Associations between Immune-related Adverse Events and Prognosis in Cancer Patients Receiving Immune Checkpoint Inhibitor Therapy.

Introduction
Immunotherapy with immune checkpoint inhibitors (ICIs) has become one of the main options for cancer treatment, and the number of eligible indications has expanded in both metastatic and neoadjuvant/adjuvant settings across cancer types (1). Currently, the most commonly used ICIs are monoclonal antibodies targeting cytotoxic T lymphocyte antigen-4 (CTLA-4), programmed death-1 (PD-1), and PD-1 ligand (PD-L1).
Although these ICIs induce durable antitumor immune responses in a subset of patients, induction of non-tumor-specific immune activation by ICIs can cause a unique spectrum of organ-specific immune-mediated toxicities in many additional patients. Such toxicities are referred to as immune-related adverse events (irAEs) and often lead to the interruption or cessation of ICI treatment. Furthermore, several forms of irAEs, including pneumonitis, myocarditis, myositis, hematologic irAEs, and hepatitis, can be fatal (2).
In the real-world setting, up to 85% of patients receiving anti-CTLA-4 therapy experience irAEs (3) and a lower incidence of irAEs is caused by PD-(L)1 inhibitors (4). In addition, combined therapy with PD-(L)1 and CTLA-4 inhibitors increases the incidence of all-grade and high-grade irAEs compared with either agent alone (5-7). To improve the care of patients with irAEs, several management guidelines have been published by key oncology societies, including the American Society of Clinical Oncology (8), European Society for Medical Oncology (9), Society for Immunotherapy of Cancer (10), and the National Comprehensive Cancer Network (11).
The exponential rise in ICI use together with the high frequency of irAEs has led to increased interest in understanding the immunological and clinical relationships between ICI efficacy and irAEs. Reports of positive associations between irAE incidence and therapeutic responses have been reproducible, suggesting that immune activation by ICIs may be associated with their antitumor activity (1). However, while irAEs can affect any organ system, a knowledge gap exists regarding the impact of organ-specific irAEs on patient outcomes.
This review describes the clinical implications of individual irAEs in the context of ICI treatment efficacy.

Overview of Association between irAEs and Treatment Outcomes During ICI Therapy
Numerous studies have shown that irAE development is associated with improved survival outcomes during ICI therapy in patients with a range of cancers; however, the magnitude and significance of these relationships vary and not all evidence is supportive (3,12,13). For example, several systematic reviews and meta-analyses have reported positive relationships between irAE occurrence and ICI efficacy in patients with lung cancer (14), melanoma (15), renal cell/urothelial carcinoma (16), and in pan-cancer cohorts (17-20). Nevertheless, a randomized trial-based meta-analysis found only weak correlations between total or specific irAE development and the overall survival (OS) in several cancer types, suggesting that caution should be exercised when considering whether irAE development is a valid surrogate for survival outcome (21).
A recent study demonstrated that setting a more stringent irAE definition that discriminates between “pure” irAEs and non-immune-driven AEs based on an “irAE likelihood” score could successfully classify patient outcomes across multiple tumors (22). In that study, hazard ratios (HRs) were significantly lower for patients with high-confidence irAEs (high irAE likelihood scores) than for those with low-confidence irAEs with respect to both the progression-free survival (PFS) and OS. This study highlights the importance of establishing an immunological etiology to determine the association between irAEs and improved outcomes. Eggermont et al. (23) performed a secondary analysis of the Phase 3 EORTC 1325/KEYNOTE-054 trial, which compared pembrolizumab therapy and placebo for the treatment of patients with high-risk stage III melanoma. In the pembrolizumab arm, the occurrence of irAEs was significantly associated with a prolonged recurrence-free survival [RFS; HR, 0.61; 95% confidence interval (CI), 0.39-0.95]. Notably, the reduction in RFS risk was greater after irAE onset than either before onset or in the absence of irAEs in the pembrolizumab arm versus the placebo arm. One possible explanation for this observation is tumor neoantigen-reactive T-cell cross-reactivity against the corresponding wild-type peptide in normal tissue (24,25). Alternatively, dead tumor cells can release antigens, including neoantigens, that prime lymphocytes against wild-type antigens (26). These mechanisms could potentially underlie or contribute to the relationship between irAE development and tumor mutational burden (26-28) or tumor-specific neoantigens (29).
Increased irAE count may also be associated with better tumor responses and more favorable outcomes in patients with melanoma and non-small-cell lung cancer (NSCLC) (13,30-33). A pooled analysis of the atezolizumab monotherapy arms from four NSCLC clinical trials demonstrated an improved OS in the 5.4% of patients with multiorgan irAEs compared to that in the cohort without irAEs (34). Less clear is whether the severity of irAEs, rather than their breadth across organ systems, could be indicative of a better antitumor response and more favorable patient outcomes. Unfortunately, relatively few studies have addressed this question (13,23,32,35-42), and the results have been inconsistent, due in part to the small number of patients with severe irAEs and the difficulty in accurately assessing whether an association with tumor response exists in patients in generally poor condition. In addition, aggressive immunosuppressive therapy, often required to treat severe irAEs, may compromise their beneficial effects on the survival (43).
The timing of the irAE onset is another variable of interest. In patients with metastatic or locally advanced urothelial cancer who developed irAEs and showed a respondse to anti-PD-(L)1 antibodies, more than half (57%) developed AEs of special interest before documentation of the tumor response (44). In a single-center prospective and multicenter retrospective cohort study, a stronger association between irAEs and the efficacy of PD-1 inhibitors was noted for early-onset irAEs (within two weeks of treatment initiation) than for later-onset irAEs (within six weeks) (45). However, an analysis of the prospective registry of patients with solid tumors or hematological cancers treated with ICIs showed that very severe (grade 4 or 5) irAEs, mainly affecting the respiratory and hematological systems and the heart, occurred significantly earlier than grade 1-3 irAEs (median 41 vs. 91 days) after the initiation of ICI therapy (46). Another study found that late-onset irAEs, defined as occurring more than three months after ICI initiation, were associated with a greater magnitude of tumor response and survival benefit than earlier-onset irAEs across cancer types (47). Additional large prospective studies are required to fully understand the impact of the breadth, magnitude, and timing of irAE development on the outcomes of ICI-treated patients.
Finally, it is important to note another time-dependent variable in evaluating the association between irAEs and treatment outcomes: by their very nature, irAEs take time to develop and are less likely to emerge and/or be documented in patients who have rapidly progressing or imminently fatal disease (48). Therefore, statistical analyses should be included that accommodate ‘immortal time bias' (i.e. the event cannot occur unless the patient survives during the observation period), such as landmark analyses, extended Cox model with time-varying covariates, and the cloning approach (49). Some, but not all, studies discussed here have considered this potential confounder when confirming the links between the occurrence of irAEs and longer patient survival during ICI treatment (22,23,27,31-33,37,39-42,47,48,50-74).

Associations between Patient Outcomes and Organ-specific irAEs During ICI Treatment
Despite many studies documenting associations between improved responses to ICIs and the overall development of irAEs, whether or not certain irAEs are more strongly related to patient outcome than others, or whether they tend to associate more strongly with a favorable or unfavorable prognosis is not yet clear. This section discusses the prognostic implications of organ-specific irAEs.

Pneumonitis
ICI-induced pneumonitis is one of the most common irAEs and often serious or lethal. In a meta-analysis of clinical trials of anti-PD-1 inhibitors for melanoma, NSCLC, and renal cell carcinoma, the incidence of pneumonitis was 2.8% for monotherapy and 6.6% for combination immunotherapy (75). However, higher incidences of ICI-induced pneumonitis have been reported in real-world settings than in highly selected clinical trial populations (76-80).
The impact of pneumonitis on ICI efficacy has been extensively explored (Table). Several retrospective studies have identified significant positive associations between pneumonitis and benefits to the PFS (33,56,81) and OS (33). However, no significant association between the PFS and pneumonitis was observed in a 2-month landmark analysis of patients in a 15-center retrospective study conducted in Japan (82). Notably, five retrospective studies that adjusted for immortal time bias in calculating HRs for the survival showed a deleterious association with pneumonitis (54,78,80,83,84). A time-dependent Cox regression analysis was used to directly compare the prognostic performance of pneumonitis and other irAEs in a large multicenter retrospective study of ICI-treated lung cancer patients and showed a two-fold higher probability of death for patients with pneumonitis than non-pneumonitis irAEs (HR, 2.01; 95% CI, 1.23-3.28) (63). A recent meta-analysis of 106 studies with 41,050 patients with advanced or recurrent lung cancer identified pulmonary irAEs as an unfavorable prognostic factor for the OS (HR, 1.31; 95% CI, 1.06-1.61), despite being associated with better ICI treatment responses (14); in contrast, beneficial associations were identified for skin irAEs (HR, 0.45; 95% CI, 0.38-0.53) and endocrine irAEs (HR, 0.51; 95% CI, 0.41-0.62), which highlights the uniqueness of pulmonary irAEs in the context of the relationship with patient outcomes. In that meta-analysis, the severity of pneumonitis was also associated with distinct outcomes, namely, a favorable prognosis for patients with (vs. without) mild (grade 1 or 2) pneumonitis (HR for PFS, 0.29; 95% CI, 0.15-0.57) and an unfavorable prognosis for patients with severe (grade ≥3) (vs. mild) pneumonitis (HR for OS, 2.48; 95% CI, 1.04-5.92). In addition, the prognostic association of ICI-induced pneumonitis with outcomes is likely to differ according to radiologic patterns, such as organizing pneumonia, nonspecific interstitial pneumonia, hypersensitivity pneumonia, and diffuse alveolar damage (63). Future studies should evaluate the impact of pneumonitis heterogeneity, including treatment setting, severity, radiologic pattern, and management, on the association with outcomes in patients treated with ICIs.

Cutaneous irAEs
Many studies have reported favorable associations between cutaneous irAEs and survival outcomes in patients receiving ICIs (Table), with the most representative irAE being vitiligo in melanoma patients (31,70,85). Using a time-dependent Cox model, Kelly et al. evaluated 1,783 patients with advanced cancers receiving avelumab in the JAVELIN Solid Tumor and Merkel 200 trials and identified favorable associations between the OS and rash irAEs and thyroid irAEs (72). Another study evaluated data from the ECOG-ACRIN E1609 trial on the efficacy and safety of adjuvant ipilimumab compared with those of interferon-α in patients with resected high-risk melanoma (42) and found that, among specific irAEs, mild (grade 1 or 2) rash was most significantly associated with an improved RFS and OS. Beneficial relationships between cutaneous irAEs and a prolonged survival have also been observed among patients with multiple other types of cancers (68,69,84) including NSCLC (25,33,39,45,50,53,55,58,64,65) and hepatocellular carcinoma (41).
Associations between individual cutaneous eruptions and the survival were assessed in a large retrospective population-level analysis of 7,008 patients with multiple types of cancers treated with anti-PD-(L)1 therapy (66). In this analysis, significant protection against death was observed for many types of cutaneous irAEs, including pruritus, drug eruption, xerosis, nonspecific rashes, psoriasis, and lichen planus/lichenoid dermatitis (66), which is interesting considering that a recent report described distinct immunologic endotypes among cutaneous irAE phenotypes (86). A meta-analysis of 23 studies with 22,749 cancer patients treated with ICIs showed that the development of cutaneous irAEs and their subtypes was significantly associated with an improved OS (87). Other meta-analyses have also reported beneficial relationships between cutaneous irAEs and tumor response (17) and survival outcomes (18,19).

Thyroiditis
Thyroid irAEs are among the most common side effects of ICI treatment, and the risk of thyroid dysfunction is greater after inhibition of PD-(L)1 than after inhibition of CTLA-4 (88). ICI-associated thyroiditis typically follows one of three courses (89): new-onset hypothyroidism or subclinical hypothyroidism; overt thyrotoxicosis or subclinical hyperthyroidism; or acute elevation of thyroid-stimulating hormone in patients previously diagnosed with hypothyroidism that requires a 50% increase in the levothyroxine dose. The most frequent pattern is transient hyperthyroidism, followed by hypothyroidism over four to six weeks (88). Growing evidence suggests a positive association between thyroid irAE occurrence and improved ICI efficacy (Table). A retrospective analysis of 6,596 patients treated with ICIs for multiple cancer types showed a 20% reduction in mortality risk in patients who developed thyroiditis after accounting for immortal time bias (61); this relationship was strongest for patients with lung cancer (HR, 0.56; 95% CI, 0.40-0.79) and weakest for those with breast, melanoma, and genitourinary tumors (61). Similar findings were reported in a large meta-analysis (47 studies), of which 21 studies and 12,158 patients were included in the meta-analysis of OS (90). A significant association between thyroid irAEs and a longer OS (HR, 0.52; 95% CI, 0.43-0.62) was detected, and the relationship was again strongest for lung cancer (NSCLC) patients (HR, 0.37; 95% CI, 0.24-0.57), although insufficient data were available to perform subgroup analyses on other cancer types (90). In another retrospective study, the survival benefit of nivolumab-induced thyroid irAEs was better for lung cancer patients than for melanoma patients (73); interestingly, this study also reported that OS was significantly longer for patients in the overt thyroid irAE group, but not for those in the subclinical thyroid irAE group, than for those in the non-thyroid irAE group (73). Similarly, a prospective cohort study of cancer patients receiving anti-PD-1 inhibitors reported that patients who developed overt thyroid toxicity had an OS benefit (HR, 0.18; 95% CI, 0.04-0.76) but not those with subclinical toxicity (HR, 1.54; 95% CI, 0.79-2.98) or pre-existing thyroid toxicity (HR, 0.92; 95% CI, 0.44-1.90) compared to patients without thyroid toxicity (74). In another retrospective study of 1,781 cancer patients, pre-existing thyroid abnormalities were associated with worse OS outcomes (HR, 1.62; 95% CI, 1.30-2.02), which contrasted with the benefit observed with ICI-induced thyroid dysfunction (HR, 0.62; 95% CI, 0.44-0.88) (91). The improved survival of patients treated with ICIs may be related to the presence of anti-thyroglobulin antibodies (92) and/or an increase in thyroid volume after ICIs (93).

Adrenal insufficiency and hypophysitis
Hypophysitis is a rare irAE in patients treated with PD-(L)1 inhibitors, but it is more common (incidence of 5-13%) in patients receiving anti-CTLA-4 antibodies than anti-PD-(L)1 antibodies (71,94,95). A systematic review of 48 trials (6,938 patients) reported a higher incidence of hypophysitis in patients receiving anti-CTLA-4 antibodies than in patients receiving anti-PD-1 antibodies [odds ratio (OR), 6.5; 95% CI, 3.0-14.3] (96), in part due to direct antibody binding to CTLA-4 expressed by the pituitary gland, resulting in increased complement-mediated inflammation (97). Only a few studies have explored the prognostic significance of adrenal insufficiency and hypophisitis in patients receiving ICIs (Table), likely because most studies have assessed the prognostic value of these irAEs together with other endocrine irAEs. One retrospective study found no significant relationship between hypophysitis and mortality risk (98), whereas two other retrospective studies demonstrated significant protection against mortality in patients who developed adrenal insufficiency (59,99).

Hepatitis
Hepatitis is one of the most frequent ICI-related AEs with an incidence of up to 16% (Table). Liver-related irAEs have histological characteristics that are distinct from those of autoimmune liver diseases, suggesting that their underlying immunological mechanisms differ (100). The results of studies evaluating the prognostic implications of ICI-induced hepatitis are inconsistent (Table). An analysis of 2,561 melanoma patients with grade 3 or 4 AEs registered in the prospective nationwide Dutch Melanoma Treatment Registry found no significant differences in the PFS or OS of patients who developed hepatitis (101). Similarly, a recent study reported neutral associations between ICI-related hepatitis and OS at the 6-month landmark time point in a large multi-institutional cohort and population-level database of cancer patients (84). However, a reduced risk of mortality (HR, 0.53; 95% CI, 0.30-0.93) was reported in a retrospective analysis of NSCLC patients with ICI-related hepatic toxicity using a time-dependent Cox regression analysis (39). This favorable prognostic association was supported by another retrospective study of patients treated with pembrolizumab (58). In contrast, an unfavorable association between severe liver irAEs (grade ≥3) and the OS was reported in a retrospective study of patients with NSCLC treated with PD-(L)1 inhibitors (102). In a meta-analysis of 48 clinical trials involving 7,936 patients treated with nivolumab or nivolumab plus ipilimumab, the objective response rate did not correlate with hepatic irAEs (17). Similarly, a meta-analysis of 30 studies with 4,971 cancer patients showed no significant association between hepatobiliary irAEs and either the PFS or OS (19).

Colitis/diarrhea
Colitis was also one of the most frequently observed irAEs in patients receiving anti-CTLA-4 antibodies with an OR of 8.7 (95% CI, 5.8-12.9) compared with anti-PD-1 antibodies (96). In the phase III KEYNOTE-598 trial comparing pembrolizumab plus ipilimumab with pembrolizumab monotherapy in patients with untreated metastatic NSCLC and a PD-L1 tumor proportion score ≥50%, the incidence of any grade colitis was 6.4% for combination therapy and 1.1% for monotherapy (7). Nevertheless, relatively few studies have addressed the prognostic significance of ICI-related colitis alone (Table), as most studies have categorized colitis or diarrhea as gastrointestinal irAEs and evaluated them in combination with other digestive tract irAEs. For example, a recent retrospective study of data obtained from a large cohort of cancer patients and a population-level database identified patients with distinct clusters of irAEs but found no significant prognostic association in the group with gastrointestinal irAEs (84). A similar lack of significant relationships between immune-related colitis and ICI efficacy has been reported in other studies on melanoma and NSCLC patients (31,38,103,104). However, there is also evidence of a favorable association between the patient survival and the development of ICI-related colitis, as reported in 2 retrospective studies that included 410 patients with lung cancer, melanoma, or urologic cancer (71) and 195 patients with NSCLC (33), as well as in a meta-analysis of 5 studies that showed a significant OS benefit with the development of gastrointestinal irAEs (HR, 0.68; 95% CI, 0.51-0.89) (19).

Renal irAEs
The overall incidence of acute kidney injury (AKI; defined as an increase in serum creatinine at least 1.5 times from the baseline within 12 months) in patients treated with ICIs has been reported to be 17% (105); however, the incidence of directly ICI-related renal irAEs is considered to be much lower (＜5%) (106,107). Among the spectrum of renal irAEs, acute interstitial nephritis (AIN) is more common than glomerular diseases (107). Baker et al. assessed mortality in ICI-treated patients after the development of either AKI or estimated AIN [eAIN; defined as those who had ＞90th percentile probability of AIN using a validated diagnostic model for biopsy-proven AIN (108)] using time-varying Cox hazards models (109). In that study, patients with eAIN tended to have more severe AKI than those with non-eAIN AKI; however, the mortality of patients with eAIN was similar to that of patients without AKI (HR, 1.41; 95% CI, 0.70-2.87), whereas patients with non-eAIN AKI had a 3-fold higher risk of death than those without AKI (HR, 3.11; 95% CI, 2.55-3.80), suggesting that AIN may be a marker of therapeutic response to ICIs. Indeed, in a retrospective cohort of patients with multiple types of cancer, patients who developed ICI-associated AKI [defined as proposed by Gupta et al. (110)] had a better survival than those who did not develop AKI (adjusted HR, 0.60; 95% CI, 0.38-0.97), whereas no such association was observed in those with non-ICI-associated AKI (adjusted HR, 1.06; 95% CI, 0.86-1.32) (111). Similarly, an Australian retrospective study of ICI-treated NSCLC patients found that immune-related AKI was associated with a better prognosis (112). In contrast, other retrospective studies have reported that patients with AKI (113) or renal irAEs (114) had a reduced survival compared to those without AKI and ICI-related renal AEs, respectively, and additional studies reported no significant association between ICI-induced AKI development and the survival (115,116) (Table).
A meta-analysis of 5 studies showed that patients who developed ICI-related AKI had a 51% increase in risk of all-cause mortality compared to those who did not develop AKI (HR, 1.51; 95% CI, 1.07-2.14) (117). In another meta-analysis of 48 clinical trials of 7,936 patients with advanced solid tumors who received nivolumab or nivolumab plus ipilimumab, the development of renal irAEs did not correlate with objective response rates with either treatment regimen (17). Collectively, these results suggest that a clear understanding of the prognostic associations of renal irAEs will likely require a more precise differentiation between ICI-associated or immune-related AKI/AIN from other causes of AKI.

Myositis and myocarditis
ICI-related myositis is a rare irAE, but it is the most frequent neuromuscular irAE (118), with an incidence of approximately 0.4% (119,120). Myositis can be fatal (121), particularly in the presence of myasthenia gravis or myocarditis (119,120), yet there is a dearth of knowledge about its association with ICI efficacy, which is likely due to the rarity of myositis as an irAE, as well as its high mortality rate (122) and permanent sequelae (118). In several retrospective studies, myositis was assessed as a collective group of musculoskeletal irAEs. A retrospective study of pembrolizumab-treated NSCLC patients identified musculoskeletal irAEs as a favorable factor for the PFS but not for the OS (58). Two additional studies found no significant association between musculoskeletal irAEs and ICI efficacy (84,104), and a retrospective study of patients with melanoma treated with nivolumab found that the development of myalgias had a neutral effect on the OS (31).
ICI-related myocarditis is also rare (0.04-1.14%) (123) but often develops with generalized myositis (122). This serious irAE is a major cause of ICI-associated death, with mortality rates in the range from 32.5% to 46% (124,125). Accordingly, one retrospective study showed that cardiotoxicity was significantly associated with a poor prognosis in ICI-treated cancer patients (126). Further studies are required to explore the long-term associations with ICI efficacy in patients who have recovered from myocarditis.

Cytokine release syndrome (CRS)
ICI-related CRS is a systemic inflammatory response characterized by massive release of cytokines. Although ICI-related CRS has been considered extremely rare (approximately 0.07%) (127), it has likely been overlooked because of diagnostic difficulties and potential misdiagnoses with other concurrent AEs, such as sepsis. However, the reporting of ICI-related CRS has increased over time with increasing awareness in the context of ICI treatment (127). In a recent large, single-center cohort study of 2,672 ICI-treated cancer patients, CRS was identified in 1.1% of patients within 30 days of ICI administration and had a mortality rate of 3.6% (128). Furthermore, the results of a multicenter, randomized phase III trial (JCOG2007) of NSCLC patients treated with chemotherapy in combination with nivolumab plus ipilimumab suggest that the incidence of CRS may be even higher and carry a substantial risk of mortality (129). In the single-center cohort study of 2,672 ICI-treated cancer patients described above, treatment responses following ICI-induced CRS were evaluable in 22 patients, showing complete response in 4.5%, partial response in 23%, stable disease in 4.5%, and progressive disease in 68% (128). The authors reported that the outcomes were consistent with previously reported responses among patients with solid tumors receiving ICIs (130). Several case reports also suggest an association between CRS or hemophagocytic lymphohistiocytosis, a variant of CRS, and improved efficacy of ICIs (131-133), warranting further study.

Future Challenges and Conclusions
The development of irAEs seems inevitable in most patients receiving ICIs; thus, a deeper understanding of the prognostic implications of each irAE would be beneficial in the management of cancer patients. In this review, we outlined our current knowledge of the potential associations between the development of specific irAEs and ICI treatment efficacy (Figure). However, it is important to note that several roadblocks exist to obtaining an accurate assessment of such associations: first, the difficulty in recognizing and characterizing irAEs (79), due in part to the lack of solid diagnostic criteria that enable discrimination between irAEs and other treatment-related AEs; second, the time-dependent nature of irAE development, which introduces immortal time bias into survival analyses; third, the relative rarity of some irAEs (e.g. myositis and CRS); fourth, the varying incidence and severity of irAEs according to cancer type and the specific ICI targets of therapy, including combination therapy; and finally, the potential influence of irAE treatments, such as corticosteroids (43,134) and immunosuppressants (43,135), on ICI efficacy. Further studies are warranted to explore the prognostic association of irAEs while considering these points, which will require large sample sizes and thoughtful study designs that include robust statistical methodology. Such analyses will also need to be updated as improvements in irAE management are introduced to minimize the detrimental influence of irAE treatment on ICI efficacy, which may include improved peak dose (134) and duration (136,137) of corticosteroid therapy and the use of more appropriate immunosuppressants (138,139). Data from such studies will provide further insights into the mechanisms underlying the development of individual irAEs in association with the mechanisms of antitumor immunity.

The authors state that they have no Conflict of Interest (COI).