Development and validation of a modified albumin–bilirubin grade and α-fetoprotein score (mALF score) for hepatocellular carcinoma patients receiving atezolizumab and bevacizumab

Predicting the survival of hepatocellular carcinoma (HCC) patients receiving atezolizumab and bevacizumab (Atez/bev) remains a challenge. This study aims to validate the modified albumin–bilirubin grade and α-fetoprotein score (mALF score). This retrospective, multicenter study included 426 HCC patients receiving Atez/Bev. Each patient was randomized 3:2 to a training set (n = 255) and a validation set (n = 171). We investigated prognostic factors in the training set and developed an easily applicable mALF score, which was evaluated in the validation set. We built the mALF score using baseline mALBI grade 2b or 3 (HR 2.36, 95% CI 1.37–4.05, p = 0.002) and α-fetoprotein ≥ 100 ng/ml (HR 2.61, 95% CI 1.49–4.55, p < 0.001), which were identified as unfavorable prognostic factors in a multivariate analysis. The 1-year OS rates were 82.7% (95% CI 68.9–90.8) in patients who meet neither of the criteria (mALF 0 points, n = 101), 61.7% (95% CI 44.5–74.9) in patients who meet either of the two criteria (mALF 1 point, n = 109), and 24.6% (95% CI 9.0–44.3) in patients who meet both criteria (mALF 2 points, n = 45); the difference was statistically significant (p < 0.001). The median PFS in patients with mALF 0, 1, and 2 points was 9.5 months (95% CI 4.3-NA), 6.6 months (95% CI 6.0–8.0), and 3.8 months (95% CI 3.0–5.2), respectively, which amounted to a significant difference (p < 0.001). These results were confirmed in the validation set (1-year OS rates, 0/1/2 points = 94.2%/62.1%/46.3%, p < 0.001; median PFS, 0/1/2 points = 9.3/6.7/4.7 months, p = 0.018). The mALF score can reliably predict the prognosis of HCC patients receiving Atez/Bev.


Introduction
The choices of systemic therapies for advanced hepatocellular carcinoma (HCC) have increased. Although three regimens, namely atezolizumab and bevacizumab (Atez/ Bev) [1], lenvatinib [2], and sorafenib [3,4], have already been approved as first-line treatments, Atez/Bev is recommended for first-line treatment according to recent guidelines [5,6], and is commonly used worldwide. However, predicting the survival of HCC patients receiving immune checkpoint inhibitor (ICI) treatment remains a challenge. While the expression of PD-L1 [7], and activated Wnt/βcatenin signaling [8,9] may be promising biomarkers to predict the clinical outcome of immune checkpoint inhibitors, reliable biomarkers are still lacking. Furthermore, a histological specimen is required to evaluate the expression of PD-L1 and activated Wnt/β-catenin signaling. Therefore, it is important to establish a simple scoring system that reflects the preserved liver function and oncologic prognostic factors, which can be easily measured in clinical settings. Therefore, the aim of current study was to newly develop and validate a simple score that can be applied to predict survival in patients treated with Atez/Bev.

Participants
The present retrospective study included 426 HCC patients who received atezolizumab (1200 mg/body) and bevacizumab (15 mg/kg body weight) intravenously every 3 weeks in 22 institutions in Japan. Between September 2020 and January 2022, a total of 426 patients were included. We did not exclude any patients. These eligible patients were randomized 3:2 to a training set (n = 255) and a validation set (n = 171) (Fig. 1).
Baseline characteristics, including age, sex, body mass index, chronic liver disease, biochemical parameters, liver function, and tumor stage were collected. The liver function was assessed according to the Child-Pugh score, and modified albumin-bilirubin (mALBI) grade [10]. The tumor stage was determined based on the Barcelona Clinic Liver Cancer (BCLC) system [6].

Evaluation on therapeutic outcome of Atez/bev treatment
The Atez/Bev treatment was continued until the presence of disease progression or unacceptable adverse events were found. The tumor response was assessed according to the Response Evaluation Criteria in Solid Tumors version 1.1 (RECIST ver.1.1). The best radiological response was classified as a complete response (CR), partial response (PR), stable disease (SD), or progressive disease (PD) based on local review. Progression-free survival (PFS) was computed from the date on which treatment with Atez/Bev was initiated to the date of disease progression or death from any cause, whichever came first. OS was calculated from the initiation date on which treatment with Atez/Bev was initiated to the date of death from any cause. Adverse events were graded based on The Common Terminology Criteria for Adverse Events version 5.0. Interruption or discontinuation of each drug was carried out according to the guidelines for Atez/ Bev treatment provided by the manufacturer.

Statement of ethics
All research procedures were approved by the Institutional Ethics Committee of Ehime Prefectural Central Hospital (IRB No. 30-66) (UMIN000043219) and were performed in accordance with the Declaration of Helsinki. All patients included in the present study agreed and gave their written consent for the anonymous use of their clinical data for scientific research.

Statistical analyses
Categorical variables were reported as the number (percentage) and were compared using the chi-squared or Fisher's exact test, as appropriate. Continuous variables were reported as the median (interquartile range) and compared using Mann-Whitney U test. Because the development of a scoring system based on results in same cohort may fail to be overoptimistic and replication of results in another cohort is important, we randomly divided our cohort into two groups. First, we newly built a simple score based on the results of multivariate analyses in the training set. We then replicated and tested its performance in a validation set. We used Cox proportional hazards regression models to investigate prognostic factors. The number of explanatory variables depended on the number of events. The following factors were included as explanatory variables: age, sex, BCLC stage (0 or A or B vs. C or D), baseline mALBI grade (1 or 2a vs. 2b or 3), and AFP (< 100 ng/ml vs. ≥ 100 ng/ml). We adapted the BCLC staging system but not Atez/Bev atezolizumab and bevacizumab, HCC hepatocellular carcinoma the CLIP score or JIS score because the BCLC staging system was used in recent phase 3 studies, such as IMbrave150 [1] and REFLECT study [2]. The cut-off value of AFP was compatible with previous studies related to immune classification [9], gene mutation [11], and ICI treatment [12,13]. The PFS and survival curves were drawn using the Kaplan-Meier method and analyzed by a log-rank test. The concordance index was used to evaluate the predictive abilities for prognoses. All reported p values were 2-sided and p values of < 0.05 were considered statistically significant. All statistical analyses were conducted using EZR Ver. 1.55 (Saitama Medical Center, Jichi Medical University, Saitama, Japan) [14].

Patient profile in training and validation set
The patient profiles of the training and validation sets are shown in Table 1
The multivariate analysis revealed that baseline mALBI grade 2b or 3 and AFP ≥ 100 ng/ml were unfavorable factors for OS (mALBI grade 2b or 3, hazard ratio [HR] 2.36, 95% CI 1.37-4.05, p = 0.002: AFP ≥ 100 ng/ml, HR 2.61, 95% CI 1.49-4.55, p < 0.001: Table 2). Given that both the mALBI grade and AFP were prognostic factors and the numerical value of the HR was similar in the multivariate analysis, we developed a simple score named the modified Albumin-biLirubin grade and α-Fetoprotein score (mALF score). We assigned 1 point for mALBI grade 2b or 3 and 1 point for baseline AFP ≥ 100 ng/ml. The scores of the patients were determined as follows: mALF score 0 points (mALBI grade 1 or 2a and AFP < 100 ng/ml), 1 point (either mALBI grade 2b or 3, and AFP ≥ 100 ng/ml), and 2 points (both mALBI grade 2b or 3, and AFP ≥ 100 ng/ ml). The survival curves according to the baseline mALBI grade, AFP, and cause of liver disease were shown in Supplemental Figs. 2a-c.

The performance of mALF score in the validation set
The mALF scores of the validation cohort were as follows: 0 points, n = 60 (35.1%), 1 point, n = 77 (45.0%); and 2 points, n = 34 patients (19.9%). The ORR in patients with a mALF score of 0 points was numerically higher in comparison to those with those with scores of 1 point or 2 points, but the difference was not statistically significant (p = 0.89: supplemental Table 2). The median PFS in patients with mALF scores of 0, 1, and 2 points was 9.3 months (95% CI 5.8-NA), 6.7 months (95% CI 3.8-9.0), and 4.7 months (95% CI 2.9-6.4), respectively. The 1-year OS rates in patients with a mALF score of 0 points were the highest (94.2%, 95% CI 78.5-98.5%), followed by patients with 1 point (61.7%, 95% CI 44.5-74.9) and patients with 2 points (46.3%, 95% CI 19.6-69.4). The PFS and the survival curve were well stratified by the mALF score (PFS, p < 0.018 concordance index 0.579; OS, p < 0.001 concordance index 0.602: Fig. 3a and b). Post-progression treatment according to the mALF score is shown in Supplemental Table 3. The incidence of any-grade fever and any-grade of hepatic edema showed significant differences according to the mALF score (p = 0.02, and p = 0.04, respectively; Table 3).  , respectively, which amounted to a statistically significant difference (p < 0.001, concordance index 0.675). CI confidence interval, mALF score modified albumin-bilirubin grade and α-fetoprotein score, OS overall survival, PFS progression-free survival

Discussion
The major findings of the present study were the identification of baseline mALBI grade 2b or 3, and AFP ≥ 100n g/ ml as independent unfavorable prognostic factors in a multivariate analysis. Based on this, we built an easy, and widely applicable scoring system named the mALF score. The concordance index for the mALF score was higher than that for the mALBI grade and AFP, indicating that the prognostic capability of the mALF score was better than that for the mALBI grade and AFP. The PFS and survival curve were well stratified by the mALF score in the training set and these results were confirmed in the validation set. Staging systems, including the Liver Cancer Study Group of Japan staging system [15] have been used to evaluate pure tumor factors. Other staging systems, such as the BCLC staging system [6], CLIP score [16], and JIS score [17], have been used to evaluate both liver function and tumor factors. However, the inclusion of both tumor factors and liver function is generally complicated. New evaluation methods for patients receiving systemic therapy are needed because many therapeutic agents have been developed. The liver function plays an important role in maintaining sequential systemic therapies, resulting in prolonged survival. A precise oncologic evaluation is necessary to predict the prognosis. Moreover, although Atez/ Bev is recommended as a first-line treatment according to recent guidelines, there has been no simple survival estimation model for patients receiving this regimen [5,6]. Accordingly, we built a simple applicable score that includes both the liver function and tumor factors.
The ALBI score can be calculated based on only two variables: serum albumin and bilirubin [18]. According to data from a Japanese nationwide survey, the mALBI grade can evaluate the preserved liver function more precisely and accurately in comparison to the Child-Pugh classification, especially in patients with a well-preserved liver function [10]. In addition, the mALBI grade can predict and stratify the prognosis of HCC patients [10]. In fact, the liver function at the initial diagnosis of HCC in Japanese patients was improved due to the development of direct-acting antiviral therapy for HCV [19]. In addition, one of the eligible key criteria of the Imbrave150 trial [1] was Child-Pugh class A. Accordingly, Atez/bev treatment as well as other systemic therapies were likely to be introduced for Child-Pugh class A patients. Indeed, the liver function of about 90% of patients was Child-Pugh class A in the present cohort. Accordingly, a more detailed evaluation for cases with a well-preserved liver function is required. We also analyzed the prognostic capabilities of the mALBI grade and Child-Pugh class A in the training set (shown in Supplemental Fig. 3a and b), showing that the C index for the mALBI grade was higher than that for the Child-Pugh score (0.653 vs. 0.646). Given these findings, we adapted the mALBI grade as good prognostic abilities in the present cohort.
The ALBI score plays an important role in predicting survival in advanced HCC patients treated with systemic agents, including atezolizumab plus bevacizumab [20], lenvatinib [21], sorafenib [22], and ramucirumab [23,24]. These findings were in line with the present results. Recently, we reported that the early interruption of Bev due to AEs was relevant to PFS and OS in patients receiving Atez/Bev, and that Bev treatment was likely to be interrupted in patients with mALBI grade 2b [25]. Accordingly, a relatively poor liver function leads to the interruption of Bev treatment, resulting in poor PFS and OS, which supports the present results.
AFP is a well-known tumor marker and is widely applied in the management of HCC for uses, such as surveillance, diagnosis, treatment response monitoring, and prognostic factors [26]. AFP elevation is associated with a poor prognosis across all stages of HCC [27]. AFP elevation was also related to a high risk of tumor recurrence after surgical resection [28] and liver transplantation [29]. The analysis of transcriptome data, whole-exome sequencing data, and DNA methylome profiling demonstrated that AFP-high tumors showed a different phenotype, which was characterized by poor differentiation, enrichment of progenitor features and enhanced proliferation in comparison to AFP-low tumors [30]. This analysis also showed VEGF pathway enrichment in AFP-high tumors [30]. VEGF hampered the benefit and durability of the response of ICI via certain mechanisms [31]. Given these previous reports, HCC patients with AFP elevation are less responsive to Atez/Bev treatment and show a poor prognosis, which was consistent with the present results.
Another well-known tumor marker, des-gamma-carboxy prothrombin (DCP), is widely used in the screening and diagnosis of HCC, particularly in Japan. We conducted a multivariate analysis (Supplemental Table 4), which showed that DCP was not a predictive factor. One possible reason for this finding is that the serum DCP level is elevated due not only to tumor progression but also to hypoxia stimulation [32] by previous treatments.
The objective response rate (ORR) was numerically the highest in patients with mALF score of 0 points in both the training and validation set, followed by those with 1 point and 2 points; however, the differences were not significant. The reason for the lack of significance is probably due to the low statistical power. In comparison to tyrosine kinase inhibitor treatment, a longer treatment period is required to achieve a tumor response in patients receiving ICI treatment. Accordingly, with a longer observation period and a larger number of cases, significant differences may be observed among the mALF scores.
One thing to note is that the percentage of post-progression treatment was lowest among patients with an ALF of 2 points. Post-progression treatment helped prolong the survival in advanced HCC patients. While the details of post-progression treatment after Atez/Bev have not yet been established, regorafenib [33], cabozantinib [34], and ramucirumab [35] were the established treatments after sorafenib based on phase 3 randomized controlled studies. In fact, sorafenib-regorafenib sequential therapies achieved a good survival, with a median OS of 26 months [36]. Given that the liver function is associated with the transition to postprogression treatment after sorafenib [37] and lenvatinib [38], the baseline liver function might affect the post-progression treatment after Atez/Bev treatment. Accordingly, patients with mALF score of 2 points who had a relatively poor liver function received post-progression treatment less frequently than those with mALF scores of 0 or 1 point.
There are significant differences in any-grade decreased appetite and any-grade fever in the training set, and in anygrade fever, and any-grade hepatic edema in the validation set. Any-grade fever was the only of these factors that seemed to be confirmed in the validation set. However, the incidence of any-grade fever was highest in patients with mALF score of 1 point in the training set, followed by patients with 2 points and patients with 0 points. On the other hand, it was most frequently found in patients with mALF score of 2 points in the validation set, followed by patients with 1 point and 0 points. That is, data concerning to any-grade fever in the training set were not confirmed in the validation set. A further study may be required to confirm whether or not the mALF score can predict the development of AEs during Atez/Bev treatment.
While a previous study [39] reported that the development of sorafenib-related skin AEs and hypertension was related to a favorable prognostic outcome, we did not find any relationship between the mALF score and the development of Atez/Bev-related AEs. A future study with greater focus on prognostic AEs is therefore warranted.
About 10% of patients with BCLC very early-and earlystage HCC were included in the present study. Curative treatments, including surgical treatment and radiofrequency ablation, were recommended for such patients according to guidelines [5,6]. The main reasons why they received systemic therapies but not curative treatments are that they were at a high risk of postoperative liver failure and of mortality after liver resection, and the targeted tumors were adjacent to major vessels and other organs, which were considered to be at high risk for ablation therapies.
In this connection, the Child-Pugh score was ≥ 7 in 7% of patients in the present cohort. Treatment options for such patients might be limited due to their liver function deterioration [40]. While liver transplantation is the only way to cure HCC and improve a poor liver function, it is indicated only for relatively young HCC patients without severe comorbidities who meet the Milan criteria [40]. Regarding Atez/Bev treatment for HCC patients with Child-Pugh class B, two real-world studies [41,42] reported that the PFS and OS in Child-Pugh class B patients were worse than in Child-Pugh class A patients, although the number of Child-Pugh class B cases included was small. Further studies are warranted to investigate the efficacy of Atez/Bev treatment for Child-Pugh class B patients.
The present study was associated with some limitations. First, the study population was relatively small. Second, the median OS was not reached at the time of the analysis due to the relatively short observation period. A longer observation period may affect the present results. Third, this study conducted in a retrospective manner and no patients were excluded. Therefore, patients with poor PS (≥ 2), a poor liver function (Child-Pugh score ≥ 7 or mALBI grade 3), and all BCLC stages were included.
In conclusion, the mALF score can reliably predict the prognosis of HCC patients receiving Atez/Bev.