Immune-Related Adverse Events Associated With Immune Checkpoint Inhibitors for Advanced Non-small Cell Lung Cancer: A Network Meta-Analysis of Randomized Clinical Trials

Objective: This network meta-analysis will provide a complete toxicity profile, toxicity profile, and safety ranking of immune checkpoint inhibitors (ICIs) for treatment of advanced non-small cell lung cancer (NSCLC). Methods: We found 12 phase II or III randomized clinical trials (RCTs) including 8,453 patients with NSCLC by searching Pubmed, Embase, and ClinicalTrials.gov. Risk ratios (RRs) and 95% confidence interval (CI) were used to compare the rate of immune-related adverse events (irAEs) for different ICIs-based treatments using pairwise and network meta-analysis with random effects. Results: For dermatologic irAEs, the corresponding ranking of incidences of the seven groups from high to low was: nivolumab + ipilimumab (97.4%), pembrolizumab (80.1%), nivolumab (67.1%), pembrolizumab + platinum (43.3%), atezolizumab + platinum (39.9%), durvalumab (17.5%), platinum-based chemotherapy (4.7%). For colitis, the corresponding ranking of incidences of the six groups from high to low was: atezolizumab + platinum (77.1%), nivolumab (67.3%), pembrolizumab (60.5%), durvalumab (45.2%), pembrolizumab + platinum (41.4%), platinum-based chemotherapy (8.5%). For endocrine irAEs, the corresponding ranking of incidences of the seven groups from high to low was: nivolumab + ipilimumab (79.1%), durvalumab (69.1%), pembrolizumab (61.9%), atezolizumab + platinum (60.4%),nivolumab (45.7%), pembrolizumab + platinum (33.5%), platinum-based chemotherapy (0.3%). For pneumonitis, the corresponding ranking of incidences of the seven groups from high to low was: pembrolizumab (99.3%), pembrolizumab + platinum (65.1%), durvalumab (62.2%), atezolizumab + platinum (56%), nivolumab (35.9%), platinum-based chemotherapy (18.1%),atezolizumab (13.3%). For hepatitis, the corresponding ranking of incidences of the six groups from high to low was: pembrolizumab (71.2%), pembrolizumab + platinum (64.3%), durvalumab (56.4%), atezolizumab + platinum (53.8%), nivolumab (44.5%), platinum-based chemotherapy (9.8%). Conlusion: In addition to platinum-based chemotherapy, durvalumab for dermatologic and liver irAEs, pembrolizumab for gastrointestinal irAEs, pembrolizumab + platinum for endocrine irAEs, and atezolizumab for pneumonitis may be associated with lower rates of irAEs than other immune-based regimens. Nivolumab + ipilimumab for dermatologic and endocrine irAEs, atezolizumab + platinum for colitis, and pembrolizumab for pneumonitis and hepatitis may be associated with higher rates of irAEs.


INTRODUCTION
According to the latest statistics, lung cancer is still the leading cause of cancer deaths, although the incidence of cancer has declined in recent years (Goldstraw et al., 2016;Siegel et al., 2020). Nearly 70% of patients with lung cancer are diagnosed with locally advanced or metastatic disease (Torre et al., 2016), and the 5-years survival rate is only 5% (Molina et al., 2008;Bironzo and Di Maio, 2018). However around 80-85% of lung cancer cases are classified as NSCLC (Torre et al., 2015). Platinum-based chemotherapy remains the standard first-line treatment for NSCLC without target gene mutations, with a response rate of only 15% (Sandler et al., 2006;Scagliotti et al., 2008).
In recent years ICIs, which constitute an inhibitory pathway detected in a variety of malignant tumors, have opened a new era of cancer treatment (Ettinger et al., 2017;Hanna et al., 2017). Tumor cells form immune escape through immune checkpoints, however ICIs are to prevent the immune escape caused by the tumor cells by combining with the immune checkpoint, thus enabling the immune cells to resume their killing effect on the tumor cells. Currently, programmed cell death receptor 1(PD-1) and cytotoxic T lymphocyte-associated antigen-4(CTLA4) are widely studied immune checkpoints. PD-1 is a type I transmembrane glycoprotein (He et al., 2015), which is an inhibitory receptor of T cells. It can bind and interact with the specific ligands programmed cell death ligand-1(PD-L1) and programmed cell death ligand-2 (PD-L2). Tumor cells can activate PD-1, thereby promoting it to bind to PD-L1 and PD-L2 on the surface of antigen presenting cells, inhibiting the proliferation of effector T cells and preventing T cells from recognizing dangerous molecules in a timely and effective manner, thus enabling tumor cells to evade the pursuit of immune cells. PD-1/PD-L1 inhibitors promote the activation and proliferation of T cells by inhibiting the expression of PD-1/PD-L1, thus stimulating the killing of tumor cells (Pedoeem et al., 2014). CTLA-4 is a leukocyte differentiation antigen and a transmembrane receptor on T cells. It shares the B7 molecular ligand with CD28, while CTLA-4 induces T cells to be inreactive after binding to B7 molecule, and participates in the negative regulation of immune response. However CTLA-4 inhibitors can inhibit the molecule CTLA-4, allowing T cells to proliferate and attack tumor cells. Data from a series of randomized clinical trials suggest that ICIs alone or in combination as first-line treatment for advanced NSCLC patients provides better clinical benefits and fewer side effects than conventional platinum-based chemotherapy (Corey et al., 2016;Reck et al., 2016;Carbone et al., 2017;Barlesi et al., 2018;Gandhi et al., 2018;Paz-Ares et al., 2018;Hellmann et al., 2019;Tony et al., 2019;West et al., 2019;Jotte et al., 2020;Naiyer et al., 2020;Roy et al., 2020). However, ICIs enhance self immune functions against cancer cells through a unique mechanism that blocks negative regulators expressed on immune or tumor cells, while yielding higher rates of irAEs than platinum-based chemotherapy (Remon et al., 2020). According to statistics, the occurrence of grade 3 or higher irAEs varied from 8 to 10% among patients with advanced NSCLC receiving ICIs (Remon et al., 2020). Common target organs of irAEs included dermatologic irAEs (pruritus and rash), endocrine irAEs (hypothyroidism and hyperthyroidism), colitis, pneumonitis and hepatitis (Wang et al., 2018). Any of these irAEs, if not properly treated and managed in clinical practice, irAEs will lead to treatment termination, failure, and may even be life-threatening. Therefore, it was very necessary for clinicians to better grasp the irAEs most likely to be caused by each treatment regime, and to prevent, detect and treat them as early as possible. However, which treatment regime was more likely to induce irAEs was controversial.
Here, to help clinicians improve early prediction, early detection and early treatment of irAEs, a network metaanalysis was conducted to compare the incidences of irAEs and rank the safety of ICI + chemotherapy, ICI alone, and dual ICIs combination.

Data Sources and Searches
This network meta-analysis (NMA) was based entirely on the preferred reporting items for systematic reviews and metaanalysis (PRISMA) (Liberati et al., 2009;Hutton et al., 2015) and PRISMA extended guidelines for an NMA. RCTs on ICIs versus platinum-based chemotherapy as the first-line treatment of advanced NSCLC from 2015 to 2020 were searched in PubMed, Embase. To search for data not explicitly given in the RCTs, the National Institutes of Health ongoing Trial Registry (Clinicaltrials.gov) was also searched. A combination of MeSH and free-text words was searched according to the PICOS principle.Search terms and their combinations used in the search strategy included (PD-1 OR PD-L1 OR CTLA-4 inhibitor OR nivolumab OR pembrolizumab OR durvalumab OR atezolizumab OR ipilimumab) AND (Non-Small-Cell Lung Carcinoma OR Non-Small Cell Lung Cancer) AND chemotherapy AND (randomized controlled trial). Two independent reviewers (Jingjing Gu and Weidong Zhang) conducted a preliminary screening of the searched topics and abstracts, and if they did not meet the criteria, further read the full text, and all references were evaluated as potentially relevant articles.

Inclusion and Exclusion Criteria
Inclusion criteria for RCTs of this NMA were as follows: 1) study type: only Phase II or III double blind RCTs for advanced NSCLC; 2) participants: included patients that were pathologically diagnosed as advanced NSCLC; 3) experimental group: patients were treated with immunotherapy alone or an immune-based combination as first-line treatment, control group: patients were treated with only platinum-based chemotherapy as first-line treatment; 4) outcome indicators: there was at least one irAE in the RCTs or searched on Clinicaltrials.gov.
Exclusion criteria: non-English articles, studies without valid data, reviews, meta-analyses, editorials, commentary letters, repeat studies.

Data Extraction
By reading the titles, abstracts and full texts, the two authors screened out the articles that met the predetermined inclusion criteria. 1) Trial information, including first author, study year, trial id; 2) Stage information, study endpoint, and sample size of treatment; 3) Patient characteristics at baseline included median age, sex, and the numbers of patients with dermatologic irAEs, colitis, endocrine irAEs, pneumonitis and hepatitis (grade 1-5).
FIGURE 1 | The Cochrane risk bias assessment tool was used to evaluate bias from seven key sources: 1. Random sequence generation; 2. Allocation concealment; 3. Blindness of subjects and researchers; 4. Blindness of outcome evaluation; 5. Incomplete data; 6. Selective reporting of results; 7. Other biases. Green represents low risk, yellow represents unclear risk, and red represents high risk. And the two authors checked the extracted data before data analysis.

Risk-Of-Bias Assessment
The Cochrane risk bias assessment tool (Higgins and Green, 2011) (Figure1)were used to evaluate the quality of the literature by two reviewers (Jingjing Gu and Weidong Zhang). The seven major sources of biases (random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting and other bias) were classified into three grades including "yellow represents unclear risk", "green represents low risk" and "red represents high risk" which were assessed using the Cochrane risk bias assessment tool.

Outcome Measures
The study endpoints were any grade of dermatologic irAEs including pruritus and rash, colitis, endocrine irAEs including hypothyroidism and hyperthyroidism, pneumonitis and hepatitis. Review criteria were based on the National Cancer Institute Common Terminology Criteria for Adverse Events (Basch et al., 2016). The probability of associated adverse events for each treatment regimen was assessed by the combined RRs and 95% CI.

Data Synthesis and Statistical Analysis
The loops to illustrate the network geometry were generated using Stata13.0. In this NMA, summary RRs and 95% confidence intervals were used as the results of the effect size of ICI-based drugs on the risk of irAEs in NSCLC patients. RR greater than 1 represented a low probability of irAEs in the control group.
Markov chain Monte Carlo (MCMC) methods were used to establish random-effects and consistency models calculating the RR and 95% confidence intervals within the Bayesian framework by using R (version 4.0.1) (CoreTeam 2019, Vienna, Austria) and JAGS (version 4.3.0) with the package "getmtc" (version 0.8.2) (Salanti et al., 2008;Sutton et al., 2008). Heterogeneity of the included studies was assessed by I 2 statistics. I 2 values below 25%, between 25 and 50%, and above 50% represented low, medium, and high heterogeneity, respectively (Mills et al., 2013). If the heterogeneity is low, the fixed effects model would be selected, otherwise the random effects model would be selected. In order to explore the interstudy heterogeneity of each outcome comparison, the values of different parameters of a log-normal distribution were fitted as a prior distribution (Higgins et al., 2003). To obtain the posterior distribution, 10,000 burn-ins and 50,000 iterations of 4 each chain and a thinning interval of 10 were generated for each outcome by using MCMC methods. We judged whether each MCMC chain reached a stable and good iteration during the calculation process through the Brookse-Gelmane-Rubin diagnostic plot with a cutoff value of 1, so as to determine whether the degree of convergence of the model was satisfactory. The surface under the cumulative ranking curve (SUCRA) metric was a tool to rank the probabilities of irAEs of each treatment and identify the best treatment. The SUCRA value was ranged from 0 to 1, the closer to 1 this value approached, the higher its incidence of an immunerelated adverse event was in this network meta-analysis (Salanti et al., 2011).A funnel plot was used to assess the publication bias and symmetrical distribution in funnel plot suggests no publication bias (Higgins et al., 2012). Sensitivity analysis was performed to assess the stability of the results which were considered stable if there was significant consistency between direct and indirect results (Dias et al., 2010). We conducted a inconsistency test (that is, the comparison of the differences between direct and indirect comparisons) using the nodal analysis and p > 0.05 indicated that there is no inconsistency (Sterne and Egger, 2001;Furukawa et al., 2016).

Convergence, Heterogeneity and Publication Bias
As shown in Supplementary figure S2, all comparisons of different irAEs all suggested that the contraction factor in Brookse-Gelmane-Rubin diagnostic plots were equal to the predefined cut-off value 1, suggesting that the study model had good convergence. In this network meta-analysis the comparisons showed low, medium or high heterogeneity (Supplementary figure S3), then the random effects model was selected. The funnel plots were all symmetrically distributed, suggesting no publication bias ( Figure 6).

DISCUSSION
In recent years, a number of RCTs had successively evaluated ICI + chemotherapy, ICI alone, dual ICIs combination versus chemotherapy in terms of efficacy and safety in the first-line treatment of advanced NSCLC (Corey et al., 2016;Reck et al., 2016;Carbone et al., 2017;Barlesi et al., 2018;Gandhi et al., 2018;Paz-Ares et al., 2018;Hellmann et al., 2019;Tony et al., 2019; Frontiers in Pharmacology | www.frontiersin.org October 2021 | Volume 12 | Article 686876 9 West et al., 2019;Jotte et al., 2020;Naiyer et al., 2020;Roy et al., 2020). These RCTs suggested that the efficacy and treatmentrelated side effects of ICI + chemotherapy, ICI alone or dual ICIs were better than chemotherapy alone. Because of this, ICIs including nivolumab (Kazandjian et al., 2016), pembrolizumab (Pai-Scherf et al., 2017) and atezolizumab (Weinstock et al., 2017) were approved for NSCLC by FDA and had became an important part of treatment options in advanced NSCLC (Antonia et al., 2017). However with the widespread use of ICIs, the irAEs showed a tendency to be more prevalent in ICIs than in Frontiers in Pharmacology | www.frontiersin.org October 2021 | Volume 12 | Article 686876 10 chemotherapy. IrAEs represented an entirely new toxicity spectrum of ICIs, and it could affect any tissue or organ, if mishandled, could reduce the survival benefit of antitumor efficacy, which might be related to the mechanism by which ICI alters the balance of immune cells in the body, inducing damage to some organ systems. The incidence and severity of irAEs were usually related to the drug type, tumor type, irAEs history and other immune-related medical history (Chen et al., 2015), and different ICIs had different immune mechanisms, and even belonging to the same mechanism had different tolerability to different irAEs. However which treatment regime could be better tolerated was worth studying.
Here, we indirectly compared the probability of irAEs caused by ICI + platinum, ICI alone, dual ICIs combination for advanced NSCLC through NMA. Data on irAEs from published studies and Clinicaltrials.gov from 2015 to 2020 were collected. Finally, 12 head-to-head phase II and III RCTs (Corey et al., 2016;Reck et al., 2016;Carbone et al., 2017;Barlesi et al., 2018;Gandhi et al., 2018;Paz-Ares et al., 2018;Hellmann et al., 2019;Tony et al., 2019;West et al., 2019;Jotte et al., 2020;Naiyer et al., 2020;Roy et al., 2020)including 8,453 patients with advanced NSCLC were included in this network meta-analysis. Notably, this NMA concluded that for dermatologic irAEs, colitis, endocrine irAEs, pneumonitis and hepatitis, ICI-based therapy showed a higher incidence of irAEs than platinum-based chemotherapy. Qiang Su (Su et al., 2019) concluded that compared with chemotherapy, PD-1/PD-L1 inhibitors showed significant increase in grade 1-5 and grade 3-5 pneumonitis, which was consistent with the results of ours. And for dermatologic irAEs, colitis, endocrine irAEs, pneumonitis and hepatitis, pembrolizumab + platinum showed a higher incidence of irAEs than pembrolizumab, which was consistent with Slater et al.'s (Slater et al., 2002) conclusion that chemotherapy may produce immunosuppression. Interestingly, only dermatologic and endocrine irAEs involved nivolumab + ipilimumab and we concluded that nivolumab + ipilimumab showed a higher incidence of dermatologic and endocrine irAEs than ICI + platinum and ICI alone and the explanation for this might be that the combination of the dual ICIs increased the imbalance of immune cells and led to more irAEs. And we found something new that except for chemotherapy, durvalumab for dermatologic irAEs, pembrolizumab + platinum for colitis, pembrolizumab + platinum for endocrine irAEs, atezolizumab for pneumonitis and nivolumab for hepatitis had the least incidence of irAEs, which were different from the conclusion of Cheng Xu et al. (Xu et al., 2018) and Xinru Chen et al. (Chen et al., 2020). Cheng Xu et al. (Xu et al., 2018) compared the occurrences of irAEs with different immunotherapy regimens for all cancers in 2018, and they compared the efficacy of ICI monotherapy, ICI monotherapy plus chemotherapy, and chemotherapy in lung cancer and concluded that atezolizumab had the best overall safety, while nivolumab had the best overall safety in the treatment of lung cancer with a combined approach. Xinru Chen et al. (Chen et al., 2020) investigated different immunotherapy regimens for immune-related pneumonia (IRP), and they concluded that ICIs increased the risk of IRP compared to chemotherapy and that ICI + chemotherapy was associated with a lower risk of IRP than dual ICIs combination and ICI monotherapy. However the reason why my results were different from theirs might be: firstly, compared to previous studies, we only included patients with advanced NSCLC and excluded other cancer types in this NMA, which was relatively specific, and all data were up to date. Secondly, the studies of us that only platinum-based chemotherapy was included, and those involving docetaxel were excluded. Thirdly, instead of integrating different drugs with similar mechanisms into the same arm, we analyzed each of the treatment regimens separately.
The current analysis has several strengths. Firstly, to our knowledge, this NMA analyzed the probability of irAEs caused by different treatment regimens for advanced NSCLC more specifically. Secondly, some conclusions this study drew were innovative compared to previous studies, which will have novel significance for guiding clinical treatment and subsequent studies. Thirdly, all of the articles included in this NMA were RCTs and the Cochrane risk bias assessment tool were used to examine the quality of included studies and to ensure their inherent authenticity by assessing the potential risk of bias in various aspects of RCT design, implementation, and outcome evaluation. Fourthly, in this NMA, in order to detect the heterogeneity of the included literature and data, we used heterogeneity test, and in order to detect publication bias, we used funnel plots, and the results showed that no significant publication bias was found. Fifthly, we conducted sensitivity analysis on all studies to investigate inconsistencies between direct and indirect comparisons by inconsistency test, and the results all suggested that p＞0.05, indicating that our research results are stable. All of these suggested that our study is quite reliable.
Nonetheless, there were a few unavoidable limitations to our article. Firstly, in all of the studies of irAEs included, almost all comparisons showed low heterogeneity except for some comparisons. Secondly, although we found that the rating systems and terminology used in the reports were consistent and compatible, the diagnosis of each irAE was based solely on the experience of each clinician, rather than on a centralized review, which can lead to bias in irAE evaluation. Thirdly, some irAEs were delayed diseases, and clinicians could not have observed the symptoms of patients within the clinical timeframe, leading to the loss of clinical data, which would bring some potential heterogeneity to this study. Fourthly, the median follow-up time for each RCT was different, which would increase the frequency of irAEs associated with immunotherapy and increase the confounding factors for these events. Fifthly, although we included the latest RCT data that could be retrieved, some studies had a small sample size, which might be the reason why some comparisons showed moderate or high heterogeneity.
Despite there are some limitations in our study, we still have some prospects for the prediction and treatment of irAEs. According to the reports , most irAEs have been reported to be mild, such as cutaneous and endocrine, but death from moderate, severe or life-threatening irAEs have been reported in 1-2% of patients, such as pneumonia and colitis and T. W. Chen et al. (Chen et al., 2015) suggest that seventeen percent of studies over stated the safety of the experimental Frontiers in Pharmacology | www.frontiersin.org October 2021 | Volume 12 | Article 686876 regimen, and although glucocorticoids and steroids (Weber et al., 2009) can be used for irAES, some patients still die because of late diagnosis, which emphasize the importance of determining the predictors of irAEs and early management and prevention. Studies have suggest that after treatment of ICI peripheral blood eosinophil is associated with irAEs (Schindler, 2014). Circulating IL-17 (Callahan, 2011), expression of CD177 and CEACAM1 (Shahabi et al., 2013), and neutrophil infiltration in the lamina propria of the colon (Berman et al., 2010) is suggested to be associated with gastrointestinal toxicity, however the occurrence mechanism of other irAEs is not clear, and at present, the evaluation system of irAEs is not perfect. In the future, more effective predictors of irAEs will be needed and with the extensive application of ICIs treatment and the development of related trials, people will have a fuller understanding of the mechanism of irAEs, and the management measures will be more standardized.

CONCLUSION
This systematic review and NMA suggests that, in addition to platinum-based chemotherapy, durvalumab for dermatologic and liver irAEs, pembrolizumab for gastrointestinal irAEs, pembrolizumab + platinum for endocrine irAEs, and atezolizumab for pneumonitis may be associated with lower rates of irAEs than other immune-based regimens. Nivolumab + ipilimumab for dermatologic and endocrine irAEs, atezolizumab + platinum for colitis, and pembrolizumab for pneumonitis and hepatitis may be associated with higher rates of irAEs. These conclusions will be helpful for clinical management, early prediction, early detection and early treatment. However a large number of multicenter clinical trials and studies are still needed to balance the therapeutic efficacy and the occurrence of irAEs of ICIs to maximize patient benefit.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author.