Association between immune‐related adverse events and prognosis in patients treated with immune checkpoint inhibitors in melanoma: A surrogacy analysis

Immune checkpoint inhibitors (ICI) represent a breakthrough in oncology in terms of prognosis and safety. They now constitute a cornerstone in the management of metastatic melanoma. However, a new kind of adverse event called immune‐related adverse events (irAE) has emerged. These irAE could be conceptually considered as an indicator of the antitumoral immune response, but the association between irAE and prognosis is still a matter of debate.


| INTRODUCTION
Melanoma is the most aggressive skin cancer and a real health concern because of its growing incidence in the last decades.Before 2011, the therapeutic panel was restricted to dacarbazine, an alkylating agent with a dismal outcome.Immune checkpoint inhibitors (ICI) have constituted a breakthrough in oncology because of their long-term effectiveness and their fewer side effects compared with chemotherapy [1,2].The ICI development is the direct application of the concept of cancer immuno-editing described by Schreiber et al. [3,4].Programmed cell death 1 (PD1), programmed cell death-ligand 1 (PDL1), and cytotoxic T-lymphocyte associated protein 4 (CTLA4), which are the main actors of the lymphocyte negative regulation, are targeted by immunotherapies with the aim to restore the antitumoral immune response.Metastatic melanoma is the first neoplasia that has benefited from this kind of treatment [5], and nowadays, ICI have become the cornerstone of the therapeutic strategy in melanoma both as adjuvant and metastatic setting [6].
As ICI activate the immune system, they are responsible for new side effects called immune-related adverse events (irAE) [7].These can affect all the organs; notably, skin, gastrointestinal, endocrine, rheumatologic, and hepatic irAE have been reported [8].Most irAE are reversible if detected early and adequately treated, but they still can be severe and sometimes even deadly.Thus, a proper management of irAE is essential and should involve multidisciplinary teams [9][10][11][12].Moreover, understanding the underlying mechanisms of irAE can help to improve their treatment and care management [13].Different hypotheses regarding the development of irAE have been proposed, such as self-tolerance breaking, cross antigen reactivity, cytokine production, off-target effects, and interactions with the microbiome [14].
Conceptually, these irAE could be an indicator of the antitumoral immune response and have been reported as associated with a better prognosis.Notably, antigen-specific adverse events such as vitiligo have been reported as potentially more strongly associated with the prognosis than other irAE in observational studies on melanoma [15][16][17].This association could result from a skin lymphocyte activation against melanoma-associated antigens expressed also by healthy melanocytes.The initial mechanistic hypothesis leading to this study was the absence of relationship between all irAE and prognosis but a strong relationship between antigen-specific adverse events and prognosis because it is an indicator of a specific anti-tumor response.In the present study investigating melanoma, the antigen-specific adverse event was cutaneous irAE, specifically vitiligo.For example, we could apply the same concept to lung cancer, the antigen-specific adverse event would then be pulmonary irAE.However, this point remains debated because of numerous confounding factors [18].In addition, considering adverse events as a beneficial marker is counterintuitive from a clinical point of view.
Thus, using the results of phase 3 trials on ICI use in melanoma, the present study aimed to estimate the relationship between survival parameters (overall survival [OS] and progression-free survival [PFS]) and the number of irAE, including dermatological irAE.
Additional articles were searched for in the reference lists of relevant articles identified through the electronic search.The method used was consistent with the recommendations of the Cochrane Handbook and Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) [19].

| Study selection
The first level of selection was performed by a single reviewer (RE) using titles and abstracts.The second level was performed by two independent reviewers (RE and MR) reading the full text of the selected publications.The inclusion criteria for studies were prospective randomized trials, double-blind, or open-label design, assessing ICI (anti-PDL1, anti-PD1, and anti-CTLA4 therapies) in the context of melanoma irrespective of stage and treatment setting.Trials including combinations of ICI were also included.Only the full publications that met all the inclusion criteria underwent data extraction.If the description of irAE was missing, we contacted the authors, when possible, to collect missing data.
Exclusion criteria were non-randomized controlled trials, and randomized trials that did not report results allowing data extraction.

| Data extraction and quality assessment
Two reviewers (RE, MR) independently extracted data from all included articles.Collected data included study design, patient characteristics (PDL1 tumoral expression, B-RafV600E status, age) and OS, PFS, drug class (ipilimumab, nivolumab, pembrolizumab, atezolizumab, tremelimumab), length of follow-up, and the number and type of irAE.The severity of irAE was evaluated in accordance with the Common Terminology Criteria for Adverse Events scale (CTCAE); severe irAE were defined as scoring 3 or more on the CTCAE scale.
In case of disagreement between the two reviewers, consensus was sought through discussion.The quality of the studies was assessed using the Cochrane riskof-bias tool for randomized trials (RoB 2) in accordance with the Cochrane guidelines.The RoB 2 tool (August 2019 version) included five key domains.

| Statistical analysis
The standard method of surrogacy evaluation proposed by Buyse et al. was used [20].A weighted regression was performed to estimate the relation between the treatment effect estimate (i.e., the logarithm of relative risk or hazard ratio of OS) and the logarithm of the absolute number of irAE to determine whether it might influence the size of intervention effect.The percentile bootstrapping method (resampling 1000 times) was calculated to compute the 95% confidence intervals (95% CI) for R 2  trial with high accuracy.The thresholds of lower bounds of confidence intervals set to identify good surrogates were >0.8 for correlation coefficients (ρ) or >0.65 for coefficients of determination (R 2 trial ).Results were presented graphically, including the relative risk, and corresponding 95% CI.Sensitivity analyses were conducted including non-weighted regression, PFS as dependent variable, number of patients with at least one irAE as dependent variable, and absolute number of dermatological irAE as dependent variable.
The pooled incidence of irAE and their 95% CI were estimated using generalized random-effects linear mixed model with log transformation.Heterogeneity between study-specific estimates was assessed using an inconsistency index (I 2 ).The effect of ICI on allcause mortality was estimated using pooled hazard ratio and a random-effect model.The risk of publication bias was evaluated according to the visual inspection of the funnel plot and the Egger's regression test.A pvalue below 0.05 was considered statistically significant.All analyses were performed using R v3.6.3 (R foundation for Statistical Computing, Vienna, Austria) with the packages metafor (version 1.9-5), metaboot (version 1.3-24), and stats (version 3.6.2).The research protocol was defined before the start of the study but not published on a platform.
Furthermore, Eastern cooperative oncology group (ECOG) and lactate dehydrogenase (LDH) were not reported in the adjuvant trial because these data have no value outside of a metastatic context.All these data are consistent with melanoma epidemiology [34].
The visual inspection of the funnel plot of prevalence of irAE was not in favor of publication bias, whereas the Egger's test was significant (p < 0.0001; Figure 2).One trial did not provide the confidence interval of the hazard ratio that precluded the calculation of standard error [23].The quality of studies was consistent with a low risk of bias (see Table S1 in supplementary material).

| Occurrence of irAE
A total of 5503 irAE was reported among the 4551 patients in the ICI treatment arms.The risk of all-cause death was reduced by 0.73 [95%CI 0.66-0.80] in patients receiving ICI compared with controls using random-effects model, with substantial heterogeneity (I 2 = 34%) (Figure 3).Fatal irAE occurred in 38 patients (0.83%; Table S2) but 28 deaths (73.7% of ICI related deaths) were attributable to anti-CTLA4 therapy alone.The incidence of irAE and severe irAE according to pharmacological class was lower for anti-CTLA4 therapy (10.1 and 3.7 per 100 patients.year, respectively) and higher for the association anti-CTLA4 and anti-PD1 (65.7 and 18.3 per 100 patients.year, respectively; Table 3 and Figures S1-7).The incidence of endocrine irAE was lower for anti-CTLA4 therapy (1.3 per 100 patients.year; Table 3), than anti-PD1 therapy (6.0 per 100 patients.year; Table 3).While hypophysitis were more frequent with anti-CTLA4 therapy (62.5% of endocrine irAE), dysthyroidism were more frequent with anti-PD1 (or PDL1) therapy (96.5% of endocrine irAE) (Table S2).F I G U R E 2 Funnel plot of irAE risk (using hazard ratio of overall survival) for the search of publication bias.The visual inspective is not in favor of publication bias.One trial did not provide data to estimate standard error.Abbreviation: irAE, immune-related adverse events.
F I G U R E 3 Forest plot of treatment effects on all-cause mortality of included studies.The confidence interval of the hazard ratio (HR; 0.88) for the trial of Ribas et al was not provided by the publication.The p-value was nonsignificant (p = 0.12) for this trial.Abbreviations: HR, hazard ratio; irAE, immune-related adverse events.
T A B L E 3 Incidence rate of immune-related adverse events (irAE) for 100 patients.Year according to pharmacological class.Abbreviations: CI, confidence interval; irAE, immune-related adverse events; k, number of studies.

| Association between irAE occurrence and survival parameters
A total of 14 trials were included in the regression analysis.No relationship was found between the number of irAE and OS considering all events (beta coefficient 0.078; R 2 3%; p = 0.52; Figure 4a and Table 4).The result was similar using unweighted regression.
Considering the number of patients with at least one irAE (data available in 10 trials), no relationship was found between irAE occurrence and OS (beta coefficient 0.051; R 2 2%; p = 0.66; Table 4; Figure 4b).No relationship was found between the absolute number of irAE and PFS (beta coefficient 0.040; R 2 0%; p = 0.83; Table 4; Figure 4c).This survival parameter is available only for 12 trials.In addition, absolute number of cutaneous irAE was not associated with OS (beta coefficient 0.080; R 2 6%; p = 0.33; Table 4; Figure 5).Because of the lack of reported data, the association between vitiligo risk and OS was not estimated.

| DISCUSSION
In the present study-level analysis, there was no significant relationship between the number of irAE and the survival parameters, including OS and PFS.Thus, an elevated risk of irAE was not associated with an increase of ICI efficacy.
Even though the patients were included in clinical trials, the final population included in this present analysis was adequate to perform a surrogacy analysis.First, the included patients in the present study were representative of melanoma epidemiology [34].This first point is probably explained by the large-scale clinical trials with a robust methodology.Second, most patients were naïve of treatment limiting interactions with previous drugs.Third, there were only few missing data, specifically about B-RafV600E mutation and PDL1 status, but these variables were not used in the analyses and have only a descriptive value.Another point of note in the present study is that the incidence of irAE was higher with combination of ICI than monotherapy, whatever the subcategory.The increased toxicity of this combination has been already well documented [35], and the present study underlines that this combination of ICI are not without risk even if it is more effective.It is important to consider the risk-benefit ratio of this combination before initiation, particularly in an adjuvant context.Considering the toxicity by pharmacological class of ICI, herein, the incidence of all categories of irAE were less frequent with anti-CTLA4 than anti-PD1 therapy.Nevertheless, these class-organ categories were difficult to compare between these two classes of drug because the spectrum of symptoms is different [36].For example, the endocrine irAE in the present study due to anti-CTLA4 were mostly hypophysitis while those due to anti-PD1 therapy were more frequently dysthyroidism.This was already reported in a real-world study showing that the most frequent endocrine irAE with anti-CTLA4 antibodies was hypophysitis, whereas anti-PD1 (or PDL1) antibodies induced thyroiditis more frequently [37].However, although these irAE have different clinical consequences for the patient each contributed to endocrine irAE in the present study Regression of the logarithm of treatment effect on immune-related adverse events (irAE) according to the logarithm of overall survival (OS) using absolute number of irAE (a), number of patients with at least one irAE (b) and according to the logarithm of progression free survival (PFS) using absolute number of irAE (c).The blue area represents the 95% confidence interval, whereas the red dotted lines correspond to the bounds of the 95% prediction interval.
further complicating the comparison between anti-CTLA4 and anti-PD1 therapy.
The present study found no significant relationship between the number of irAE, including dermatological irAE, and the survival parameters (OS and PFS).An association between these two parameters has been reported in cohort studies including melanoma and other patients with cancer [38][39][40][41][42][43] but remains controversial.This discrepancy might be due to a dilution of the association in the present study including all class of ICI and not only anti-PD1 therapy.irAE seem to be more associated with anti-PD1 and PDL1 response than to anti-CTLA4 antibody [44].There is a survivor bias because of time exposure of ICI; long-term survivors are more exposed to ICI and are prone to experience irAE.In addition, when a patient develops an irAE, the time to irAE occurrence may influence OS; for example, in a series including 436 consecutive patients treated with anti-PDL1 antibodies, OS and PFS were not significantly reduced by the occurrence of late-irAE (i.e., >12 months) compared with early-irAE (≤12 months) [45].These results were also confirmed in a recent study with a better survival in late-onset irAE in patients treated for non-small cell lung cancer [46].In the present study, the mean follow-up was limited to 3 years, with probably an underestimation of late-onset irAE explaining, at least partially, the absence of association.Concerning the dermatological irAE, considering our initial hypothesis, unexpectedly no relationship was found.Nevertheless, because of missing data from the included trials, we could not analyze the association between vitiligo occurrence and OS in the present analysis.We therefore used an alternate variable, cutaneous irAE, including a large panel of symptoms from pruritus to severe toxic skin adverse events.However, no association was found between the number of cutaneous irAE and OS, which could be explained by the heterogeneity of this variable.
Because irAE were new adverse events, their screening and management evolved over time [14].The present study includes all phase 3 trials concerning melanoma over a 10-year period.Some irAE were described after the first included study such as the interstitial pneumonia related to anti-PD1 therapy that subsequently appeared [47].In addition, the effects of corticosteroid exposure for the treatment of irAE, difference in monitoring, and management (including ICI interruption, less and less frequent) may have changed the overall prognosis of the patient complicating the management of the cancer [48].For example, when irAE are first described, treatment was often corticosteroids and even sometimes ICI interruption worsening the course of cancer.Furthermore, the imputability of ICI in the occurrence of an event in an oncologic patient F I G U R E 5 Regression of the logarithm of treatment effect on absolute number of dermatological immune-related adverse events (irAE) according to the logarithm of overall survival (OS).The blue area represents the 95% confidence interval, whereas the red dotted lines correspond to the bounds of the 95% prediction interval.
T A B L E 4 Sensitivity analysis of the association between the treatment effects on the risk of irAE and the risk of mortality.

Outcomes
Number  Abbreviation: irAE, immune-related adverse events.The correlation of the effects of treatment on the relative risk of clinical worsening and the relative risk of mortality was calculated by using a weighted linear regression model.The strength of association was assessed by using the coefficient of determination R 2 trial , with the 95% confidence interval (CI) calculated with the percentile bootstrapping method (resampling 1000 times).a The analysis was conducted using absolute number of irAE.
is difficult to establish.For example, the occurrence of an hepatic dysfunction during the treatment could be an hepatic irAE but it could be due to a local progression of the disease, which is a real difficulty for event adjudication.This heterogeneity may influence the estimate of irAE incidence rate.However, it is unlikely that this limitation have significantly biased our surrogacy analyses.
In the present study, the death rate attributable to ICI was 0.83%, which is consistent with previous studies.A meta-analysis including 112 trials has reported toxicityrelated fatality rates of 0.36% (anti-PD1 antibodies), 0.38% (anti-PDL1 antibodies), 1.08% (anti-CTLA4 antibody), and 1.23% (anti-PD1 or PDL1 plus anti-CTLA4 antibodies) [49].The impact of irAE on mortality varied widely according to organ involvement.Typically, an endocrine or cutaneous irAE has a different clinical weight on organ damage and sequela compared with a respiratory irAE [50].For example, the ICI-induced rheumatic events were associated with a fatality rate ranging from 24% for myositis to 0-7% for other rheumatic and musculoskeletal events [51].Therefore, the heterogeneity in rate and prognosis of irAE according to ICI exposure may explain the absence of association between irAE occurrence and OS.
The main limitation of the present study lies in its low statistical power related to the number of included studies.Indeed, the power of the regression analysis is estimated to be 61% for detecting surrogacy considering the number of included studies, the type 1 error, and the threshold of 0.65 suggested by Buyse et al. [20].Furthermore, because we did not have access to individual patient data for the included trials, the study was limited by a trial-level analysis using aggregate data.A patient-level analysis is needed to confirm the absence of association between prognosis and irAE occurrence.Finally, the analysis of funnel plot and Egger's test were limited by lack of power and high heterogeneity of included studies.
In conclusion, the present trial-level analysis was not in favor of an association between irAE occurrence and survival parameters in the context of melanoma.In addition, the study confirms severe irAE were more frequent with combination of ICI than ICI monotherapy, and emphasizes the importance of considering irAE primarily as safety defects and not a surrogacy marker.

T A B L E 2 4 )
Abbreviations: ECOG, Eastern cooperative oncology group; IQR, interquartile range; LDH, lactate dehydrogenase; M0, localized melanoma; M1, metastatic melanoma.a Among 4645 patients because of missing data for five trials.b Among 6695 patients, data not available in the adjuvant setting trial.

2
of the literature search and study inclusion.Summary of the trials included in the analysis.