Meta-analysis of the GALAD model for diagnosing primary hepatocellular carcinoma

Abstract

BACKGROUND:

Ever since the GALAD (gender-age-Lens culinaris agglutinin-reactive alpha-fetoprotein-alpha-fetoprotein-des-gamma-carboxy prothrombin) logistic regression model was established to diagnose hepatocellular carcinoma (HCC), there has been no high-level evidence that evaluates and summarizes it.

OBJECTIVE:

This meta-analysis was performed to assess the diagnostic ability of the GALAD model.

METHODS:

The following databases were systematically searched for original diagnostic studies on HCC: PubMed, Embase, Medline, the Web of Science, Cochrane Library, China National Knowledge Infrastructure Wanfang (China), Wiper and the Chinese BioMedical Literature Database. After screening the search results according to our criteria, the Quality Assessment of Diagnostic Accuracy Studies 2 tool was used to evaluate the methodologic qualities, and statistical software were used to output the statistics.

RESULTS:

Ultimately, 10 studies were included and analyzed. The results revealed the pooled sensitivity and specificity of the GALAD model to be 0.86 (95% confidence interval [CI]: 0.82, 0.90) and 0.90 (95% CI: 0.87, 0.92), respectively, for all-stage HCC. The area under the curve (AUC) was 0.94. For early-stage HCC, the pooled sensitivity and specificity of the GALAD model were 0.83 (95% CI: 0.78, 0.87) and 0.81 (95% CI: 0.78, 0.83), respectively. The AUC was 0.90.

CONCLUSION:

This meta-analysis confirmed that the GALAD model has excellent diagnostic performance for early-stage and all-stage HCC and can maintain high sensitivity and specificity in early-stage HCC. Therefore, the GALAD model is qualified for screening early-stage canceration from chronic liver disease.

Keywords

HCC logistic models biomarkers diagnostic efficiency early detection of cancer

1. Introduction

In 2018, approximately 841,000 patients worldwide were newly diagnosed with hepatocellular carcinoma (HCC), and about 781,000 patients died of the disease. The World Health Organization estimated that there were 905,677 new cases of HCC worldwide in 2020 [1]. Today, HCC is the sixth most common malignant tumor in the world, and the number of new cases of HCC is expected to exceed 1 million annually by 2025 [1, 2]. This increase may be correlated to the population’s increased exposure to risk factors.

There are several proven risk factors associated with HCC. For instance, viral hepatitis or liver cirrhosis caused by the hepatitis B virus (HBV) or the hepatitis C virus (HCV), excessive aflatoxin intake, fatty liver caused by long-term alcoholism, obesity, smoking, diabetes mellitus type 2, and other factors all can lead to HCC. The main risk factors vary from region to region. In most of the world’s high-risk areas (e.g., China and East Africa), long-term aflatoxin intake and chronic HBV infection are the primary factors, while in some countries (e.g., Japan and Egypt), chronic HCV infection may be the main cause [3].

The 5-year survival rate of patients with early-stage HCC (per the Barcelona Clinic Liver Cancer [BCLC] staging classification 0–A) who receive timely radical treatment can be improved from 5%–30% to 69.0%–86.2% compared with patients receiving late treatment [4, 5]. The key to increasing the overall survival rate of HCC is to scientifically determine the high-risk population and formulate a hierarchical monitoring strategy for early detection and diagnosis [6].

At present, the screening and detection methods of HCC include imaging methods, serological biomarkers, epigenetic markers, and pathological biopsy. In terms of imaging, the American Association for the Study of Liver Diseases recommended in its 2018 guidelines [7] that regular abdominal ultrasound should be the primary monitoring method for high-risk individuals. However, the benefits of regular ultrasound examination for the early detection of HCC are limited, whether combined with serum alpha-fetoprotein (AFP) or not, and several studies have demonstrated that the efficacy of ultrasound in detecting early HCC is unsatisfactory. For liver cancers $<$ 2 cm, the sensitivity of ultrasound is 39%–65% [8, 9, 10, 11]. Even when the common serum biomarker AFP is used in combination with ultrasound, the sensitivity for HCC reaches only about 63% [12].

There is no doubt that computed tomography (CT) and magnetic resonance imaging (MRI) are more reliable than ultrasound for accurately diagnosing HCC. The detection of lesions by CT is still regarded as one of the necessary criteria in the Chinese diagnosis standard, with the total sensitivity and specificity of the technique reaching 70%–81% and 79%–94%, respectively [13, 14, 15]. However, MRI and positron emission tomography are more accurate than ultrasound, but since they take longer and are more expensive, they are not generally used for primary screening [13].

In terms of biomarkers, including serum markers and cancer gene testing, gene testing requires technologies such as DNA sequencing to measure the concentration of telomerase reverse transcriptase, microRNAs, tumor protein 53, etc. to make the diagnosis clear or to assess treatment efficacy [12, 16]. However, cancer gene testing involves an immature laboratory test with poor reproducibility, while the use of serum biomarkers, including AFP, Lens culinaris agglutinin-reactive AFP (AFP-L3), carcinoembryonic antigen (CEA), and des-gamma-carboxy prothrombin (DCP, also known as PIVKA-II), is an established technology [6]. Liebman [17] reported DCP as a remarkably specific indicator of HCC; its diagnostic value has been verified globally, and it has been included in the Chinese diagnosis and treatment guidelines for HCC and approved for risk stratification in the United States [18]. However, although the specificity of DCP can reach 81%–98%, its sensitivity reaches only 48%–62%, which is not conducive to the detection of early-stage HCC [19].

Several studies have shown that the combination of DCP and other serum biomarkers, such as AFP, CEA, and AFP-L3, can effectively raise the sensitivity of the diagnosis and maintain high specificity [18]. In 2014, Johnson [20] established the GALAD (gender-age-Lens culinaris agglutinin-reactive alpha-fetoprotein-alpha-fetoprotein-des-gamma-carboxy prothrombin) logistic regression model, which combined gender, age, AFP-L3, AFP, and DCP. They found that the GALAD model had a high correct diagnosis rate for all-stage and early-stage HCC. The GALAD model has been preliminarily verified in China, Germany, The U.K., The U.S., Japan, and many other countries, most of which reported excellent diagnostic efficacy [21, 22, 23].

Consequently, this meta-analysis aimed to evaluate the diagnostic accuracy of the GALAD model for all-stage and early-stage HCC by integrating previous relevant high-quality studies on the application of the model for diagnosing HCC to provide reference data for accurate screening, diagnosis, and prognosis assessments for HCC in clinical practice.

2. Methods

2.1 Study retrieval strategies

The timescale for database retrieval was from the time of the databases’ establishment to June 2022. English periodical databases included PubMed, Embase, Medline, the Web of Science, and Cochrane Library, while Chinese periodical databases included the China National Knowledge Infrastructure, Wanfang, Wiper and the Chinese BioMedical Literature Databases. The retrieval mode of each database was determined by prior discussion. Our search string combined synonyms for each medical term to reduce the probability of missing studies.

2.2 Study inclusion criteria

The inclusion criteria were studies that (1) used methods of pathological or histopathological biopsy or imaging examination to formulate a definite diagnosis for HCC, (2) used the GALAD model to diagnose HCC and analyze its diagnostic efficacy (patients with HCC were compared with control groups), (3) provided true-positive (TP), true-negative (TN), false-positive (FP), and false-negative (FN) data, or these could be determined from the statistical data of the paper, and (4) included articles on prospective or retrospective randomized controlled studies.

2.3 Study exclusion criteria

The exclusion criteria were studies that (1) were not human clinical trials, (2) were expert consensuses, conferences, monographs, books, reviews, and interviews, (3) were repeatedly reported or searched in different databases, (4) did not present specific key data, (5) did not use the diagnostic gold standard tests of biopsy or radiography, (6) had an interval of more than 6 months between gold standard tests and blood sampling, and (7) had less than 50 cases in total.

2.4 Study quality assessment

Two reviewers independently read the title, abstract, and full text of the retrieved results to determine which records to include and exclude. Disagreements arising from this process were resolved by discussion. The quality and inclusion qualifications of the alternative studies were assessed using The Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) tool [24].

2.5 Data analysis

(1) Key data extracted by two reviewers from each study independently were summarized in two-by-two tables of TP, FP, FN, and TN values based on the formulas for sensitivity (TP rate) $=$ TP/(TP $+$ FN), specificity (TN rate) $=$ TN/(TN $+$ FP), and diagnostic odds ratio (DOR) $=$ (TP/FP)/(FN/TN). Other extracted data included the name of primary authors, country of publication, publication time (year), sample size, number of controls and patients with HCC, cut-off values of each biomarker, methods of tumor staging, and the number of patients at each stage.

(2) When diagnosing the same disease, the sensitivity and specificity of the same index will differ owing to diverse cut-off values; this results in the threshold effect, which is the main source of heterogeneity. Therefore, in our study, Spearman’s correlation coefficient, which was accessed by the Moses-Shapiro-Littenberg model (i.e., the Moses model), was used in Meta-DiSc statistical software (version 1.4) [25] to check for heterogeneity caused by the threshold effect. Just as the $P$ -value in Spearman’s correlation coefficient and the $B$ -value in Moses’ model were $>$ 0.05, respectively, there was no threshold effect, so unified statistical data from different studies were combined with statistical software.

(3) An inconsistency test and Cochran’s Q statistics were used to evaluate the heterogeneity among studies induced by the non-threshold effect. In this test, $I^{2}\geqslant$ 50 and $P\leqslant$ 0.05 statistically indicated significant heterogeneity induced by the non-threshold effect. In such situations, a random-effects model was used in Stata software (version 16) for statistical analysis. A fixed-effects model was used in situation of $I^{2}\leqslant$ 50. Furthermore, meta-regression in Stata 16 was used to explore possible heterogeneity sources induced by the non-threshold effect. Deeks’ funnel plot asymmetry test was performed on the included studies to identify publication bias, and a funnel plot was created [26]. A value of $P\leqslant$ 0.05 in the funnel plot indicated significant publication bias.

(4) Forest plots related to sensitivity and specificity for each biochemical index and summary receiver operating characteristic (SROC) curves were drawn in Stata 16 software. By comparing the SROC curve of each indicator and their area under the curve (AUC), the advantages and disadvantages of each indicator as a diagnostic marker of HCC were obtained.

3. Results

3.1 Retrieval results

According to the jointly agreed retrieval scheme, 340 records were obtained through preliminary searches. Seventeen records of repeated retrieval were deleted, and 276 records irrelevant to the diagnosis of HCC were excluded after reading the abstract. After reading the full text, 25 records without GALAD model analysis were excluded along with 12 records without key data. Ultimately, a total of 10 studies [20, 21, 22, 23, 27, 28, 29, 30, 31, 32] were retained for consideration for further analysis. Figure 1 shows the retrieval strategies we discussed and agreed on jointly, along with the overall retrieval flow chart and selection process.

Figure 1.

Part of the retrieving strategy in PubMed (A) and studies retrieval flow chart (B).

All 10 studies contained data on using the GALAD model to diagnose HCC and included 10,251 patients with HCC. The GALAD value calculation methods used in the included studies were the same as those of Johnson [20]:

[Z $=$ $-$ 10.08 $+$ 0.09 $\times$ age $+$ 1.67 $\times$ sex $+$ 2.34 log10 (AFP) $+$ 0.04 $\times$ AFP $-$ L3 $+$ 1.33 $\times$ log10 (DCP)].

In this equation, the “sex” value for males $=$ 1 and females $=$ 0, and the same unit of measurement (ng/mL) was used for both AFP and DCP. Ultimately, the $Z$ -value was used to distinguish patients with HCC and patients without HCC. In addition to the GALAD model, the data of other biomarkers were also included in the 10 studies. Six studies contained a combination of AFP, AFP-L3, and DCP for the diagnosis of all-stage HCC, eight studies contained AFP diagnostic data, and seven studies contained diagnostic data of DCP and AFP-L3 biomarkers. For early-stage HCC, five studies contained GALAD model diagnostic data on BCLC stage 0–A, two of which did not calculate specific values. Finally, three studies were included in the meta-analysis for early-stage HCC.

3.2 Quality assessment data and extraction

Table 1 provides the quality evaluation results for all 10 studies through the application of the QUADAS-2 tool. Two reviewers used the QUADAS-2 tool independently for evaluation before summarizing the data. In the case of dispute, we re-evaluated the data through discussion or by specifying more detailed criteria. Table 2 lists the regions, the total number of participants, the diagnostic indicators used, and the nature of the control group in each included study; additionally, it presents the final data extracted from each study. For cases of early-stage HCC, the data calculation results of the GALAD model are shown in Table 3.

Table 1
QUADAS-2 tool for study quality evaluation

Study		Risk of bias				Applicability concerns
		Patient selection	Index text	Reference standard	Flow and timing	Patient selection	Index text	Reference standard
Philip J. Johnson [20]	2014	Low	Low	Low	Low	Low	Low	Low
J. Best [22]	2016	?	Low	Low	?	Low	Low	Low
Sarah Berhane [21]	2016	Low	Low	Low	?	Low	Low	Low
Qingbiao Lin [30]	2018	High	Low	Low	Low	Low	Low	Low
Ju Dong Yang [23]	2018	Low	?	Low	Low	Low	Low	Low
Lin Tong [31]	2019	?	Low	Low	Low	Low	Low	Low
Lingyan Xu [32]	2020	?	Low	Low	Low	Low	Low	Low
Chenjun Huang [27]	2021	Low	Low	Low	Low	Low	Low	Low
Clemens Schotten [28]	2021	Low	Low	Low	Low	Low	Low	Low
Xiaqin He [29]	2021	Low	Low	Low	Low	Low	Low	Low

QUADAS-2 tool for study quality evaluation. Low: Low risk; High: High risk; ?: Unclear risk.

To enhance the comparability of the data, AFP, AFP-L3, DCP, and their combination were also tested by Spearman’s correlation coefficient and Moses’ model in addition to the GALAD model to verify whether there was heterogeneity of threshold effect sources. However, the cut-off values of AFP and AFP-L3 in Lin’s study [30] were 186 ng/mL and 1%, respectively, while in Huang’s research [27], the cut-off value of AFP was 7.75 ng/mL. In other studies, the values were 20 ng/mL and 10%, respectively. Therefore, the above data were not included in the threshold effect test and the final analysis of the results.

Figure 2.

Funnel plot: GALAD model (A); AFP $+$ AFP-L3 $+$ DCP (B); AFP (C); AFP-L3 (D); DCP (E).

Table 4 shows Spearman’s correlation coefficient and Moses’ model test results, which revealed that there was no threshold effect for each serum biomarker. Funnel plots were obtained using Deeks’ funnel plot asymmetry test. The funnel plots are presented in Fig. 2 and indicate that there was no publication bias in the included studies for each biomarker. The heterogeneity test conducted in Meta-DiSc software showed that heterogeneity existed among different studies; hence, the random-effects model in Stata software was used in this meta-analysis.

3.3 Data analysis

A meta-analysis of the GALAD model for the diagnosis of all-stage and early-stage HCC was conducted, and the three serum biomarkers mentioned above and their combination for the diagnosis of

Table 2
The basic information features and the specific data extracted of studies

	Author	Year	Region	Total	Index	Controls	GALAD model				Combination				AFP				AFP-L3				DCP				Quality evaluation
							TP	FP	FN	TN	TP	FP	FN	TN	TP	FP	FN	TN	TP	FP	FN	TN	TP	FP	FN	TN	Asia	Europe	NC	Trial	Follow-up	Number $>$ 500
1	Philip J. Johnson	2014	The UK	816	GALAD	CLD	356	49	26	385	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	0	1	0	1	1	1
2	J. Best	2016	Germany	687	AFP (10, 20 ng/mL)AFP-L3DCPAFP $+$ AFP-L3 $+$ DCPGALAD	CLD	244	27	41	375	254	78	31	324	166	24	119	378	183	34	102	368	163	20	102	368	0	1	0	1	0	1
3	Sarah Berhane	2016	Germany, Japan, The UK	6587	AFP (20 ng/mL)AFP-L3DCPAFP $+$ AFP-L3 $+$ DCPGALAD	CLD	1836	486	347	3918	1854	1024	329	338	1172	157	1011	4247	1117	563	1066	3841	1268	463	915	3941	0	1	01	0	1	1
4	Qingbiao Lin	2018	China	550	AFP (186 ng/mL)AFP-L3DCPAFP $+$ AFP-L3 $+$ DCPGALAD	CLD	199	21	76	254	206	2	69	255	152	11	123	264	273	126	2	149	220	56	55	219	1	0	0	1	0	1
5	Ju Dong Yang	2018	American	291	GALADUltrasound	CLD	101	27	10	153	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	–	0	0	0	1	1	0
6	Lin Tong	2019	China	7664	AFP (20 ng/mL)AFP-L3DCPAFP $+$ AFP-L3 $+$ DCPGALAD	BLDs	544	230	479	1515	5191	204	728	1541	3491	168	2428	1577	2445	31	3474	1714	4872	187	1047	1558	1	0	0	1	0	1
7	Lingyan Xu	2020	China	271	AFP (20 ng/mL)AFP-L3DCPAFP $+$ AFP-L3 $+$ DCPGALAD	BLDsNC	80	11	18	162	85	34	13	139	61	3	37	143	63	16	35	157	76	23	22	150	1	0	1	1	0	0
8	Chenjun Huang	2021	China	1525	AFP (7.75 ng/mL)AFP-L3DCPAFP $+$ AFP-L3 $+$ DCPGALAD	CLDNC	469	109	133	814	–	–	–	–	396	99	206	824	395	179	207	744	497	139	105	784	1	0	1	1	1	1
9	Clemens Schotten	2021	European	573	AFP (20 ng/mL)GALAD	CLD	173	19	23	358	–	–	–	–	119	15	77	362	–	–	–	–	–	–	–	–	0	1	0	0	0	1
10	Xiaqin He	2021	China	332	AFP (20 ng/mL)AFP-L3DCPAFP $+$ AFP-L3 $+$ DCPGALAD	CLDNCICC	177	30	23	102	170	8	3	124	155	17	45	115	134	4	66	128	162	16	38	116	1	0	1	1	0	0

Total: Total number of cases and controls; AFP: Alpha-fetoprotein. The cut-off values are in parentheses. The cut-off value of AFP used in the fourth paper was 186 ng/mL, and 7.75 ng/mL was used in the eighth paper, and 20 ng/mL in the rest; AFP-L3: The Percentage of Lens Culinaris Agglutinin-reactive Alpha-fetoprotein; DCP: Des- $\gamma$ -carboxy prothrombin; AFP $+$ AFP-L3 $+$ DCP: Combination of AFP, AFP-L3 and DCP were used for diagnosis; CLD: Chronic Liver Disease; NC: Healthy Subjects as Negative Controls; BLDs: Benign Liver Diseases; ICC: Intrahepatic Cholangioc Arcinoma. The results of GALAD Model, Combination, AFP, AFP-L3, and DCP calculated by formula respectively. TP: True Positive, FP: False Positive, FN: False Negative, TN: True Negative. Part of the Quality Evaluation is used for meta-regression analysis. 1: Yes; 0: No; Asia: Whether the case is from Asia; Europe: Whether the case came from Europe; NC: Whether there are healthy subjects as negative controls in the control group; Trial: 1 means retrospective randomized controlled trial, and 2 means retrospective cohort study or two-paired sample trial; Follow-up: Whether the control group was followed to confirm the absence of HCC. Number $>$ 500: 1 represents the total number of people included in the study is more than 500.

Table 3

Data results for early-stage HCC

Author	Year	Nationality	Total	BCLC 0-A	Control	TP	FP	FN	TN
J. Best [22]	2016	Germany	463	61	402	41	29	20	373
Ju Dong Yang [23]	2018	American	240	60	180	55	38	5	142
Chenjun Huang [27]	2021	China	857	208	649	176	173	32	476

GALAD model was used to diagnose patients with early liver cancer (BCLC 0/A). Total: Total number of cases and controls in early liver cancer patients.

Table 4

Result of spearman correlation coefficient and Moses’ model test

	Spearman correlation coefficient	$P$ -value	Moses’ model $T$ -value	$b$ -value
AFP	0.771	0.072	1.839	0.140
AFP-L3	0.257	0.623	2.316	0.082
DCP	0.536	0.215	0.392	0.711
AFP $+$ AFP-L3 $+$ DCP	0.143	0.787	1.399	0.234
GALAD	0.442	0.200	0.200	0.831

It can be considered that there is no threshold effect if the $P$ -value of Spearman Correlation Coefficient $>$ 0.05 or the $b$ -value of Moses’ Model Test $>$ 0.05.

all-stage HCC were analyzed for comparison. The preliminary analysis revealed that the pooled sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of the GALAD model were 0.86 (95% confidence interval [CI]: 0.82, 0.90), 0.90 (95% CI: 0.87, 0.92), 8.4 (95% CI: 6.6, 10.7), and 0.15 (95% CI: 0.12, 0.20), respectively, for all-stage HCC. The DOR was 56 (95% CI: 39, 79), and the AUC was 0.94 (95% CI: 0.92, 0.96). For early-stage HCC classified by BCLC, the pooled sensitivity, specificity, and DOR of the GALAD model were 0.83 (95% CI: 0.78, 0.87), 0.81 (95% CI: 0.78, 0.83), and 22.49 (95% CI: 12.84, 39.37), respectively, and the AUC was 0.899.

The analysis results of the three serum markers that constitute the GALAD model were as follows, with all results relating to all-stage HCC. (1) The pooled sensitivity, specificity, PPV, NPV, and DOR of AFP were 0.62 (95% CI: 0.56, 0.68), 0.92 (95% CI: 0.88, 0.95), 8.1 (95% CI: 5.3, 12.3), 0.41 (95% CI: 0.36, 0.47), and 20 (95% CI: 13, 29), respectively. The AUC was 0.82 (95% CI: 0.78, 0.85). (2) The pooled sensitivity, specificity, PPV, NPV and DOR of AFP-L3 were 0.59 (95% CI: 0.51, 0.66), 0.93 (95% CI: 0.86, 0.96), 7.9 (95% CI: 4.3, 14.5), 0.45 (95% CI: 0.38, 0.53), and 18 (95% CI: 10, 33), respectively. The AUC was 0.79 (95% CI: 0.75, 0.82). (3) The pooled sensitivity, specificity, PPV, NPV and DOR of DCP were 0.76 (95% CI: 0.68, 0.82), 0.88 (95% CI: 0.84, 0.91), 6.4 (95% CI: 5.1, 8.0), 0.28 (95% CI: 0.21, 0.36), and 23 (95% CI: 17, 32), respectively. The AUC was 0.90 (95% CI: 0.87, 0.93). (4) The pooled sensitivity, specificity, PPV, NPV, and DOR of the combination of the three biomarkers were 0.85 (95% CI: 0.81, 0.88), 0.86 (95% CI: 0.80, 0.91), 6.3 (95% CI: 4.3, 9.2), 0.17 (95% CI: 0.14, 0.21), and 37 (95% CI: 24, 55), respectively. The AUC was 0.92 (95% CI: 0.89, 0.94).

Figure 3.

Forest plot of sensitivity and specificity for GALAD model (A) and combination of AFP, AFP-L3 and DCP (B).

Figure 4.

SROC curve (A), meta regression for looking for sources of heterogeneity (B), and Fagan nomogram of GALAD model (C).

In Fig. 3A is the forest plot of the GALAD model and shows the sensitivity and specificity of each included and pooled study. The forest plot of the combination of AFP, AFP-L3, and DCP is shown in Fig. 3B. Figure 4 presents an overview of the SROC curves for all indicators and their AUCs.

In addition, meta-regression in Stata 16 was applied to assess the source of heterogeneity caused by non-threshold effects. According to the information given in the included studies, we divided them by “yes” and “no” per the criteria of the region of the study case (Europe and Asia), the types of control groups selected, the type of study, whether the controls were followed up, and the number included. The “Quality Evaluation” section in Table 2 shows the results of our classification based on the information in each paper. Figure 4B shows the analysis results of the meta-regression. It can be seen that none of our classifications were the source of heterogeneity.

4. Discussion

Hepatic carcinoma is one of the world’s major cancers, with China being one of the world’s hardest-hit areas. According to China’s 2020 cancer statistics, HCC is the fourth most commonly diagnosed malignant tumor in the country, but it causes the second highest number of deaths among all malignant diseases [33]. In China, chronic HBV infection that can cause cirrhosis is still the main cause of HCC [34]. It is thought that monitoring carcinogenesis in patients with hepatitis cirrhosis is a particularly important measure for its prevention and control. The cost of serum biomarker monitoring is low, and its accuracy is considerable, which provides convenience for long-term and multiple dynamic monitoring of patients with chronic liver disease (CLD). Furthermore, it can not only reduce medical costs but also decrease the mortality of high-risk patients with HCC [35]. In this meta-analysis, the diagnostic efficacy of AFP, AFP-L3, DCP, and their combination (AFP $+$ AFP-L3 $+$ DCP is used below) corresponded with previous results [36, 37, 38, 39]. However, their combination was slightly higher than that used in previous studies [37, 40], which may be due partly to the fact that the linear regression analysis of the three was used for diagnosis in the included studies, thus improving the diagnosis efficacy of their combination. The GALAD model added two factors of age and gender based on a logistic regression model to further improve their diagnostic ability. In our study, the pooled sensitivity, specificity, and the AUC of the GALAD model were 0.86, 0.90, and 0.94, respectively, while diagnosis using the three biomarkers in combination reached 0.85, 0.86, and 0.92, respectively, indicating that both have superior diagnostic value for patients with all-stage HCC. However, the GALAD model was slightly better than the AFP $+$ AFP-L3 $+$ DCP combined diagnosis. It can be seen from the results that when a patient was given a 50% HCC pretest probability, a percentage of 89% was acquired for the post-test probability; this is visually illustrated in Fig. 4C. Analogously, a negative test lowered the post-test probability of HCC from 50% to 13% for a low negative likelihood ratio of 0.15. For patients with BCLC stage 0–A HCC, the pooled sensitivity, specificity, and AUC of the GALAD logistic model were 0.83, 0.80, and 0.90, respectively, showing that it has a nearly perfect diagnostic ability, as reported by Johnson [20].

In Zhou’s [39] meta-analysis, the AUC of AFP-L3 was 0.755. In a meta-analysis by Tzartzeva [9], the pooled sensitivity of ultrasound alone for early-stage HCC (according to Milan Criteria staging) was just 47%. When it was combined with AFP, the sensitivity reached 63%, while the sensitivity of CT or MR was higher, reaching 84%; however, these advantages were lower than the 85% sensitivity of the GALAD model.

Although the GALAD model may not be the optimal solution for the diagnosis and screening of all-stage patients with HCC, compared with other indicators for the diagnosis of early-stage patients, the GALAD model’s significant advantage means that it can play a better role in the early detection of HCC in regular screening for CLD. Currently, clinical screening methods for HCC are mainly ultrasound and AFP, which have been used for a long time and seem reliable, but we should not stop there. Today’s patients are busy and visit the hospital only occasionally, needing a quick and reliable diagnosis. Previous studies have shown that compared with combined ultrasound and AFP, the combination of AFP, AFP-L3, and DCP can significantly improve diagnostic efficiency [37]. However, the GALAD model improves diagnostic accuracy on the basis of the combination and fully extracting the information left by patients when they register for blood testing. Additionally, computers can quickly perform the model’s series of algorithm calculations to obtain rapid results and help establish the best algorithm suitable for different laboratories.

It is worth mentioning that on the one hand, changes in serum AFP and AFP-l3 levels are used as an indicator to evaluate HCC recurrence; the studies of Xu et al. [32] and Huang [27] showed that the GALAD model can be regarded as an indicator to evaluate the prognosis of HCC as well as to estimate the survival rate after resection. Statistically, the survival time in patients with low GALAD model scores was 32 to 36 months longer than that of patients with high GALAD model scores, on average. Furthermore, the diagnosis of HCC by this model is independent of HBV-DNA status, which is helpful for screening HCC from chronic viral hepatitis cirrhosis. On the other hand, the studies of Tong et al. [31] and He et al. [29] showed that the GALAD model has a good verification effect for the microvascular invasion. It is indicated that this model can guide the staging of HCC and help decide whether to expand the surgical scope. Furthermore, in Yang’s [23] study, when ultrasound was incorporated into the GALAD model and combined into a GALADUS model, the AUC for all-stage HCC reached 0.98, while the AUC for early-stage HCC remained at 0.97. If this conclusion is further confirmed, the non-invasive diagnosis of HCC can be raised to close to the gold standard. However, additional regions and further studies are needed for further meta-analyses and confirmation of this high diagnostic value.

The GALAD model has both merits and shortcomings. In terms of advantages, it has a high cost-performance ratio. In a study by Piratvisuth [41] 50 HCC-related biomarkers were used to explore the diagnostic ability of all-stage and early-stage HCC, and 3–5 biomarkers were combined in an attempt to improve diagnostic efficacy. The combination of AFP, DCP, matrix metalloproteinase-3 (MMP3), insulin-like growth factor-binding protein-3 (IGFBP3), cartilage oligomeric matrix protein (COMP), as well as gender and age, can achieve the best diagnostic efficacy. The corresponding calculation method and its diagnostic efficacy are both similar to that of the GALAD model; however, the biomarkers of IGFBP3, COMP, and MMP3 are not yet technically mature enough, and their low penetration and high detection cost limit their application. In contrast, the GALAD model is a promising method for HCC screening because of its low cost, high prevalence, and nearly perfect diagnostic efficiency.

Logistic regression algorithms also need to be improved. In this meta-analysis, according to the sensitivity $+$ specificity maximization principle, the cut-off value of the GALAD model varied from study to study, which posed a challenge to the development of unified diagnostic criteria. In the calculation formula of the GALAD model, each index coefficient used is consistent. However, the average values of biomarkers in different regions may be inconsistent. As a result, the use of the same coefficient may reduce the diagnostic efficacy of the entire logistic regression model. However, adapting calculations to local conditions in each laboratory is one way in which diagnostic efficacy can be improved.

The GALAD model also performs well in assessing HCC prognosis. However, the BALAD-2 model combined with glutamic pyruvic transaminase and aspartate aminotransferase seemed to have a better predictive effect. Therefore, it remains to be studied which indicators are most closely related to prognosis [42]. In addition, some biomarkers with high specificities, such as Golgi transmembrane glycoprotein 73 [43], have been reported recently but have not been included in logistic regression model studies. When additional effective serum biomarkers are discovered and popularized, logistic regression analysis may be used to develop more accurate diagnostic indicators for HCC screening and diagnosis, such as the GALAD model.

This meta-analysis has some limitations. One of the main ones is that only 10 studies were included, and the small number prevented us from investigating all the potential reasons for heterogeneity. Second, the inclusion of healthy negative controls in studies may have led to a larger final result of the diagnostic efficacy of the meta-analysis. Third, the included cases covered a long period, and the changes in the prevalence and detection rate of patients with HCC may also have affected the results. Fourth, most of the included studies did not conduct long-term follow-ups for the control group and did not exclude pregnancy and reproductive-derived tumors, which may have raised the positive rate of women and thus reduced the diagnostic efficacy results of our meta-analysis. Finally, the serum DCP concentration is affected by oral anticoagulants, such as warfarin [44], and oral warfarin will lead to high DCP concentration, with FPs in the GALAD model. However, some of the included studies did not mention the related exclusion criteria.

5. Conclusion

In conclusion, our meta-analysis revealed that the GALAD model can maintain excellent sensitivity while ensuring high specificity for all-stage and early-stage HCC diagnoses. Therefore, it is suitable for the regular screening of HCC in patients with CLD.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Ethical approval

The study did not require ethical board approval because it did not contain human or animal trials.

Footnotes

Conflict of interest

The authors declare no potential conflicts of interest with respect to the research, authorship, and publication of this article.

References

Global Cancer Observatory (2022.03). http://gco.iarc.fr.

Philips

Rajesh

Nair

Ahamed

Abduljaleel

Augustine

. Hepatocellular carcinoma in 2021: An exhaustive update. Cureus. 2021 Nov 5; 13(11): e19274. doi: 10.7759/cureus.19274. PMID: 34754704; PMCID: PMC8569837.

Bray

Ferlay

Soerjomataram

Siegel

Torre

Jemal

. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2018 Nov; 68(6): 394-424. doi: 10.3322/caac.21492. Epub 2018 Sep 12. Erratum in: CA Cancer J Clin. 2020 Jul; 70(4): 313. PMID: 30207593.

Allemani

Matsuda

Di Carlo

Harewood

Matz

Nikši

Bonaventure

Valkov

Johnson

Estève

Ogunbiyi

Azevedo

Silva

Chen

Eser

Engholm

Stiller

Monnereau

Woods

Visser

Lim

Aitken

Weir

Coleman

; CONCORD Working Group. Global surveillance of trends in cancer survival 2000-14 (CONCORD-3): Analysis of individual records for 37513025 patients diagnosed with one of 18 cancers from 322 population-based registries in 71 countries. Lancet. 2018 Mar 17; 391(10125): 1023-1075. doi: 10.1016/S0140-6736(17)33326-3. Epub 2018 Jan 31. PMID: 29395269; PMCID: PMC5879496.

Tsilimigras

Bagante

Sahara

Moris

Hyer

Ratti

Marques

Soubrane

Paredes

Lam

Poultsides

Popescu

Alexandrescu

Martel

Workneh

Guglielmi

Hugh

Aldrighetti

Endo

Pawlik

. Prognosis after resection of barcelona clinic liver cancer (BCLC) stage 0, A, and B hepatocellular carcinoma: A comprehensive assessment of the current BCLC classification. Ann Surg Oncol. 2019 Oct; 26(11): 3693-3700. doi: 10.1245/s10434-019-07580-9. Epub 2019 Jul 2. PMID: 31267302.

Professional Committee for Prevention and Control of Hepatobiliary and Pancreatic Diseases of Chinese Preventive Medicine Association. Guideline for stratified screening and surveillance of primary liver cancer (2020 edition). J Clin Hepatol. 2021; 37(02): 286-295. doi: 10.3969/j.issn.2095-7815.2021.01.001.

Heimbach

Kulik

Finn

Sirlin

Abecassis

Roberts

Zhu

Murad

Marrero

. AASLD guidelines for the treatment of hepatocellular carcinoma. Hepatology. 2018 Jan; 67(1): 358-380. doi: 10.1002/hep.29086. PMID: 28130846.

Yang

Kim

. Surveillance for hepatocellular carcinoma in patients with cirrhosis. Clin Gastroenterol Hepatol. Jan 2012; 10(1): 16-21. doi: 10.1016/j.cgh.2011.06.004.

Tzartzeva

Obi

Rich

Parikh

Marrero

Yopp

Waljee

Singal

. Surveillance imaging and alpha fetoprotein for early detection of hepatocellular carcinoma in patients with cirrhosis: A meta-analysis. Gastroenterology. 2018 May; 154(6): 1706-1718.e1. doi: 10.1053/j.gastro.2018.01.064. Epub 2018 Feb 6. PMID: 29425931; PMCID: PMC5927818.

10.

Singal

Conjeevaram

Volk

Fontana

Askari

Lok

Marrero

. Effectiveness of hepatocellular carcinoma surveillance in patients with cirrhosis. Cancer Epidemiol Biomarkers Prev. 2012 May; 21(5): 793-9. doi: 10.1158/1055-9965.EPI-11-1005. Epub 2012 Feb 28. PMID: 22374994; PMCID: PMC5640437.

11.

Samoylova

Mehta

Roberts

Yao

. Predictors of ultrasound failure to detect hepatocellular carcinoma. Liver Transpl. 2018 Sep; 24(9): 1171-1177. doi: 10.1002/lt.25202. PMID: 29781162.

12.

Lim

. Role of tumor biomarkers in the surveillance of hepatocellular carcinoma. Korean J Gastroenterol. 2021 Nov 25; 78(5): 284-288. doi: 10.4166/kjg.2021.137. PMID: 34824186.

13.

Nakamura

Higaki

Honda

Tatsugami

Tani

Fukumoto

Narita

Kondo

Akagi

Awai

. Advanced CT techniques for assessing hepatocellular carcinoma. Radiol Med. 2021 Jul; 126(7): 925-935. doi: 10.1007/s11547-021-01366-4. Epub 2021 May 5. PMID: 33954894.

14.

Lee

Park

Kim

Han

Choi

. Hepatocellular carcinoma: Diagnostic performance of multidetector CT and MR imaging-a systematic review and meta-analysis. Radiology. 2015 Apr; 275(1): 97-109. doi: 10.1148/radiol.14140690. Epub 2015 Jan 5. PMID: 25559230.

15.

Guo

Seo

Ren

Hong

Lee

Kim

Jiang

. Diagnostic performance of contrast-enhanced multidetector computed tomography and gadoxetic acid disodium-enhanced magnetic resonance imaging in detecting hepatocellular carcinoma: Direct comparison and a meta-analysis. Abdom Radiol (NY). 2016 Oct; 41(10): 1960-72. doi: 10.1007/s00261-016-0807-7. PMID: 27318936; PMCID: PMC5018023.

16.

Zhang

Shao

Liu

Zheng

. The crosstalk between regulatory non-coding RNAs and nuclear factor kappa B in hepatocellular carcinoma. Front Oncol. 2021 Nov 5; 11: 775250. doi: 10.3389/fonc.2021.775250. PMID: 34804980; PMCID: PMC8602059.

17.

Liebman

Furie

Tong

Blanchard

Lee

Coleman

Furie

. Des-gamma-carboxy (abnormal) prothrombin as a serum marker of primary hepatocellular carcinoma. N Engl J Med. 1984 May 31; 310(22): 1427-31. doi: 10.1056/NEJM198405313102204. PMID: 6201741.

18.

Zhou

Yan

Yuan

Wang

Chang

Zhang

Shi

Han

Hou

Wei

Zhang

. The diagnostic value of PIVKA-II, AFP, AFP-L3, CEA, and their combinations in primary and metastatic hepatocellular carcinoma. J Clin Lab Anal. 2020 May; 34(5): e23158. doi: 10.1002/jcla.23158. Epub 2019 Dec 10. PMID: 31821607; PMCID: PMC7246383.

19.

Zhang

Chu

Cui

Song

. Des-γ-carboxy prothrombin (DCP) as a potential autologous growth factor for the development of hepatocellular carcinoma. Cell Physiol Biochem. 2014; 34(3): 903-15. doi: 10.1159/000366308. Epub 2014 Aug 21. PMID: 25200250.

20.

Johnson

Pirrie

Cox

Berhane

Teng

Palmer

Morse

Hull

Patman

Kagebayashi

Hussain

Graham

Reeves

Satomura

. The detection of hepatocellular carcinoma using a prospectively developed and validated model based on serological biomarkers. Cancer Epidemiol Biomarkers Prev. 2014 Jan; 23(1): 144-53. doi: 10.1158/1055-9965.EPI-13-0870. Epub 2013 Nov 12. PMID: 24220911.

21.

Berhane

Toyoda

Tada

Kumada

Kagebayashi

Satomura

Schweitzer

Vogel

Manns

Benckert

Berg

Ebker

Best

Dechêne

Gerken

Schlaak

Weinmann

Wörns

Galle

Yeo

Chan

Reeves

Cox

Johnson

. Role of the GALAD and BALAD-2 serologic models in diagnosis of hepatocellular carcinoma and prediction of survival in patients. Clin Gastroenterol Hepatol. 2016 Jun; 14(6): 875-886.e6. doi: 10.1016/j.cgh.2015.12.042. Epub 2016 Jan 13. PMID: 26775025.

22.

Best

Bilgi

Heider

Schotten

Manka

Bedreli

Gorray

Ertle

van Grunsven

Dechêne

. The GALAD scoring algorithm based on AFP, AFP-L3, and DCP significantly improves detection of BCLC early stage hepatocellular carcinoma. Z Gastroenterol. 2016 Dec; 54(12): 1296-1305. English. doi: 10.1055/s-0042-119529. Epub 2016 Dec 9. PMID: 27936479.

23.

Yang

Addissie

Mara

Harmsen

Dai

Zhang

Wongjarupong

Ali

Hassan

Lavu

Cvinar

Giama

Moser

Miyabe

Allotey

Algeciras-Schimnich

Theobald

Ward

Nguyen

Befeler

Reddy

Schwartz

Harnois

Yamada

Srivastava

Rinaudo

Gores

Feng

Marrero

Roberts

. GALAD score for hepatocellular carcinoma detection in comparison with liver ultrasound and proposal of GALADUS score. Cancer Epidemiol Biomarkers Prev. 2019 Mar; 28(3): 531-538. doi: 10.1158/1055-9965.EPI-18-0281. Epub 2018 Nov 21. PMID: 30464023; PMCID: PMC6401221.

24.

Whiting

Rutjes

Westwood

Mallett

Deeks

Reitsma

Leeflang

Sterne

Bossuyt

; QUADAS-2 Group. QUADAS-2: A revised tool for the quality assessment of diagnostic accuracy studies. Ann Intern Med. 2011 Oct 18; 155(8): 529-36. doi: 10.7326/0003-4819-155-8-201110180-00009. PMID: 22007046.

25.

Zamora

Abraira

Muriel

Khan

Coomarasamy

. Meta-DiSc: A software for meta-analysis of test accuracy data. BMC Med Res Methodol. 2006 Jul 12; 6: 31. doi: 10.1186/1471-2288-6-31. PMID: 16836745; PMCID: PMC1552081.

26.

Wang

Mou

Zhai

Zong

Zhao

. Application of stata software to test heterogeneity in meta-analysis method. Zhonghua Liu Xing Bing Xue Za Zhi. 2008 Jul; 29(7): 726-9. Chinese. PMID: 19031771.

27.

Huang

Fang

Xiao

Wang

Gao

Liu

Wang

Gao

. Validation of the GALAD model for early diagnosis and monitoring of hepatocellular carcinoma in Chinese multicenter study. Liver Int. 2022 Jan; 42(1): 210-223. doi: 10.1111/liv.15082. Epub 2021 Oct 31. PMID: 34679250.

28.

Schotten

Ostertag

Sowa

Manka

Bechmann

Hilgard

Marquardt

Wichert

Toyoda

Lange

Canbay

Johnson

Wedemeyer

Best

. GALAD score detects early-stage hepatocellular carcinoma in a european cohort of chronic hepatitis B and C patients. Pharmaceuticals (Basel). 2021 Jul 27; 14(8): 735. doi: 10.3390/ph14080735. PMID: 34451832; PMCID: PMC8401792.

29.

Wang

Zhang

. Diagnosis value of galad model in primary hepatocellular carcinoma and predictive effect on microvascular invasion. Biotechnology. 2021; 31(02): 166-170. doi: 10.16519/j.cnki.1004-311x.2021.02.0025.

30.

Lin

Dong

Lin

Kong

Xiong

Chen

. Validation of the diagnostic value of GALAD model in hepatocellular carcinoma patients. Zhonghua Gan Zang Bing Za Zhi. 2018 Aug 20; 26(8): 621-623. Chinese. doi: 10.3760/cma.j.issn.1007-3418.2018.08.012. PMID: 30317796.

31.

Lin

Zhiyuan

Chenjun

, et al. Role of GALAD serological model in the clinical diagnosis of primary hepatocellular carcinoma. Chin J Lab Med. 2019; (12): 1037-1041. doi: 10.3760/cma.j.issn.1009-9158.2019.12.012.

32.

. Clinical value of GALAD model on diagnosis and prognosis of primary hepatocellular carcinoma. The Journal of Evidence-Based Medicine. 2020; 20(05): 277-281. doi: 10.12019/j.issn.1671-5144.2020.05.006.

33.

Department of Medical Administration, National Health and Health Commission of the People’s Republic of China. Guidelines for diagnosis and treatment of primary liver cancer in China (2019 edition). Zhonghua Gan Zang Bing Za Zhi. 2020 Feb 20; 28(2): 112-128. Chinese. doi: 10.3760/cma.j.issn.1007-3418.2020.02.004. PMID: 32164061.

34.

Zhang

Niu

. Prevalence of primary liver cancer in Eastern countries and related influencing factors. J Clin Hepatol. 2018; 34(07): 1399-1402. DOI:CNKI:SUN:LCGD0.2018-07-007.

35.

Yang

Liu

Kong

Liu

. Progression of prothrombin induced by vitamin K Absence-II in hepatocellular carcinoma. Front Oncol. 2021 Nov 10; 11: 726213. doi: 10.3389/fonc.2021.726213. PMID: 34900676; PMCID: PMC8660097.

36.

Gao

Tian

Liu

. Diagnostic performance of des-γ-carboxy prothrombin (DCP) for hepatocellular carcinoma: A bivariate meta-analysis. Neoplasma. 2012; 59(2): 150-9. doi: 10.4149/neo_2012_020. PMID: 22248272.

37.

Wang

Zhang

Yang

Tao

Kou

Jiang

. Evaluation of the combined application of AFP, AFP-L3%, and DCP for hepatocellular carcinoma diagnosis: A meta-analysis. Biomed Res Int. 2020 Sep 17; 2020: 5087643. doi: 10.1155/2020/5087643. PMID: 33015170; PMCID: PMC7519464.

38.

Bao

. Alpha-fetoprotein-L3 in hepatocellular carcinoma: A meta-analysis. Clin Chim Acta. 2013 Oct 21; 425: 212-20. doi: 10.1016/j.cca.2013.08.005. Epub 2013 Aug 13. PMID: 23954771.

39.

Zhou

Wang

Zhang

. AFP-L3 for the diagnosis of early hepatocellular carcinoma: A meta-analysis. Medicine (Baltimore). 2021 Oct 29; 100(43): e27673. doi: 10.1097/MD.0000000000027673. PMID: 34713864; PMCID: PMC8556013.

40.

Tian

Sun

Meng

. Evaluation of individual and combined applications of serum biomarkers for diagnosis of hepatocellular carcinoma: A meta-analysis. Int J Mol Sci. 2013 Dec 2; 14(12): 23559-80. doi: 10.3390/ijms141223559. PMID: 24317431; PMCID: PMC3876063.

41.

Piratvisuth

Tanwandee

Thongsawat

Sukeepaisarnjaroen

Esteban

Bes

Köhler

Swiatek-de Lange

Morgenstern

Chan

. Multimarker panels for detection of early stage hepatocellular carcinoma: A prospective, multicenter, case-control study. Hepatol Commun. 2022 Apr; 6(4): 679-691. doi: 10.1002/hep4.1847. Epub 2021 Nov 19. PMID: 34796691; PMCID: PMC8948551.

42.

Wongjarupong

Negron-Ocasio

Mara

Prasai

Abdallah

Ahn

Yang

Addissie

Giama

Harmsen

Therneau

Roberts

. BALAD and BALAD-2 predict survival of hepatocellular carcinoma patients: A North American cohort study. HPB (Oxford). 2021 May; 23(5): 762-769. doi: 10.1016/j.hpb.2020.09.014. Epub 2020 Oct 3. PMID: 33023823; PMCID: PMC8018983.

43.

Dai

Chen

Liu

Peng

Meng

Dai

. Diagnostic value of the combination of golgi protein 73 and alpha-fetoprotein in hepatocellular carcinoma: A meta-analysis. PLoS One. 2015 Oct 6; 10(10): e0140067. doi: 10.1371/journal.pone.0140067. PMID: 26441340; PMCID: PMC4595485.

44.

Kondo

Nakagawa

Ito

Shimomura

Hatanaka

Yamamoto

Nakatani

Iwai-Kanai

Matsuo

. Influence of warfarin therapy on prothrombin production and its posttranslational modifications. J Appl Lab Med. 2020 Nov 1; 5(6): 1216-1227. doi: 10.1093/jalm/jfaa069. PMID: 32594109.

Meta-analysis of the GALAD model for diagnosing primary hepatocellular carcinoma

Abstract

BACKGROUND:

OBJECTIVE:

METHODS:

RESULTS:

CONCLUSION:

Keywords

1. Introduction

2. Methods

2.1 Study retrieval strategies

2.2 Study inclusion criteria

2.3 Study exclusion criteria

2.4 Study quality assessment

2.5 Data analysis

3. Results

3.1 Retrieval results

Table 1 QUADAS-2 tool for study quality evaluation

Table 2 The basic information features and the specific data extracted of studies

5. Conclusion

Funding

Ethical approval

Footnotes

Conflict of interest

References

Table 1
QUADAS-2 tool for study quality evaluation

Table 2
The basic information features and the specific data extracted of studies