The aMAP score combined with plasma cfDNA methylation testing for risk stratification in hepatitis population: a case-control study

Abstract

Effective surveillance of the hepatitis populations reduces liver cancer incidence and improves prognosis. This study aimed to develop an easy and non-invasive risk assessment model for this population. A total of 392 subjects were included in the present study, comprising 72 cases of chronic hepatitis, 31 cases of liver fibrosis, 152 cases of cirrhosis and 137 cases of liver cancer from diverse ethnic groups in China. Concurrent evaluation of the methylation levels of the GNB4 and Riplet genes in plasma, alongside the age-male-albumin-bilirubin-platelet (aMAP) score and serum AFP, was performed for each subject. Head-to-head comparisons demonstrated that methylation is a superior diagnostic biomarker for liver cancer relative to AFP and aMAP scores. The sensitivity and specificity of GNB4/Riplet methylation for detecting liver cancer were 87.59% (95% confidence interval [CI]: 80.61%-92.40%) and 93.33% (95% CI: 89.25%-95.95%), respectively. For TNM stage I and II liver cancer, methylation showed sensitivities of 69.70% (95% CI: 51.13%-83.79%) and 91.18% (95% CI: 75.19%-97.69%), respectively. The aMAP score yielded an area under the curve (AUC) of 0.745 (95% CI: 0.679–0.812) for differentiating hepatitis from fibrosis/cirrhosis/cancer samples, and an AUC of 0.764 (95% CI: 0.710–0.818) for differentiating hepatitis/fibrosis from cirrhosis/cancer samples. When combining the aMAP score with methylation, the corresponding AUCs increased to 0.825 (95% CI: 0.774–0.875) and 0.867 (95% CI: 0.829–0.905), respectively. The combination of aMAP score and plasma GNB4/Riplet gene methylation shows promise as a tool for risk stratification, which needs external validation in the future.

Introduction

Primary liver cancer (PLC) is the fourth most prevalent malignant tumor worldwide. The predominant pathological type is hepatocellular carcinoma (HCC), accounting for 75%−85% of cases. The remaining types include intrahepatic cholangiocarcinoma (ICC) and combined hepatocellular-cholangiocarcinoma (cHCC-ICC)¹. China reported the highest number of liver cancer cases in 2022, accounting for over 40% of new cases worldwide^2,3. Nevertheless, the five-year survival rate of liver cancer patients in China remains extremely low at only 14.4%⁴. This is primarily due to the lack of systematic risk stratification and effective surveillance of high-risk populations to detect early-stage lesions. More than 70% of liver cancer patients in China are diagnosed at an advanced stage⁵.

In China, the leading risk factor for liver cancer is the hepatitis B virus (HBV) infection, with other risk factors including the hepatitis C virus (HCV) infection, alcoholic liver disease (ALD) and non-alcoholic fatty liver disease (NAFLD)^6–8. The progression from chronic hepatitis to liver fibrosis, cirrhosis, and liver cancer typically occurs over several years or even decades, which provides a favorable opportunity for early screening. Given China’s substantial at-risk population of over 70 million HBV carriers, risk stratification of high-risk groups and the implementation of tailored surveillance approaches for different risk levels appear to be the optimal strategy⁹.

The combined use of ultrasonography and alpha-fetoprotein (AFP) testing is currently recommended in guidelines for the screening and surveillance of liver cancer in high-risk populations. This approach plays a significant role in reducing the incidence and mortality of liver cancer^10,11. However, there is still room for improvement in the current strategies for detecting early-stage HCC. The aMAP score, which is based on age, male sex, albumin, bilirubin, and platelet count, can predict the risk of developing HCC and can be used for individuals with hepatitis caused by various factors¹². Additionally, the methylation of circulating cell-free DNA in plasma shows considerable potential for the early screening and diagnosis of HCC^13–15. For example, a combined plasma GNB4 and Riplet gene methylation assay has been shown to have an overall sensitivity of 84.4% for tumor-node-metastasis (TNM) stage I HCC¹⁵.

While prior studies have used methylation or aMAP independently to predict HCC risk, no study has combined these markers to stratify risk across the full spectrum of liver disease stages (hepatitis-fibrosis-cirrhosis-PLC) in a multi-ethnic Chinese population. This gap motivated the present investigation. This study involved a parallel comparison of three biomarker levels (the aMAP score, AFP, and GNB4/Riplet gene methylation) across different liver disease samples (hepatitis, liver fibrosis, cirrhosis, and liver cancer). The aim was to validate the individual and combined efficacy of these biomarkers in risk stratification and early detection, and to construct models for risk stratification of hepatitis populations and early liver cancer diagnosis. This approach seeks to provide new insights into surveillance high-risk populations and screening for liver cancer at an early stage in China.

Materials and methods

Participants and samples

This was a single-center, cross-sectional study conducted at the Department of Hepatology at the People’s Hospital of Qiannan. From February 2024 to November 2025, consecutive sampling was used to enroll all patients aged over 18. The study adhered to the Declaration of Helsinki guidelines. Approval for this study was granted by the hospital’s Ethics Committee (approval number 2024-qnzy-20), and informed consent was obtained from all participants. The inclusion criteria for the sample were as follows:

Individuals highly suspected of having PLC, based on imaging and/or biopsy results, who had not undergone any treatment such as surgery, radiotherapy, or ablation. Case samples were mainly included from this group of patients, who needed to exhibit typical imaging features or be diagnosed with PLC by histopathological examination. All cancer cases had to ultimately be confirmed by surgical pathology. Cancer TNM staging was performed in accordance with the 2017, 8th edition criteria of the American Joint Committee on Cancer (AJCC).

A broad spectrum of control samples were required in order to evaluate the specificity. Therefore, patients with chronic hepatitis, liver fibrosis, and cirrhosis triggered by hepatitis viruses, alcohol, obesity and other etiologies were used as control samples. As the control samples in this study were heterogeneous, analysis of confounding factors was necessary when evaluating model performance. Diagnosis of various types of liver disease was based on relevant clinical guidelines, including virology, imaging (ultrasound, computed tomography, magnetic resonance imaging and elastography), and histopathological results. Transient elastography was used to determine liver stiffness measurements (LSM) for diagnosing liver fibrosis in patients. Patients exhibiting evidence of coarse echotexture and a nodular liver surface on ultrasound, or demonstrating clinical manifestations of portal hypertension (such as splenomegaly, ascites, or varices), or presenting with typical histological changes identified through a liver biopsy were classified as having cirrhosis. Sample exclusion criteria: (1) postoperative pathology showing non-PLC in the case group, (2) the presence of other malignant tumors, (3) specimens with missing clinicopathological information and (4) the receipt of a blood transfusion or bone marrow transplantation within the prior six months. Since the renewal cycle of blood components (e.g., white blood cells, platelets, and red blood cells) typically ranges from a few days to around 120 days, it is theoretically safe to exclude samples from individuals who have received blood transfusions from external sources within the past six months, as this eliminates their impact on the results.

Sample size estimation

The one-sample Z-test for proportions method was used to calculate the sample size. The model’s anticipated sensitivity for diagnosing liver cancer was set at 85.0%, with a minimum target sensitivity of 75.0% deemed acceptable. The anticipated specificity was 90.0%, with a minimum target specificity of 80.0% deemed clinically acceptable. With a two-tailed significance level of α = 0.05 and a test power of 1-β = 0.8, the calculated sample sizes were 132 cases for the cancer group and 107 cases for the non-cancer group. The actual enrollment exceeded calculated requirements, particularly for controls (n = 255 actual vs. 107 required), to ensure adequate representation across diverse disease stages (hepatitis, fibrosis, compensated/decompensated cirrhosis) and improve specificity estimation precision.

aMAP score calculation

The bilirubin, albumin and platelet concentration/number values in the serum were measured separately using an automatic biochemistry analyzer, and the aMAP score for each sample was calculated according to the following formula¹²: aMAP = ({0.06 × age + 0.89 × sex (male = 1, female = 0) + 0.48 × [(log10 bilirubin × 0.66) + (albumin × −0.085)] − 0.01 × platelets} + 7.4)/14.77 × 100. Here, age is in years, bilirubin is in µmol/L, albumin is in g/L, and platelets are in 10³/µL.

AFP assay

The concentration of AFP in serum samples was measured using the Roche Cobas e801 system. Samples for which AFP test results were missing were excluded.

Methylation testing

Ten milliliters of venous blood were extracted from each subject and stored in cfDNA preservation tubes (Cwbio-tech, Jiangsu, China). After centrifuging the fresh, anticoagulated blood twice (at 2,000 rcf for 10 min per spin), 2 mL of the upper plasma layer was aspirated. The cfDNA was then extracted using the cfDNA nucleic acid extraction kit supplied by Wuhan Ammunition Life Technology Co. Ltd., China. In the final purification step, 50 µL of elution solution was used.

The methylation status of the GNB4 and Riplet genes in the plasma samples was detected using a commercially available kit (Wuhan Ammunition Life Technology Co., Ltd., China). This test has undergone preclinical studies at a single center^16,17, as well as multicenter clinical trial¹⁵, and has been approved by the China National Medical Products Administration. Triple-fluorescent PCR amplification was performed using the transformed cfDNA as the template and the ACTB gene as the internal reference. Positive and negative controls were set up for each PCR: the positive control was a synthetic plasmid containing the target amplified fragment, and the negative control was purified water. The negative control was used alongside the samples in the DNA extraction and transformation process. The fluorescence quantitative PCR instrument used was an ABI7500, and the amplification procedure was as follows: hold at 95 °C for 5 min (pre-denaturation), hold at 95 °C for 15 s (denaturation), hold at 60 °C for 30 s (annealing and extension), carry out 45 cycles of denaturation, annealing and extension steps, collecting fluorescence after each cycle.

Once the PCR had been completed and the baseline and threshold were adjusted, the cycle threshold (Ct) value of each gene in each sample was read. The test was only considered valid if the positive control amplified normally, the negative control did not amplify, and the Ct value of the sample’s internal reference gene (ACTB) was ≤ 35.

The repeatability of methylation detection was evaluated using four clinical plasma samples with known methylation status: one case with dual-gene methylation positivity, one case with GNB4 single-gene methylation positivity, one case with Riplet single-gene methylation positivity, and one case with dual-gene methylation negativity. Each sample underwent 20 replicate tests and the coefficient of variation (CV) for each gene’s Ct value in each sample was calculated. The CV value was required not to exceed 5%.

Data analysis

This study employed a double-blind design, meaning that the physicians diagnosing patients and the laboratory technicians analyzing samples were unaware of each other’s results in advance.

Receiver operating characteristic (ROC) curves were used to evaluate the diagnostic performance of individual biomarker and combined-marker models for disease detection. A binary logistic regression model was constructed for multi-parameter diagnosis, with disease status as the dependent variable. The model was fitted using maximum likelihood estimation to obtain the predicted probability for each subject. Ten-fold cross-validation was used to validate the area under the ROC curve (AUC) and avoid overfitting.

The diagnostic accuracy was quantified by calculating the AUC and its 95% confidence interval (CI). The 95% CIs were calculated using the Wilson score method with continuity correction - a refined normal approximation that performs well even with small samples or extreme proportions. The optimal cut-off value was determined by maximizing the Youden index (Youden index = sensitivity + specificity – 1).

Statistical testing

Descriptive statistics were used to summarize continuous variables. Variables conforming to a normal distribution were described using the mean and standard deviation, while skewed variables were described using the median and interquartile range (IQR). All analyses in this study were pre-specified. Multiple comparisons were corrected using the Bonferroni method. The DeLong test was used to calculate the difference in the AUC between different ROC curves. Count data were described using frequencies and percentages, and comparisons between groups were performed using the chi-squared test or Fisher’s exact test. Statistical analyses were performed using IBM SPSS Statistics software (version 22.0, IBM Corporation, USA) and R4.4.1.

Results

Sample information

A total of 392 clinical plasma samples were collected for this study, comprising 137 carcinoma cases and 255 control samples. Only two of the PLC samples were ICC, the remainder were HCC. The control samples were categorized into four groups: chronic hepatitis (n = 72), hepatic fibrosis (n = 31), compensated cirrhosis (n = 59) and decompensated cirrhosis (n = 93). The baseline characteristics of the samples are presented in Table 1. Of the study participants, 73.21% were from male patients, with more than half (50.77%) belonging to ethnic minorities other than Han Chinese. The largest proportion of the four control sample types was decompensated cirrhosis samples (23.72%). In terms of etiology, 53.32% of individuals were infected with HBV, accounting for the largest proportion. Of the carcinomas, 48.91% were in TNM stage I or II (early stage) according to the TNM classification, and 35.04% were single tumors measuring less than 3 cm in size.

Table 1.

Demographic and pathologic information of samples enrolled in this study.

Characteristic	Group	Number
Age	Median (IQR)	55 (46,64)
Sex	Male, No (%)	287 (73.21)
	Female, No (%)	105 (26.79)
Ethnic group	Han, No (%)	193 (49.23)
	Buyi, No (%)	98 (25.00)
	Miao, No (%)	48 (12.24)
	Shui, No (%)	38 (9.69)
	Others, No (%)	15 (3.83)
Disease type	Chronic hepatitis, No (%)	72 (18.36)
	Liver fibrosis, No (%)	31 (7.91)
	Compensated cirrhosis, No (%)	59 (15.05)
	Decompensated cirrhosis, No (%)	93 (23.72)
	Liver cancer, No (%)	137 (34.95)
Aetiology	HBV, No (%)	209 (53.32)
	HCV, No (%)	22 (5.61)
	ALD, No (%)	44 (11.22)
	NAFLD, No (%)	21 (5.36)
	Other (cholestasis, autoimmune diseases, other hepatitis viruses, drugs), No (%)	12 (3.06)
	Multiple factors, No (%)	55 (14.03)
	NA, No (%)	29 (7.40)
Pathological type	HCC, No (%)	132 (96.35)
	ICC, No (%)	5 (3.65)
TNM stage	I, No (%)	33 (24.09)
	II, No (%)	34 (24.82)
	III, No (%)	45 (32.85)
	IV,, No (%)	25 (18.25)
Tumour number and size	Single, ≤ 3 cm, No (%)	48 (35.04)
	Single, > 3 cm, No (%)	71 (51.82)
	Multile, any size, No (%)	18 (13.14)

IQR: Interquartile range; NA: Not available.

Comparison of methylation and AFP for liver cancer diagnosis

The results of the precision analysis of methylation testing are presented in Table S1. For the four samples with known methylation statuses, the CV of the gene Ct values remained below 1.5% following 20 replicate tests, indicating the high reproducibility of the methylation assay. The distribution of the AFP concentrations, aMAP scores, GNB4 and Riplet gene Ct values in samples from patients with hepatitis, liver fibrosis, compensated cirrhosis, decompensated cirrhosis, and PLC was analyzed, and the results are shown in Fig. 1. Figure 1A and B demonstrate that the Ct values for the GNB4 and Riplet genes differed significantly between the hepatitis and PLC groups, the liver fibrosis and PLC groups, the compensated cirrhosis and PLC groups, and the decompensated cirrhosis and PLC groups (all P < 0.001). However, no significant differences were observed when comparing the hepatitis, liver fibrosis, compensated cirrhosis and decompensated cirrhosis groups. Similarly, AFP concentration values exhibited variation only between the cancer and non-cancer groups, with no variation observed among the various non-cancer subgroups (Fig. 1C). However, as shown in Fig. 1D, the aMAP score increased progressively across the hepatitis-to-compensated cirrhosis-to-decompensated cirrhosis sample spectrum. Although there was no statistically significant difference between the hepatitis and fibrosis groups, or between the compensated cirrhosis and carcinoma groups (P ≥ 0.05), significant differences were observed in all other pairwise comparisons (P < 0.001 or P < 0.01). Furthermore, aMAP scores were found to be significantly higher in decompensated cirrhosis patients than in carcinoma patients in this study (P < 0.001). This finding may reflect poorer hepatic reserve function in decompensated cirrhosis patients.

Fig. 1.

Distribution of different indexes in the AFP, aMAP and methylation assays. (A) Boxplot showing the Ct values of the GNB4 gene in samples from patients with hepatitis, liver fibrosis, compensated cirrhosis, decompensated cirrhosis and liver cancer. (B) Boxplot showing the Ct values of the Riplet gene in samples from patients with hepatitis, liver fibrosis, compensated cirrhosis, decompensated cirrhosis and liver cancer. (C) Boxplot showing the AFP concentrations in samples from patients with hepatitis, liver fibrosis, compensated cirrhosis, decompensated cirrhosis and liver cancer. (D) Boxplot showing the aMAP score in samples from patients with hepatitis, liver fibrosis, compensated cirrhosis, decompensated cirrhosis and liver cancer. ‘Ns’ denotes ‘Not Significant’, indicating that the P-value is greater than or equal to 0.05, ‘**’ denotes a P-value less than 0.01, and ‘***’ denotes a P-value less than 0.001. Statistical tests were performed using ANOVA analysis with the Bonferroni correction.

Genetic methylation, AFP, and aMAP scoring were performed simultaneously on 137 liver cancer samples and 255 non-liver cancer samples. Following analysis, 10-fold cross-validation was employed to generate ROC curves to distinguish between PLC and non-PLC samples for each biomarker. The results are presented in Fig. 2A (methylation), Fig. 2B (AFP) and Figure S1 (aMAP score). The AUC values for GNB4/Riplet methylation and AFP were 0.901 (95% CI: 0.857–0.945) and 0.696 (95% CI: 0.635–0.758), respectively, with optimal cutoff values of 0.182 and 21.522, respectively. The cut-off for gene methylation testing is a probability value, while the cut-off for AFP testing is a concentration value (ng/mL). The AUC value of methylation biomarkers was significantly higher than that of aMAP (DeLong test, P value < 0.001). The AUC value for distinguishing PLC from non-PLC samples using the aMAP score was relatively low at 0.544 (Figure S1). These results suggest that GNB4/Riplet gene methylation is the most effective PLC diagnostic biomarker, with an overall sensitivity and specificity of 87.59% (95% CI: 80.61%−92.40%) and 93.33% (89.25%−95.95%), respectively (Fig. 2C and D). The overall sensitivity of methylation is significantly higher than AFP’s at 51.09% (42.45%−59.67%), though the specificity is not significantly different. Notably, adjusting the methylation ROC curve for confounding factors, including age, gender, ethnicity, and etiology, showed no significant change in the AUC value compared to the unadjusted curve (DeLong test, P-value = 0.182), as detailed in Figure S2.

Fig. 2.

ROC and overall performance analysis of methylation and AFP for distinguishing PLC from non-PLC. (A) ROC curve showing the ability to distinguish between PLC and non-PLC samples (including hepatitis, liver fibrosis, compensated cirrhosis, and decompensated cirrhosis), constructed using a logistic model based on the methylation of the GNB4 and Riplet genes, with no covariate adjustment; (B) ROC curve showing the ability to distinguish between PLC and non-PLC samples (including hepatitis, liver fibrosis, compensated cirrhosis and decompensated cirrhosis), based on AFP concentration, with no covariate adjustment. Methylation AUC 0.901 vs. AFP AUC 0.696, DeLong test, P = 1.55 e-07; (C) Overall sensitivity of methylation and AFP. Values in bars represent mean sensitivity and 95% CI. Differences were assessed for significance using a chi-squared test. ‘***’ indicates P < 0.001; (D) Overall specificity of methylation versus AFP. Bar values represent the mean specificity and 95% CI. Differences were assessed for significance using a chi-squared test. ‘NS’ denotes not significant (P ≥ 0.05).

The sensitivity and specificity data for each subgroup are displayed in Table 2. The sensitivity of methylation detection exhibited an increasing trend with TNM progression, with values of 69.70% (95% CI: 51.13%−83.79%) in stage I and 91.18% (95% CI: 75.19%−97.69%) in stage II. Meanwhile, the sensitivity of AFP detection was 39.39% (95% CI: 23.43%−57.76%) and 41.18% (95% CI: 25.13%−59.17%) for stage I and II carcinomas, respectively. These findings suggest that methylation is superior to AFP for the detection of early-stage (TNM Stages I and II) liver cancer. Regarding specificity, no significant differences were observed for either methylation or AFP across different disease types (P = 0.791 for methylation and 0.404 for AFP), ethnicities (P = 0.212 for methylation and 0.060 for AFP) or age subgroups (P = 0.576 for methylation and 0.302 for AFP). Notably, while the specificity of the methylation marker was lowest in decompensated cirrhosis samples at 91.40% (95% CI: 83.28%−95.94%), this difference was not significant when compared to the other three groups (P = 0.791).

Table 2.

Sensitivity and specificity analysis of cfDNA methylation detection.

Category			Number	Methylation	P Valuea	AFP	P Valuea
				Point estimate (%,95 CI)		Point estimate (%,95 CI)
Sensitivity	Overall		137	87.59 (80.61-92.40)	NAb	51.09 (42.45-59.67)	NAb
	TNM stage	Stage I	33	69.70 (51.13-83.79)	0.004	39.39(23.43-57.76)	0.089
		Stage II	34	91.18 (75.19-97.69)		41.18 (25.13-59.17)
		Stage III	45	91.11 (77.87-97.11)		64.44 (48.73-77.71)
		Stage IV	25	100.00 (83.42 -100.00)		56.00 (35.27-74.98)
Specificity	Overall		255	93.33 (89.25-95.95)	NAb	91.37 (87.06-94.39)	NAb
	Type of liver disease	Chronic hepatitis	72	94.44 (85.65-98.21)	0.791	90.28 (80.42-95.67)	0.404
		Fibrosis	31	96.77 (81.49-99.83)		87.09 (68.23-95.78)
		Compensated cirrhosis	59	93.22 (82.73-97.81)		96.61 (87.25-99.41)
		Decompensated cirrhosis	93	91.40 (83.28-95.94)		90.32 (81.97-95.20)
	Ethnic group	Han	121	91.74 (84.95-95.74)	0.212	93.38 (86.98-96.89)	0.060
		Buyi	65	95.38 (86.24-98.80)		92.31 (82.25-97.13)
		Miao	34	100.00 (87.36-100.00)		91.18 (75.19-97.69)
		Shui	24	87.50 (66.54-96.71)		91.67 (71.53-98.54)
		Others	11	90.91 (57.12-99.52)		63.63 (31.61-87.63)
	Age	≤60 years old	192	92.71 (87.82-95.80)	0.576	90.10 (84.76-93.78)	0.302
		>60 years old	63	95.24 (85.83-98.76)		95.24 (85.83-98.76)

^bThe abbreviation “NA” stands for “not applicable”.

^a P values were calculated by the Fisher’s exact test.

DNA methylation combined with aMAP score improves risk stratification efficacy in hepatitis population

Although the aMAP score is not effective in distinguishing between PLC and non-PLC samples (Figure S1), it shows an upward trend in the progression from hepatitis to liver fibrosis to compensated cirrhosis to decompensated cirrhosis (Fig. 1D), suggesting its potential as a risk stratification tool. We therefore evaluated the ROC curves for the aMAP score in distinguishing the hepatitis group from the liver fibrosis, cirrhosis and carcinoma group, and for distinguishing the hepatitis and liver fibrosis group from the cirrhosis and carcinoma group, using 10-fold cross-validation. The results are shown in Fig. 3A and B, with respective AUC values of 0.745 (95% CI: 0.679–0.812) and 0.764 (95% CI: 0.710–0.818). Notably, when adjusting for confounding factors, including age, gender, ethnicity, and etiology, both AUC values remained unchanged (Figure S3, DeLong test, P values 0.886 and 0.851).

Fig. 3.

GNB4/Riplet genes methylation combined with aMAP score for differentiating between different liver diseases. (A) ROC curves for hepatitis versus fibrosis, cirrhosis and cancer samples. (B) ROC curves for hepatitis and fibrosis versus cirrhosis and cancer.

A binary logistic regression model was constructed by combining the methylation of the GNB4 and Riplet genes with the aMAP score. The relevant parameters of the logistic regression models are presented in Tables S2 and S3. ROC curves were then plotted to differentiate between hepatitis and fibrosis/cirrhosis/cancer groups, as well as between the hepatitis/fibrosis and cirrhosis/cancer groups, using the ten-fold internal validation method. The respective AUC values were 0.825 (95% CI: 0.774–0.875) (Fig. 3A) and 0.867 (95% CI: 0.829–0.905) (Fig. 3B). After adjusting for confounding factors, the AUC value also showed no significant change (see Figure S3, DeLong test, P-values 0.322 and 0.926). Compared to using the aMAP score alone, the combined test improved the accuracy of diagnosing fibrosis, cirrhosis and cancer samples (Fig. 3, DeLong test, P-value 0.000 and 0.000). In particular, the diagnostic accuracy for end-stage liver disease (cirrhosis and cancer) reaches as high as 0.867.

Discussion

There are three main problems with early screening for PLC:1) The risk group cannot be identified. Due to requirements such as protecting the privacy of hepatitis virus-infected patients, it is difficult to accurately identify the infection risk group in a non-clinical scenario (e.g. in the community or at physical examination centers). Therefore, a model that enables risk assessment without prior knowledge of hepatitis virus infection status would be highly valuable. 2) It is difficult to quantify the risk level of the risk group. The causes of PLC are numerous and complex, and the risk of cancer varies depending on the cause or the stage of the disease. While there are risk assessment models for individual cause (e.g., the REACH-B model for patients with HBV infection not receiving antiviral therapy¹⁸, the PAGE-B model for patients with HBV infection receiving antiviral therapy¹⁹, and the THRI model for patients with cirrhosis²⁰, there are fewer generalized cancer risk models for various liver disease types. 3) The available surveillance tools for at-risk populations (mainly ultrasound and serum AFP testing) have diagnostic value for liver cancer, but have limitations in terms of risk stratification and providing an early warning. The aMAP combined with a DNA methylation strategy, as proposed in this study, is applicable across multiple etiologies. It enables quantitative risk assessment and provides an early warning of liver fibrosis, cirrhosis, and liver cancer.

The aMAP score and DNA methylation are two relatively independent yet interconnected processes. Prolonged and repeated inflammatory responses can lead to damage and regeneration of hepatocytes. During this process, DNA methylation patterns may be altered, thereby affecting physiological processes such as cell proliferation and apoptosis and increasing the risk of cancer. While some DNA methylation assays show promise in the diagnosis of PLC, one such method targeting 20 differentially methylated regions has demonstrated sensitivities of 84.4% and 70.8%, and specificities of 90.3% and 89.5% for BCLC stage 0/A in the training and validation sets, respectively¹⁴. Another method targeting six methylated genes demonstrated a sensitivity of 75% for BCLC stage 0 HCC and 93% for BCLC stage A HCC, with a specificity of 92%²¹. To the best of our knowledge, no previous studies have evaluated gene methylation for distinguishing hepatitis, fibrosis or cirrhosis from PLC in a multi-ethnic Chinese cohort, nor have they examined its combined application with aMAP. Furthermore, prior research has primarily focused on differentiating PLC from non-PLC, but this study, however, is the first time to combine aMAP scores with the detection of plasma GNB4 and Riplet gene methylation to construct a diagnostic model for end-stage liver disease (including cirrhosis and liver cancer). The model’s AUC value for diagnosing cirrhosis and PLC patients was as high as 0.867 (Fig. 3).

The methylation of the GNB4 and Riplet (also known as RNF135) genes for diagnosing HCC has been reported in several papers^15–17,22. This study investigated the methylation of these two genes at different stages of liver disease. The results showed that the GNB4 and Riplet gene methylation assays were less specific for decompensated cirrhosis samples (91.40%) (Table 2). These findings demonstrate that both genes are highly correlated with PLC. However, partial hypermethylation was observed in 8.60% of decompensated cirrhosis samples (see Table 2). This may reflect either the “field effect” wherein epigenetic alterations associated with liver cancer are initiated by chronic inflammation and fibrosis during the precancerous stage, or the presence of occult preclinical malignancies that are not yet detectable by imaging. Importantly, we lack longitudinal follow-up data to determine which “false-positive” cirrhosis patients will progress to HCC. This limits our ability to distinguish true false positives from early detection. Future prospective studies with serial sampling and extended follow-up are essential to clarify this phenomenon and optimize surveillance strategies for patients with high-risk cirrhosis. The average sensitivity of the combined GNB4 and Riplet methylation assay for TNM stage I liver cancer was 69.70 (Table 2). The value was lower than the respective figure of 84.4% reported in a previous study¹⁵. Consistent with the previous study¹⁵, the overall accuracy of AFP for diagnosing PLC in this study was lower than that of plasma GNB4 and Riplet gene methylation assays (Table 2). The sensitivity of TNM Stage I PLC at 69.70% remains inadequate for independent use in early screening, with nearly one-third of early-stage cancers likely to be missed. While combining methylation with other biomarkers (AFP, DCP, AFP-L3) may enhance sensitivity, prospective studies are needed to determine the optimal multi-biomarker combination and validate whether combined testing can achieve the ≥ 80% early-stage sensitivity threshold recommended for effective surveillance programs.

The aMAP score incorporates only a limited number of objective metrics that demonstrate efficacy in the full life cycle management of PLC. For instance, the model has been employed to predict the risk of HCC^12,23,24, to predict HCC recurrence^25,26 and to suggest prognosis^27,28. In the published study in which the aMAP score was initially established¹², the annual incidence of HCC was found to be 0.0–0.2% in the low-risk group (aMAP score 0–50) and 0.4–1.0% in the intermediate-risk group (aMAP score 50–60). In the high-risk group (aMAP score 60–100), the annual incidence of HCC was 1.6–4.0%. In contrast to the preceding studies, this research utilized the aMAP score to diagnose two categories of end-stage liver disease: cirrhosis and PLC. The optimal threshold for distinguishing end-stage liver disease from non-end-stage liver disease (i.e. hepatitis and fibrosis) was found to be 53.233 (see Fig. 3B), which is consistent with the score observed in the intermediate-risk group in the previous study¹². Consequently, it can be deduced that samples exhibiting aMAP scores that surpass this threshold are more prone to manifest cirrhosis or PLC. Moreover, when aMAP was combined with GNB4/Riplet, the AUC for diagnosing end-stage liver disease (cirrhosis and HCC) increased to 0.867. In clinical practice, the diagnosis of liver fibrosis, cirrhosis and early HCC typically necessitates diagnostic confirmation by one or more of the following methods: ultrasound, computed tomography, magnetic resonance imaging, transient elastography and liver biopsy²⁹. This study proposes a non-invasive and convenient method for hepatitis risk stratification, achieved by dynamic real-time monitoring of the risk of developing end-stage liver disease. The approach effectively enriches high-risk populations, making it suitable for large-scale population screening or risk population surveillance. Our research indicates that the aMAP score combining with methylation holds promise for guiding clinical decision-making. For instance, patients diagnosed with hepatitis who exhibit elevated aMAP levels (≥ 53) and positive methylation may be eligible for a monitoring protocol involving imaging and AFP assessments conducted every six months. Conversely, low-risk patients (aMAP < 53 and methylation-negative) may proceed with annual monitoring. However, the integration of this approach with existing guideline-recommended monitoring regimens (e.g., the AASLD’s 6-month ultrasound/AFP monitoring regimen³⁰ necessitates prospective validation.

The limitations of this study are as follows: (1) The single-center design is susceptible to selection bias. (2) The utilization of case-control analysis as opposed to prospective screening may result in an overestimation of accuracy. (3) Potential pre-analytical variability (e.g. sample handling, storage, and batch effects) was not evaluated. (4) Although the ten-fold cross-validation method employed in this study could mitigate overfitting and the model was not influenced by factors such as age, gender, ethnicity, and etiology (see Figures S2 and S3), further validation is required in other regions (e.g., other areas of China or Southeast Asia), across different etiologies (e.g., pure non-alcoholic fatty liver disease) and within diverse healthcare settings to validate the model’s generalizability. It is particularly noteworthy that the ethnic composition of this study cohort (50.77% ethnic minorities from Guizhou Province) limits its generalizability to regions with Han majority populations and non-Chinese populations. Although data indicate that ethnic background does not significantly influence methylation specificity (Table 2, P = 0.212), this conclusion requires validation in larger, ethnically diverse cohorts. (5) Assessments of cost-effectiveness and real-world feasibility are lacking.

Conclusion

The findings indicated that the combination of aMAP score and gene methylation has the potential to serve as a diagnostic tool for identifying hepatic cirrhosis and carcinoma. This study’s main contribution is demonstrating that combining aMAP and cfDNA methylation helps to categorize hepatitis patients according to their risk of developing advanced liver disease (cirrhosis, PLC). It may help to identify high-risk individuals who could benefit from closer surveillance, which in turn might improve stage at diagnosis and survival in future prospective studies. Nevertheless, its clinical application necessitates validation through multicenter, larger-scale cohort studies.

Abbreviations

AFP

Alpha-fetoprotein

AJCC

American joint committee on cancer

ALD

Alcoholic liver disease

aMAP

Age-male-albumin-bilirubin-platelet

ANOVA

Analysis of variance

AUC

Area under the curve

BCLC

Barcelona clinic liver cancer

cfDNA

Cell-free DNA

cHCC-ICC

Combined hepatocellular-cholangiocarcinoma

CI

Confidence interval

Ct

Cycle of threshold

HBV

Hepatitis B virus

HCC

Hepatocellular carcinoma

HCV

Hepatitis C virus

ICC

Intrahepatic cholangiocarcinoma

IQR

Interquartile range

NAFLD

Non-alcoholic fatty liver disease

PLC

Primary liver cancer

ROC

Receiver operating characteristic

TNM

Tumor- node- metastasis

Author contributions

DY.C.: Conceptualization, Data curation, Formal analysis, Investiagtion, Methodology, Resources, Writing-original draft, Writing-review & editing.K.Z.: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Supervision, Writing-original draft, Writing-review & editing.XZ.J.: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Resources, Writing-review & editing.TY.L.: Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Writing-review & editing.J. Z.: Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Writing-review & editing.YH.A.: Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Writing-review & editing.DY.R.: Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Writing-review & editing.F.L.: Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Writing-review & editing.SD.L.: Funding acquisition, Conceptualization, Supervision, Resources, Investiagtion, Methodology, Writing-original draft, Writing-review & editing.

Funding

This work was supported by the Guizhou Qiannan Infectious Disease Control Talent Base Project (NO.RCJD2020-67), the Guangzhou Science and Technology Plan Project (NO. 202103000060) and Qiannan Liver Disease Clinical Medical Research Center Project (NO. Qiannan ke he zi(2020)05).

Data availability

The raw data used in this study have been uploaded to figshare (https://doi.org/10.6084/m9.figshare.31554157).

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

The Institutional Review Board of the The People’s Hospital of Qiannan granted ethical approval to the study with the following ethical approval number: 2024-qnzy-20.

Consent for publication

Written informed consent was obtained from all participants.