Immune checkpoint inhibitors (ICIs) targeting programmed death-(ligand) 1 [PD-(L)1] have become the cornerstone of treatment for advanced non–small cell lung cancer (NSCLC). By inhibiting tumor immune evasion, ICIs are able to reactivate antitumor immunity, but their use may also lead to a broad spectrum of immune-related adverse events (irAEs). Mechanically, the generation of irAEs is likely to be linked with destroyed immune tolerance, overactivated lymphocytes, and increased pro-inflammatory cytokines and autoantibodies [1-5], which appears to represent an enhanced antitumor response due to the overactivation of the immune system triggered by ICIs. Therefore, there has been growing concern about whether irAEs could be a predictive clinical biomarker of ICI efficacy.

Previous studies have correlated irAEs with increased ICI efficacy in NSCLC [1,6-8]. Nevertheless, since the development of irAEs is a time-varying covariate that would lead to immortal-time bias (ITB) if analyzed conventionally as a fixed covariate, these studies that did not take this bias into account may have overestimated the survival difference in favor of the irAE group [9]. To eliminate ITB, appropriate statistical methods, such as landmark or time-dependent Cox regression analysis, have been used in multiple studies but have demonstrated conflicting results. For instance, several studies manifested a significantly greater survival benefit in NSCLC patients with irAEs than in those without [10,11], while others indicated no significant difference in progression-free survival (PFS) or overall survival (OS) between irAE and non-irAE groups [12,13]. Furthermore, this inconsistency was more prominent when both methods were simultaneously applied to limit the bias. Kfoury et al. [14] detected that the occurrence of irAEs was positively associated with PFS and OS in the time-dependent Cox model, which was absent in the landmark analysis. Conversely, a recent pooled analysis of NSCLC patients from three randomized phase III trials reported that regardless of statistical methods, patients with irAEs had a longer OS than those without irAEs in the atezolizumab arm [15]. In addition, available data have shown that not all specific irAEs are equal in terms of improving ICI response, which is an intricate balance of the integrational effects of irAE characteristics and subsequent management. Different types and numbers of organ-specific irAEs, different grades of irAEs, and different subsequent management strategies, including whether to discontinue or resume ICIs or to use systemic steroids, may selectively exert differing effects (beneficial, harmful, or uncertain) [1,11,15-17], but the evidence has been limited and controversial.

Given the above unresolved questions, we retrospectively investigated the association between irAEs and survival outcomes in advanced NSCLC patients who received anti–PD-(L)1 monotherapy, using appropriate statistical methods to adequately limit ITB. Furthermore, due to the complexity and differing effects of different irAEs, we aimed to further analyze the impact of specific irAEs on survival in real-world settings, especially organ-specific irAEs.

Our study suggests that the presence of overall irAEs was associated with improved survival, which was independent of ITB and potential confounders. Furthermore, patients with skin, thyroid, or hepatic irAEs have better survival benefits, particularly those with single-organ involvement. To our knowledge, this study was the largest cohort study to date specifically focusing on such an association in real-world NSCLC patients receiving anti–PD-(L)1 monotherapy, and it had a longer follow-up than any other previous study.

The incidence of overall irAEs (29.9%) in our study was slightly lower than that in most retrospective studies on NSCLC (28.9% to 51%) [11,12,17-19], which might be attributed to differences in selection criteria that we excluded nonspecific irAEs (such as fatigue, nausea, and vomiting) and patients with combination therapy as additional chemotherapy could result in an overestimation of irAEs [20]. The spectrum of irAEs was similar to those previously reported, and no new safety signal was observed. Interestingly, patients who presented with irAEs received more ICI doses than those who did not (median, 12.0 vs. 4.5). This could be interpreted by the significant difference in ORR between the irAE and non-irAE groups (37.0% vs. 17.8%), suggesting that patients with irAEs had a greater response to ICI than those without irAEs and therefore received more ICI doses. However, in the non-irAEs group, a significant proportion of patients (19.1%, 57/298) rapidly progressed and even died with only 2 doses of ICI, potentially reducing the duration of treatment in this group. This further adds to the evidence that the occurrence of irAEs could be a predictive clinical biomarker of ICI efficacy.

Regardless of the statistical methods, we observed that the occurrence of irAEs was significantly associated with improved outcomes, similar to most studies using the time-dependent Cox model [10,14,18,20] but contrary to some studies using landmark analysis [12,14]. Such heterogeneous results might be explained by the limitations of the methodology. Landmark analysis excluded patients who experienced early progression and/or death as well as late-onset irAEs, dramatically reducing the sample size and thus impairing statistical power, particularly in small-size analyses. Additionally, the variations in the selection of landmark time points could also strongly influence the results. In most studies, landmark times ranged from 4 weeks to 12 months after ICI initiation. We set the landmarks at 2, 3, 6, and 9 months, aiming to maximize the number of patients developing irAEs while minimizing the number of patients having progression/death prior to the landmarks. The time-dependent Cox analysis adequately included all events and follow-ups since cohort entry, avoiding the two important drawbacks of landmark analysis, so it appeared to be a better methodology to constrain the ITB. Larger prospective studies are warranted to definitively confirm our findings.

Regarding organ-specific irAEs, our results were partially consistent with some retrospective studies [11,14,21,22], suggesting that skin and thyroid irAEs were significantly correlated with ICI efficacy; unlike the controversial relationships [8,13,23], we revealed that hepatic irAEs were associated with better outcomes, while pneumonitis tended to be detrimental. However, why different organ-specific irAEs are related to different outcomes remains unsolved. A possible explanation from a clinical perspective is that different irAE grades and management may have differing effects on survival. Indeed, we noted that grade 1-2 irAEs were associated with superior survival benefits than grade 3-5 or no irAEs and that treatment discontinuation appeared to counteract ICI efficacy, both of which are in line with other studies [15,24,25]. Compared to the other three types of irAEs in our study, pneumonitis exhibited the lowest rate of grade 1-2 (66.7%) and the highest rate of treatment discontinuance (77.8%) and even resulted in three deaths; thus, it seemed logical that immune-related pneumonitis could impair survival. This result might be related to the fact that pneumonitis is generally dangerous because of its poorly tolerated hypoxia, particularly in lung cancers that often present with underlying lung damage and limited lung function. In contrast, hepatic irAEs in our study mostly manifested as an increased alanine aminotransferase or aspartate aminotransferase without symptoms or other manifestations, which may be responsible for its positive effect on survival. Altogether, survival differences between different organ-specific irAEs may be attributed to irAE severity and whether treatment was discontinued or not. Alternatively, another assumption from a biological perspective was that organ-specific irAEs might be driven by specific immunopathogenic mechanisms, contributing to distinct responses to ICIs [26,27], which merits further study in detail.

Patients with multi-organ irAEs have previously been shown to have better survival outcomes than those with single-organ or no irAEs [17]. Mechanistically, the development of multi-organ irAEs could be interpreted as ICIs effectively targeting several organs and reflecting a greater antitumor response. However, enhanced immune activation may also be more likely to cause severe irAEs, offsetting the efficacy of ICIs and increasing the risk of death. In our analysis, patients with single-organ irAEs had superior PFS and OS than those with multi-organ or no irAEs. This different result might be attributed to the fact that patients with multiorgan irAEs (12/38, 31.6%) experienced higher rates of grade ≥ 3 irAEs than those with single-organ irAEs (15/89, 16.9%), and even led to two deaths. Notably, this survival benefit of single-organ irAEs persisted in subgroups of skin, thyroid, and hepatic irAEs. Insights into mechanisms from basic and translational research are needed.

After severe irAEs are resolved, it may be feasible to resume treatment with ICIs. Simonaggio et al. [28] reported tolerable safety and similar PFS regardless of whether ICIs were resumed or not. Another study by Santini et al. [29] demonstrated a significant improvement in PFS and OS in patients who resumed ICIs, and that subsequent irAEs were mostly mild and manageable. In our study, ICI resumption showed a trend of improved PFS and OS compared to treatment withdrawal, with an acceptable incidence and severity of subsequent irAEs. Overall, ICI resumption appears to be a promising therapeutic option in some patients, but the decision should involve careful balancing of the benefits and risks by physicians. Large prospective trials are warranted to complement the limited available evidence from small-scale retrospective studies.

There were some limitations to our study. First, this was a single-institutional retrospective study. Second, even though two methods, with their own advantages, were applied to eliminate the ITB, they may also create new biases. This is inevitable, but we tried to minimize this bias. Additional limitations include the small subgroup size of patients, especially in the landmark analysis. Finally, it has been hypothesized that the development of irAEs may capture only part of the mechanism of action by which ICIs improve survival, with potential interference from other factors that can affect immune efficacy, such as PD-L1 expression and tumor mutational burden (TMB). However, we were unable to determine the prognostic value of irAEs independently of PD-L1 expression and TMB because both were not routinely tested. Considering these limitations, we should carefully interpret the current results and conduct prospective studies to verify the findings of the association between irAEs and ICI efficacy.

In conclusion, our study indicates that the occurrence of overall irAEs is associated with the efficacy of anti–PD-(L)1 monotherapy in patients with advanced NSCLC after adequately controlling for ITB with landmark and time-dependent Cox analysis. Specifically, certain organ-specific irAEs, such as skin, thyroid, or hepatic irAEs, seem to be associated with better clinical outcomes, particularly in patients with single-organ involvement. In patients experiencing irAEs, the presence of severe grade irAEs and irAE-related treatment discontinuation were related to a detrimental effect on survival, but the use of systemic steroids was not.