Diagnostic accuracy of circulating long noncoding RNAs in hepatocellular carcinoma: a systematic review and meta-analysis

Abstract

Background

Hepatocellular carcinoma (HCC) remains a leading cause of cancer-related mortality worldwide, largely due to late diagnosis and the limited sensitivity of current screening biomarkers such as alpha-fetoprotein (AFP). Circulating long noncoding RNAs (lncRNAs) have recently emerged as promising, noninvasive biomarkers that may reflect the molecular mechanisms underlying hepatocarcinogenesis. This systematic review and meta-analysis aimed to comprehensively evaluate the diagnostic accuracy of circulating lncRNAs for detecting HCC.

Methods

A systematic search of PubMed, Scopus, Web of Science, and Embase was performed up to March 2025 following the PRISMA-DTA guidelines. Eligible studies reporting the diagnostic performance of circulating lncRNAs for HCC were included. Pooled sensitivity, specificity, and diagnostic odds ratio (DOR) were calculated using a bivariate random-effects model. Subgroup analyses were conducted according to sample type, detection method, and reference standard.

Results

Eighty-one studies encompassing over 6,000 participants were included. The pooled sensitivity and specificity of circulating lncRNAs were 0.83 (95% CI: 0.80–0.86) and 0.80 (95% CI: 0.75–0.84), respectively, with an area under the summary receiver operating characteristic (SROC) curve of 0.88. Serum-based assays showed slightly higher accuracy than plasma-based assays. lncRNAs also demonstrated good diagnostic performance in discriminating HCC from cirrhotic patients, as well as from patients with HBV–positive and HCV–positive status.

Conclusion

Circulating lncRNAs exhibit high diagnostic accuracy and hold significant potential as complementary biomarkers to AFP for early HCC detection. Their mechanistic roles in tumor proliferation and immune regulation underscore their value in molecular diagnosis and personalized management of liver cancer.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12876-026-05049-0.

Introduction

Liver cancer ranks as the sixth most commonly diagnosed malignancy and the third major cause of cancer mortality worldwide, posing a significant global health concern impacting millions of people [1]. Hepatocellular carcinoma (HCC) is the most prevalent form of liver cancer, representing approximately 90% of cases, and is generally associated with a poor prognosis [2]. HCC develops due to the combined effects of genetic and epigenetic factors, which interact at all stages of liver cancer. Primary causes include chronic hepatitis B and C infections, continued exposure to alcohol and harmful substances, and non-alcoholic fatty liver disease [3]. This multifactorial etiology has been further supported by recent clinical studies [4, 5]. Although understanding of the molecular mechanisms of HCC has advanced and treatment options have improved, patient survival rates remain suboptimal. In fact, HCC typically presents without clear symptoms, causing 70% of patients to be diagnosed at advanced stages, when radiotherapy and chemotherapy are generally less effective, and surgical treatment is no longer an option [6, 7].

Currently, HCC detection relies on abdominal ultrasonography (US) and elevated serum alpha-fetoprotein (AFP) levels. Although the US is accurate, it may fail to identify small nodules. AFP is the most widely utilized serological marker. However, serum AFP levels remain normal in up to 40% of HCC patients, especially during the early stages of disease. Both methods, therefore, have significant limitations [6, 8, 9]. To improve the limitations of AFP, several multimarker models have been developed. The GALAD score, which combines gender, age, AFP, AFP-L3, and PIVKA-II, has shown promising results for the detection of HCC, including early-stage tumors. Nevertheless, novel biomarkers are still being investigated to further improve diagnostic accuracy and support early detection [6, 10, 11]. Over the past few decades, long noncoding RNAs (lncRNAs) have been identified as key regulators in various biological processes [12]. LncRNAs are defined as transcripts longer than 200 nucleotides that do not code for proteins and are broadly expressed across multiple organs and tissues throughout the body [13–15]. They play regulatory roles in numerous physiological processes, including cell proliferation, differentiation, and apoptosis [16]. Circulating lncRNAs are present in various biological samples, including serum and plasma. However, their measured levels can be affected by pre-analytical factors. For instance, the coagulation process can cause blood cells and extracellular vesicles to release the RNA, while remaining platelet contamination can affect plasma samples [17]. Such differences can impact the concentration and stability of circulating lncRNAs and may contribute to the variability observed in the diagnostic performance of serum and plasma lncRNAs.

Some studies have identified that unusually elevated or reduced levels of lncRNAs in body fluids may function as biomarkers for the early detection of cancer [18]. Extensive research has emphasized the important role of lncRNAs in HCC development. More than 74 lncRNAs with altered expression have been identified in HCC, including 52 upregulated and oncogenic, and 22 downregulated and tumor-suppressive [19]. This widespread dysregulation has led to the investigation of numerous lncRNAs as potential biomarkers for HCC [12, 18, 19]. Although these molecules differ in sequence and biological function, many are associated with common molecular changes underlying hepatocarcinogenesis. Thus, despite their biological heterogeneity, circulating lncRNAs may collectively provide valuable diagnostic information for HCC. Prominent lncRNAs, such as HULC (Highly Upregulated in Liver Cancer), MALAT1 (Metastasis-Associated Lung Adenocarcinoma Transcript 1), HOTAIR (HOX Transcript Antisense RNA), and HEIH (HBV Enhancer-Induced lncRNA), have been linked to various clinicopathological characteristics and patient prognoses [19–21].

Nevertheless, there has been debate regarding the sensitivity and specificity of various lncRNAs for early detection in HCC patients. Recent evidence has revealed that dysregulated lncRNAs are key regulators of hepatocarcinogenesis. Aberrant expression of specific lncRNAs such as HULC, MALAT1, HOTAIR, and LINC00152 has been implicated in promoting cell proliferation, angiogenesis, and metastasis through modulation of oncogenic pathways, including Wnt/β-catenin, PI3K/AKT, and TGF-β signaling. These lncRNAs can also influence epigenetic regulation and immune microenvironment remodeling in the liver. Such mechanistic insights provide a biological rationale for exploring circulating lncRNAs as noninvasive diagnostic biomarkers for hepatocellular carcinoma. Therefore, this systematic review and meta-analysis aimed to comprehensively evaluate the diagnostic accuracy of circulating lncRNAs for HCC detection.

Methods

Protocol and registration

In the current study, we adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for diagnostic test accuracy (PRISMA-DTA). A PRISMA DTA 2020 checklist is available in the Supplementary Table 1. The article was registered in the International Prospective Register of Systematic Reviews (PROSPERO) under the registration number CRD420251069028.

Eligibility criteria

Studies reporting the diagnostic accuracy of circulating lncRNAs expression for distinguishing HCC cases from healthy controls, patients with cirrhosis, hepatitis C virus (HCV), or hepatitis B virus (HBV), were included in our study. The language of the included studies was limited to English. Studies with case reports or case series designs, as well as abstracts, were excluded.

Information sources and search strategies

We systematically searched four main databases, including PubMed, Scopus, Embase, and Web of Science, up to June 2025. The following Keywords and syntaxes were used in the PubMed search: (lncRNA OR long non-coding RNA OR Long Untranslated RNA OR Long Non-Protein-Coding RNA) AND (cancer OR tumor OR neoplasm OR malignant OR metastasis OR carcinoma OR hepatocellular carcinoma OR HCC) AND (liver or hepatocyte OR hepatic) AND (Serum OR blood OR plasma OR sensitivity OR specificity OR AUC OR areas under the curve). The detailed search strategies of each database are available in Supplementary Table 2.

Selection and data collection process

Two independent reviewers (B.S. and N.B.) independently selected articles by evaluating their titles and abstracts, and/or full texts, against the eligibility criteria using EndNote Computer Software (Version 21, Clarivate). Any disagreement in the selection process was resolved by the third author (E.T.). The data collection was performed using a standard spreadsheet by two reviewers (B.S. and N.B.). The primary index test was lncRNA expression, and the reference standard was the presence of HCC. The following study characteristics and population details were extracted from the included studies: first author’s name, year of publication, location, type of lncRNA, specimen, detection method of lncRNA, comparison group (healthy, cirrhotic, HCV, and HBV status), average age of participants, true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN), AUC, sensitivity, and specificity of LncRNA expression.

Study risk of bias assessment

We evaluate the risk of bias within the included studies using the QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies) tool [22]. This tool evaluates articles in four main domains: patient selection, index test, reference standard, and flow and timing. The assessment was independently performed by two reviewers (F.J. and F.O.S.), with any discrepancies adjudicated by a third reviewer (M.P.).

Statistical analysis

First, we constructed a 2 by 2 table containing TP, TN, FP, and FN for each study to synthesize the pooled sensitivity, specificity, and 95% CIs of lncRNAs in detecting HCC. We conducted a meta-analysis of diagnostic test accuracy (DTA) using the bivariate random-effects approach and created the summary receiver operating characteristic (SROC) plots [23]. The size of the point estimate for each study was calculated by its relative weight in the random-effects univariate diagnostic odds ratio (DOR) model. For each analysis, the AUC and its confidence interval were estimated using a bivariate model with 2,000 bootstrap samples to ensure robust and reliable interval estimation. We further evaluated the clinical utility of lncRNAs using Fagan plots and likelihood ratio scattergrams. A positive likelihood ratio (PLR) exceeding 10 was considered strong evidence supporting the confirmation of HCC, while a negative likelihood ratio (NLR) less than 0.1 was suggested as evidence for ruling out HCC. Fagan nomograms were plotted for pre-test probabilities of 20% based on the bivariate model estimates [24].

The I² metric, based on the method proposed by Holling et al., was used to quantify between-study heterogeneity [25]. The observed heterogeneity was classified as low for I² < 25%, moderate for 25% < I² < 75%, and high for I² > 75% in our analysis. To evaluate the risk of publication bias, we used a visual assessment of paired funnel plots for sensitivity and specificity to find asymmetry, as well as the Generalized Egger’s statistical test using 2000 sample bootstraps [26]. All statistical analyses were conducted in R software (Version 4.5.0) using “Mada”, “MVPBT” [27], “dmetatools” [28], “Metafor” [29], and “meta” [30] packages.

Results

Search results

Our primary search identified 7125 relevant papers from PubMed, Web of Science, Scopus, and Embase libraries. After removing duplicate articles (n = 2915) and excluding studies based on title and abstract screening (n = 4098), the remaining studies were evaluated by the full text (n = 132), and eligible studies were included in our review (n = 81) [31–111]. The PRISMA flow diagram shows detailed information about the screening process (Fig. 1). A list of excluded studies, along with the reasons for their exclusion, is provided in Supplementary Table 3.

Fig. 1.

PRISMA flowchart of the screening process

Study characteristics

A total of 6196 participants with HCC, 3947 healthy controls (HC), 1167 patients with HBV infection, 919 with HCV infection, and 1273 with liver cirrhosis (LC) were evaluated across the 81 studies included in our review. Most studies were conducted in China (n = 46), followed by Egypt (n = 30), South Korea (n = 2), Japan (n = 1), Spain (n = 1), and Thailand (n = 1), and were published between 2014 and 2024. A summary of study characteristics is provided in Supplementary Table 4. Almost all studies employed quantitative real-time polymerase chain reaction (qRT-PCR) to measure circulating lncRNAs, except for three studies that used q-PCR [57, 64, 100], and two studies that used Droplet digital PCR (ddPCR) [61, 62]. Forty-eight studies evaluated the diagnostic value of serum lncRNA, 26 examined plasma, five assessed whole blood [58, 70, 73, 78, 79], one examined peripheral blood mononuclear cells (PBMCs) [60], and one investigated saliva [100]. The details of the quality assessment of each study based on the QUADAS-2 tool are presented in Supplementary Tables 5, and the overall risk of bias among the included studies is demonstrated in Fig. 2.

Fig. 2.

Risk of bias of the included studies according to QUADAS-2

Meta-analysis

Of the 81 studies included, 55 studies were utilized in the meta-analysis. Forty-one studies were used for the comparison between HCC and HC, 19 for HCC and LC, nine for HCC and HCV+ patients, and five for HCC and HBV+ patients. The findings of the 26 studies not included in the meta-analysis are summarized in Table 1. Seven of these studies were excluded because they used specimens other than serum or plasma, such as whole blood, PBMCs, or saliva (Table 1, Study ID 1–7) [58, 60, 70, 73, 78, 79, 100]. Another seven studies were excluded due to comparing HCC with mixed or alternative groups, such as a combination of HC and LC groups, or hepatitis (Table 1, Study ID 8–14) [49, 59, 61, 69, 87, 99, 101]. Finally, 12 studies were excluded because they only reported AUC without providing sensitivity or specificity data (Table 1, Study ID 15–26) [41, 46, 51, 53, 55, 64, 66, 68, 93, 94, 105, 106].

Table 1.

Summary of Findings (Studies not included in the meta-analysis)

Study ID.	Author	Year	Country	Specimen	Type of lncRNA	Group 1 (N)	Group 2 (N)	AUC	Sensitivity %	Specificity %	Cut-off
1	Khalil et al.	2024	Egypt	Whole Blood	HOST2	HCC (60)	LC (60)	0.844	88.33	78.33	0.8765
1	Khalil et al.	2024	Egypt	Whole Blood	HOTAIR	HCC (60)	LC (60)	0.964	91.67	96.67	1.999
1	Khalil et al.	2024	Egypt	Whole Blood	HOXA-AS2	HCC (60)	LC (60)	0.875	88.33	70	1.5019
1	Khalil et al.	2024	Egypt	Whole Blood	MALAT-1	HCC (60)	LC (60)	0.985	96.67	90	4.632
2	Kunadirek et al.	2022	Thailand	PBMCs	MIR-4435-2HG	HCC HBV+ (100)	HC / CHB (200)	0.8	75	75	1.5781
2	Kunadirek et al.	2022	Thailand	PBMCs	SNHG9	HCC HBV+ (100)	HC / CHB (200)	0.68	66	67	1.3579
2	Kunadirek et al.	2022	Thailand	PBMCs	LCP2-1	HCC HBV+ (100)	HC / CHB (200)	0.61	64.29	42.64	1.5105
2	Kunadirek et al.	2022	Thailand	PBMCs	POLD3-2	HCC HBV+ (100)	HC / CHB (200)	0.78	74	75	0.036
3	Mo et al.	2022	China	Whole Blood	LINC01793	HCC HBV+ (52)	HC (30)	0.824	67.3	100	NR
4	Mostafa et al.	2021	Egypt	Whole Blood	AF085935	HCC HCV+ (60)	HC / HCV+ (120)	0.966	100	90	3.2
5	Rashad et al.	2021	Egypt	Whole Blood	AFAP1-AS1	HCC (12)	LC (15)	0.816	83.3	94	104
6	Refai et al.	2019	Egypt	Whole Blood	TUG1	HCC HCV+ (30)	HC / HCV+ (40)	NR	93.3	100	20.6
6	Refai et al.	2019	Egypt	Whole Blood	CASC2	HCC HCV+ (30)	HC / HCV+ (40)	NR	67	78	1.018
7	Xie et al.	2018	China	Saliva	Lnc-PCDH9-13:1	HCC HBV+ (100)	HC (50)	0.898	85	98	NR
7	Xie et al.	2018	China	Saliva	Lnc-PCDH9-13:1	HCC HBV+ (100)	CHB (50)	0.896	87	98	NR
7	Xie et al.	2018	China	Saliva	Lnc-PCDH9-13:1	HCC HBV+ (100)	LC (50)	0.881	87	92	NR
8	Golam et al.	2024	Egypt	Serum	MALAT1	HCC HCV+ (36)	HC / HCV+ (20)	NR	72.2	55.4	2.65
8	Golam et al.	2024	Egypt	Serum	CASC2	HCC HCV+ (36)	HC / HCV+ (20)	NR	97.2	60.5	0.995
9	Konishi et al.	2016	Japan	Plasma	MALAT1	HCC HCV + HBV+ (880)	HBV+ / HCV+ / NAFLD / NASH (28)	0.66	51.1	89.3	1.6
10	Li et al.	2020	China	Serum	NONHSAT053785	HCC (112)	HC / CHB (195)	0.801	73.2	75.4	NR
11	Matboli et al.	2020	Egypt	Serum	RP11-583F2.2	HCC (60)	HC / CHC (60)	0.946	96.7	91.7	5.02
12	Shaker et al.	2020	Egypt	Serum	HOTAIR	HCC CHC+ (50)	HC / CHC (95)	0.781	80	68	3.13
13	X. Ma et al.	2016	China	Plasma	DANCR	HCC (52)	HC / CHB / LC (94)	0.868	83.8	72.7	1.65
13	X. Ma et al.	2016	China	Plasma	DANCR	HCC (52)	CHB / LC (51)	0.864	80.8	84.3	1.64
14	Xu et al.	2019	China	Serum	LINC00978	HCC (58)	Hepatitis / LC (94)	0.91	0.76	0.96	NR
15	Dong et al.	2019	China	Serum	MEG3	HCC (54)	HC (54)	0.8865	NR	NR	NR
16	Fu et al.	2022	China	Serum	AC005332.5, ELF3-AS1, LINC00665	HCC HBV+ (76)	HC (36)	0.809 0.815 0.852	NR	NR	NR
17	Gong et al.	2018	China	plasma	nc-HOXC8-143, XLOC_000667, AK123675	HCC (200)	CHB (100)	0.821 0.660 0.728	NR	NR	NR
18	Huang et al.	2018	China	Serum	LINC-ITGB1	HCC (80)	HC (44)	0.852	NR	NR	NR
19	Jiang et al.	2020	China	Serum	HAND2-AS1	HCC (44)	HC (32)	0.88	NR	NR	NR
19	Jiang et al.	2020	China	Serum	HAND2-AS1	HCC (44)	HBV+ (38)	0.8792	NR	NR	NR
20	Liu et al.	2020	China	Plasma	MACC1-AS1	HCC (80)	HC (80)	0.91	NR	NR	NR
21	Lu et al.	2014	China	Serum	uc003wbd	HCC (137)	HC (138)	0.86	NR	NR	NR
21	Lu et al.	2014	China	Serum	uc003wbd	HCC (137)	HBV+ (104)	0.7	NR	NR	NR
21	Lu et al.	2014	China	Serum	AF085935	HCC (137)	HC (138)	0.96	NR	NR	NR
21	Lu et al.	2014	China	Serum	AF085935	HCC (137)	HBV+ (104)	0.86	NR	NR	NR
22	Ma et al.	2019	China	Plasma	PAPAS	HCC (74)	HC (52)	0.88	NR	NR	NR
23	Wang et al.	2015	China	Serum	uc001ncr	HCC HBV+ (61)	HC / HBV+ (120)	0.9466	NR	NR	NR
23	Wang et al.	2015	China	Serum	AX800134	HCC HBV+ (61)	HC / HBV+ (120)	0.8877	NR	NR	NR
24	Wang et al.	2020	China	Plasma	SENP3-EIF4A1	HCC (50)	HC (50)	0.8028	NR	NR	NR
25	Yao et al.	2020	China	Serum	lnc-FAM72D-3, lnc-EPC1-4, lncZEB2-19, lnc-GPR89B-15	HCC (45)	HC (45)	0.584 0.576 0.852 0.717	NR	NR	NR
26	Yuan et al.	2017	China	Plasma	LINC00152, RP11-160H22.5, XLOC014172	HCC (100)	HC (100)	0.869 0.884 0.759	NR	NR	NR
26	Yuan et al.	2017	China	Plasma	LINC00152, RP11-160H22.5, XLOC014172	HCC (100)	CHB (100)	0.826 0.859 0.735	NR	NR	NR

Abbreviations: AUC Area Under the Curve, HBV+ Patients with Hepatitis B Virus infection, HC Healthy Controls, HCC Hepatocellular Carcinoma, HCV+ Patients with Hepatitis C Virus infection, lncRNAs Long Non-Coding RNA, NAFLD Non-alcoholic Fatty Liver Disease, NASH Non-alcoholic Steatohepatitis, LC Liver Cirrhosis, CHB Chronic Hepatitis B, CHC Chronic Hepatitis C, PBMCs Peripheral Blood Mononuclear Cells;

Diagnostic value of lncRNAs in differentiating HCC from healthy controls

The summary of meta-analysis findings is presented in Table 2. For the diagnosis of HCC compared with HC, circulating lncRNAs demonstrated a pooled sensitivity of 82.8% (95% CI 79.5–85.7%) and specificity of 79.6% (95% CI 75.0–83.6%) (Fig. 3). The heterogeneity based on the Holling sample size unadjusted approaches ranged from 64.2% to 82%. The summary ROC curve (SROC) revealed an AUC of 0.88 (95% CI 0.84–0.89) (Fig. 4). Subgroup analysis based on the type of specimen used for lncRNA detection showed that serum lncRNAs had a significantly greater AUC (0.90 vs. 0.81, p-value < 0.01), sensitivity (83.8% vs. 79.4%), and specificity (83.6% vs. 69.3%) compared to plasma lncRNAs (Figs. 3 and 4). Meta-regression analyses on several covariates, including country (Egypt vs. China), reference gene (Beta-actin, 18 S rRNA, and cel-miR-39 vs. GAPDH), mean age (< 56 vs. ≥ 56 years), and proportion of male participents (< 75% vs. ≥ 75%), revealed that none of these variables accounted for the observed heterogeneity (all p-values > 0.05). Fagan plot analysis demonstrated that, given a pre-test probability of 20%, the corresponding positive post-test probabilities were 56.28% for serum and 39.49% for plasma lncRNAs, while the negative post-test probabilities were 4.65% and 7.02%, respectively (Fig. 5A). Based on the likelihood ratio (LR) scatter plot presented in Fig. 5B, serum lncRNAs showed a moderate ability to rule-in or rule-out HCC in a positive or negative test (LR+ between 5 and 10, LR– between 0.1 and 0.2), whereas plasma lncRNAs exhibited a low ability (LR+ between 2 and 5, LR– between 0.2 and 0.5). Funnel plot visual assessment and the Generalized Egger’s test indicated the presence of publication bias (Fig. 6).

Fig. 3.

Forest plots showing the sensitivity and specificity of circulating lncRNAs in plasma and serum

Fig. 4.

Summary ROC (SROC) curve presenting the pooled AUC of circulating lncRNAs in plasma and serum

Fig. 5.

(A) Fagan nomogram plots and (B) likelihood scatter plot of circulating lncRNAs in plasma and serum

Fig. 6.

Funnel plots for the sensitivity and specificity of circulating lncRNAs in plasma and serum

Diagnostic value of lncRNAs in differentiating HCC from liver cirrhosis, HCV, and HBV

The summary of meta-analysis findings is presented in Table 2. Circulating lncRNAs demonstrated the ability to differentiate HCC from LC with an AUC of 0.84, a sensitivity of 73.9%, and a specificity of 79.8%. Heterogeneity was between 54.2% and 65.7%, and Egger’s test confirmed publication bias (Supplementary Figs. 1–4). There was no significant difference between the AUCs of serum 2and plasma lncRNAs (p-value = 0.316). For distinguishing HCC from HCV+ individuals, lncRNAs showed an AUC of 0.88, sensitivity of 79.8%, and specificity of 83.3%. Heterogeneity ranged from 53.6% to 77.6%, and Egger’s test indicated publication bias (Supplementary Figs. 5–8). Again, no difference between the AUCs of serum of plasma lncRNAs was found (p-value = 0.085). Finally, lncRNAs differentiated HCC from HBV+ patients with an AUC of 0.81, sensitivity of 76.7%, and specificity of 77.5%. Heterogeneity was between 63% and 76.8%, and no publication bias was detected (Supplementary Figs. 9–12). It should be noted that among the nine studies comparing lncRNAs between HCC and HBV+ patients, eight evaluated serum lncRNAs, while one assessed plasma lncRNA and was therefore excluded from the meta-analysis.

Table 2.

Summary of meta-analysis findings

Diagnostic Test Accuracy of lnaRNAs in HCC vs.	Number of Tests	Sample Size HCC / Control	AUC	Sensitivity %	Specificity %	DOR	posLR	negLR	Heterogeneity %a	Generalized Egger’s test P-value
Healthy Controls	59	5133 / 3843	0.88 (0.84, 0.89)	82.8 (79.5, 85.7)	79.6 (75.0, 83.6)	19.0 (13.8, 25.7)	4.09 (3.32, 5.02)	0.22 (0.18, 0.26)	64.2–82.0	< 0.001
Plasma	16	1142 / 1062	0.81 (0.73, 0.85)	79.4 (71.6, 85.5)	69.3 (60.5, 76.9)	8.98 (5.24, 14.4)	2.61 (2.01, 3.40)	0.3 (0.21, 0.41)	64.2–80.9
Serum	43	3991 / 2781	0.90 (0.86, 0.91)	83.8 (80.1, 86.9)	83.6 (78.5, 87.7)	26.7 (18.4, 37.6)	5.15 (3.92, 6.73)	0.20 (0.16, 0.24)	61.3–80.5
Cirrhotic Patients	36	1970 / 1483	0.84 (0.80, 0.85)	73.9 (68.4, 78.8)	79.8 (75.9, 83.1)	11.3 (8.65, 14.4)	3.66 (3.13, 4.28)	0.33 (0.27, 0.39)	54.2–65.7	< 0.001
Plasma	19	1018 / 800	0.82 (0.77, 0.84)	73.8 (66.2, 80.2)	78.4 (74.3, 82.0)	10.4 (7.07, 14.8)	3.42 (2.86, 4.07)	0.34 (0.26, 0.43)	65.5–72.6
Serum	17	952 / 683	0.86 (0.82, 0.88)	74.9 (66.4, 81.8)	84.3 (76.9, 89.7)	16.2 (11.8, 21.7)	4.84 (3.46, 6.67)	0.30 (0.23, 0.38)	22.1–34.4
HCV+ Patients	15	1066 / 970	0.88 (0.80, 0.91)	79.8 (68.3, 87.8)	83.3 (75.2, 89.2)	20.9 (10, 38.7)	4.87 (3.24, 7.2)	0.25 (0.15, 0.38)	53.6–77.6	< 0.001
Plasma	5	270 / 344	0.94 (0.77, 0.97)	81.4 (57.3, 93.5)	89.9 (79.9, 95.2)	55.4 (7.5, 203)	8.64 (3.39, 18.1)	0.23 (0.07, 0.5)	46.6–70
Serum	10	796 / 626	0.85 (0.76, 0.89)	81.2 (65.8, 90.6)	77.8 (68.9, 84.8)	16.3 (7.14, 32)	3.68 (2.64, 5.1)	0.25 (0.13, 0.43)	60.2–81.6
HBV+ Patients										0.15
Serum	8	447 / 463	0.81 (0.72, 0.86)	76.7 (67.6, 83.9)	77.5 (45.6, 93.4)	13.7 (3.41, 37.7)	4.09 (1.48, 10.8)	0.32 (0.23, 0.45)	63.0–76.8

Abbreviations: AUC Area Under the Curve, DOR Diagnostic Odds Ratio, HBV Hepatitis B Virus, HCC Hepatocellular Carcinoma, HCV Hepatitis C Virus, lncRNAs Long Non-Coding RNAs, negLR Negative Likelihood Ratio, posLR Positive Likelihood Ratio

^a I² estimate based on the Holling sample size unadjusted approaches

Discussion

Main finding

Our systematic review and meta-analysis synthesized findings from 81 studies to evaluate the diagnostic performance of long non-coding RNAs (lncRNAs) in distinguishing hepatocellular carcinoma (HCC) from various control groups. Overall, lncRNAs demonstrated high diagnostic accuracy, with pooled sensitivity and specificity of 82.8% and 79.6%, respectively, and an AUC of 0.88 when differentiating HCC patients from healthy individuals. Among sample types, serum-derived lncRNAs yielded superior performance compared to plasma assays. When compared with cirrhotic, HCV-positive, and HBV-positive controls, diagnostic accuracy remained robust (AUC 0.84, 0.88, and 0.81, respectively), indicating discriminatory value even in populations with underlying liver disease. Based on positive and negative likelihood ratios (LR + = 4.09; LR− = 0.22; DOR = 19), serum lncRNAs demonstrated a moderate ability to rule in or rule out HCC in positive and negative tests among the healthy population, whereas plasma lncRNAs exhibited a low diagnostic ability. This evidence suggests that circulating lncRNAs, particularly serum-based assays, represent promising noninvasive biomarkers for HCC detection and could work alongside existing diagnostic approaches, such as α-fetoprotein (AFP) and imaging modalities, highlighting their growing relevance in liver cancer diagnosis.

Diagnostic performance in HCC vs. Healthy Controls

lncRNAs are transcripts exceeding 200 nucleotides, and they are involved in RNA stability as well as interactions with proteins and DNA [112]. They regulate gene expression at multiple levels, including chromatin remodeling, transcription, and post-transcriptional processing [113, 114]. Acting as molecular signals, scaffolds, or miRNA sponges, they play crucial roles in both physiological processes and cancer development, including hepatocellular carcinogenesis [113, 115, 116]. HCC development is a complex, multistep process driven by aberrant gene expression, dysregulation of key signaling pathways, and remodeling of the tumor microenvironment (TME) [117]. Numerous proteins contribute to tumor progression and modulation of the TME, and their activity is regulated by molecular networks that include transcriptional and post-transcriptional mechanisms mediated by non-coding RNAs, such as microRNAs and lncRNAs [117].

In the majority of the included studies, circulating lncRNAs demonstrated high diagnostic accuracy for distinguishing HCC from healthy individuals, with a pooled sensitivity and specificity of around 80% and an overall AUC of 0.88. This is consistent with the findings of a recent network meta-analysis by Jiang et al., which evaluated the diagnostic performance of lncRNAs in distinguishing HCC from healthy controls and reported pooled sensitivity and specificity values of 0.839 and 0.748, respectively, with an AUC of 0.867 [118].

Serum-based assays consistently outperformed plasma-based tests (AUC 0.90 vs. 0.81). Similarly, Hao et al. reported that the plasma-based lncRNA profile showed lower diagnostic accuracy compared to the serum-based assay, suggesting that sample type may affect lncRNA performance and that serum may be the preferred matrix for analysis [119]. This finding aligns with previous molecular profiling research showing that the choice of biofluid affects extracellular RNA composition. As demonstrated by Max et al., substantial differences exist between serum and plasma RNA profiles due to platelet contamination in plasma and coagulation-related RNA release or degradation in serum [120].

Our findings indicate that circulating lncRNAs provide diagnostic accuracy comparable to the alpha-fetoprotein (AFP), which has reported pooled AUCs ranging from 0.73 to 0.93 in previous meta-analyses [121–124]. Several individual studies further support the potential of specific lncRNAs to match or surpass AFP in HCC detection. The study by Wang et al. demonstrated that Lnc-MyD88 exhibited superior diagnostic accuracy compared to AFP when distinguishing HCC patients from either healthy individuals (AUC 0.776 vs. 0.725) or those with liver cirrhosis (AUC 0.753 vs. 0.727) [125]. Similarly, Golam et al. showed that ROC analysis of CASC2 had a higher diagnostic accuracy at the chosen cutoff for HCC than AFP (94.6% vs. 90.9%) [49].

In contrast to AFP alone, multiparametric models that combine AFP with other tumor markers show substantially higher diagnostic performance. The GALAD score, which integrates gender, age, AFP, AFP-L3, and PIVKA-II, achieved pooled sensitivity and specificity of 0.86 and 0.90, respectively, with an AUC of 0.94 for all-stage HCC in recent meta-analyses [126]. Similarly, adding only PIVKA-II to AFP yields an AUC close to 0.86 [127]. Although the observed AUC for circulating lncRNAs suggests a diagnostic performance that is broadly comparable to AFP alone, this does not clearly surpass multimarker scores such as GALAD or AFP+PIVKA-II. Importantly, none of the included studies directly compared lncRNA panels with GALAD or other validated biomarker algorithms in the same patients, and only a minority reported head-to-head comparisons with AFP. Therefore, based on current evidence, we cannot determine whether circulating lncRNAs offer an incremental diagnostic advantage beyond established serum markers. Rather, they should be considered promising complementary candidates for future multi-marker strategies.

Diagnostic performance in HCC vs. liver disease controls

Accurate discrimination between HCC and patients with underlying liver disease is clinically critical, as these populations represent the main at-risk groups for hepatocarcinogenesis. In our subgroup analyses, lncRNAs held strong diagnostic accuracy when compared with cirrhotic, HCV-infected, and HBV-infected controls (AUC 0.84, 0.88, and 0.81, respectively). The ability to distinguish HCC from cirrhosis is particularly important because cirrhotic patients constitute the primary target population for HCC surveillance. Compared with healthy controls, this distinction is more challenging due to overlapping inflammatory and fibrotic changes, which can also reduce the specificity of traditional biomarkers such as AFP. In our study, circulating lncRNAs achieved an AUC of 0.84 for differentiating HCC from cirrhosis, indicating meaningful diagnostic performance despite the biological similarities between these conditions. These findings suggest that circulating lncRNAs may serve as valuable adjunctive biomarkers for the early detection of HCC in high-risk patients with chronic liver disease. Similar findings were observed by Chen et al., who investigated the diagnostic value of lncRNAs as biomarkers between patients with HCC and non-HCC controls (healthy participants, patients with liver cirrhosis, and patients with chronic hepatitis B) and reported a pooled sensitivity, specificity, and DOR of 0.87, 0.829, and 23.085, respectively [114]. This finding also aligns with a recent meta-analysis by Lumkul et al., which reported that the serum HULC, HOTAIR, and UCA1 achieved an AUC of 0.86 and a DOR of 20 when distinguishing HCC from liver-disease controls [116]. HULC expression is associated with lower AFP levels and indicates a tendency toward longer relapse-free survival, though the difference is not statistically significant. Bioinformatic analyses suggested that elevated HULC expression may predict a better prognosis in HCC [128]. HOTAIR is a key regulator of epigenetic modifications across numerous genes and has been associated with tumor metastasis and resistance to therapy [129]. Comparable findings were reported by Huang et al. and Alemayehu et al., who observed that RNA-based biomarkers maintained reliable performance even among chronic hepatitis C and liver cirrhosis [130, 131]. However, evidence in HBV-related populations remains limited and warrants further validation.

Clinical implications

LncRNAs demonstrate strong potential as non-invasive diagnostic biomarkers for HCC, particularly for early detection in the healthy population and in differentiating HCC from cirrhosis or viral hepatitis. Almost all of them have shown consistent diagnostic value in HCC patients and potential to complement AFP testing or to be incorporated into multi-lncRNA diagnostic panels.

However, the diagnostic accuracy of a single lncRNA is often limited. Combining multiple lncRNAs or integrating lncRNAs with traditional markers such as AFP may improve diagnostic performance. Multi-lncRNA panels are expected to provide superior diagnostic performance compared with single markers by capturing distinct oncogenic pathways. For example, a combined panel of lncRNA-UCA1 and lncRNA-WRAP53 demonstrated higher accuracy than either lncRNA alone [59]. Han et al. reported that ROC analysis of SCARNA10 yielded an AUC of 0.82 for diagnosing HCC, whereas combining SCARNA10 with AFP increased the AUC to 0.92, compared with 0.83 for AFP alone (p < 0.01) [132], suggesting that multi-marker approaches may enhance robustness and clinical utility for HCC screening and early detection. The observed diagnostic performance of circulating lncRNAs is biologically plausible, considering their mechanistic involvement in hepatocarcinogenesis. Several lncRNAs identified in the included studies, such as HULC, MALAT1, and HOTAIR, modulate tumorigenic signaling pathways that contribute to hepatocyte transformation, epithelial–mesenchymal transition (EMT), and tumor vascularization. For example, HULC promotes cell proliferation and angiogenesis through CREB and PI3K/AKT activation, whereas MALAT1 and HOTAIR are involved in EMT regulation and metastasis via the TGF-β and Wnt/β-catenin pathways. These findings collectively suggest that the dysregulation of circulating lncRNAs reflects underlying molecular events in liver tumor biology, supporting their potential role as functional biomarkers bridging basic molecular hepatology and clinical diagnosis. Standardization of circulating lncRNA assays is a prerequisite for clinical translation. Variability in sample handling, processing time, storage conditions, RNA extraction, normalization, and qRT-PCR platforms can introduce significant measurement bias. Harmonized pre-analytical and analytical protocols are required to ensure reproducibility and inter-laboratory comparability. Further research is needed to confirm the diagnostic value of multi-lncRNA panels in clinical settings and to establish optimized scoring systems that combine lncRNAs with established biomarkers like AFP. These panels could enable real-world HCC screening, especially for high-risk populations with liver cirrhosis and chronic viral hepatitis.

Strengths and limitations

This study has several notable strengths. First, it represents a comprehensive and up-to-date synthesis of circulating lncRNAs for HCC diagnosis, incorporating 81 studies, of which 12 were published after 2022, substantially expanding the available sample size. Earlier analyses, such as that by Lumkul et al., focused on only three specific lncRNAs (HULC, HOTAIR, and UCA1) and demonstrated that combinatorial panels can distinguish HCC from liver-disease controls [116]. The present study encompasses multiple clinically relevant control subgroups, including healthy individuals, patients with cirrhosis, and those with HBV- or HCV-related liver disease. A key strength is that subgroup analyses were conducted to assess diagnostic performance separately for HBV and HCV etiologies, yielding a level of detail that previous reviews did not offer. The inclusion of both serum and plasma data provided valuable comparative insight into sample-type performance, revealing that serum-based assays yielded a higher accuracy for the detection of HCC in the healthy population. The application of a bivariate random-effects model enabled simultaneous pooling of sensitivity and specificity while accounting for between-study variability.

Nevertheless, several limitations should be acknowledged. Significant heterogeneity was observed among studies, likely due to differences in design, population characteristics, lncRNA types, and reference standards. A major contributor to this heterogeneity is the absence of methodological standardization in circulating lncRNA research. Pre-analytical factors, including the choice of biofluid (serum vs. plasma), centrifugation protocols, and storage conditions, are known to affect extracellular RNA profiles [120, 133]. In addition, analytical inconsistencies, such as variation in RNA extraction kits and qRT-PCR conditions, further limit reproducibility and cross-study comparability [133, 134]. Evidence of publication bias was identified, but due to the diagnostic nature of the meta-analysis, conventional methods such as the trim-and-fill approach [2] were not applicable to further evaluate its impact [3]. Future research should prioritize larger, prospective, and standardized studies, including validation of high-performing lncRNAs in diverse populations and the development of multiplex panels or scoring systems to enhance diagnostic accuracy. Establishing a standardized workflow for lncRNA analysis is crucial to ensure reproducibility. At the same time, a deeper understanding of their biological mechanisms may aid in identifying the most reliable biomarkers for clinical application.

Conclusion

This systematic review and meta-analysis demonstrated that lncRNAs have strong diagnostic accuracy for hepatocellular carcinoma in the healthy population, with a pooled sensitivity of 82.8%, specificity of 79.6%, and an AUC of 0.88. Serum-based assays showed superior performance to plasma, emphasizing serum as the preferred analytical base. Circulating lncRNAs were also valuable in differentiating HCC from high-risk populations such as patients with cirrhosis, HBV, or HCV infection. With further validation and methodological standardization, lncRNA-based assays could be integrated into routine HCC screening protocols to enhance early diagnosis and patient outcomes.

Abbreviations

HCC

hepatocellular carcinoma

LncRNA

long noncoding RNA

HBV

hepatitis B virus

HCV

hepatitis C virus

AFP

alpha-fetoprotein

DOR

diagnostic odds ratio

ROC

receiver operating characteristics

Authors’ contributions

Study concept and design: E.T., A.D. Data acquisition: N.B., B.S. Data analysis and interpretation: B.S. Drafting of the manuscript: N.B., B.S., F.OS., SM.P., F.J., Critical revision of the manuscript for important intellectual content: E.T., A.D.

Funding

None received.

Data availability

Data presented in this study may be available on request from the corresponding author. Restrictions may apply due to legal reasons.

Declarations

Ethics approval

This work was approved by the Research Ethics Committees of the National Institute for Medical Research Development (IR.NIMAD.REC.1404.063).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.