Circulating tumor DNA as a predictive biomarker for prognosis in hepatocellular carcinoma: a systematic review and meta-analysis

Abstract

Objective

The prognosis of patients with hepatocellular carcinoma (HCC) is often poor, making prediction of prognosis and risk stratification highly significant. However, existing indicators are insufficient for accurately predicting the prognosis of patients with HCC. This study aimed to systematically summarize the value of circulating tumor DNA (ctDNA) as a prognostic biomarker for HCC patients.

Methods

PubMed, Web of Science, Embase, Cochrane Library, Scopus, and clinical trials.gov databases were searched to collect observational studies and randomized clinical trials from January 2016 to November 2024. Studies focusing on ctDNA status or ctDNA methylation and prognostic outcomes in HCC patients were included. Pooled hazard ratios (HRs) were calculated for the primary outcomes: relapse-free survival (RFS) and overall survival (OS). Random-effects models were applied considering the potential heterogeneity.

Results

A total of 25 studies involving 2490 HCC patients were included. Positive ctDNA status both before and after surgery were significantly associated with shorter RFS (before surgery: HR = 3.88, 95% confidence interval [CI]: 1.46–10.33, P = 0.007; after surgery: HR = 5.08, 95% CI: 3.35–7.72, P < 0.001). Compared with ctDNA-negative groups, ctDNA-positive before and after surgery groups both exhibited shorter OS (before surgery: HR = 6.59, 95% CI: 2.47–17.55, P < 0.001; after surgery: HR = 7.01, 95% CI: 2.21–22.27, P = 0.001). Sensitivity analyses yielded results similar to the main analysis. Additionally, ctDNA may detect recurrence 2–5 months earlier than radiographic imaging.

Conclusions

ctDNA detection was significantly associated with poorer prognosis in HCC patients. The potential applications of ctDNA in prognostic prediction are promising, and the predictive value of ctDNA dynamic change warrants further exploration.

1. Introduction

Liver cancer is the third leading cause of cancer-related mortality worldwide¹, posing a significant threat to human health. Hepatocellular carcinoma (HCC) is the most common type of primary liver cancer, accounting for 75%–85% of cases.^1,2 Due to the absence of early symptoms, most HCC patients are diagnosed at an advanced stage, leading to a poor prognosis. Even after surgical resection, recurrence within 2 years occurs in 30–50% of patients following primary resection.^3,4 Accurately identifying patients at high risk of recurrence or mortality and implementing appropriate therapeutic strategies is crucial for improving prognosis.

Unfortunately, currently available clinical indices remain suboptimal for predicting prognosis in patients with HCC. Among them, only serum alpha-fetoprotein (AFP) levels have strong supporting evidence and are widely used for tumor assessment in therapeutic evaluation and prognosis prediction.⁵ However, AFP has limited sensitivity and specificity, as increased levels are observed in only a small proportion of patients—approximately 10% in early-stage HCC, increasing to 40% in advanced stages.⁶ Moreover, tissue biopsy is invasive and impractical for continuous monitoring of disease progression.⁷ Consequently, there is an urgent need for novel biomarkers that can provide more timely, sensitive, and accurate prognostic assessments.

In recent years, liquid biopsy has emerged as a non-invasive and highly sensitive tool in oncology research.^8,9 In particular, circulating tumor DNA (ctDNA) has gained considerable attention for prognostic prediction in various solid tumors, such as non-small cell lung cancer^10,11 and colorectal cancer.^12,13 ctDNA consists of small fragments of DNA released mainly by apoptotic or necrotic tumor cells into the circulation.¹⁴ Mutations can be detected in ctDNA, providing insight into the mutational landscape of tumors. Furthermore, its short half-life allows real-time assessment of disease progression.¹⁵ Notably, ctDNA enables the prediction of tumor recurrence several months ahead of imaging evidence.¹⁶

Several studies have explored the prognostic value of ctDNA status (positive or negative) or ctDNA methylation in HCC. For instance, Sogbe et al. found that ctDNA detection was associated with inferior overall survival (OS) and progression-free survival (PFS) in HCC patients receiving systemic treatment.¹⁷ Additionally, in patients with early- and intermediate-stage HCC, postoperative ctDNA was also an independent prognostic predictor of OS and disease-free survival (DFS).¹⁸ However, these studies are limited by small sample sizes, and there are significant variations in ctDNA sampling time points across different studies. As a result, the role of ctDNA in prognostic prediction for HCC remains a topic of debate, and a comprehensive systematic review of this field is still lacking.

In this study, we systematically searched for and included all studies on the association between ctDNA status or ctDNA methylation and prognosis among patients with HCC. We evaluated the prognostic value of ctDNA at different time points and various outcomes, so as to provide scientific basis for precise prognostic prediction of HCC patients.

2. Materials and methods

2.1. Protocol and registration

This study was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta Analyses (PRISMA)¹⁹, A MeaSurement Tool to Assess systematic Reviews (AMSTAR) 2 checklist²⁰, and Transparency In The reporting of Artificial INtelligence (TITAN) guideline checklist²¹ (Supplementary Tables 1–2 and Supplementary Fig.1). The study protocol was prospectively registered on PROSPERO (CRD42022323474). No artificial intelligence was used in the research and manuscript development.

2.2. Search strategy

PubMed, Web of Science, Embase, Cochrane Library, Scopus, and clinicaltrials.gov were searched from January 2016 to November 2024. Detailed search strategies for each platform are available in Supplementary Methods. The search results were imported into EndNote X9 and duplicates were removed automatically by the software.

2.3. Study selection

After removing duplicates, the titles and abstracts were carefully screened according to predefined inclusion and exclusion criteria. Subsequently, potentially eligible studies were thoroughly reviewed. Study selection and data collection were conducted independently and simultaneously by two researchers, with any discrepancies resolved through discussion.

The inclusion criteria were as follows: (i) original articles published from 2016–2024 or conference abstracts published in 2022, 2023, and 2024; (ii) observational or randomized controlled trials involving patients with HCC; (iii) documented collection and measurement of ctDNA status (categorized as positive or negative) or ctDNA methylation; (iv) reporting of clinical prognostic outcomes, such as relapse-free survival (RFS) and OS; and (v) written in English. All time points and measurement methods of ctDNA were allowed, including transitions in ctDNA status (e.g., from positive to negative). We did not impose any additional restrictions on how ctDNA positivity was defined (e.g. variant allele frequency/tumor fraction cut-off), as long as the original study dichotomized patients into ctDNA-positive and ctDNA-negative groups.

The exclusion criteria were: (i) non-original studies or those with unavailable results (e.g., reviews, editorials, comments, letters, or case reports); (ii) studies where ctDNA status was not categorized as a positive or negative variable; (iii) studies focusing on diseases other than HCC; (iv) studies addressing diagnostic or screening outcomes; and (v) conference abstracts published prior to 2022.

2.4. Outcomes

The primary outcomes of interest were RFS (composite endpoints including RFS and DFS, depending on the study design) and OS. RFS was defined as the time from inclusion or treatment to relapse or death, and OS was defined as the time from inclusion or treatment to death from any cause. Secondary outcomes included recurrence and lead time, which was defined as the interval from the detection of positive ctDNA to the confirmation of clinical recurrence through imaging tests.

According to the included studies, the following ctDNA measurement timepoints were defined: before surgery (before surgical resection or liver transplantation), during percutaneous ablation, and after surgery (after surgical resection or liver transplantation), which can be categorized into just after surgery (within 1 to 4 weeks after surgery) and long-term post-treatment surveillance. However, due to the limited number of studies, we combined them as “after surgery” for meta-analysis. In some studies, ctDNA was also measured before any anti-tumor treatment (including surgery, locoregional, and systemic therapy).

2.5. Data extraction

The following variables were extracted from the included studies: (i) general information: first author, year of publication, country where the study was conducted, number of centers (single or multi), and total sample size; (ii) population characteristics: age, sex, and follow-up duration; (iii) disease characteristics: clinical stage and the presence of hepatitis B virus (HBV) infection; (iv) ctDNA measurement details: sources of the samples, sources of the tests, methods, detection platforms, and sampling timepoints; and (v) outcome data: effect estimates with 95% confidence intervals (CIs) for RFS and OS for ctDNA status, ctDNA methylation and AFP, the number of events for recurrence and median RFS or OS in the ctDNA positive and negative groups, and the time difference between ctDNA testing and imaging for lead time.

2.6. Assessment of risk of bias

The risk of bias in observational studies and single-arm clinical trials was assessed using the Newcastle-Ottawa Scale (NOS). This process was carried out independently by two authors and any discrepancies were resolved by discussion.

2.7. Statistical analysis

Meta-analyses were conducted separately for each ctDNA measurement timepoint. If the study only provided hazard ratio (HR) and P values, we used RevMan 5.4 software to calculate the corresponding 95% CI. A random-effects model was adopted to account for potential heterogeneity across studies. The pooled HRs with 95% CIs for RFS and OS, and risk ratios (RRs) with 95% CIs for recurrence were calculated separately, using the inverse-variance weighting method. The significance of pooled results was assessed using a Z-test, and heterogeneity was evaluated and reported using I² statistics, with values greater than 50% indicating significant heterogeneity.

In the main analysis, we prioritized HRs from the univariable Cox analysis; however, if only HRs from the multivariable analysis were reported in a study, we included those instead. As a sensitivity analysis, we only incorporated all multivariable HRs. We also conducted the leave-one-out method. Additionally, we excluded conference abstracts and included only original research as an additional sensitivity analysis. We reanalyzed the data after excluding studies that used droplet digital polymerase chain reaction (ddPCR) or did not clearly specify the ctDNA detection method, and we repeated the analysis after excluding studies that defined ctDNA positivity as ≥2 mutations or failed to provide a clear definition of ctDNA positivity. Furthermore, we conducted sensitivity analyses restricted to studies that included only patients with stages I–III disease, as well as analyses limited to studies that used commercially available sources of ctDNA testing.

A two-sided P-value of <0.05 was considered statistically significant. Funnel plots and Egger's test were used to detect publication bias. All analyses were conducted using R software version 4.4.2 (with the metafor and meta packages).

3. Results

3.1. Characteristics of included studies

A total of 25 studies,^15,17,18,22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43 encompassing 2490 patients were included, of which 7 were conference abstracts (Fig. 1). Reports assessed for eligibility and subsequently excluded, along with the reasons for exclusion, were provided in Supplementary Table 3. Among the included studies, 19 evaluated ctDNA status as the primary exposure, while six investigated ctDNA methylation markers. Eighteen studies reported the country of the study population, with most studies conducted in China (66.67%). The sample sizes ranged from 7 to 1098, the proportion of male ranged from 75.16% to 94.44%, and the median age ranged from 50 to 73 years old. The proportion of HBV-positive cases ranged from 11.54% to 100.00%. Seventeen studies reported whether the study population was from a single-center or multi-center, and 82.35% were single-center studies. The median follow-up period ranged from 6.8 to 45.3 months (Table 1). We also recorded the exact definition of ctDNA positivity or ctDNA methylation, as well as sources of the tests and detection platforms (Supplementary Tables 4–5). Thirteen studies specified the definition of positivity of ctDNA and five studies gave the definition of ctDNA methylation positivity. As shown in Table 2, eighteen studies reported the source of ctDNA status, all of which were derived from blood. Eighteen studies reported ctDNA measurement methods, with next-generation sequencing (NGS) being the most common (77.78%).

Fig. 1.

Flow diagram of studies inclusion and exclusion. ctDNA, circulating tumor DNA; HCC, hepatocellular carcinoma.

Table 1.

Characteristics of the included studies.

Study IDRef	Article type	Study type	Country	Multi center	No. of Samples	Clinical stage	HBV infection (positive/negative)	Sex (male/female)	Age, median (range)/mean±SD, years	Follow-up, median (range), months
ctDNA status
An, 201922	Article	Cohort	China	N	26	I, II, III	25/1	23/3	51 (range 27–86)	-
Cai 201923	Article	Cohort	China	N	34	I, II, III	8/26	27/7	53±12	-
Wang, 202024	Article	Cohort	China	N	81	I, II	69/12	69/12	55 (range 28–78)	-
Ge, 202125	Article	Cohort	The Netherlands	-	26	BCLC A, B, C	3/23	23/3	67±12	-
Ye, 202218	Article	Cohort	China	N	96	BCLC 0/A, B, C	92/4	85/11	50 (range 18–75)	-
Zhu, 202226	Article	Cohort	China	N	41	I, II	30/11	35/6	Mean 73	17.7 (range 2.1–19.3)
Xia, 202227	Article	Single- arm clinical trial	China	N	18	II, III	15/3	17/1	55 (range 34–76)	Around 12
Lee, 202228	Article	Cohort	The Republic of Korea	N	102	BCLC A, B, C	71/31	87/15	Mean 60	19.1
Xu, 202329	Article	Cohort	China	N	20	I, III	14/6	17/3	58 (range 42–76)	14.1 (range 5.5–22.1)
Campani, 202415	Article	Cohort	France	N	103	BCLC 0/A, B, C	19/84	84/19	64 (IQR 56–74)	>45.3
Huang, 202430	Article	Cohort	China	N	74	I, II, III, IV	62/12	57/17	54 (range 35–76)	17.1
Hong, 202431	Article	Cohort	-	N	7	-	-	6/1	70 (range 32–75)	15
Sogbe, 202417	Article	Cohort	Spain	N	31	-	-	-	-	37.4 (range 0.5–65.6)
Abdelrahim, 202232	Conference abstract	Cohort	-	-	10	I, II, III, IV	-	-	-	-
Jiang, 202233	Conference abstract	Cohort	-	-	45	-	-	42/3	51 (range 31–76)	-
Pan, 202234	Conference abstract	Cohort	-	-	20	-	14/6	-	-	13.6
Marron, 202335	Conference abstract	Cohort	-	-	21	-	-	-	Median 68	23 (IQR 20–26)
Abdelrahim, 202436	Conference abstract	Cohort	-	-	42	-	-	-	-	-
Guo, 202437	Conference abstract	Cohort	China	N	136	-	-	-	-	24
ctDNA methylation
Xu, 201738	Article	Cohort	China	Y	1098	I, II, III, IV	-	905/130	55 (range 15–81)	-
Liu, 202039	Article	Cohort	China	N	105	I, II, III, IV	105/0	86/19	59 (IQR 55–64)	6.8 (IQR 2.6–9.7)
Saeki, 202340	Article	Cohort	Japan	Y	157	BCLC A, B, C	26/131	118/39	Training cohort: 73.0 (IQR 68.0–80.0); Validation cohort: 72.0 (IQR 64.0–76.0)	Training cohort: 13.9 (5.7–21.7); Validation cohort: 13.8 (5.8–24.3)
Angeli-Pahim, 202441	Article	Cohort	The United States	-	21	-	-	17/4	63±10	-
Guo, 202442	Article	Cohort	China	Y	103	I, II, III	-	88/15	-	-
Yang, 202443	Conference abstract	Cohort	-	-	73	-	-	-	-	11.77

Abbreviations: -, not reported; BCLC, Barcelona Clinic Liver Cancer; HBV, hepatitis B virus; IQR, interquartile range; N, no; SD, standard deviation; Y, yes.

Table 2.

Methodology and outcome information of the included studies about ctDNA status.

Study IDRef	Source of ctDNA	ctDNA detection method	Outcome	NOS score
An, 201922	Blood	NGS	RFS/recurrence	6
Cai 201923	Blood	NGS	Lead time	5
Wang, 202024	Blood	ddPCR	RFS/OS/recurrence	7
Ge, 202125	Blood	ddPCR and NGS	OS	5
Ye, 202218	Blood	NGS	RFS/OS/recurrence	7
Zhu, 202226	Blood	NGS	RFS/recurrence	8
Xia, 202227	Blood	NGS	RFS	5
Lee, 202228	Blood	NGS	OS	7
Xu, 202329	Blood	NGS	RFS/recurrence	6
Campani, 202415	Blood	ddPCR and NGS	RFS/OS	9
Huang, 202430	Blood	NGS	RFS/recurrence	9
Hong, 202431	Blood	NGS	Recurrence	5
Sogbe, 202417	Blood	ULP-WGS	OS/PFS	7
Abdelrahim, 202232	-	NGS	Recurrence/lead time	-
Jiang, 202233	Blood	-	Recurrence/RFS	-
Pan, 202234	Blood	NGS	RFS	-
Marron, 202335	Blood	NGS	RFS/Lead time	-
Abdelrahim, 202436	Blood	NGS	RFS	-
Guo, 202437	Blood	NGS	Lead time	-

Abbreviations: -, not reported; ctDNA, circulating tumor DNA; ddPCR, droplet digital Polymerase Chain Reaction; NGS, next-generation sequencing; NOS, Newcastle-Ottawa Scale; OS, overall survival; PFS, progression-free survival; RFS, relapse-free survival.

3.2. Risk of bias

The overall risk of bias score among the included 18 original articles ranged from 4 to 9. Eleven studies were classified as low-risk (NOS scores: 7–9) and seven studies were moderate-risk (NOS scores: 4–6). The majority of moderate-risk studies were limited by insufficient information of outcomes and details of follow-up (Supplementary Table 6).

3.3. ctDNA status and RFS

The status of ctDNA measured at different timepoints were significantly associated with worse RFS (Fig. 2 and Supplementary Table 7). Compared with the ctDNA-negative before surgery group, the RFS of ctDNA-positive group was shorter (HR = 3.88, 95% CI: 1.46–10.33, P = 0.007; I² = 56.7%) for two studies. A total of seven studies reported the association between ctDNA measured after surgery and RFS (HR = 5.08, 95% CI: 3.35–7.72, P < 0.001; I² = 0). Among them, six studies measured ctDNA just after surgery, while one study did not specify the exact time of measurement after surgery. Additionally, Campani et al. found that positive ctDNA at first percutaneous ablation was also significantly associated with shorter RFS (HR = 1.73, 95% CI: 1.09–2.75, P = 0.002).¹⁵

Fig. 2.

Forest plot of the association between ctDNA detection at different timepoint and relapse-free survival. "Before surgery" refers to the period prior to surgical resection or liver transplantation; "after surgery" indicates the period following surgical resection or liver transplantation, which can be further divided into "just after surgery" (within 1–4 weeks post-surgery) and "long-term post-treatment surveillance." CI, confidence interval; ctDNA, circulating tumor DNA; HR, hazard ratio.

3.4. ctDNA status and recurrence

In the four studies that reported the association between ctDNA and the rate of recurrence, compared with patients with negative ctDNA, positive ctDNA before surgery was associated with a higher rate of recurrence (RR = 2.01, 95% CI: 1.22–3.32, P = 0.006; I² = 0, Fig. 3). For ctDNA measured after surgery, the rate of recurrence was also higher in ctDNA-positive group (RR = 2.50, 95% CI: 1.40–4.45, P = 0.002; I² = 51.7%) among seven studies.

Fig. 3.

Forest plot of the association between ctDNA detection at different timepoint and recurrence. "Before surgery" refers to the period prior to surgical resection or liver transplantation; "after surgery" indicates the period following surgical resection or liver transplantation, which can be further divided into "just after surgery" (within 1–4 weeks post-surgery) and "long-term post-treatment surveillance." CI, confidence interval; ctDNA, circulating tumor DNA.

3.5. ctDNA status and OS

Similar with RFS, ctDNA measured at different timepoints were significantly associated with shorter OS (Supplementary Fig. 2 and Supplementary Table 7). The HRs were 6.59 (95% CI: 2.47–17.55, P < 0.001; I² = 0) for ctDNA measured before any anti-tumor treatment, 7.01 (95% CI: 2.21–22.27, P = 0.001) for ctDNA measured after surgery, and 2.18 (95% CI: 1.26–3.74, P = 0.005) for ctDNA measured during percutaneous ablation, respectively.

3.6. ctDNA status and AFP

A total of nine studies simultaneously reported the associations of both ctDNA status and AFP with prognostic outcomes (Supplementary Table 8). Among them, five studies demonstrated a significant association between ctDNA and RFS, but not between AFP and RFS, whereas two studies identified significant associations of both ctDNA and AFP with RFS. Regarding OS, four studies revealed a significant association between ctDNA and OS but not between AFP and OS, while one study reported that both ctDNA and AFP served as independent prognostic predictors for patients with HCC.

3.7. Lead time

Four studies reported lead time from ctDNA detection to radiographic recurrence. Cai et al. discovered that the lead time was 4.6 months on average.²³ Guo et al. found that the median interval from the first positive longitudinal ctDNA to imaging recurrence was four months.³⁷ Marron et al. identified relapse based on ctDNA with a median lead time of five months before imaging recurrence.³⁵ Abdelrahim et al. found that one patient tested ctDNA positive two months prior to imaging.³²

3.8. ctDNA dynamics

Xia et al. found that patients from positive or negative at 7–10 days after surgery to negative at 1–7 days after adjuvant therapy have superior RFS than those without ctDNA clearance (from positive or negative to positive) (not reached vs 165.5 days, P < 0.001).²⁷ However, patients with favorable ctDNA change during neoadjuvant treatment and during surgery did not achieve longer RFS than those without ctDNA clearance, and the median RFS were 326 days vs not reached (P = 0.720) and 326 days vs 205 days (P = 0.930), respectively. Campani et al. found that compared to patients who remained ctDNA-negative before and after percutaneous ablation (ctDNA−/−), those with ctDNA+/+ had lower RFS and OS (HR = 2.66, 95% CI: 1.52–4.63; HR = 4.76, 95% CI: 2.58–8.78, respectively). However, the differences of RFS and OS between the ctDNA−/− group and the ctDNA+/− or ctDNA−/+ patients were not statistically significant.¹⁵

3.9. Sensitivity analysis and publication bias

Sensitivity analysis of RFS and recurrence were shown in Supplementary Figs. 3–14. Whether replacing HRs from the univariable analyses with the multivariable HRs or excluding conference abstracts, the results of RFS and recurrence remained similar to the main analysis (Supplementary Figs. 3–5). We reanalyzed the studies that explicitly employed NGS (Supplementary Fig. 6). For postoperative samples, the findings were consistent with the main analysis. However, for preoperative samples, no statistically significant association was observed (HR = 1.79, 95% CI: 0.81–3.95). Additionally, we performed three sensitivity analyses, each incorporating only studies that met specific criteria: those defining ctDNA positivity as ≥1 mutation, those utilizing commercially available ctDNA testing sources, and those including only patients with stage I–III disease. As shown in Supplementary Figs. 7–12, the results for RFS across different time points remained consistent with the main analysis. For recurrence, the associations were no longer statistically significant, although the HRs consistently remained >1. Using leave-one-out methods, the associations between ctDNA measured after surgery with RFS and recurrence were also robust (Supplementary Figs. 13,14). Risks of publication bias were graphically summarized in funnel plots in Supplementary Fig.15 and Fig.16, and the results showed no apparent potential publication bias in the RFS and recurrence analysis of ctDNA after surgery (RFS: P_{Egger test} = 0.313, recurrence: P_{Egger test} = 0.912).

3.10. ctDNA methylation

Six studies analyzed the prognostic value of ctDNA methylation in HCC. Xu et al. demonstrated that a combined prognosis score (cp-score) incorporating ctDNA methylation markers served as an independent predictor of survival (HR = 2.41, 95% CI: 1.90–3.04 in the training set; HR = 1.55, 95% CI: 1.25–1.92 in the validation set), whereas AFP was no longer significant as a risk factor after adjustment for cp-score and clinical covariates.³⁸ Similarly, Liu et al. reported that CCND1 methylation status was independently associated with both OS (HR = 0.27, 95% CI: 0.09–0.80) and PFS (HR = 0.11, 95% CI: 0.03–0.38), while AFP showed no significant association with either outcome.³⁹ In line with these findings, Guo et al. developed the HepaAiQ assay based on 20 HCC-specific differentially methylated regions, and confirmed that postoperative HepaAiQ status was an independent predictor of recurrence, whereas preoperative AFP failed to show predictive value.⁴² Yang et al. applied two methylation-based approaches for minimal residual disease (MRD) detection, both of which were strongly associated with recurrence risk.⁴³

In contrast, Saeki et al. identified both methylated SEPT9 gene (HR = 2.01, 95% CI: 1.13–3.59) and AFP (HR = 3.95, 95% CI: 2.07–7.55) as independent predictors of OS.⁴⁰ Similarly, Angeli-Pahim et al. calculated the tumor methylation score (TMS) by measuring the difference between the amount of methylation in the plasma and buffy coat. They found that both ΔTMS and ΔAFP were significantly different between tumor progression group, stability group, and response group (P = 0.035 and 0.034, respectively).⁴¹

4. Discussion

In this systematic review and meta-analysis, our findings suggest that positive ctDNA, whether detected preoperatively or postoperatively, and ctDNA methylation, can serve as a minimally invasive biomarker that provides clinically relevant prognostic information for patients with HCC. ctDNA detection also has the potential to predict disease recurrence months in advance compared to imaging examinations, which helps in more accurate and timely risk stratification, treatment planning, and surveillance.

The results of this study align with those of a previous systematic review evaluating the prognostic value of ctDNA in Asian HCC patients, although the earlier study did not differentiate the time points of ctDNA sampling.⁴⁴ Genomic analyses have established TERT, TP53, and CTNNB1 as major driver mutations in HCC, serving as key molecular indicators. The studies included in our review identified high frequencies of mutations in TP53, TERT, CTNNB1, and other genes in ctDNA^15,26,28,29, and ctDNA positivity was generally defined as the detection of at least one mutation among the monitored genes. When compared to tissue samples, the mutation profiling of ctDNA showed similar mutation frequencies in these known HCC driver genes.⁴⁵ Among all somatic mutations, TP53 is the most frequently mutated gene in HCC and its mutations were significantly associated with worse OS in HCC patients.⁴⁶ In addition, the detection of TERT promoter mutations in plasma may reflect an abundance of immortal and rapidly proliferating cells, which contributes to poor prognosis.⁴⁷ Taken together, these findings provide a biological rationale and clinical feasibility for using ctDNA testing as a prognostic tool in HCC, whereby patients with ctDNA positivity generally exhibit shortened RFS and OS.

Our study found that both preoperative and postoperative ctDNA testing were significantly associated with RFS or recurrence. The HR for postoperative ctDNA was higher than for preoperative ctDNA, suggesting that the postoperative time point may be a more indicative one. Zhu et al. observed that the ctDNA positivity rate decreased from 63.4% before surgery to 46% after surgery, which may be due to the reduced circulation of ctDNA in the bloodstream following curative surgery, with some patients transitioning to negative ctDNA results.²⁶ For those who remained ctDNA-positive postoperatively, the risk of tumor recurrence was higher. However, a single ctDNA measurement may not be sufficient to fully characterize MRD or reliably predict long-term prognosis, as ctDNA shedding from small residual lesions is intermittent or below the assay’s detection threshold at certain time points. As illustrated by the two studies described in our results, persistent or recurrent ctDNA positivity, rather than a single isolated measurement, appears to better capture MRD and may help identify patients who could benefit from intensified surveillance, adjuvant therapy, or early interventions. Nevertheless, the available data on ctDNA dynamics are limited and heterogeneous with respect to sampling schedules and definitions. Studies with standardized sampling time points and harmonized definitions of ctDNA dynamics are needed to validate the role of ctDNA dynamics in MRD monitoring and treatment decision-making in HCC. ctDNA has also been investigated in other liver-related malignancies. In patients with colorectal liver metastases (CRLM), several studies have shown that postoperative ctDNA is strongly associated with an increased risk of recurrence and poorer survival.48, 49, 50 Serial ctDNA monitoring can further refine risk stratification and predicts prognosis more accurately than imaging or serum markers.^51,52 Emerging data also indicate that, in patients with extrahepatic cholangiocarcinoma, ctDNA status and dynamics can predict recurrence during adjuvant therapy and outperform traditional biomarkers.⁵³ These observations are highly consistent with our findings in HCC and may indirectly suggest that liver tumors exhibit organ-specific characteristics in terms of tumor shedding patterns and perioperative ctDNA dynamics. Liver resection and transplantation can markedly alter hepatic blood flow and may thereby contribute to the rapid and pronounced perioperative changes in ctDNA levels,⁵⁴ although the exact mechanisms remain incompletely understood. Therefore, when interpreting the perioperative value of ctDNA in liver-related malignancies, it is necessary to consider the liver’s unique hemodynamic and histologic context.

Beyond AFP, several advanced AFP-derived biomarkers have been developed, including AFP-L3 (the fraction of AFP bound to Lens culinaris agglutinin). HCC cells supplied by arterial blood differentially express uridine diphosphate alpha(1,6)-fucosyltransferase, which attaches fucose residues to AFP, resulting in differential binding to Lens culinaris agglutinin, termed AFP-L3.⁵⁵ Multiple studies have shown that AFP-L3 can predict HCC diagnosis and poor prognosis, and it appears to be more powerful than AFP in predicting early HCC recurrence.^56,57 Nonetheless, AFP-L3 also has limitations. Tumors that express AFP-L3 are thought to be more aggressive,⁵⁸ with the potential for rapid growth and early metastasis, so AFP-L3 may not be suitable for risk assessment in all patients with HCC. In addition, test availability and cost may limit its routine use in some regions. The prognostic value of ctDNA and AFP-L3 requires further direct comparison.

Some studies have identified other liquid biopsy markers, such as circulating tumor cells (CTCs) and non-coding RNAs (ncRNAs, e.g., microRNAs and long non-coding RNAs), as predictors of recurrence or progression risk.59, 60, 61 CTCs are cells that migrate from a primary or secondary tumor into nearby vasculature, thereby entering the systemic circulation.⁶² The isolation and characterization of CTC are technically challenging and expensive, and the detection rate of CTCs is relatively low in some studies, especially at early stages.^25,63,64 Ge et al. evaluated the association of ctDNA and CTC counts with OS.²⁵ Their study found that ctDNA positivity was a significant risk factor for OS, whereas CTC count (<2 vs. ≥2) showed no significant association. Another drawback of CTCs is their apparent lack of specificity, as its detection relies on the use of pan-cancer markers such as epithelial cell adhesion molecule.⁶⁵ These factors make it challenging to integrate CTC-based techniques into routine HCC surveillance. NcRNAs function as key regulators in critical processes such as chromatin alterations, and DNA transcription, rather than being translated into proteins. MicroRNAs have been proposed as promising biomarkers for early detection in other malignancies, such as lung and pancreatic cancer.^66,67 However, few studies have specifically evaluated the prognostic value of ncRNAs in HCC, and robust comparative data versus ctDNA are lacking. Furthermore, clinical implementation of ncRNAs remains restricted by heterogeneous detection platforms, the absence of standardized normalization and reporting methods, and an incomplete understanding of their biological contribution to HCC.⁶²

Compared to other biomarkers, ctDNA has a higher detection rate in patients with HCC and can be analyzed relatively easily for mutations using NGS or ddPCR. Nevertheless, pre-analytical factors (e.g., choice of collection tubes, processing delays), intra- and extrahepatic tumor heterogeneity, and assay-specific limits of detection can lead to false-positive or false-negative results. Standardized vein-to-analysis workflows will be essential to minimize these sources of error. Overall, ctDNA are generally more scalable and easier to implement than CTC or ncRNA profiling, although they remain more costly and less widely available than AFP and imaging in many settings. Formal cost-effectiveness analyses and real-world implementation studies will therefore be essential to define how ctDNA-based strategies can be integrated into routine HCC surveillance and management.

In the absence of a clinical gold standard or consensus on how ctDNA should be used in patients with HCC, there is substantial heterogeneity across studies in the methods for ctDNA measurement, the assay panels employed, limits of detection, genes covered, and the operational definition of ctDNA positivity. This lack of harmonized cut-offs complicates clinical decision-making, because a “positive” result does not consistently trigger the same escalation strategy (e.g., adjuvant therapy, intensified imaging) across centers. Consensus thresholds that are explicitly linked to predefined clinical actions are therefore urgently needed.

In the included studies, NGS was more commonly used than ddPCR, likely due to differences in their technical characteristics. DdPCR improves upon conventional allele-specific PCR amplification by partitioning DNA samples into a large number of smaller reactions, thereby enabling absolute quantification with enhanced sensitivity.⁶⁸ Although ddPCR is highly sensitive and easy to operate, its low throughput and inability to detect novel mutations limit its clinical applicability for HCC.⁶⁹ In contrast, NGS can cover a broader range of mutations by examining the whole sequences of genes of interest.⁷⁰ The high sequencing depth of NGS allows for the detection of more low-frequency mutations. Additionally, the definition of ctDNA positivity varied, with some studies considering the presence of at least one mutation, while others required the presence of at least two mutations (Supplementary Table 4). In specific clinical applications, it is essential to carefully consider the gene panel used for MRD monitoring. Including more genes reduces the risk of false negatives but significantly increases costs, limiting its practical use in clinical settings. Balancing a comprehensive mutation profile with reasonable cost constraints remains a key challenge in the clinical application of ctDNA detection.

This study has several strengths. First, to our knowledge, it represents the first systematic review and meta-analysis to integrate all available evidence on the relationship between ctDNA detection, ctDNA methylation, and prognostic outcomes. Second, instead of analyzing all time points collectively, we examined the association between ctDNA and prognostic outcomes at multiple treatment time points. Third, we performed various sensitivity analyses to validate the robustness of our results.

However, our study also has some limitations. First, the number of available studies is limited, and the sample sizes are relatively small. Due to the limited number of included studies, subgroup analysis based on treatment strategy was not feasible. Although we compiled the study-specific definitions of ctDNA positivity, these criteria varied widely and we were unable to harmonize these thresholds across studies. This heterogeneity in positivity definitions, along with differences in ctDNA assays, HBV prevalence, age, tumor stage, and treatment strategies, likely contributes to residual between-study variability and may affect pooled HRs; thus, our findings should be interpreted cautiously, especially for specific subgroups. Additionally, we included only studies that reported qualitative ctDNA results to ensure interpretability. The prognostic impact of quantitative changes in ctDNA levels remains unclear. Furthermore, there is likely to be residual lead time bias, particularly with respect to other modalities such as AFP and imaging, and the limited available data make it difficult to adequately adjust for this. Moreover, the current evidence is insufficient to conclusively demonstrate that the prognostic value of ctDNA is superior to that of AFP. Lastly, given that the U.S. Food and Drug Administration's first approval of liquid biopsy tests occurred in 2016⁷¹, we included only studies published between 2016 and 2024. However, another study found that most relevant studies were published after 2018⁷², suggesting that this inclusion criterion likely had minimal impact on our results.

5. Conclusions

This study demonstrates that both preoperative and postoperative ctDNA testing can effectively detect MRD and identify HCC patients with poor prognosis. Our findings highlight the potential clinical value of ctDNA-based prediction of recurrence in patients with HCC, providing both clinicians and patients with a reliable tool for prognostic assessment. Future prospective validation studies should adopt harmonized sampling and analytic pipelines, prespecify ctDNA positivity criteria tied to clinical endpoints, and test integrated models that fuse serial ctDNA with imaging findings and clinicopathological covariates to refine risk-adapted management.

CrediT author contribution statement

Meng Zhang: Conceptualization, Methodology, Data curation, Visualization, Formal analysis, Writing-original draft. Xiaowei Chen: Data curation, Methodology, Writing-Review & Editing. Qingxin Zhou: Data curation, Formal analysis, Writing-Review & Editing. Nana Guo: Data curation, Writing-Review & Editing. Baoshan Cao: Conceptualization, Supervision, Writing-Review & Editing. Hongmei Zeng: Conceptualization, Supervision, Writing-Review & Editing. Wanqing Chen: Conceptualization, Supervision, Writing-Review & Editing. Feng Sun: Conceptualization, Funding acquisition, Supervision, Writing-Review & Editing.

Declaration of competing interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.