The major findings of this study include: (1) durvalumab is effective in anti-PD-1 refractory HCC; (2) a prior response to anti-PD-1 predicts the response and outcomes to durvalumab; (3) the irAEs from prior anti-PD-1 therapy showed no association with subsequent irAEs from durvalumab.

In HCC, the evidence supporting the strategy of rechallenging with a second ICI in patients who were previously refractory to an initial ICI is limited. Wong et al. reported a retrospective cohort study involving 25 HCC patients who had progressed on anti-PD-(L)1 therapy and were subsequently treated with a combination of ipilimumab and anti-PD-1 (nivolumab or pembrolizumab), demonstrating an ORR of 16%. This study found no significant difference in prognosis between cases of primary and acquired resistance to prior anti-PD-(L)1 treatment, marking it as the first to show that combining anti-CTLA-4 with anti-PD-1 could overcome resistance to prior anti-PD-(L)1 treatments [12]. Similarly, Alden et al. reported a retrospective study delivering salvage therapy with ipilimumab plus nivolumab to 32 patients who had progressed on prior anti-PD-(L)1 treatment, which showed an ORR of 22% [13]. Furthermore, Scheiner et al. described an international, retrospective multicenter study involving 58 HCC patients who received at least 2 lines of ICI-based therapies at 14 institutions. The therapies included ICI monotherapy, dual ICI regimens, and ICI combined with targeted therapies or anti-vascular endothelial growth factor (anti-VEGF) agents. The ORR for second ICI-based therapies reached up to 26%, despite the heterogeneity of the regimens used [14]. In summary, in HCC patients resistant to prior anti-PD-(L)1 treatments, rechallenge with dual immunotherapy comprising anti-PD-1 and anti-CTLA-4 has shown some evidence of antitumor efficacy. However, rechallenge with different anti-PD-(L)1 agents still lacks definitive data.

As for rechallenging with a second ICI in other cancer types, data remain sparse and the patient contexts are typically heterogeneous. Fujita et al. retrospectively analyzed the limited efficacy of salvage anti-PD-L1 with atezolizumab in 18 non-small lung cancer (NSCLC) patients previously treated with anti-PD-1 (nivolumab, pembrolizumab), noting that none achieved tumor shrinkage [15]. Similarly, Kitagawa et al. reported an ORR of 5.9% among 17 NSCLC patients undergoing interchange between anti-PD-1 (nivolumab and pembrolizumab) and anti-PD-L1 (atezolizumab) as rechallenge therapy, where 41.2% of cases showed PD-L1 expression greater than 1% [16]. Furthermore, a meta-analysis incorporating 18 studies of NSCLC involving patients previously treated with anti-PD-(L)1 with or without anti-CTLA-4, and subsequently rechallenged with anti-PD-(L)1 after disease progression, demonstrated a pooled ORR of 8%, indicating a modest benefit [17]. In metastatic renal cell carcinoma, another meta-analysis that included ten studies from prospective, retrospective, or ambispective designs revealed therapeutic benefits from reintroducing second-line ICI-containing treatments in patients previously treated with ICIs, with a pooled ORR of 19% [18]. In summary, evidence regarding the interchange between anti-PD-1 and anti-PD-L1 is limited and the available data appear controversial.

The mechanisms underlying primary and acquired resistance to ICIs are complex, and overcoming this resistance remains a significant challenge. Any disruption in the key steps of the immune response against cancer diminishes the effectiveness of ICIs. These key processes include the release of neoantigens from cancer cells, their capture and presentation by antigen-presenting cells, the recognition of these neoantigens and subsequent activation of CD8 + cytotoxic T cells, and the interaction of these T cells with malignant cells [19,20,21]. Among the critical elements of the immune mechanism, the upregulation of immune checkpoints such as CTLA-4, PD-(L)1, lymphocyte activation gene 3 (LAG-3), T cell immunoglobulin and mucin domain 3 (TIM-3), and V-domain Ig suppressor of T cell activation (VISTA) can inhibit interactions between T cells and malignant cells, contributing to the loss of T cell effector function [22]. Theoretically, when cancer becomes refractory to one type of ICI, other ICIs may still be effective. Clinical data reviewed earlier demonstrate that anti-CTLA-4 can effectively treat PD-(L)1 refractory cancers. In addition, relatlimab, a novel ICI targeting the LAG-3, was shown to have an ORR of 12.0% when combined with nivolumab in melanoma previously treated with anti-PD-(L)1, according to the phase I/IIa RELATIVITY-020 trial [23]. This trial offers a rationale for rechallenging with different ICI targets in resistant cases.

Anti-PD-1 and anti-PD-L1 inhibitors target the same signaling pathway and are approved for many of the same cancer types, often leading to their consideration as equivalent therapies. However, anti-PD-1 inhibitors block signals through both PD-L1 and PD-L2 ligands, adding a layer of complexity to their mechanism of action [24]. In contrast, Linhares et al. demonstrated that anti-PD-L1 inhibitors have lower half maximal effective concentrations (EC50) and are more potent than anti-PD-1 inhibitors, as evidenced by functional assays conducted in vitro [25]. Therefore, despite their similarities, significant differences between these drugs exist. These differences may partially explain why durvalumab, a PD-L1 inhibitor, shows some therapeutic effects in patients with HCC previously refractory to anti-PD-1 therapy in our study.

Regarding predictive markers, immunotherapy has shown particular effectiveness in treating HBV-related HCC. Subgroup analysis from the IMbrave150 study indicated that HBV-related HCC generally has better outcomes compared to the non-HBV group [26]. Similarly, the HIMALAYA study demonstrated that the STRIDE regimen resulted in more favorable outcomes for HBV-HCC [27]. However, our study did not find a significant difference in efficacy between HBV-related and non-HBV-related HCC, possibly due to the heavily pretreated patient population. As for liver function, several studies have identified the ALBI grade as a predictive marker in systemic treatments for HCC [28]. Our findings align with this, underscoring the importance of ALBI grade as a critical prognostic indicator in our cohort.

Currently, there are no standard therapies recommended by guidelines for HCC patients who have failed multiple lines of treatment. In our cohort, 76.9% of patients had previously received more than two lines of therapy, with all patients having been treated with anti-PD-1 therapy and at least one molecular targeted therapy. Although the combination of ICI and molecular targeted therapies lacks robust evidence from randomized controlled trials—evidenced by the failures of studies like LEAP-002 and COSMIC-312—meta-analyses have shown potential benefits from these combinations [29]. For example, a meta-analysis that included both randomized controlled trials and single-arm studies demonstrated that ICI combined with targeted therapy resulted in higher ORR and DCR, as well as longer PFS and OS, compared to ICI monotherapy. Similar findings were reported in meta-analyses focusing on lenvatinib combined with ICI and HCV-related HCC [30,31,32]. This evidence supports the rationale for using combination therapy in the real-world treatment of our heavily pretreated patients. In addition, the combination therapy of a TKI with an anti-PD1/PD-L1 agent is also being evaluated in the IMBRAVE-251 study. While the results are awaited, the trial design suggests that this regimen could be considered as a salvage treatment for patients previously treated with ICIs.

Our study is subject to several limitations. First, as a retrospective analysis, it may be affected by information bias and selection bias. However, in the absence of prospective clinical trials addressing this specific issue, retrospective analysis is the only feasible approach. Second, patient heterogeneity, particularly regarding the number of prior systemic treatments and the combinations with different targeted therapies, may introduce bias into the analysis of therapeutic efficacy. However, our analysis indicated that durvalumab combined with various targeted therapies consistently showed efficacy across the cohort. Moreover, the Cox regression model did not reveal any significant differences in efficacy between the different targeted therapies, suggesting that the treatment effects observed are consistent and not strongly influenced by the specific choice of targeted therapy. Nonetheless, these findings should be interpreted with caution, given the inherent challenges posed by patient heterogeneity. Finally, our study focused on anti-PD-L1 rechallenge for anti-PD-1 refractory HCC; whether the reverse sequence of administration could offer benefits remains unknown.