Introduction

Hepatocellular carcinoma (HCC) is a rapidly increasing and highly fatal cancer, leading to 865,000 new cases and 757,948 deaths per year. It ranks sixth in terms of cancer burden worldwide [1]. Chronic hepatitis B virus (HBV) infection is a major cause of HCC development [2]. Therefore, the World Health Organization (WHO) has urgently called for efforts to reduce HCC mortality [3]. As reported, the curative potential of treatment for HCC depends heavily on the stage at diagnosis, with substantially longer survival in patients diagnosed at an early stage than in those diagnosed at an advanced stage [4-6]. However, HCC is often not detected early because of its asymptomatic or nonspecific presentation in the early stages. Consequently, it is of great importance to deepen our knowledge of potential approaches for early identification of HCC, not only to improve individuals’ education and counselling, but also to raise awareness among high-risk groups and ultimately facilitate their enrollment in surveillance programs [7].

Because HCC occurs predominantly in high-risk populations, can be detected early, and may be amenable to curative treatment when diagnosed at an early stage, it is considered a malignancy well suited for screening [8]. Accordingly, in recent decades, considerable efforts have been made to develop novel and effective scoring systems in order to stratify at-risk population for HCC monitoring, including the Chinese University of Hong Kong (CUHK) clinical scoring system [9]; the Risk Evaluation of Viral Load Elevation and Associated Liver Disease/Cancer-Hepatitis B Virus (REVEAL-HBV) nomograms [10]; and the Age, Diabetes, Race, Etiology of cirrhosis, Sex and Severity of liver dysfunction (ADRESS)-HCC risk models [11]. However, these models have a significant disadvantage because their applicability may be restricted to populations with specific etiologies or ethnic backgrounds. As a result, there is a need for a broadly applicable model that can be used across different etiologies and ethnic groups. To meet this demand, the aMAP risk model was developed using data from a large population and has demonstrated good discrimination in predicting HCC risk [12]. The model may be particularly useful for risk stratification in patients with cirrhosis, helping physicians identify those who may require closer monitoring Two novel models, aMAP-2 and aMAP-2 Plus, were constructed recently based on multifactorial longitudinal data and circulating free DNA profiles, which are expected to further improve the accuracy of HCC risk prediction [13]. However, considering the ease of application and technical constraints, the aMAP risk model may be more practical for routine clinical use. Currently, the aMAP risk model has been formally recommended in the Chinese national guidelines for the diagnosis and treatment of primary liver cancer as an effective risk stratification tool for HCC surveillance, and has been extensively validated in multinational cohorts across diverse ethnic populations [14].

Despite the aMAP risk model having been developed primarily in Asian populations, its time-varying predictive performance remains unclear in Chinese hospitalized CHB patients. Understanding the dynamic predictive performance of the model can help assess model stability, and also predict the dynamic risk of HCC in HBV patients, so as to conduct early screening of HCC patients. Here, we conducted a retrospective cohort study using time-dependent ROC analysis to evaluate the dynamic predictive ability of the aMAP risk model for HCC development, with the aim of clarifying the clinical utility of the aMAP risk score. To the best of our knowledge, this is the first retrospective analysis to investigate the time-dependent predictive performance of the aMAP risk score for incident HBV-related HCC in hospitalized CHB patients.

Material and methods

Study population

A hospital-based cohort was established by screening 4700 hospitalized patients with chronic HBV infection, who received treatment in Ganzhou Fifth People’s Hospital from 2016 to 2021. We placed no restrictions on patients’ age, gender, or disease stage in enrollment, but strict exclusion criteria were applied as follows: (i) patients with HCC at baseline. For patients with equivocal ultrasound findings or AFP levels above 400 ng/ml, contrast-enhanced CT or MRI was performed to confirm HCC status; those with confirmed HCC were excluded from the study; (ii) patients with presence of other malignancies, or other diseases (e.g. HCV, NAFL, or HIV infection); (iii) patients who were pregnant or nursing women; (iv) patients younger than 18 years; (v) patients who suffered from severe diseases of other systems; (vi) patients without complete clinical data or at least 1 year of follow-up; (vii) patients with diseases affecting total bilirubin levels, such as inherited enzyme defects and other hemolytic disorders; (viii) patients who had a prior history of HCC or who developed HCC within 6 months. Finally, 2914 patients with CHB were included in this study.

Treatment

All patients were receiving antiviral therapy at enrollment. The specific regimens included entecavir (ETV, n = 2722), tenofovir disoproxil fumarate (TDF, n = 118), and pegylated interferon alfa (PegIFNα, n = 74).

Data collection

A standardized data collection form was designed to extract all relevant demographic and clinical information, which was obtained for each patient through medical chart review and/or consultation with the treating physicians. Demographic variables included gender, age, smoking status, drinking status, family history of cancer, and cirrhosis. Clinical parameters at baseline included antiviral treatment status, laboratory test (albumin, total bilirubin, platelet count, AFP, prothrombin time), clinical presentations (such as ascites, encephalopathy, and gastrointestinal bleeding) and imaging examinations (such as abdominal ultrasound [US], CT or MRI), recorded at the initial visit or within the preceding 3 months. AFP levels were categorized as either above or below 20 ng/ml, the common cut-off used for HCC surveillance [15]. The aMAP score was calculated using a formula that assigns weights to five predictive variables (age, gender, albumin, total bilirubin, and platelet count), as previously described [12].

Etiological testing was used to ascertain HBV infection status, and the CHB patients were defined those with HBV infection lasting at least 6 months. Liver cirrhosis was determined mainly through imaging examinations, supplemented by evidence of clinical presentations and laboratory data. HCC diagnosis was based on liver biopsy findings and imaging results.

Follow-up

The endpoint of this study was the occurrence of HCC. After the initial hospitalization, each patient was followed up with AFP testing and ultrasonography every 6 months for further identification of HCC development. At each of these follow-up visits, a detailed record was made of the date, examination results, and clinical presentations. Time to HCC development was defined as the time from enrollment to the date of HCC diagnosis or the date of the last follow-up for patients who remained free of HCC. Patients who did not develop HCC during follow-up were censored for analysis.

Statistical analysis

Statistical analysis was carried out using R software (version 4.1.2) and IBM SPSS Statistics (version 24.0; IBM Corp., Armonk, NY, USA). The median (interquartile range [IQR]) was used to describe continuous data and the count (percentage) for categorical variables. Differences in the demographic and clinical variables between the HCC group and the non-HCC group were compared using the Wilcoxon rank-sum and chi-square tests. Univariate and multivariable Cox proportional hazards regression models were applied to evaluate the associations of patients’ characteristics and the aMAP score with HCC risk, with hazards ratios (HRs) and 95% confidence intervals (CIs) reported. The multivariable HRs were obtained by adjusting for variables that were significant in univariate analysis (p value < 0.10), including smoking status, drinking status, family history of cancer, cirrhosis, prothrombin time, and the aMAP score. The optimal cutoffs for the aMAP score were generated by X-tile plots, and were used to separate patients into low-, medium-, and high-risk groups. The Kaplan-Meier method was used to evaluate the cumulative incidence of HCC in patients of different risk groups, and the log-rank test was used to compare the groups. Receiver operating characteristic (ROC) curves were employed to assess the discriminatory ability of different variables for HCC prediction. Time-dependent ROC curves were generated using the “timeROC” package to assess discrimination at different follow-up times. A calibration plot generated to assess agreement between predicted and observed HCC probabilities, with 1,000 bootstrap resamples. All statistical tests were two-sided, and a p value <0.05 was considered statistically significant.

Results

Baseline characteristics

A total of 135 patients (4.63%) developed HCC among the 2,914 patients in the retrospective cohort during follow-up, with a median follow-up of 40.8 (range: 12.2-72.5) months. The baseline characteristics are summarized in Table 1. The majority of HCC patients were older (89.0%), male (88.9%), ever smokers (51.9%), never drinkers (60.0%), and had cirrhosis (73.3%). In addition, patients who developed HCC had lower albumin levels and platelet counts, higher aMAP scores, and longer prothrombin times than patients who did not develop HCC. Except for total bilirubin, AFP level and family history of cancer, the other variables differed significantly between the two groups (p < 0.05).

Association of patient characteristics with HCC risk in CHB patients

As shown in Table 2, several variables, including smoking status, drinking status, cirrhosis, and prothrombin time, were associated with the hazard of HCC development in univariate analysis. In multivariable analysis, drinking status and cirrhosis remained significantly associated with the hazard of HCC development, with HRs of 1.84 (95% CI: 1.27-2.66) and 1.81 (95% CI: 1.17-2.80).

Association of aMAP score with HCC risk in CHB patients

In order to evaluate the distribution of the aMAP score, we divided the patients into low-, medium-, and high-risk groups according to optimal cutoff values (60 and 70) calculated by the X-tile method and then conducted a tertile analysis to assess the association between the aMAP score and HCC risk. Notably, the two cutoffs were both higher than previously determined [12]. The results of the univariate and multivariable model in Table 3 suggested that, with the low-risk group as reference, patients in medium- and high-risk groups both had a higher hazard of HCC development. Furthermore, the hazard of HCC development increased with increasing aMAP score, with a statistically significant P value for trend. The C-indexes of the univariate and multivariable models were 0.786 and 0.879, respectively.

Consistent with these results, the cumulative incidence of HCC estimated using the Kaplan-Meier method was significantly higher in the medium- and high-risk groups than in the low-risk group (log-rank p < 0.0001; Fig. 1).

Discrimination accuracy of the aMAP score

We generated an ROC curve and calibration plots to illustrate the discriminatory ability of the aMAP score. The AUC was 0.837 (95% CI: 0.808-0.867). Figure 2 shows similar predicted and observed probabilities of HCC development during both 2-year and 5-year follow-up periods. These findings suggested that the aMAP score had good discriminatory ability for predicting HCC development.

Dynamic discrimination abilities of the aMAP score

To evaluate the time-dependent discriminatory ability of the aMAP score, we performed time-dependent ROC analysis, which showed that the AUC values increased over the follow-up period. The lowest AUC was 0.782 for HCC prediction at 24 months. These findings suggest that the aMAP score may have good discriminatory ability over longer follow-up periods. The results are shown in Figure 3.

Discussion

In the present study, we validated the predictive role of the aMAP risk score in a large, retrospectively followed cohort of HBV-related hospitalized patients, demonstrating favorable discrimination capability. Similar to previous studies, the aMAP risk score remained dominant after adjustment for other significant variants such as drinking status and cirrhosis, indicating higher hazard among CHB patients with a higher aMAP risk score by survival analysis. An effective opportunity for HCC counselling based on specific characteristics was provided, which could not only reduce excessive detection in a low-risk population, but also guide personal surveillance programs for medium-risk and high-risk stratifications. Nevertheless, the inclusive cut-offs of medium- and high-risk grouping in our study were both higher than those indicated by the previous study, which was mainly caused by the subjects included with more severe condition of disease. It implied that appropriate cut-offs would be tested at a specific population.

The convenient construction of the aMAP risk score, which incorporates three common laboratory parameters – albumin, total bilirubin, and platelet count – as well as age and male gender, facilitates its widespread clinical application in early screening. ALBI, which comprises albumin and bilirubin, was initially developed to reflect liver function [16], and platelets play an essential role in the progression of liver fibrosis [17]. These factors have been identified as useful indicators of HCC initiation in previous studies [18-20]. Consistent with our result, the aMAP risk model was also validated in a large cohort of patients with CHB-related cirrhosis, with better predictive performance than other risk models [21]. As reported, a good evaluation of an HCC risk score was typically defined by its ability to reliably identify subgroups of CHB patients with an annual HCC risk exceeding 3-5% [22]. In our study, 68 of the 668 patients with CHB in the high-risk group developed HCC during the 72.5-month follow-up period, in which the annual incidence was calculated as 3.36%. Additionally, the AUC value of the aMAP risk score was 0.837, indicating good discriminatory ability. Consequently, the aMAP risk score may be useful for risk stratification in the population studied, although its cost-effectiveness was not assessed. In addition, the aMAP risk score has been reported to have prognostic value for HCC, liver failure, and chronic liver diseases [23-25]. Furthermore, its predictive performance was reported to improve when it was combined with tumor number and ALT status [23, 26]. However, in our study, we found no clear evidence that combining the aMAP score with other significant parameters, such as cirrhosis and drinking status, improved its predictive performance.

Another major finding of our study was that the aMAP risk score demonstrated favorable discriminatory performance over time for HCC development, with AUC values increasing at later follow-up time points. Thus, the aMAP risk score may be a useful tool for informing HCC surveillance over longer follow-up periods, although its clinical utility requires further evaluation. Meanwhile, the combined use of the aMAP and ΔaMAP scores has been proposed as a potentially more effective approach, given the ease of calculating the scores [23], and this approach warrants further validation.

There were several strengths of our study. In addition to the relatively large sample size, the relative homogeneity of the well-defined CHB population reduced potential confounding factors. Another notable strength was the use of time-dependent ROC analysis to assess the discriminatory performance of the aMAP risk score over time. However, several limitations should be considered. First, bias may have arisen from the retrospective, single-center design, as well as from missing data. Second, the study population was restricted to hospitalized patients with CHB, and no external validation was performed. Further studies in other settings and populations with different etiologies are needed to assess the generalizability of the findings and to validate the optimal cutoffs for high-risk stratification and the time-dependent predictive performance of the score. Third, as HBV DNA testing was not uniformly performed in this retrospective cohort, complete data on treatment duration and virological response (HBV DNA < 20 IU/ml) were unavailable. Consequently, we could not report virological suppression rates or perform stratified analyses according to treatment duration. Lastly, the analysis included only a subset of demographic and clinical variables. Future studies could evaluate additional variables to develop and assess more comprehensive noninvasive scoring systems [27, 28].

Conclusions

A hospital-based cohort study was conducted to evaluate the discriminatory ability of the aMAP risk score for predicting HBV-related HCC development. The risk score demonstrated good discriminatory ability for stratifying HCC risk and may help inform HCC surveillance strategies. Future larger-scale, multi-center studies are needed to confirm these findings.

Disclosures

This work was supported by the Key Project of Jiangxi Provincial Natural Science Foundation (Grant No. 20244BAB28047), the Jiangxi Provincial Natural Science Foundation (Grant No. 20244BAB28030), the funds of the Medical Leading Discipline of Ganzhou City and the funds for the Training Object of Medical Academic Leader of Ganzhou City.

This study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Ganzhou Fifth People’s Hospital (approval number: GZWY-EC-2023001). Written informed consent was obtained from all participants (or their next of kin if the participants were deceased).

The authors declare no conflict of interest.

References