Early alpha-fetoprotein response predicts prognosis of immune checkpoint inhibitor and targeted therapy for hepatocellular carcinoma: a systematic review with meta-analysis

ABSTRACT Background The prognostic value of alpha-fetoprotein (AFP) response for efficacy of targeted therapy or immune checkpoint inhibitors (ICIs) has not been established. The purpose of this meta-analysis is to elucidate the relationship between AFP response and survival outcomes in hepatocellular carcinoma (HCC) patients who received targeted therapy or ICIs. Methods The hazard ratio (HR) with 95% confidence interval (CI) was used to evaluate the relationship between AFP response and overall survival (OS)/progression-free survival (PFS). Results Twenty-six articles containing 3056 HCC patients were finally included in the study. The pooled results showed that after targeted therapy or ICIs, patients with decrease in AFP had better prognosis (OS:HR = 0.48, 95%CI:0.40–0.56; PFS:HR = 0.39, 95%CI:0.33–0.46), while patients with increase in AFP had worse prognosis (OS:HR = 2.30, 95%CI:1.82–2.89; PFS:HR = 2.34, 95%CI = 1.69–3.24). Subgroup analysis revealed that compared to AFP decrease >50%, AFP decrease >20% can better predict the prognosis of patients who received targeted therapy (OS:HR = 0.51, 95%CI:0.41–0.62; PFS:HR = 0.39, 95%CI:0.30–0.51) or ICIs treatment (OS:HR = 0.34, 95%CI:0.16–0.71; PFS:HR = 0.22, 95%CI:0.10–0.47), and 8 weeks after targeted therapy may be the appropriate time point for AFP assessment. Conclusion AFP decrease >20% within 8 weeks may be the appropriate definition for early AFP response which probably works best in predicting the efficacy of therapy. Registration The review was not registered.


Introduction
Hepatocellular carcinoma (HCC) is the sixth most common tumor and the third leading cause of cancer-related death worldwide, with approximately 906,000 new cases per year [1]. Approximately 60-70% of patients are diagnosed with intermediate or advanced HCC [2] and for the vast majority of patients with advanced HCC and some intermediate HCC patients, systemic therapy is the main treatment measure [3]. Sorafenib, a multikinase inhibitor, is the first drug used in systemic therapy for advanced HCC and was the only first-line drug approved by the FDA until 2017 [4]. In recent years, lenvatinib has become a new first-line drug for targeted therapy of HCC, and its clinical role has also been confirmed. At the same time, in order to solve the problem of acquired resistance to first-line drugs, second-line drugs such as regorafenib, cabozantinib, and ramucirumab have emerged [5]. In addition, increasing clinical studies have shown the anticancer efficacy of immune checkpoint inhibitors (ICIs), which target immune checkpoint molecules, such as cytotoxic T-lymphocyte-associated protein 4 (CTLA-4), programmed cell death 1 (PD-1), and programmed cell death ligand-1 (PD-L1) [6]. Therefore, immunotherapy has become a new area of intense research in the systemic therapy of advanced HCC, and atezolizumab (anti-PD-L1 antibody) combined with bevacizumab (anti-VEGF antibody) has become the first-line treatment for advanced HCC [3].
For advanced HCC patients who receive targeted therapy, especially sorafenib therapy, many studies have been performed to investigate if there are biomarkers, such as plasma interleukin-8, tumor necrosis factor-α, amphiregulin, vascular endothelial growth factor and monocyte-to-lymphocyte ratio, which can predict the effect of treatment [7][8][9]. However, there are still no validated prognostic markers of the response to sorafenib in HCC [10]. PD-L1 expression is the most commonly used biomarker for predicting prognosis after ICIs treatment [11]. At the same time, tumor lymphocytic infiltration, CTNNB1 mutation status, circulating tumor cell counts, and circulating tumor cells expressing PD-L1 are also used as predictive biological indicators for HCC after ICIs treatment [12][13][14][15]. However, the effectiveness of these biological indicators needs to be further validated. So far, for patients with advanced HCC, the role of biomarkers remains limited in guiding and helping decision-making treatment, and it is necessary to find better indicators to help the decisionmaking for advanced HCC [3].
Alpha-fetoprotein (AFP) is a glycoprotein, serving as a serological indicator for the diagnosis and follow-up of HCC initially [16]. In addition, AFP plays a variety of roles in HCC development, so it is also used as a classical biomarker for judging the efficacy of clinical treatment [17]. Despite many new biomarkers, AFP is still the most commonly used diagnostic, prognostic, and predictive biomarker for HCC [18]. Elevated AFP is a biomarker for poor prognosis of HCC, and the serum level of AFP may be used as a noninvasive predictor of targeted therapy for HCC in clinical trials [18,19]. For patients with elevated baseline AFP levels, changes in AFP levels are associated with the prognosis of sorafenib therapy, and decreased AFP levels are related to prolonged progression-free survival (PFS) and overall survival (OS) [20,21]. Preliminary meta-analysis has found that AFP response (AFP decrease after therapies) was significantly correlated with the prognosis of HCC after treatment [22]. However, it did not evaluate the effect of AFP response on the prognosis of ICIs, and there was no consensus on the definition of AFP response and the prognostic value of early AFP response in advanced HCC. The purpose of this meta-analysis is to explore the association between AFP response and survival outcomes in HCC patients who received targeted therapy or ICIs. At the same time, we hope to guide the decision-making process for treatment decisions by finding the best indicator of early AFP response during the systemic therapy of HCC.

Methods
This study was based on the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) statement guideline [23] to design, organize, and report our metaanalytic findings.

Data sources and search methods
We searched four main databases including PubMed, EMBASE, Cochrane library, and Web of Science from the time of their inception to 20 April 2022. We searched for related articles by searching subject terms plus keywords, including alphafetoproteins, HCC, hepatocellular carcinoma, immune checkpoint inhibitors, targeted therapy, etc. A detailed search strategy for different databases is presented in Supplementary Retrieval Methods.

Inclusion and exclusion criteria
Articles included in the study must meet the following conditions: (1) Patients were diagnosed with HCC and received relevant ICIs or targeted therapy regardless of whether they were combined with other therapies; (2) These studies need to report indicators related to AFP response; (3) These studies need to report indicators related to survival outcomes including OS or PFS; (4) These studies need to provide hazard ratio (HR) with 95% confidence interval (95% CI) that can be directly obtained or calculated. Exclusion criteria include: (1) Case report, review, meta-analysis and letter; (2) Conference and abstract; (3) Animal trials; (4) Non-English language articles; (5) For repeated publications and studies that include overlapping populations, we only include the latest and most comprehensive studies.

Data extraction and quality assessment
Two researchers (TBW & YLJ) independently extracted the following content from the included studies: first author, year, country, enrollment period, intervention, patient number, data collection, baseline AFP levels, (early) AFP response definition, time of AFP response, study type, HR, and 95%CI. Some of the included studies reported data from different experimental groups or data related to different AFP response indicators, and we included them as different studies into the analysis or subgroup analysis. At the same time, some studies have reported the response to increased AFP, and we also included the study for analysis. When both univariate analysis and multivariate analysis data were available, we gave priority to the results in multivariate analysis. We used the Quality in Prognosis Studies (QUIPS) tool to assess the risk of bias of the included studies [24]. The QUIPS tool consists of six bias domains: study participation, study attrition, prognostic factor measurement, outcome measurement, study confounding and statistical analysis, and reporting. For each domain, QUIPS tool classifies the risk of bias as low, moderate, or high.

Statistical analysis
Statistical analyses were performed using Stata 16.0 statistical software (Statacorp LP, College Station, TX), P < 0.05 was considered statistically significant, and the random-effect model would be used. OS and PFS were used to evaluate the AFP response of HCC patients receiving ICIs or targeted therapy. OS and PFS were estimated by using HR with 95% CI. The inverse variance approach was used to construct study weights. Sensitivity analyses were performed to determine the stability of the pooled results and assess the robustness of the pooled effect. If the removal of one study outcome in the sensitivity analysis results in a significant bias of the pooled HR and 95%CI, the outcome will be eliminated.
We used Cochran Q and the inconsistency index (I2) statistic to assess the statistical heterogeneity of the studies. Either I2 greater than 50% or P < 0.10 were considered substantial or significant heterogeneity. Subgroup analyses were performed by treatment measures (targeted therapy and ICIs), sample size (<100 and ≥100), country (China and others), definition of AFP response (AFP decrease >20% and AFP decrease >50%) and time to AFP response (≤4 weeks, ≤8 weeks, and ≤12 weeks), in order to determine the potential sources of heterogeneity. We used Funnel plots, Begg's rank correlation test, and Egger's regression asymmetry test to examine the potential publication bias [25]. Good symmetry of the funnel plot indicated that there was no obvious publication bias, while Begg test and Egger test were used to assess the symmetry of the funnel plots.

Study selection and characteristics
The flowchart of the literature search process is shown in Figure 1. In total, 3462 records were initially identified from the four databases, 1264 of them were repeatedly excluded through the Endnote20 software. After browsing the title and abstract, 2099 records were excluded. After reviewing the remaining 99 full texts, 73 were excluded due to the lack of valid data (OS, PFS, HR, 95% CI), studies sharing the same participants and treatment methods were not related to the study. Finally, 26 articles from 2009 to 2022 were included in the analysis [20,21,.
The main characteristics of the articles that fit the research are summarized in Table 1. Of the 26 articles, only three were prospective, and the rest were retrospective. In total, our study included 3056 patients with HCC who received ICIs or targeted therapy. Among them, 18 studies were on targeted therapy, and the other 8 were on ICIs. Most of studies defined the AFP response as AFP decrease >20% or AFP decrease >50%. Therefore, in the subgroup analysis, we decided to use AFP decrease >20% and AFP decrease >50% as the grouping criteria. There were also four studies reporting an increase in AFP, and three studies reporting an AFP ratio (post-treatment AFP/baseline AFP). We categorized this part of the data into the AFP progress group for a meta-analysis. One study divided the AFP response into Class I-IV [45]. According to the characteristics of its Class I group (AFP decrease ≥50% of baseline at week 4), we decided to include it in the AFP decrease >50% group for analysis. In addition, one study did not compare AFP response with AFP nonresponse directly, so we did not include it in our meta-analysis [47]. Most articles defined the AFP Figure 1. Flow diagram of study selection process.

The correlation of AFP response or progress with PFS
Fourteen studies provided data on the relationship between PFS and AFP response. The pooled HR data showed that patients with AFP response after targeted therapy or ICIs treatment had better PFS (HR = 0.39, 95%CI:0.33-0.46, P < 0.001; Figure 4; Table 3) with mild heterogeneity (I2 = 26.9%, P = 0.166). Sensitivity analysis showed that after excluding the studies one by one, the pooled result did not change (Supplementary Fig 4). Subgroup analysis showed that patients with AFP response from targeted therapy (HR = 0.40, 95%CI:0.31-0.51) or ICIs treatment (HR = 0.37, 95%CI:0.29-0.48) had better PFS than the patient without AFP response. It was noteworthy that significant differences had been shown in the subgroup analysis based on country (P = 0.030) and sample size (P = 0.029). Heterogeneity was reduced when studies were grouped by country and sample size.
Only three studies reported data on the relationship between AFP progress and PFS. Based on these three studies, AFP progress was significantly associated with poor PFS (HR = 2.34, 95%CI:1.69-3.24, P < 0.001; Table 3).

Publication bias
We used funnel chart, Egger's and Begg's linear regression tests to detect the potential publication bias (P < 0.05 was considered a significant publication bias). The funnel plots of the AFP decrease OS and PFS results were approximately symmetrical ( Supplementary Fig 7, 8). There was no evidence of publication bias for the vast majority of results. The Egger's test figures showed there was no evidence of publication bias for the AFP decrease OS and PFS results. (Supplementary Fig 9, 10) For the subgroup with the study number less than 10, we did not perform publication bias analysis. The Begg's and Egger's test results are presented in Table 2 and Table 3, respectively.

Discussion
For advanced HCC patients receiving targeted therapy or ICIs, judging the patient's response to the drug as soon as possible and predicting the therapeutic effect are very important for clinicians to guide the patient's treatment plan and choose the right time to replace the drug. However, there is currently no consensus on the predictive indicators of the prognosis of patients with advanced HCC receiving targeted therapy or ICIs. The value of various indicators is still in the exploratory stage.
As the earliest biological indicator used for the diagnosis and prediction of HCC, AFP is one of the most easily obtained clinical biological indicators for HCC patients. Previous metaanalysis has proved that AFP response as a biomarker was helpful in predicting the OS and PFS of HCC after treatment [22]. However, the predictive role of AFP response in ICIs remains unknown. According to our meta-analysis, AFP response is significantly related to the prognosis of HCC patients receiving ICIs or targeted therapy, especially in patients receiving ICIs. This makes AFP response a good biological indicator in predicting the efficacy of ICIs. Meanwhile, AFP response cannot be observed in all HCC patients, and some patients do not respond or even demonstrate AFP progress after receiving targeted therapy or ICIs. Our meta-   analysis and previous studies have revealed that AFP progress was strongly associated with poor prognosis in HCC patients receiving systemic therapy [18]. Notably, heterogeneity remained significant in some subgroup analyses. In subgroups China and ICIs, we observed a significant reduction in heterogeneity, which may suggest that the main source of heterogeneity derives from populations in different regions (with differences in the main causes of HCC) and different treatment measures. The mechanism by which AFP response can predict patient prognosis may be explained by the role of AFP in promoting HCC growth, proliferation, and metastasis [48]. Therefore, AFP decline is reflective of a patient responding well to treatment.
In order to find a better cutoff value for the AFP response and an earlier time to determine whether the treatment plan needs to be changed, we conducted an in-depth analysis of the data obtained. Compared with the cutoff value of AFP decrease >50%, AFP decrease >20% performs better at predicting OS and PFS, especially for patients who are treated with targeted therapy. At the same time, this prevents many potentially treatment-sensitive patients from being misjudged as insensitive to current treatment. In Hsu's study, early AFP response was defined as AFP decrease >10% after 4 weeks or decrease >20% after 8 weeks treatment [47]. Similarly, Lee and his colleagues also used AFP decrease >10% at 4 weeks as the AFP response [40]. Our study found that 4 weeks time after treatment is too early to determine the AFP response for patients receiving targeted therapy, and it would take even longer time (usually 2-3 months) for ICIs to take effect. According to our analysis, 'AFP decrease >20% within 8 weeks' may be appropriate to judge the AFP response after treatment, which can help clinicians to identify patients who will benefit from targeted or ICIs therapy as soon as possible. Recently, some studies have pointed out that the prognosis prediction of ICIs treatment should adopt the '10-10 rule' or '50-10' rule [40,45]. In addition, it is still unknown whether the predictive value of AFP response in HCC patients with different AFP initial values is consistent. More high-quality prospective studies are warranted in future study.
Based on our meta-analysis, if the patient's AFP level drops by more than 20% within 8 weeks, it will be better to continue the current treatment measures. For patients whose AFP level has increased, it is better to change the existing treatment, as our analysis has revealed that AFP progress was associated with poor prognosis. Patients whose AFP level changes within this range can continue to be observed for up to 12 weeks. If the patient's AFP level drops by more than 20% during this period, they may continue the existing treatment. If not, they may need to consider other drugs. At the same time, it is necessary to make a comprehensive judgment based on imaging changes and other biological indicators, especially for patients whose baseline AFP is within normal ranges.
Compared with other biological indicators, AFP is routinely monitored clinically and relatively cheap. Two other commonly used indicators in clinical practice are Lens culinaris-agglutininreactive fraction of AFP (AFP-L3) and des-gammacarboxyprothrombin (DCP) [50]. Some meta-analyses have proved that a combined assay for multiple serum biomarkers has higher clinical value than individual assay for single biomarkers [51,52]. However, there are relatively few studies on the dynamic changes of AFP-L3 and DCP to predict the efficacy of treatment. Using liquid biopsy to detect circulating tumor cells and circulating tumor DNA also serve as new biomarkers for therapeutic monitoring in HCC patients [53,54]. The relatively complex and expensive testing technique limits their large-scale clinical applications. The main limitation of AFP response is that only about 70% HCC patients' AFP value exceeds the normal range. For the remaining HCC patients, the clinical significance of judging the decline of AFP value may not be significant, so the application of other indicators is more critical for judging the therapeutic effect and prognostic significance of these patients.
Although our meta-analysis provides a comprehensive summary of the current data on the AFP response to targeted therapy or ICIs, our research still has some limitations. First, the included studies have relatively few data on ICIs treatment, and the number of patients included in the study is also relatively small. Second, most of the included studies are retrospective studies, which are prone to cause some bias. Third, the publication bias of studies may occur in some subgroup analysis with a small number of studies. Only articles published in English were included, which may result in certain publication bias. Fourth, in some subgroup analyses, there is still significant heterogeneity, and more research is needed across various subgroups. Last but not least, our definition of early AFP response and guidance for treatment in accordance with AFP response are based on currently available data. For some of the newly proposed schemes, more work is required for further verification, which can contribute to reach the full potential of the guiding role of AFP response.

Conclusion
In summary, AFP response has the prognostic value for the efficacy of targeted or ICIs therapy, especially for ICIs. 'AFP decrease >20% within 8 weeks' may be appropriate to judge early response after treatment, which can help clinicians to identify patients who will benefit from targeted or ICIs therapy as soon as possible.