Diagnostic and prognostic performances of GALAD score in staging and 1-year mortality of hepatocellular carcinoma: A prospective study

Abstract

BACKGROUND

The GALAD score has improved early hepatocellular carcinoma (HCC) detection rate. The role of the GALAD score in staging and predicting tumor characteristics or clinical outcome of HCC remains of particular interest.

AIM

To determine the diagnostic/prognostic performances of the GALAD score at various phases of initial diagnosis, tumor features, and 1-year mortality of HCC and compare the performance of the GALAD score with those of other serum biomarkers.

METHODS

This prospective, diagnostic/prognostic study was conducted among patients with newly diagnosed HCC at the liver center of Vajira Hospital. Eligible patients had HCC staging allocation using the Barcelona Clinic Liver Cancer (BCLC) categorization. Demographics, HCC etiology, and HCC features were recorded. Biomarkers and the GALAD score were obtained at baseline. The performance of the GALAD score and biomarkers were prospectively assessed.

RESULTS

Exactly 115 individuals were diagnosed with HCC. The GALAD score increased with disease severity. Between BCLC-0/A and BCLC-B/C/D, the GALAD score predicted HCC staging with an area under the curve (AUC) of 0.868 (95%CI: 0.80–0.93). For identifying the curative HCC, the AUC of GALAD score was significantly higher than that of Alpha-fetoprotein (AFP) (0.753) and Lens culinaris agglutinin-reactive fraction of AFP-L3 (0.706), and as good as that of Protein induced by vitamin K absence-II (PIVKA-II) (0.897). For detecting aggressive features, the GALAD score gave an AUC of 0.839 (95%CI: 0.75–0.92) and significantly outperformed compared to that of AFP (0.761) and AFP-L3 (0.697), with a trend of superiority to that of PIVKA-II (0.772). The performance to predict 1-year mortality of GALAD score (AUC: 0.711, 95%CI: 0.60–0.82) was better than that of AFP (0.541) and as good as that of PIVKA-II (0.736). The optimal cutoff value of GALAD score was ≥ 6.83, with a specificity of 72.63% for exhibiting substantial reduction in the 1-year mortality.

CONCLUSION

The GALAD model can diagnose HCC at the curative stage, including the characteristic of advanced disease, more than that by AFP and AFP-L3, but not PIVKA-II. The GALAD score can be used to predict the 1-year mortality of HCC.

Key Words: Alpha-fetoprotein; Barcelona clinic liver cancer; GALAD score; Hepatocellular carcinoma; Lens culinaris agglutinin-reactive fraction of alpha-fetoprotein; Protein induced by vitamin K absence-II

Core Tip: The GALAD score performance showed a benefit not only in the accuracy of curative hepatocellular carcinoma (HCC) staging but also in the characteristic of advanced disease. Incorporating the GALAD model may increase the opportunity for prognosis prioritization for patients with HCC to predict the 1-year mortality.

INTRODUCTION

Hepatocellular carcinoma (HCC) is one of the leading causes of cancer-related mortality worldwide. In Thailand, it was the leading cause of cancer-related deaths among males and the third-highest cause among females[1]. Despite the rapid introduction of novel and efficient HCC treatments, screening for HCC using a combination of serum alpha-fetoprotein (AFP) and ultrasonography, slightly improved the early detection rate of HCC[2,3].

Due to the suboptimal improvements obtained from combining AFP with ultrasound, the European Association for the Study of the Liver and the American Association for the Study of Liver Diseases recommended only biannual ultrasound for HCC surveillance[3-5]. Other biomarkers including AFP-L3, the Lens culinaris agglutinin-reactive fraction of AFP, which appears to be more specific for HCC[6,7]; and des-carboxy-prothrombin (DCP), a prothrombin precursor produced by HCC, have been investigated as potential tumor markers for HCC in several ethnic populations[8-10]. Additionally, AFP, AFP-L3, and DCP levels have been extensively studied in relation to prognosis[11-15]. According to a meta-analysis, combining these serum biomarkers may boost sensitivity and specificity compared to that from using each biomarker alone[16].

Previous research has presented a novel diagnostic algorithm, the GALAD score, which considers gender, age, AFP-L3, AFP, and DCP. This model shows enhanced early-stage HCC detection sensitivity[17-20]. As such, the role of the GALAD score in terms of tumor stage or clinical outcomes of HCC was intriguing. Thus, this study aimed to determine the diagnostic/prognostic performances of the GALAD score at various phases of the initial HCC diagnosis, tumor features, and 1-year mortality of patients with HCC and compare the performance of the GALAD score with that of individual serum biomarkers.

MATERIALS AND METHODS

Study design and study population

This prospective study was conducted at Faculty of Medicine Vajira Hospital, Navamindradhiraj University, Bangkok, Thailand. Participants were required to be at least 18 years old and diagnosed with HCC for the first time between September 2021 and March 2022. Hepatitis B and C virus (HBV and HCV, respectively) infections were verified based on the presence of hepatitis B surface antigen and anti-HCV antibodies. Alcohol consumption was considered a cause of HCC when there was a significant and documented history of alcohol abuse in the patient[21]. Metabolic-associated fatty liver disease (MAFLD) was diagnosed based on either an increase in liver ultrasound echogenicity or steatosis, as determined by a transient liver stiffness test in the presence of metabolic syndrome[22]. Cirrhosis was defined by: (1) Histology; (2) cirrhosis characteristics in imaging studies using ultrasound or computed tomography (CT) imaging (nodular configuration of the liver, dilated portal vein (PV), splenomegaly with or without ascites); or (3) transient elastography with a cut point of liver stiffness greater than 12.5 kPa. The predictive prognosis of cirrhosis was evaluated using both the Child-Pugh and Model for End-Stage Liver Disease scores. HCC was diagnosed based on: (1) Histology; or (2) presence of cirrhosis with the radiologic characteristics of HCC on a CT scan. In chronic HBV infection, HCC diagnosis is determined by the presence of both radiologic hallmarks and an AFP level > 200 ng/mL[23-25]. Patients with other primary liver malignancies, patients with liver metastases, and patients with HCC who did not complete the informed consent form were excluded. All individuals who receive a diagnosis of HCC will be eligible for standard treatment according to the Barcelona Clinic Liver Cancer (BCLC) guideline. Each individual provided informed consent prior to enrollment, after the researchers had thoroughly explained them the research topic. Participants’ demographic information, including age, sex, body mass index, symptoms at initial presentation, HCC etiology, performance status, tumor burden and characteristics of advanced disease, was collected. The following biochemical data were collected: Total blood count, blood urea nitrogen, creatinine level, levels of electrolytes, coagulogram, liver function test, and viral markers.

Individual blood samples (10 mL) for AFP, AFP-L3, and Protein induced by vitamin K absence- II (PIVKA-II) (stored at 20 °C) were measured using the µTASWako i30 fully automated immune analyzer (Fujifilm Wako Pure Chemical Corporation, Osaka, Japan)[26]. Microfluidic chips were analyzed using liquid-phase binding assays, followed by capillary electrophoresis and fluorescence detection. This machine had lower limited detections of 0.3 ng/mL for AFP and 5 mAU/mL for PIVKA-II for each biomarker. The percentage of AFP-L3 was measured when the AFP value was more than 0.3 ng/mL.

The GALAD score was computed as follows: Z = -10.08 + 0.09 × age + 1.67 × sex + 2.34 Log (AFP) + 0.04 × (AFP-L3) + 1.33 Log (DCP). The formula × 0.012 was used to convert the DCP (ng/mL) to the PIVKA-II value (mAU/mL)[27]. https://www.mayoclinic.org/medical-professionals/model-end-stage-liver-disease/GALAD is a web-based calculator. The definition of sex was 1 for males and 0 for females. The BCLC staging system was used to categorize tumor stages[28]. Participants who were eligible received conventional HCC treatment. Mortality was measured from the initial HCC diagnosis until death or at the end of follow-up at 1 year.

Sample size calculation

The sample size determined the number of participants for estimating accuracy index formula using the area under the curve (AUC)[29]:

n = Z²_α/2V (AUC)/d²

Where n is the sample size for each group with disease/endpoint and non-disease/non-endpoint, V_(AUC)= (0.0099 x) x (6a² + 16), a = φ^-1_(AUC)x 1.414, and φ^-1 is the inverse of standard cumulative normal distribution or Z_AUC.

The reference AUC values of GALAD score for calculating sample sizes for staging patients with HCC into each of the five stages of BCLC (0/A/B/C/D) and predicting other endpoints were not obtained from any research through our review. Therefore, we used the values from previous research that studied the GALAD performance to diagnose HCC. The reference AUC values were obtained from a systematic review and meta-analysis of the performance of GALAD score for diagnosing HCC in patients with chronic liver diseases, with an AUC of 0.86 for detecting early-stage HCC (BCLC 0/A)[20], and from a multicenter case-control study with an AUC of 0.933 for advanced HCC (BCLC B/C/D)[30]. The reference value used to calculate the sample size for 1-year mortality was obtained from a recent study with an AUC of 0.792[31]. Z_α/2 is the Z-score corresponding to a normal distribution defined as 1.96 (α = 0.05) with 95% confidence; and the degree of precision of estimate being about 0.129[20], 0.14[30], and 0.1188[31] (d was set at 15% error of AUC) for statistical significance, with V_(AUC) = 0.092483. Therefore, the required sample size obtained by inserting the formula[29] was 44 participants for detecting early-stage HCC, 18 participants for detecting advanced-stage HCC, and 68 participants for predicting 1-year mortality of patients with HCC.

Ethical approval

The institutional review board of the Faculty of Medicine at Vajira Hospital (COA 165/2564) authorized the study protocol, which was conducted in accordance with the ethical norms of the 1975 Declaration of Helsinki. All participants provided written informed consent prior to enrollment in the trial.

Statistical analysis

STATA version 13.0 (Stata Corporation, College Station, TX, United States) was used for statistical analyses. Pearson’s chi-squared or Fisher’s exact tests were used to assess comparable categorical data between HCC stages. A one-way analysis of variance (one-way ANOVA) or Kruskal–Wallis test was used to compare continuous variables. Statistical significance was set at P value < 0.05. The diagnostic and prognostic performance of GALAD score and biomarkers were estimated using the c-statistic, commonly referred to as the area under the receiver operating characteristic (ROC) curve analysis. The area under the ROC curve, sensitivity, specificity, positive and negative predictive values, positive and negative likelihood ratio, correctly classified of the GALAD score and each tumor marker for staging HCC, clinical features of the advanced disease, and 1-year mortality were obtained.

The diagnostic and prognostic performance of the GALAD score and other tumor markers were measured using the AUC that reflects the overall discriminative value of the test. The AUC ranged from 0 (at 0.5 representing “the probability of a false and true diagnosis is both 50%” and such the test is no better than flipping a coin) to 1.0 (indicating excellent discrimination)[32]. Generally, an AUC of 0.75 is considered high enough for use in clinical practice. Therefore, the best cutoff point was the point closest-to- (0,1) corner in the ROC plane (Euclidean distance), that reached 100% sensitivity and 100% specificity. Hence, the cut-points for GALAD score and tumor markers were selected to optimize the values of both sensitivity and specificity based on the Euclidean distance[32,33]. The AUC of the GALAD score and each tumor marker were compared for very early stage (BCLC-0) and early stage (BCLC-A) HCC, curative HCC stage (BCLC-0 to A), and non-curative HCC stage (BCLC B to D), clinical features of poor prognosis, and 1-year mortality.

RESULTS

Tumor characteristics, biomarkers, GALAD score and staging of HCC

A total of 115 individuals were diagnosed with HCC, of which 98 (85.2%) were male. The most prevalent symptom was abdominal pain (38.3%), followed by gastrointestinal hemorrhage (10.4%). More than one-third (33.9%) of the patients had no symptoms. The most common cause of chronic hepatitis was hepatitis B (37.4%), followed by chronic hepatitis C (CHC) (12.2%), alcoholic hepatitis (14.8%), and MAFLD (7.8%). The remaining 27.8% of the patients were classified as having mixed etiologies. Eighty percent of the patients with HCC were diagnosed with cirrhosis. Child-Pugh scores A, B, and C were distributed as follows: 56.5%, 34.8%, and 8.7%, respectively. Most participants (68.1%) were in good physical condition. Sixty-four individuals (55.7%) had tumors > 5 cm in size. Fifty patients (43.5%) were diagnosed with a single lesion. The percentage of each stage (0, A, B, C, and D) according to the BCLC criteria was 11.3%, 23.47%, 31.3%, 13.3%, and 20%, respectively. Approximately 65.2% of the patients were in the non-curative stage. Characteristics of advanced-stage HCC (PV thrombosis, macrovascular invasion, and metastasis) were observed in 21.7%, 13.9%, and 8.0% of the cases, respectively. The median levels of AFP, AFP-L3 (%), and PIVKA-II at the time of HCC diagnosis were 38.9 ng/mL (5.4-3305), 7% (1-34.1), and 604 mAU/mL (54-21878), respectively. The median GALAD score for BCLC stages 0, A, B, C, and D were -2.27 (-3.9, 1.2), -0.62 (-1.81, 1.86), 4.15 (1.21, 8.81), 9.59 (6.17, 13.33), and 7.22 (3.7, 10.12), respectively (Table 1). There were no significant differences in median GALAD score and the various etiology of HCC (Supplementary Table 1).

Table 1 Patient characteristics and biomarkers, n (%).

               

	Total	Stage 0	Stage A	Stage B	Stage C	Stage D	P valuea
	(n = 115)	(n = 13)	(n = 27)	(n = 36)	(n = 16)	(n = 23)
Age (mean ± SD)	60.83 ± 12.93	60.69 ± 11.64	61.07 ± 11.34	58.67 ± 15.03	62.00 ± 12.73	63.22 ± 12.48	0.754
Male	98 (85.2)	11 (84.6)	22 (81.5)	31 (86.1)	15 (93.8)	19 (82.6)	0.849
BMI (kg/m2)	23.12 ± 3.5	24.31 ± 3.12	23.43 ± 2.74	23.06 ± 3.33	22.98 ± 5.06	22.26 ± 3.53	0.543
Symptom at first presentation
Abdominal pain	44 (38.3)	1 (7.7)	7 (25.9)	16 (44.4)	10 (62.5)	10 (43.5)	0.02
Jaundice	7 (6.1)	1 (7.7)	0 (0)	1 (2.8)	1 (6.3)	4 (17.4)	0.106
Anemia	1 (0.9)	0 (0)	0 (0)	1 (2.8)	0 (0)	0 (0)	0.697
Ascites	9 (7.8)	2 (15.4)	1 (3.7)	1 (2.8)	2 (12.5)	3 (13)	0.368
Weight loss	7 (6.1)	0 (0)	0 (0)	5 (13.9)	0 (0)	2 (8.7)	0.102
Fever	1 (0.9)	1 (7.7)	0 (0)	0 (0)	0 (0)	0 (0)	0.095
Edema	4 (3.5)	0 (0)	1 (3.7)	0 (0)	2 (12.5)	1 (4.3)	0.223
Abdominal mass	3 (2.6)	0 (0)	0 (0)	3 (8.3)	0 (0)	0 (0)	0.154
GI bleeding	12 (10.4)	1 (7.7)	0 (0)	2 (5.6)	5 (31.3)	4 (17.4)	0.012
Asymptomatic	39 (33.9)	8 (61.5)	17 (63)	11 (30.6)	1 (6.3)	2 (8.7)	< 0.001
Etiology
CHB	43 (37.4)	8 (61.5)	7 (25.9)	18 (50)	5 (31.3)	5 (21.7)	0.043
CHC	14 (12.2)	1 (7.7)	5 (18.5)	3 (8.3)	3 (18.8)	2 (8.7)	0.615
Alcohol	17 (14.8)	2 (15.4)	4 (14.8)	5 (13.9)	2 (12.5)	4 (17.4)	0.995
MAFLD	9 (7.8)	1 (7.7)	4 (14.8)	2 (5.6)	1 (6.3)	1 (4.3)	0.64
CHB with alcohol	20 (17.4)	0 (0)	4 (14.8)	4 (11.1)	5 (31.3)	7 (30.4)	0.069
CHC with alcohol	12 (10.4)	1 (7.7)	3 (11.1)	4 (11.1)	0 (0)	4 (17.4)	0.526
Cirrhosis	92 (80)	9 (69.2)	21 (77.8)	27 (75)	12 (75)	23 (100)	0.108
CTP-A/B/C (%)	56.5/34.8/8.7	88.9/11.1/0	81/19/0	70.4/29.6/0	50/50/0	8.7/56.5/34.8	< 0.001
MELD (mean ± SD)	10.31 ± 7.28	7.54 ± 7.24	8.78 ± 6.73	8.28 ± 6.33	10.5 ± 6.98	16.74 ± 6.17	< 0.001
Performance status
0-1	79 (68.7)	12 (92.3)	26 (96.3)	36 (100)	4 (25)	1 (4.3)	< 0.001
2	2 (1.7)	0 (0)	1 (3.7)	0 (0)	0 (0)	1 (4.3)	0.613
3	15 (13)	0 (0)	0 (0)	0 (0)	10 (62.5)	5 (21.7)	< 0.001
4	19 (16.5)	1 (7.7)	0 (0)	0 (0)	2 (12.5)	16 (69.6)	< 0.001
Tumor size (cm)
≤ 2	23 (20)	12 (92.3)	8 (29.6)	1 (2.8)	0 (0)	2 (8.7)	< 0.001
2.1-3	14 (12.2)	1 (7.7)	12 (44.4)	0 (0)	0 (0)	1 (4.3)	< 0.001
3.1-5	14 (12.2)	0 (0)	7 (25.9)	4 (11.1)	0 (0)	3 (13)	0.065
> 5	64 (55.7)	0 (0)	0 (0)	31 (86.1)	16 (100)	17 (73.9)	< 0.001
Tumor number
1	50 (43.5)	13 (100)	14 (51.9)	17 (47.2)	2 (12.5)	4 (17.4)	< 0.001
2	35 (30.4)	0 (0)	10 (37)	8 (22.2)	10 (62.5)	7 (30.4)
≥ 3	30 (26.1)	0 (0)	3 (11.1)	11 (30.6)	4 (25)	12 (52.2)
Tumor characteristic
Macrovascular invasion	16 (13.9)	0 (0)	0 (0)	2 (5.6)	8 (50)	6 (26.1)	< 0.001
PV thrombosis	25 (21.7)	0 (0)	0 (0)	1 (2.8)	11 (68.8)	13 (56.5)	< 0.001
Metastasis	10 (8.7)	0 (0)	0 (0)	0 (0)	6 (37.5)	4 (17.4)	< 0.001
Liver function test
Albumin (mg/dL)	3.7	3.8	4.1	3.9	3.55	2.7	< 0.001
	(3.1, 4.2)	(3.7, 4.4)	(3.5, 4.3)	(3.45, 4.25)	(3.15, 3.7)	(2.3, 3.4)
Total bilirubin (mg/dL)	1.04	0.91	0.69	0.86	1.35	3.44	< 0.001
	(0.63, 1.84)	(0.69, 1.12)	(0.56, 1.64)	(0.55, 1.5)	(1.17, 1.83)	(0.92, 5.19)
AST (IU/L)	79 (47, 130)	34 (29, 67)	52 (36, 68)	79 (53, 116.5)	145 (94.5, 207.5)	141 (93, 479)	< 0.001
ALT (IU/L)	41 (24, 65)	24 (20, 43)	35 (20, 43)	53 (38, 71)	41 (20.5, 124)	48 (24, 97)	0.017
ALP (IU/L)	136 (92, 237)	86 (71, 108)	108 (79, 130)	143.5 (102, 271.5)	176 (152, 276.5)	201 (159, 314)	< 0.001
Biomarker (median, 25-75 quartile)
AFP (ng/mL)	38.9	3.1	10	50	22146.3	79	< 0.001
	(5.4, 3305.3)	(1.7, 50)	(2.8, 61)	(13.2, 2482.65)	(190.55, 163697)	(17.2, 23475)
AFP- L3 (%)	7	4.55	4.83	6.07	25.8	30.5	< 0.001
	(1, 34.1)	(0.5, 6.76)	(0.5, 8.1)	(0.9, 24.7)	(3.95, 63.9)	(8.16, 74.7)
PIVKA II (mAU/mL)	604	26	55	7820	20188	4807	< 0.001
	(54, 21878)	(20,35)	(37, 220)	(139, 44019.5)	(761, 91915)	(581, 203031)
GALAD score
Median (range)	3.08	-2.27	-0.62	4.15	9.59	7.22	< 0.001
	(-0.56, 9.09)	(-3.9, 1.2)	(-1.81, 1.86)	(1.21, 8.81)	(6.17, 13.33)	(3.7, 10.12)

^aP value < 0.05 is statistically significant, ANOVA test, Kruskal–Wallis test, Chi-square test, or Fisher’s exact test were used.

AFP: Alpha-fetoprotein; AFP-L3: Lens culinaris agglutinin-reactive fraction of alpha-fetoprotein; ALT: Alanine aminotransferase; AST: Aspartate aminotransferase; ALP: Alkaline phosphatase; BMI: Body mass index; CTP: Child Turcotte Pugh; CHB: Chronic hepatitis B; CHC: Chronic hepatitis C; GI: Gastrointestinal; MAFLD: Metabolic associated fatty liver disease; MELD: Model for end stage liver disease; PIVKA II: Protein induced by vitamin K absence- II; PV: Portal vein.

Comparisons of the diagnostic performance between GALAD score and serum biomarkers

For very early stage (BCLC-0) and early stage (BCLC- A) of HCC: AUC for predicting very early stage HCC was non-significantly superior with GALAD score (0.6097, 95%CI: 0.40 to 0.82) compared to that using individual AFP (0.5655, 95%CI 0.36 to 0.77, P = 0.3721) and AFP-L3 (0.5128, 95%CI 0.32 to 0.70, P = 0.2992), and was comparable with that of PIVKA-II (0.7236, 95%CI 0.54 to 0.91, P = 0.1798). PIVKA-II showed a trend of higher AUC than that of AFP (P = 0.1100) and AFP-L3 (P = 0.0409; Table 2).

Table 2 Diagnostic performances of GALAD score and other biomarkers on Barcelona Clinic Liver Cancer staging, aggressive features, and prognostic performances for 1-year mortality of hepatocellular carcinoma.

Scores	Barcelona Clinic Liver Cancer staging		Aggressive featuresa,2	1-year mortality3
	0 vs A2	0/A vs B/C/D2
GALAD
AUC (95%CI)	0.6097 (0.40–0.82)	0.8677 (0.80–0.93)	0.8385 (0.75–0.92)	0.7108 (0.60–0.82)
Cutoff values	≥ -1.95	≥ 2.65	≥ 7.22	≥ 6.83
Sensitivity/Specificity (%)	81.48/53.85	74.67/85.00	75.00/85.54	60.00/72.63
PPV/NPV (%)	78.57/58.33	90.32/64.15	66.67/89.87	31.58/89.61
Positive/Negative LR	1.77/0.34	4.98/0.30	5.19/0.29	2.19/0.55
Correctly classified (%)	72.50	78.26	82.61	70.43
PIVKA-II
AUC (95%CI)	0.7236 (0.54–0.91)	0.8970 (0.84–0.95)	0.7718 (0.68–0.86)	0.7395 (0.63–0.85)
Cutoff values	≥ 37	≥ 354	≥ 581	≥ 2959
Sensitivity/Specificity (%)	77.78/76.92	81.33/92.50	93.75/63.86	75.00/68.42
PPV/NPV (%)	87.50/62.50	95.31/72.55	50.00/96.36	33.33/92.86
Positive/Negative LR	3.37/0.29	10.84/0.20	2.59/0.10	2.38/0.37
Correctly classified (%)	77.50	85.22	72.17	69.57
AFP
AUC (95%CI)	0.5655 (0.36–0.77)	0.7525 (0.67–0.84)	0.7613 (0.65–0.87)	0.5405 (0.39–0.69)
Cutoff values	≥ 4.4	≥ 16	≥ 79	≥ 79
Sensitivity/Specificity (%)	74.07/53.85	74.67/60.00	75.00/71.08	50.00/60.00
PPV/NPV (%)	76.92/50.00	77.78/55.81	50.00/88.06	20.83/85.07
Positive/Negative LR	1.60/0.48	1.87/0.42	2.59/0.35	1.25/0.83
Correctly classified (%)	67.50	69.57	72.17	58.26
AFP-L3
AUC (95%CI)	0.5128 (0.32–0.70)	0.7058 (0.61–0.80)	0.6969 (0.58–0.82)	0.7361 (0.61–0.86)
Cutoff values	≥ 4.83	≥ 7.4	≥ 12.3	≥ 14.5
Sensitivity/Specificity (%)	51.85/53.85	64.00/77.50	65.63/71.08	65.00/72.63
PPV/NPV (%)	70.00/35.00	84.21/53.45	46.67/84.29	33.33/90.79
Positive/Negative LR	1.12/0.89	2.84/0.46	2.27/0.48	2.38/0.48
Correctly classified (%)	52.50	68.70	69.57	71.30
Comparison of AUC
GALAD and PIVKA-II	P = 0.1798	P = 0.2353	P = 0.0683	P = 0.5349
GALAD and AFP	P = 0.3721	P < 0.001a	P = 0.0136a	P < 0.001a
GALAD and AFP-L3	P = 0.2992	P < 0.001a	P = 0.0152a	P = 0.6656
PIVKA-II and AFP	P = 0.1100	P < 0.001a	P = 0.8476	P = 0.0086a
PIVKA-II and AFP-L3	P = 0.0409a	P < 0.001a	P = 0.2801	P = 0.9610

^aP value < 0.05 is statistically significant.

¹Macrovascular invasion or portal vein thrombosis or extrahepatic metastasis.

²Optimal cutoff points were determined to the point closest-to- (0,1) corner in the receiver operating characteristic plane (Euclidean distance).

³Cutoff values were adapted based on the Euclidean distance criteria to maximize specificity for predicting the 1-year mortality with acceptable sensitivity.

AFP: Alpha-fetoprotein; AFP-L3: Lens culinaris agglutinin-reactive fraction of alpha-fetoprotein; AUC: Area under the curve; LR: The likelihood ratio; NPV: Negative predictive value; PIVKA-II: Protein induced by vitamin K absence- II; PPV: Positive predictive value; ROC: Receiver operating characteristic.

For curative HCC stage (BCLC- 0 to A) and non-curative HCC stage (BCLC B to D): The diagnostic performance of the GALAD score in curative HCC stage was more accurate than that in very early stage (0.6097 to 0.8677). For the prediction of BCLC stage 0 to A, AUC of the GALAD score (0.8677, 95%CI 0.80 to 0.93) was significantly higher than that of AFP (0.7525, 95%CI 0.67 to 0.84, P < 0.001) and AFP-L3 (0.7058, 95%CI 0.61 to 0.80, P < 0.001). The performance of the GALAD score was as good as PIVKA-II: AUC at 0.8970 (95%CI 0.84 to 0.95, P = 0.2353) to predict for curative stage of HCC. For predicting curative stage of HCC, the optimal cutoff value of the GALAD score was ≥ 2.65, with 74.67% sensitivity and 85.0% specificity (Table 2 and Figure 1A).

Figure 1 Receiver operating characteristic curves displaying the diagnostic performances of GALAD and other biomarkers. A: On staging curative hepatocellular carcinoma; B: On aggressive features of hepatocellular carcinoma; C: On 1-year mortality of hepatocellular carcinoma. AFP: Alpha-fetoprotein; AFP-L3: Lens culinaris agglutinin-reactive fraction of alpha-fetoprotein; AUC: Area under the curve; PIVKA II: Protein induced by vitamin K absence – II.

For characteristics of advanced diseases and patient’s mortality: The characteristics of advanced diseases were composed of any one of the following: Macrovascular invasion, 13.9% (n = 16); PV thrombosis, 21.7% (n = 25); and extrahepatic metastasis, 8.7% (n = 10). For predicting aggressive feature of HCC, AUC of the GALAD score (0.8385, 95%CI 0.75 to 0.92) significantly outperformed that of AFP (0.7613, 95%CI 0.65 to 0.87, P = 0.0136) and AFP-L3 (0.6969, 95%CI 0.58 to 0.82, P = 0.0152); moreover, the AUC of the GALAD score showed a superior trend to that of PIVKA (0.7718, 95%CI 0.68 to 0.86, P = 0.0683). There was no significant difference in aggressive feature between PIVKA-II and AFP (P = 0.8476; Table 2 and Figure 1B).

After the diagnosis of HCC, standard therapy was commenced following the BCLC guidelines; however, after 1-year of follow-up, none of the 20 patients (17.39%) survived. Reasons for mortality included severe or terminal disease (12), HCC rupture (3), or sepsis (5). All non-survivors were in advanced or terminal stages, except one patient with BCLC stage B and ruptured HCC.

In terms of predicting the 1-year mortality, the GALAD score (0.7108, 95%CI 0.61–0.82) outperformed AFP (0.5405, 95%CI 0.39–0.69, P < 0.001), but was as good as PIVKA-II (0.7395, 95%CI 0.63–0.85, P = 0.5349). For predicting 1-year mortality of HCC, the optimal cutoff value of the GALAD score was ≥ 6.83, this cutoff value was within intermediate or advanced HCC stage, and gave 60.0% sensitivity and 72.63% specificity. The optimal cutoff value of PIVKA-II for patient’s mortality was ≥ 2959 mAU/mL, which gave a slightly lower specificity (68.42%). Even though the GALAD score gave lower sensitivity, it gave higher specificity to predict 1-year mortality than that of PIVKA-II (Table 2 and Figure 1C).

DISCUSSION

The prevalence of HCC among all cancer diagnoses in the global population is increasing. Currently, ultrasonography with AFP is the recommended screening method for diagnosing HCC. Other biomarkers, including AFP, AFP-L3, and PIVKA-II, have been shown to boost the diagnostic sensitivity for HCC[34-38]. The GALAD score was developed to enhance the utility of combining various markers derived from sex and age, which has been validated in non-alcoholic fatty liver disease[18,38].

Our investigation revealed a higher median GALAD score in CHC, which was consistent with data from the Chinese population, with the use of the GALAD score in CHC offering greater diagnostic power for HCC than for other etiologies[39]. In contrast to previous studies in which the GALAD utility was established for early HCC diagnosis[16,18], our study demonstrates the novel utility of the GALAD score for accurate HCC staging and prognosis. We emphasized that GALAD has multiple utilities, and it was demonstrated that GALAD has a high AUC for HCC at the curative stage. According to comparable patient’s age for each HCC stage, the increase of the GALAD score in parallel with higher BCLC stage, with the exception of BCLC stage D, reflects liver decompensation itself but not tumor burden.

Serum prothrombin produced by the lack of PIVKA-II is an aberrant prothrombin caused by a deficiency of gamma-glutamyl carboxylase and vitamin K[40]. PIVKA-II is not only more specific than AFP, but a positive result also increases the likelihood of micro- and macrovascular invasion[41,42]. Owing to the distinct synthesis pathway, the benefits of PIVKA-II were complementary to those of AFP[43]. Enrollment in our curative-stage HCC study showed that PIVKA-II performed much better than AFP. Some nations have suggested using both biomarkers for the initial detection of HCC. The cause of HCC in majority of our patients was chronic hepatitis B (CHB), and our PIVKA-II identification of early CHB-related HCC was comparable to that of previous reports[44,45]. According to our data, using PIVKA-II was associated with poor performance in MAFLD-HCC enrollment, which is consistent with the results of a previous study[46]. However, because only a few MAFLD cases were analyzed, our findings may not be definitive. Regarding to the enrollment of substantial number of patients with BCLC stage 0 or A, this may be of interest for prospective clinical outcome prediction research.

The characteristics of advanced disease can be evaluated by utilizing both GALAD scores and PIVKA-II levels. In the present study, after 1-year of mortality monitoring, all non-survivors were in the intermediate and advanced stages of HCC, as established by disease staging and treatment. According to a recent systematic review and meta-analysis, GALAD score was useful for diagnostic performance but not for prognosis of disease progression[20]. We highlight the updated significance of GALAD performance not only in the accuracy of curative HCC staging, but also in the characteristic of advanced diseases and predicting a decline in 1-year mortality. A combination of age and PIVKA-II in the GALAD model may offer an indirect method for determining the aggressiveness of malignancies[47]. It would be intriguing to incorporate the GALAD model in future research for increasing the staging accuracy and for the personalized prognosis of HCC. High pre-treatment serum AFP-L3% levels were also associated with a poor prognosis in our patients with HCC; however, we believe that high AFP-L3% levels may have considerable prognostic value for patients with HCC with low AFP concentrations.

This study had some limitations. Despite the ability to recruit participants based on our sample size calculations, the number of participants in each stage, particularly the initial stage, is very low. This could potentially impact the performance of the GALAD application. If more patients with HCC utilize the GALAD score, the overall advantage in proper disease staging and prognosis may become more apparent. Second, our clinical data was archived with short follow-up periods. The prognostic performance may increase with longer follow-up periods.

CONCLUSION

The GALAD model can enhance the diagnosis of HCC at the curative stage more than that by AFP and AFP-L3, but not PIVKA-II; moreover, it can also be used to predict the 1-year mortality in non-curative HCC.

ACKNOWLEDGEMENTS

We thank Dr. Chadakarn Phaloprakarn for the scientific advice and Miss Kanokwan Sansuk for the reference format.