Immune-Related Adverse Events Are Associated With Improved Outcomes After Immune Checkpoint Inhibitor Treatment in Hepatocellular Carcinoma: A Systematic Review and Meta-Analysis.

INTRODUCTION
Immune checkpoint inhibitors (ICIs) are monoclonal antibodies that block checkpoint proteins such as programmed death ligand 1, programmed cell death protein 1, or cytotoxic T-lymphocyte-associated protein 4. ICIs have become the pillar of systemic treatment of advanced-stage hepatocellular carcinoma (HCC). Although they have led to significant improvement in survival outcomes of patients with HCC, ICIs are not without risks (1).
ICI-related adverse events or immune-related adverse events (irAEs) can affect almost any organ system (e.g., thyroiditis, pneumonitis, myocarditis, hepatitis, and colitis) and have the potential to cause serious or life-threatening events (2). IrAEs are distinct from the toxicities seen with conventional systemic chemotherapy because these are inflammatory responses that arise because of an aberrant overactivation of the immune system. These irAEs often require treatment with corticosteroids or an additional immunosuppressive agent and warrant discontinuation if sufficiently severe (i.e., grade 3–4 toxicities) (3).
Previous studies have suggested that the development of irAEs is associated with increased efficacy of ICIs in various cancers including non-small cell lung cancer (NSCLC), melanoma, gastric cancer, or head and neck cancer (4–8). However, conflicting data exist regarding this relationship, with a previous systematic review of randomized trials of ICI therapy in solid tumors showing varying magnitudes of associations between treatment effects and irAEs (9).
In this study, we aimed to provide a comprehensive review of the association between irAEs and clinical outcomes in patients with HCC treated with ICIs. We also attempted to identify potential factors that might modify the strength of this association, including the grades and types of irAEs.

METHODS

Study design and search strategy
We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guideline for systematic review (10) and searched PubMed, Scopus, Web of Science, and Cochrane Central Register of Controlled Trials databases in June 2024 using a search strategy that includes keywords including “immune checkpoint inhibitor,” “hepatocellular carcinoma,” and “immune-related adverse events” (complete search strategy provided in Supplementary Digital Content, see Supplementary Tables 1-2, http://links.lww.com/AJG/D669). Additional studies were identified by screening the citations from previous related reviews and manual searches.

Inclusion and exclusion criteria and study selection
Each study was reviewed by 2 independent reviewers (S.Y.H., M.S.R.-Z., and A.M.A.) for eligibility and relevant data. Any disagreements were resolved through discussion between 2 researchers or in consultation with J.D.Y. All screening was performed using the Covidence (Covidence systematic review software, Veritas Health Innovation, Melbourne, Australia; Available at www.covidence.org) platform. Eligible studies were those that included patients with HCC who received ICI therapy and compared prognostic outcomes between patients who experienced irAE versus patients who did not experience irAE. Case reports or case series were excluded, as well as studies that did not present original data, such as a literature review. Studies lacking sufficient data comparing treatment outcomes and prognosis between patients with and without irAEs were also excluded. Duplicate cohorts were removed with the selection of the larger cohort.

Patient and public involvement
Patients were not involved in the design and conduct of this research.

Data extraction
From eligible studies, we extracted the following data: publication year, country, study design, duration of follow-up, ICIs received, non-ICI treatment received, age, sex, Barcelona Clinic Liver Cancer (BCLC) stage, Child-Pugh class, Eastern Cooperative Oncology Group performance status, etiology of HCC, presence of extrahepatic metastasis (yes/no), number of patients who developed vs. did not develop irAEs after receiving an ICI, and grades of irAEs; affected organ systems by irAEs; hazard ratio (HR) and 95% confidence interval (CI) of overall survival (OS) and progression-free survival (PFS); and odds ratio and 95% CI of disease control rate (DCR) and objective response rate (ORR). Grading of irAEs followed the Common Terminology Criteria for Adverse Events version 5.0, with mild irAEs classified as grades 1 and 2 and severe irAEs as grades 3 and 4(11).

Quality assessment
The Newcastle-Ottawa Scale (NOS) was applied to assess the quality of the included observational studies (12). An independent assessment was performed by S.Y.H. and M.S.R.-Z., and any conflict was addressed through discussion. We gave greater weight to studies that focused specifically on patients receiving only ICIs, rather than those that included patients receiving both ICIs and an additional intervention, particularly in the area of comparability. Original studies designed to explore the association between irAE (regardless of the type of irAE) and the prognosis of HCC were favored for the representativeness of the cohort.

Statistical analysis
A meta-analysis of HRs (irAE versus non-irAE group) was performed for both OS and PFS and presented as forest plots. The relative ratio (RR) of ORR and DCR between the irAE and non-irAE groups was also calculated and presented in forest plots. A random effects model and a common effects model were both used for statistical analysis, but the results were interpreted based on the random effects model. The common effects model was used as a comparison with the random effects model to provide a better understanding of the variability between the original studies. The Q and Higgins I2 statistics were computed to assess the heterogeneity across the included studies (13). To assess for publication bias, a visual inspection was performed by plotting the effect size against standard error (funnel plot), and Egger’s test was also conducted. In addition, the trim-and-fill method was applied to overcome any potential minimal publication bias. Subgroup analyses were performed by the grade of irAEs, treatment regimen (atezolizumab plus bevacizumab versus nivolumab or pembrolizumab), type of irAE (i.e., affected organ system), and study quality per the NOS scale. Sensitivity analyses on the primary outcome of OS were performed using the leave-one-out method. Metaregression analysis was performed to investigate whether any demographic or clinical factors would modify the association between irAEs and OS. All statistical analyses were conducted using the R software (https://www.r-project.org/; version 4.0.0).

RESULTS

Literature search
Of 3,028 potentially eligible studies identified in the literature search, 83 underwent full-text review. Ultimately, 24 studies (14–37) were included in the final analysis (see Supplemental Figure 1, http://links.lww.com/AJG/D669).

Study characteristics
Baseline characteristics of the included studies, including age and sex, etiology of HCC, treatment regimen, alpha-fetoprotein levels, Eastern Cooperative Oncology Group performance status, Child-Pugh score, and BCLC staging, are presented in Table 1. All studies were observational studies (3 prospective and 21 retrospective cohort studies). Three studies were multinational, 2 were from the United States, 1 was from Europe, and 18 were performed in Asia. Ten studies included atezolizumab plus bevacizumab as the treatment regimen, 5 studies included nivolumab or pembrolizumab, 1 study included sintilimab, and 1 study included tremelimumab with or without durvalumab. Other treatment regimens included toripalimab, camrelizumab, or tislelizumab.

Association between irAEs and HCC clinical outcomes
Of 4,127 patients with HCC treated with an ICI, 31.0% (n = 1,114) had at least 1 irAE, while the remaining 2,483 patients did not have irAEs. Further characteristics of the irAEs including the grade and type of irAE and overall prognosis, reported as median OS/PFS, in the irAE group and non-irAE group are presented in Table 2. IrAEs were associated with a longer OS, although the difference did not reach statistical significance (pooled HR: 0.84; 95% CI: 0.63–1.12, I2 = 73%) (Figure 1). IrAEs were associated with longer PFS (pooled HR: 0.66; 95% CI: 0.52–0.84, I2 = 71%) (Figure 2), significantly increased ORR (pooled RR: 1.73; 95% CI: 1.36–2.21, I2 = 41%), and increased DCR (pooled RR: 1.45; 95% CI: 1.21–1.74, I2 = 74%) (Figure 3a,b). All pooled outcomes were limited by high heterogeneity, which was explored in subgroup and sensitivity analyses.

Subgroup analyses
Subgroup analysis showed a survival benefit for patients with grade 1–2 irAEs (pooled HR: 0.50; 95% CI: 0.36–0.67, I2 = 0%) but not for those with grade 3–4 irAEs (pooled HR: 0.95; 95% CI: 0.24–3.72, I2 = 84%) (Figure 4a). For subgroup analysis by type of irAE, there was an association with improved survival for patients with endocrine irAEs (pooled HR: 0.63; 95% CI: 0.48–0.83, I2 = 12%) but not hepatobiliary irAEs (pooled HR: 0.94; 95% CI: 0.55–1.63, I2 = 66%), dermatological irAEs (pooled HR: 0.52; 95% CI: 0.25–1.10, I2 = 60%), gastrointestinal irAEs (pooled HR: 0.74; 95% CI: 0.37–1.45, I2 = 0%), and pulmonary irAEs (pooled HR: 1.16; 95% CI: 0.01–106.65, I2 = 89%) (Figure 4b). Subgroup analysis based on treatment regimen showed a pooled HR of 0.95 (95% CI: 0.64–1.40, I2 = 69%) for atezolizumab plus bevacizumab and a pooled HR of 0.76 (95% CI: 0.39–1.51, I2 = 73%) for nivolumab or pembrolizumab (see Supplementary Figure 2, http://links.lww.com/AJG/D669). No significant difference was observed in the quality of studies assessed using the NOS scale (Figure 3, see Supplementary Table 3, http://links.lww.com/AJG/D669).

Metaregression
Metaregression showed a statistically significant association between the percentage of patients with advanced HCC (≥BCLC stage C) and OS benefit (coefficient β = —0.0187; 95% CI: −0.0351, −0.0023, I2 = 71.63%, R2 = 25.74%), with a greater OS benefit observed in studies with a higher proportion of advanced-stage patients (see Supplementary Figure 4, http://links.lww.com/AJG/D669). On metaregression, age, the proportion of men, viral etiology, hepatitis B, extrahepatic metastasis, and Child-Pugh Class A did not modify the association between irAE and OS with metaregression coefficients of−0.0033 (P = 0.88), 0.0170 (P = 0.57), −0.0027 (P = 0.81), −0.0021 (P = 0.75), −0.0138 (P = 0.33),and0.0247 (P = 0.12), respectively (see Supplementary Figures 5-10, http://links.lww.com/AJG/D669).

Sensitivity analysis
A leave-one-out analysis was conducted to evaluate the impact of each study on the overall pooled effect estimate (see Supplementary Figure 11, http://links.lww.com/AJG/D669). The results ranged from a pooled HR of 0.80 (95% CI: 0.62–1.03, I2 = 70%) when excluding pulmonary irAEs studied by Song et al (28) to a pooled HR of 0.87 (95% CI: 0.65–1.17, I2 = 73%) when excluding Wong et al (34), but none of the exclusions altered the direction or significance of the overall findings. Song et al (28) contributed the most significant impact in terms of heterogeneity.

Publication bias
Egger’s test (P = 0.696) did not indicate significant publication bias for OS. The funnel plot (see Supplementary Figure 12, http://links.lww.com/AJG/D669) showed a minimal asymmetry, particularly in the lower right corner, suggesting underreporting of smaller studies linking irAE to worse OS. To address this, we applied the trim-and-fill method (see Supplementary Figure 13, http://links.lww.com/AJG/D669), imputing 3 potentially missing studies. After adjustment, the pooled random effects HR was 0.92 (95% CI: 0.68–1.24).

DISCUSSION
ICIs have changed the landscape of treatment strategy in advanced HCC with significant improvements in efficacy and tolerability compared with tyrosine kinase inhibitors. However, irAEs occur in a subset of patients because of the disruption of programmed cell death protein 1 and cytotoxic T-lymphocyte-associated protein 4 pathways, which are crucial for maintaining immune homeostasis (38). In our study, irAEs were associated with favorable clinical outcomes in HCC, including higher ORR and DCR and improved PFS, although there was no statistically significant difference in OS. However, we identified that mild irAEs (grade 1–2) were associated with improved OS compared with the non-irAE group, but this was not the case for severe irAEs.
Our results demonstrate a significant benefit in ORR and PFS with the development of irAEs, as we hypothesize the development of irAEs is an indicator of robust immune activation from the ICI and correlates with the efficacy of the ICI. Development of irAE showed a trend of association with improved OS but did not demonstrate a statistically significant association. This could be partly due to the lack of association between high-grade irAE and OS. Our findings suggest that mild irAEs are associated with a favorable prognosis. These are consistent with previous data from other cancers. A meta-analysis on multiple cancer types, including NSCLC, melanoma, and renal cell carcinoma, found that the occurrence of mild irAEs, but not severe irAEs, was associated with improved OS (39). Treatment discontinuation or interruption, or higher rate of secondary immunosuppressant use because of higher grade irAEs, which can be associated with significant liver dysfunction or an overall decline in performance status, may explain the lack of OS benefits with grade 3–4 toxicity (15,29).
Although corticosteroids and other immunosuppressants are commonly used to manage irAEs in HCC, their impact on OS remains controversial. One study reported that neither baseline nor concurrent corticosteroid use affected OS in patients with HOC receiving ICIs although corticosteroid use for palliation of cancer-related symptoms was predictive of worse OS (40). On the other hand, another study found that the use of systemic corticosteroids in HOC patients with irAE was associated with a trend toward a longer OS (20.7 vs. 14.3 months, P = 0.064) (22). Studies in other cancers suggest worse outcomes with corticosteroid exposure; for instance, more than 10 mg of prednisone at ICI initiation in NSCLS (41) and early high-dose corticosteroids of more than 60 mg prednisone equivalent in melanoma (42) were associated with worse outcomes. This could be due to the lower immune-mediated anticancer effect of ICIs and the side effects from steroids. High-dose steroids compared with moderate-dose steroids are thought to have more risk than benefit and are no longer recommended for the management of ICI hepatitis (43,44).
Our pooled results showed a favorable prognosis specifically in endocrine irAEs (HR 0.63, 95% CI: 0.48–0.83), and several of the included studies reported endocrine irAEs or thyroid irAEs to be associated with a better prognosis (15,21,22,29,35,37). The association of thyroid dysfunction and better prognosis was also reported in patients with NSCLC, with it being strongly linked to anti-thyroid antibodies correlating with better treatment outcomes (45,46). Increased levels of anti-thyroglobulin and anti-TPO antibodies during pembrolizumab treatment of solid tumors are associated with irAEs, and better survival and autoantibody development under ICI treatment may also serve as a potential marker for both toxicity and efficacy (47,48). Although the mechanism behind this association remains unknown, a possible cause could be related to the different frequencies in the use of high-dose corticosteroids for irAEs, which can potentially lead to worse survival. Because endocrine irAEs are less likely to be treated with high-dose corticosteroids, patients with endocrine irAEs may have better treatment outcomes than those with other types of irAEs given robust anticancer immune activity (49–51).
Skin irAEs, such as rash, were reported to be associated with a favorable prognosis in multiple studies, although our pooled results did not demonstrate statistical significance despite a large effect size (HR 0.52,95% CI 0.25–1.10), which may simply be due to a small sample size (19,22,34,35). Although the improved survival associated with cutaneous toxicities was demonstrated in various types of cancers such as melanoma, NSCLC, and renal cell carcinoma, the mechanism is not fully elucidated (52–54). Skin irAEs are also one of the earliest events to occur after starting ICI therapy, potentially serving as early predictive biomarkers for treatment response (53).
Another important area of interest is related to the prognostic impact of hepatobiliary irAEs. We found no association between hepatobiliary irAEs and OS. Celsa et al 2024 reported that grade 1–2 irAE was associated with improved OS and that hepatic decompensation, a major cause of mortality in patients with HCC, was observed in only 7% of patients with HCC who developed immune-related liver injury (14). Although patients with significant liver dysfunction would be expected to have a major detrimental effect from hepatic irAEs, our metaregression analysis suggested that the Child-Pugh class did not affect the OS benefit of irAEs. Although the impact of irAEs has yet to be fully investigated in patients with Child-Pugh B cirrhosis, Lee et al 2020 reported that the incidence of irAEs was significantly elevated in these patients (18). A recent study showed that patients with HCC with Child-Pugh class B cirrhosis had a similar safety profile and response rate to ICI treatment as those with Child-Pugh class A, despite having worse OS (55). The clinical impact of hepatic irAEs, particularly in patients with Child-Pugh B, needs further investigation.
We recognize several limitations of the study. First, there is a risk of misclassification bias, because some treatment-related adverse events may not be immune-mediated. In addition, some studies used combination regimens involving ICIs and vascular endothelial growth factor inhibitors, complicating the differentiation between immune-related and other treatment-related adverse events. Although this could bias our estimates toward the null, our results suggest a potential prognostic value of irAEs, indicating that the true magnitude of the association might be even greater. Second, we lack mechanistic evidence to support our findings. For instance, our metaregression analysis indicated that studies involving patients with more advanced-stage HCC showed a slightly greater OS benefit in the irAE group compared with the non-irAE group. Future mechanistic studies are needed to explore the underlying causes of the varying survival benefits associated with irAEs across different patient populations. In addition, the studies included in our meta-analysis are mostly retrospective cohort studies and not prospective clinical trials. Further prospective studies need to be performed to verify our results. Finally, the included studies had high heterogeneity, which resulted in wide Cis for our estimates and need to be interpreted with caution. For instance, most studies included were conducted outside of the United States, mainly in Asia, thus the applicability of our results in the US population should be further evaluated in future studies. Finally, only 2 studies (Celsa et al 2024 (14) and Hsu et al 2021 (16)) provided a number of patients with cirrhosis, thus we were unable to analyze the effect of cirrhosis in assessing the association between irAEs and prognosis. Acknowledging these limitations, we have performed additional sensitivity analyses, subgroup analyses, and metaregression, which partially identified the cause of heterogeneity.
Our study identified that irAEs are associated with favorable clinical outcomes in HCC, including higher ORR, higher DCR, and improved PFS. Specifically, patients experiencing mild irAEs (grades 1–2) demonstrated improved OS compared with those without irAEs; however, this benefit was not observed in patients with severe irAEs (grades 3–4). These data highlight the importance of early recognition and appropriate management of irAEs to maintain patients on therapy and derive the greatest benefit of ICIs in patients with HCC. Future studies are necessary to elucidate the underlying mechanisms of these associations.

Data sharing statement:
Data analyzed during the study and study protocol are available from the corresponding author by request.

Supplementary Material
Supplementary MaterialsSUPPLEMENTARY MATERIAL accompanies this paper at http://links.lww.com/AJG/D669