]. Liver fibrosis is an inevitable stage of CHB development. The timely and accurate assessment of liver fibrosis, along with effective antiviral therapies, are essential to reduce the risk of cirrhosis, hepatic failure, and hepatocellular carcinoma (HCC) [2,3]. To improve the quality of life for CHB patients, it is crucial to screen the population vulnerable to liver decompensation-related morbidity and adopt beneficial interventions [4].

Several serological biomarkers can indicate the infection status of HBV infection. Hepatitis B core antibody (HBcAb) seropositivity reveals a previous HBV infection status [5]. Studies have shown that individuals with positive-HBcAb have a higher prevalence of advanced liver fibrosis or cirrhosis than the general population, highlighting the importance of liver fibrosis screening in these individuals [6-8]. Therefore, patients with HBcAb positivity require careful monitoring to prevent liver disease deterioration.

As for screening the liver fibrosis in CHB, liver biopsy is considered the gold standard for diagnosing and evaluating liver fibrosis. However, liver biopsy has several deficiencies, such as invasiveness, compliance, sampling error, and side effects [9,10]. Due to the invasive nature of the procedure and the numerous complications that can arise, repeated liver biopsy is impractical for screening liver fibrosis in patients [11,12]. Therefore, noninvasive tests are potential alternative methods for identifying liver fibrosis. These methods can efficiently monitor the dynamic progress of liver disease without any inevitable side effects [11]. Serological markers and complex models, such as platelets and hyaluronic acid, the aspartate transaminase to platelet ratio (APRI), fibrosis-4 score (FIB-4), γ-glutamyl trans peptidase to platelet ratio (GPR), Hui score, Forms Index, and King’s Score, appear to be reliable non-invasive methods for early and rapid assessment of liver fibrosis [13–16].

The imaging test also plays a crucial role in screening for liver fibrosis. In particular, vibration-controlled transient elastography (VCTE) has been recommended as a validated non-invasive standard for assessing liver fibrosis in patients with HBV infection [17–19]. This method mainly relies on ultrasound technology and can measure liver stiffness, which can reflect the severity of fibrosis [20,21]. Patients strongly prefer VCTE over liver biopsy, which is widely used in clinical practice to contribute significantly to the invasive screening of liver fibrosis [20,21].

However, most non-invasive models for predicting liver fibrosis focused on patients with CHB who had tested positive for hepatitis B surface antigen (HBsAg) or hepatitis B surface antibody (HBsAb). These models rarely considered the risk of liver fibrosis in individuals with HBcAb and have thus ignored the necessity of early screening and prevention of advanced fibrosis in these patients. Therefore, we developed a novel prediction model for liver fibrosis in individuals with seropositive HBcAb and created a visually representative nomogram of the model. We also compared our model’s efficacy for predicting liver fibrosis with other noninvasive models.

Materials and Methods

Data Source and Study Population

Our study collected data from the National Health and Nutrition Examination Survey (NHANES), which tracks the health and nutritional status of adults and children in the United States of America over time [22]. Before participating in the NHANES, all individuals gave written informed consent. Before participating in the NHANES, all individuals provided written informed consent. Our study complied with the TRIPOD (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis) statement for reporting [23], and was conducted in accordance with the principles of the Helsinki Declaration.

NHANES data from cycle 2017.01-2020.03 were collected, and subjects with HBcAb seropositivity were enrolled in our study. A total of 641 subjects had seropositive HBcAb, but 34 subjects without valid VCTE data and two subjects under the age of 18 were excluded from our cohort. To investigate the correlation between liver fibrosis and various risk factors in patients with HBcAb seropositivity, participants with other risk factors for liver fibrosis, such as different hepatitis virus infections and obesity, were not excluded. Eventually, 605 individuals were enrolled. The number of outcome events per variable (EPV) in our study was greater than 10, which met the standard criterion for prediction model construction [24]. The enrolment process is presented in Figure 1.

Clinical Characteristics and Laboratory Tests

The definition of liver fibrosis in our study was based on the median liver stiffness (MS) obtained via VCTE (FibroScan®), with a threshold of MS ≥ 8.4 kPa indicating significant fibrosis. Various variables, including demographics (age, sex, race, and socioeconomic status), anthropometrics (hip circumference (HC), waist circumference (WC), arm circumference (AC), and body mass index (BMI)), lifestyles (alcohol consumption, smoking, and insomnia), comorbidities (hypertension, diabetes, and cardiac function), and laboratory examinations such as hemoglobin (HB), platelets (PLT), white blood cell (WBC), lymphocytes (LYM), neutrophils (NEU), monocytes (MON), alanine aminotransferase (ALT), aspartate aminotransferase (AST), γ-glut amyl trans peptidase (GGT), alkaline phosphatase (ALP), blood urea nitrogen (BUN), creatinine (CRE), albumin (ALB), globulin (GLB), triglycerides (TG), total cholesterol (TC), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), total bilirubin (TB), uric acid (UA), glycosylated hemoglobin, and high-sensitivity C-reactive protein (hs-CRP), were collected from NHANES database. Additionally, variables related to the immunity or infection of various hepatitis viruses, such as HBV vaccine status, hepatitis A antibody (HAAb), HBsAg, HBsAb, hepatitis C RNA (HCR), hepatitis D surface antibody (HDAb), and hepatitis E antibody (HEAb), were collected. The NHANES laboratory method manuals described laboratory methodology, quality assurance, and monitoring. Due to the high frequency of missing values, variables including socioeconomic status, LDL-C, cardiac function, and alcohol consumption were excluded. Finally, 39 clinically relevant variables related to liver fibrosis were analysed in our study. The proportion of missing values for each included variable was less than 5%, which were filled using multiple imputation methods in SPSS 26 software.

The existing non-invasive models

Our studies involved several efficient and widely applied noninvasive models for assessing liver fibrosis, including APRI, FIB4, and GPR. These three serological indexes were compared with our prediction model and were calculated as follows: APRI = AST (U/L) / upper limit of normal (ULN) (set at 40 U/L) × 100 / platelet count (×109/L),[25] FIB-4 = age (years) × AST (U/L)／[ALT (U/L)1/2 × platelet (×109/L)],[26] and GPR = [GGT (IU/L)/ULN (set at 50 U/L)]/platelet (×109/L) [27].

Data Analysis

Our study described continuous data using the median and interquartile range (IQR), while categorical data were described using counts and percentages. Since all continuous data had an abnormal distribution, we transformed them into categorical data based on respective criteria during the development of our clinical prediction model.

For model development, all 605 subjects were randomly divided into training and validation cohorts in a 7:3 ratio (425 and 180 subjects, respectively). Logistic regression was used to analyse the correlation between each variable and MS. We selected statistically significant variables (P<0.01) from the univariable logistic regression analysis. We performed the least absolute shrinkage and selection operator (LASSO) regression. Based on the optimal lambda.1se value in the LASSO regression, we selected variables with excellent performance to construct a multivariable logistic regression. Finally, a nomogram was developed to improve the visualisation of results and facilitate clinical application. We evaluated the odds ratio (OR) and 95% confidence interval (CI) for each variable in our model.

Regarding model validation, we performed receiver operating characteristic curve (ROC) analysis and calculated the areas under the ROC curve (AUROC). We used the net reclassification index (NRI) and integrated discrimination improvement (IDI) to compare our model with other models [28,29]. Additionally, we used the Brier score to assess the nomogram’s performance, with a lower score indicating better predictive ability. Decision curves were plotted to display the nomogram’s net benefit, reflecting the clinical benefits provided by the prediction model.

All statistical significance levels in our study were two-sided. The statistical analyses were carried out using R 4.2.2 software.

Results

Baseline demographic and clinical characteristics

Table 1 describes the baseline characteristics of our participants. The median age was 60.0 years. Based on the MS provided by VCTE, liver fibrosis was detected in 79 individuals (13.1%). Among these participants, 47 (59.5%) were male and 32 (40.5%) were female.

We randomly divided 425 participants into a training cohort and 180 into a validation cohort, using a ratio of 7:3 (Table S1). In the training cohort, 242 (56.9%) participants were male, and 58 (13.6%) had liver fibrosis. In the validation cohort, there were 96 (53.3%) male participants and 21 (11.7%) participants with liver fibrosis. The baseline characteristics of the two cohorts did not differ significantly (P > 0.05).

Identification of predictors and the construction of a nomogram

The correlation between each independent variable and MS was evaluated by univariate logistic regression in the training cohort (Table S2). Four-teen variables, including age, smoking, hypertension, ALT, AST, ALP, GGT, albumin, globulin, HDL-c, BMI, WC, HC, and hs-CRP, were found to be statistically significant (P<0.01) and were further analysed by LASSO regression (Figure 2). Among these variables, seven optimal predictors, including Age, ALT, AST, GGT, Albumin, BMI, and HC, were selected to develop a clinical prediction model for liver fibrosis through multivariate logistic regression. The logistic regression analysis results and OR values for these seven optimal predictors are shown in Table S3. These predictors were combined to develop a nomogram to quantify liver fibrosis risk (Figure 3).

Validation of the nomogram

To assess the discrimination of our model, we presented ROC curves and compared them with three other models in both training and validation cohorts. The performance of these models is exhibited in Figure 4 and Table 2.

In the training cohort, the nomogram’s AUC value for predicting liver fibrosis risk was 0.829 (95% CI 0.773-0.884), which was higher than that of APRI (0.724, 95% CI 0.650-0.798), FIB4 (0.695, 95% CI 0.619-0.772), and GPR (0.736, 95% CI 0.6690.802). The difference in AUC values between the nomogram and the other three models was statistically significant (all P < 0.05). The sensitivity and the specificity of the nomogram were 81.0% and 75.5%, respectively.

In the validation cohort, the AUC value of the nomogram was 0.782 (95% CI 0.670-0.894), while the AUC values of APRI, FIB4, and GPR were 0.660 (95% CI 0.520-0.799), 0.658 (95% CI 0.544-0.772), and 0.734 (95% CI 0.596 -0.872), respectively. The sensitivity of the nomogram was 66.7%, and the specificity was 79.9%.

Additionally, we calculated the NRI and IDI values, which showed that the nomogram had the highest NRI and IDI values among APRI, FIB4, and GPR in both training and validation cohorts, indicating that the nomogram had the best discrimination capability for identifying liver fibrosis in our participants.

The calibration plot for the probability of liver fibrosis consistently showed optimal agreement between the prediction model and actual observation in both training cohort (P = 0.328, Brier score = 0.091) and validation cohort (P = 0.877, Brier score = 0.076) (Figure S1), indicating the accurate prediction offered by our nomogram. Evaluation of the clinical utility of the nomogram

We conducted DCA to further evaluate the clinical application value of the nomogram in predicting liver fibrosis. DCA revealed that the nomogram had a better net benefit and more significant advantage compared to the other three models (Figure S2) in both training and validation cohorts.

Table 1: Baseline characteristics of participants.

Table 2: Performance of the nomogram and other existing models.

Figure 1: Flow diagram of study design.

Figure 2: Variables selection using LASSO regression. A. The ten-fold cross-validation via minimum criteria for selecting the optimal value of the tuning parameter (λ). Dotted vertical lines were drawn at the value with the minimum criteria and one standard error of the minimum criteria. B. The LASSO coefficient profiles of the 14 variables associated with liver fibrosis. A dotted vertical line was drawn at the optimal λ value identified through ten-fold cross-validation. The resulting seven predictors with non-zero coefficients were identified based on the log (λ1se) value.

Figure 3: The nomogram for predicting the risk of liver fibrosis in individuals with HBcAb seropositivity.

Figure 4: ROC curves for predicting liver fibrosis in training cohort (A) and validation cohort (B). The x-axis displayed the specificity, while the y-axis showed the sensitivity.

Table S1: Baseline characteristics of training and validation cohorts.

Table S2: The results of univariate logistic regression.

Table S3: The results of multivariate logistic regression.

Figure S1: Calibration curves for evaluating the nomogram performance in both training cohort (A) and validation cohort (B). The x-axis displayed predicted probability of liver fibrosis, while the y-axis displayed actual probability.

Figure S2: Decision curves for clinical utility of the nomogram in training cohort (A) and validation cohort (B). The x-axis displayed the threshold probability, while the y-axis displayed net benefits.