Immunotherapy with checkpoint inhibitors (CPI), i.e. anti-PD-1, anti PD-L1 and CTLA-4 antibodies has dramatically improved survival rates in various cancer types during the last decade and they are frequently used in different treatment settings (1–3). Although the introduction of CPIs has improved outcomes in several cancer types, their use cause a considerable risk for immune-related adverse events (IRAEs) which may appear in nearly every organ system and at any point during treatment and even after treatment discontinuation (4). Potential mechanisms resulting in IRAEs include increased T-cell activity against antigens present in both tumors and healthy tissue, increased levels of cytokines, and preexisting autoantibodies, and enhanced complement-mediated inflammation (4). Despite lack of knowledge considering the exact underlying pathophysiological mechanisms, the occurrence of IRAEs reflects activation of the immune system (4, 5). Since the development of IRAEs depends on the mechanism of action of CPIs, it has been assumed that patients developing IRAEs might have a better response to treatment.

The potential association between development of IRAEs and survival outcome is extensively studied. Several studies have shown a positive association between development of IRAEs and clinical benefit for different cancer types including malignant melanoma (MM), non-small cell lung cancer (NSCLC), renal cell carcinoma, urothelial cancer, head and neck cancer, and gastrointestinal cancer (6–10). Immune-related adverse events involving more than one organ, called multiple IRAEs have not been studied to the same extent. The current evidence suggests a better prognosis in patients with multiple IRAEs with a stronger effect magnitude compared to patients with single IRAEs (11–16), even if conflicting results exists (17). However, the current evidence can be questioned due to the high risk for immortal-time bias (ITB) that either has not been considered in some of the studies (14, 15) or it was dealt with in landmark analysis (12, 13, 17) that can also lead to bias compared to the more robust time-dependent Cox model (18). Besides, most of the studies only included patients with NSCLC (11–13, 16, 17), thus impacting the generalizability of study results.

The aim of the present study was to investigate the patterns of multiple IRAE occurrence and their impact on CPI effectiveness in an unselected cohort of patients with advanced cancer using time-dependent Cox models to mitigate the ITB risk.

In our study cohort of 600 patients with advanced cancer treated with CPIs, we observed that approximately one-sixth developed multiple IRAEs. The occurrence of multiple IRAEs was associated with better treatment effectiveness as demonstrated by improvements in both PFS and OS with a magnitude of benefit significantly stronger compared to patients with single IRAEs.

Current evidence concerning development of multiple IRAEs for patients treated with CPIs and its impact on treatment effectiveness is scarce. There are only few previous studies (11–13, 16, 17), most of them indicating a stronger association between the development of multiple IRAEs, as opposed to single and survival (11, 12, 16). However, the cohorts in these studies included only patients with NSCLC and, therefore, the generalizability of study results can be questioned. Two earlier studies included patients with various malignant diseases, but smaller cohorts of approximately 200 patients (14, 15) demonstrated results in line with our larger cohort that provides more convincing evidence on this potential association in a broader patient population. A major methodological drawback of previous studies is how the risk for ITB was dealt. Immortal-time bias is a challenge to consider in the association between IRAEs and clinical outcome since patients responding to therapies continue treatment for a longer time and therefore increase their risk of developing IRAEs. To eliminate the risk of ITB in observational studies of survival outcomes established methodology such as landmark analysis, Cox model with time-varying variable or inverse-probability weighed models are routinely used (18–21). These methodological approaches cannot be considered as equal in terms of the validity of results since Cox model with time-varying variable seems to outperform landmark analysis (18). Considering the current evidence on potential association between multiple IRAEs and CPI effectiveness, some studies did not deal with ITB at all (14, 15) whereas others used landmark analysis only (12, 13, 17). We found two previous studies with a proper analysis of ITB reporting positive correlation between multiple IRAEs and outcome. However, both face issues related to generalizability of results to the clinical setting since only patients with NSCLC were included (11, 16) and one of the studies only analyzed patients participating in clinical trials and not in real-world setting (16). Until today, the current study is the first investigating the association between multiple IRAEs and CPI effectiveness in a broader cohort of patients with various malignancies using the preferable Cox regression model with time-dependent covariate as methodological approach for mitigating the ITB.

Subgroup analysis in our cohort showed statistically significantly improved PFS and OS for patients with MM but, as opposed to Shankar et al, not for patients with NSCLC. The lower number of NSCLC patients in our cohort compared to the study by Shankar et al. which only included NSCLC patients (n=205 vs. n=623) may, in part, explain this discrepancy (11). Interestingly, a trend towards improved PFS and OS for patients given CPI as a first line treatment that was diminished in later lines was observed in our cohort of patients with NSCLC highlighting treatment line as a source of heterogeneity among studies that could impact the results. This information was lacking from Shankar et al. and could also contribute to the discrepancy of study results. At the same time, one could argue that the ability of IRAEs to predict CPI effectiveness might depend on cancer type, a notion that is supported by the observations on the substantial differences on tumor immunogenicity among different cancer types (22, 23). In fact, the treatment effect is not equal in patients with different cancer types and differences in adverse effects among different cancer types have also been observed. A systematic review of 48 trials reported higher risk of developing skin and gastrointestinal IRAEs and lower risk of experiencing lung IRAEs for patients with MM compared to those with NSCLC whereas higher incidence of arthralgia and hypothyroidism in MM patients compared to patients with renal cell carcinoma was observed (5). Furthermore, it has been demonstrated that specific types of IRAEs exhibit a more profound correlation to treatment effectiveness and the type of these IRAEs differ across various cancer types. For example, vitiligo has been linked to treatment response in melanoma patients and thyroid dysfunction in NSCLC and renal cell carcinoma patients (24–28).

Checkpoint inhibitor combination therapy with PD-1 and anti-CTLA4 inhibitor is, for subgroups of patients, known to have better clinical efficacy than monotherapy, but the incidence of IRAEs is higher (29, 30). In our study cohort, we confirmed that the combination was a risk factor for development of multiple IRAEs. However, subgroup analyses of monotherapy only showed significantly better PFS and OS for patients developing multiple IRAEs, indicating that the improved outcome for patients with MM is not only due to the well-known better effect of combination therapy. Performance status ≥2 was associated with lower risk for developing multiple IRAEs in our study. A conceivable explanation is that patients with worse PS are more vulnerable in general with higher risk for earlier treatment discontinuation after the first IRAE episode and thus are less likely to develop multiple IRAEs.

Immunosenescent, a term describing impaired function of the immune system that develops with aging, has been supposed to influence the effect of immunotherapies and the development of IRAEs. However, many studies did not find a higher risk for development of IRAEs in older patients (31–34), but conflicting results exist (35). In a previous study from our research group, we found that a simplified frailty score based on PS, age and comorbidity expressed as Charlson Comorbidity Index (CCI) could predict development of all grade and multiple IRAEs, whereas age, CCI or PS did not separately predict increased risk for development of IRAEs (36). Although not supported by our adjusted analyses, another potential explanation of the association between PS and the risk of IRAEs might be more related to the line of treatment rather than the PS per se. In fact, some evidence suggests a negative impact of prior chemotherapy to immune microenvironment (37) that might influence the risk for IRAEs as well. In this equation, PS could serve as a surrogate for later treatment lines rather than as an explanatory parameter for the risk of IRAEs.

In terms of occurrence patterns and outcomes of IRAEs, we found a higher frequency of IRAE grade ≥ 2 and ≥ 3 in patients with multiple IRAEs than in patients with a single IRAE which could explain why a greater proportion of patients with multiple IRAEs than patients with a single IRAE discontinued treatment due to IRAE. At the same time, multiple IRAEs were not associated with higher risk for major sequelae from IRAEs. The organ systems involved in multiple IRAEs in this study consists of combinations of common single IRAEs without identifying any specific pattern. The number of patients within each pattern of distribution was too small and precludes any further analyses for potential associations of specific patterns with prognosis.

Besides from ITB potentially being the main bias in the association between IRAEs and clinical outcome, the present study has some additional limitations associated with its retrospective nature. These include the risk for misclassification bias regarding both IRAE grading but also classification between single and multiple IRAE as well as information bias where EMRs as data sources for identifying IRAEs might not include information on low grade IRAEs. The study is restricted to tree Swedish centers thus limiting its external validity whereas the relatively lower number of patients with tumor types other than melanoma and NSCLC limits the generalizability of the results. Finally, some variables of potential interest that could be associated with both the development of IRAEs and prognosis, as ethnicity, co-medication with immunosuppressive therapy, pre-existing autoimmune disease, were not available and were, therefore, not taken into account for the analyses. For some variables of interest as prior oncological treatment, the information was available, but the subgroups were too small for relevant analyses.

In conclusion, our study results suggest a statistically significant association between development of multiple IRAEs and CPI treatment effectiveness (measured as PFS and OS) that is mainly driven by patients with MM. These results support not discontinuing immunotherapy, even upon multiple but not severe IRAEs to increase the likelihood of treatment benefit. In addition, our findings suggest that multiple IRAEs may constitute a suitable surrogate marker for treatment efficacy that might be used in clinical trials. However, further studies with larger sample size and prospective design to overcome the inherent biases of retrospective studies is essential to further address the potential interplay between the development of IRAEs and treatment outcome.

The raw data are not available due to ethical restrictions but anonymized data might be available upon request.

The study was approved by the Swedish Ethical Review Authority (reference number 019-02469 and 020-06801). The studies were conducted in accordance with the local legislation and institutional requirements. Informed consent was waived from the Swedish Ethical Review Authority since the eligible patients were already treated with checkpoint inhibitors and the current study would not influence their subsequent treatment or prognosis.

CO: Data curation, Writing – original draft. AK: Data curation, Writing – review & editing. VR: Data curation, Writing – review & editing. AI: Data curation, Writing – review & editing. ED: Data curation, Writing – review & editing. AV: Conceptualization, Formal analysis, Methodology, Project administration, Software, Supervision, Writing – original draft. GU: Conceptualization, Funding acquisition, Methodology, Supervision, Writing – original draft.