A Network Comparison on Safety Profiling of Immune Checkpoint Inhibitors in Advanced Lung Cancer

Background Immune checkpoint inhibitors (ICIs) have become one of the standard treatment options for advanced lung cancer. However, adverse events (AEs), particularly immune–related AEs (irAEs), caused by these drugs have aroused public attention. The current network meta-analysis (NMA) aimed to compare the risk of AEs across different ICI–based regimens in patients with advanced lung cancer. Methods We systematically searched the PubMed, EMBASE, and Cochrane Library databases (from inception to 19 April 2021) for relevant randomized controlled trials (RCTs) that compared two or more treatments, with at least one ICI administered to patients with advanced lung cancer. The primary outcomes were treatment–related AEs and irAEs, including grade 1–5 and grade 3–5. The secondary outcomes were grade 1–5 and grade 3–5 irAEs in specific organs. Both pairwise and network meta-analyses were conducted for chemotherapy, ICI monotherapy, ICI monotherapy + chemotherapy, dual ICIs therapy, and dual ICIs + chemotherapy for all safety outcomes. Node–splitting analyses were performed to test inconsistencies in network. Sensitivity analyses were adopted by restricting phase III RCTs and studies that enrolled patients with non–small cell lung cancer. Results Overall, 38 RCTs involving 22,178 patients with advanced lung cancer were enrolled. Both pooled incidence and NMA indicated that treatments containing chemotherapy increased the risk of treatment–related AEs when compared with ICI-based regimens without chemotherapy. As for grade 1–5 irAEs, dual ICIs + chemotherapy was associated with the highest risk of irAEs (probability in ranking first: 50.5%), followed by dual-ICI therapy (probability in ranking second: 47.2%), ICI monotherapy (probability in ranking third: 80.0%), ICI monotherapy + chemotherapy (probability in ranking fourth: 98.0%), and finally chemotherapy (probability in ranking fifth: 100.0%). In grade 3–5 irAEs, subtle differences were observed; when ranked from least safe to safest, the trend was dual ICIs therapy (60.4%), dual ICIs + chemotherapy (42.5%), ICI monotherapy (76.3%), ICI monotherapy + chemotherapy (95.0%), and chemotherapy (100.0%). Furthermore, detailed comparisons between ICI–based options provided irAE profiles based on specific organ/system and severity. Conclusions In consideration of overall immune–related safety profiles, ICI monotherapy + chemotherapy might be a better choice among ICI–based treatments for advanced lung cancer. The safety profiles of ICI–based treatments are various by specific irAEs and their severity. Systematic Review Registration https://www.crd.york.ac.uk/prospero, identifier CRD42021268650


INTRODUCTION
Lung cancer remains the leading cause of global cancer mortality, with approximately 1.8 million deaths annually (18% of the total cancer deaths) (1). Over the past decades, platinum-based chemotherapy has become the cornerstone for managing advanced lung cancer; however, its use is of concern due to inevitable resistance and intolerable adverse events (AEs) in these fragile patients (2,3). Recently, immunotherapy has revolutionised treatment approaches for advanced lung cancer by making longer survival times a reality (4). Unlike traditional therapy (chemotherapy and targeted therapy), immune checkpoint Inhibitor (ICI) therapies use monoclonal antibodies to inhibit the expression of proteins [cytotoxic T lymphocyte associated antigen (CTLA4), programmed death-1 receptor (PD-1), and its ligand (PD-L1)], thereby boosting T-cell activation against cancer (5). To date, a series of ICIs [pembrolizumab, nivolumab, ipilimumab, atezolizumab, and durvalumab registered by the U.S. Food and Drug Administration (FDA); camrelizumab, sintilimab, and tislelizumab approved by the National Medical Products Administration (NMPA)] have been successfully introduced for use in patients with advanced lung cancer. Recently, ICI monotherapy with or without chemotherapy, dual ICIs combination, even dual ICIs combined with chemotherapy have been clinically applied as standard first-line treatment options for advanced lung cancer.
With a dramatic increase in the availability of ICI drugs and their superior efficacy, a substantial proportion of patients with lung cancer are administered these agents. Nonetheless, concerns regarding unique treatment-specific toxicities owing to their pharmacological mechanisms, namely immune-related AEs (irAEs), associated with ICI regimens are growing (6). IrAEs are unintended effects following the activation of the immune system by ICI-mediation and can occur in any organ or system, including the gastrointestinal tract, lungs, endocrine, skin, heart, renal, liver, and muscles (7). In published randomised controlled trials (RCTs), patients administered ICIs experienced fewer AEs than those undergoing chemotherapy, while the incidence of any irAEs seemed to be distinctly higher in the ICI group (8). Without timely identification and proper management, irAEs can become severe complications, resulting in treatment discontinuation or failure, and even death (9,10).
Previous meta-analyses have examined the risks of irAEs associated with ICI therapy; however, most of them mainly involved patients with all types of cancers (11)(12)(13). In addition, these studies did not explicitly examine the risk of individual irAEs across different ICI regimens, which may vary according to cancer type. Recently, one network meta-analysis (NMA) (14) and six traditional meta-analyses (14)(15)(16)(17)(18)(19) addressed this issue in patients with lung cancer. However, they focused on one or two specific irAEs; therefore, the entire toxicity spectrum in these patients is yet to be described. Since, to the best our knowledge, head-to-head comparisons among ICI regimens are lacking in current literature, an indirect analysis could be performed to obtain comparative results and rank all possible treatments (20). In the present study, we conducted an NMA using up-to-date data from ICI-treated patients with advanced lung cancer to compare the risk of developing AEs during or following various treatment strategies.

Literature Search
This NMA was conducted according to the priori established protocol (PROSPERO: CRD42021268650; https://www.crd.york. ac.uk/prospero/#searchadvanced), and reported in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analyses) guidelines (21) and its extension statement for NMA ( Table S1). The PubMed, EMBASE, and Cochrane Library databases were systematically searched from inception to 19 April 2021 with language restricted to English. The search terms and their combinations used in the search strategies are shown in Table S2. We also identified potential studies listed as references in the retrieved articles and searched unpublished data from the ClinicalTrials.gov website.

Study Selection
The eligibility criteria for published studies were as follows: headto-head phase II and III RCTs comparing two or more treatments, including at least one ICI drug in patients with advanced lung cancer. Studies published only in the form of conference abstracts, posters, and presentations of ongoing RCTs were excluded. If several studies were derived from the same trial, the study that reported comprehensive safety data was involved. Two authors (Y.Y. and J.C.) independently screened all titles and abstracts, and further assessed potentially eligible full text based on the aforementioned criteria.

Study Outcomes and Data Extraction
The primary outcomes of this study were overall safety outcomes, viz. treatment-related adverse events and immune-related adverse events as defined in each study (Table S3), including grade 1-5 and grade 3-5, respectively. The secondary outcomes were grade 1-5 and grade 3-5 irAEs in specific organ systems, including the gastrointestinal system (colitis and diarrhea), pulmonary system (pneumonitis), endocrine system (hyperthyroidism, hypothyroidism, thyroiditis, hypophysitis, and diabetes), skin (pruritus, rash, and severe skin reaction), and others (myocarditis, nephritis, hepatitis, myositis, and hypersensitivity/infusion reaction). Grading of AEs was reported according to the Common Terminology Criteria for Adverse Events (CTCAE), as reported in each study. Two authors (Y.Y. and J.C.) used a pre-designed form to extract the following data: study characteristics (study ID and publication year, NCT number, cancer type, study design, arms, treatment regimens, number of patients, follow-up time, and version of CTCAE), demographics and clinical characteristics (age, sex, PS score, brain/CNS metastasis, liver metastasis, bone metastasis, current/former smoker, prior surgery, and prior radiotherapy), and data on the aforementioned outcomes. The above information was extracted from the main text and Supplementary Materials, and only accessible data were analysed.

Quality Assessment
The methodological quality of the included trials was assessed using the Cochrane Collaboration Risk of Bias Tool (22). Low, moderate, or high risk of bias was assigned to each citation within the following five aspects: random sequence generation, allocation concealment, masking, assessment of outcomes, and selective reporting. Disagreements during study selection, data extraction, and quality assessment processes were resolved by consensus following a consultation with the corresponding investigator (Z.G.).

Statistical Analyses
To illustrate the direct and indirect comparisons among the treatments, a plot of the network geometry was generated. A pairwise meta-analysis of head-to-head comparisons was conducted to make direct estimates. Results were reported as relative risks (RRs) with 95% confidence intervals (95% CIs) using a random-effect model. Statistical heterogeneity was assessed using the I 2 -test, with a value > 50% representing considerable heterogeneity (23). A sensitivity analysis was performed to assess the robustness of the results by sequentially eliminating each study from the pool (24,25). Furthermore, meta-regression analyses were performed to explore the influence of potential factors on patient outcomes (26). When a single analysis involved > 10 studies, publication bias was evaluated using funnel plots, as well as Egger's and Begg's tests (27). For outcomes with potential publication bias, the trim and fill method were used to estimate the number of missing studies and to provide an estimated intervention effect to perform adjustment for publication bias. In the network comparison among treatment regimens, chemotherapy was used as the reference comparator. Random effects and consistency models were used to calculate RRs and their 95% CIs; these models are thought to be the most conservative approach to dealing with between-study heterogeneity. Cumulative probabilities were used to provide a hierarchy of the treatments. According to the cumulative probabilities, treatment regimens were ranked from the worst (i.e., associated with the highest risk of AEs) to the best (i.e., associated with the lowest risk of AEs) (28). Transitivity was appraised in consistency and coherence: first, interaction analyses were used to assess the comparability between the consistency and inconsistency models; second, node-splitting analyses were performed to test coherence in the network. To further ensure the robustness of the findings, sensitivity analyses were adopted by restricting the following factors: phase III RCTs and studies that enrolled patients with non-small cell lung cancer (NSCLC). The incidences of grade 1-5 and grade 3-5 AEs were also pooled using meta-analysis (29). All data were analysed by using STATA version13.0 (Statacorp, College Station, Texas, United States), with p values < 0.05 indicating a statistically significant difference.

Study Selection and Characteristics
Our initial search yielded 1,725 records from databases and 914 records from the ClinicalTrials.gov platform; 2,535 records were excluded following screening of titles and abstracts. The remaining 104 full-text articles were reviewed, and 66 articles were excluded for reasons depicted in Figure 1 and Table S4. Given that only one trial (IMpower150) involved groups of ICI + targeted + chemotherapy and ICI + targeted therapy, which had no headto-head comparison with other five treatments, it was excluded in

Risk-of-Bias Assessment
The included RCTs satisfied three tool items, viz. random sequence generation, incomplete outcome data, and selective reporting. Twenty-seven open-label trials did not meet the item of allocation concealment or blinding of participants and personnel, resulting in a high risk of bias in the study assessment. Overall, 27 studies exhibited a high risk of bias, and 11 studies were considered to have a low risk of bias. The details of the quality assessment are presented in Table S6.

Pairwise Meta-Analysis Based on Head-to-Head Comparisons
Direct comparisons were conducted to assess safety profiles among five treatments options (  [3][4][5] was associated with a significantly higher risk of developing these manifestations than chemotherapy. Dual ICIs therapy had a similar risk of grade 1-5 irAEs (RR: 1.48, 95% CI: 0.77-2.84) but a superior risk of grade 3-5 irAEs (RR: 2.12, 95% CI: 1.29-3.48) compared with ICI monotherapy. Detailed data in specific irAEs are listed in Table S7. Briefly, results from organ-specific irAEs, including colitis, pneumonitis, hyper/hypothyroidism, hepatitis, and rash, were comparable with those observed in overall irAEs. Regarding heterogeneity of pairwise meta-analysis comparisons, relatively high heterogeneity was found in primary outcomes (I 2 : 38.2%-95.0%), except for two pairs (dual ICIs + chemotherapy vs. chemotherapy for grade 3-5 AEs and ICI monotherapy vs. chemotherapy for grade 3-5 irAEs). Overall, the general heterogeneity in individual irAEs was low to moderate. Sensitivity analyses were conducted by sequentially removing each study. Following this, the pooled results were in line with the set primacy safety outcomes (Table S8). In addition, metaregression analysis failed to detect any potential confounding factors affecting the primacy outcomes (Table S9). A visual inspection of the funnel plots and Begg's test showed relative symmetry, except for grade 3-5 AE in comparison of ICI monotherapy vs. chemotherapy (P = 0.018) ( Figure S1 and Table S10). However, P values of Egger's test in several outcomes were < 0.05, suggesting that publication bias existed in this study ( Table S10). The trim and fill method were adopted to mitigate publication bias, and the outcomes were consistent with our primary results (P for interaction > 0.05) ( Table S10).
The results pooled via the inconsistency model had a generally satisfactory fit compared with those calculated by the consistency model, except for minor comparisons based on   (Table S13). Likewise, a less significant inconsistency was observed following the node-splitting analysis (Table S14).

Network Meta-Analysis for Specific IrAEs
NMAs and ranking probabilities for different treatment strategies in subgroups of irAEs are depicted in Figures 4, 5 and Table 3. Although the results for individual irAEs varied by organ system and severity, traditional chemotherapy presented the lowest risk in majority of irAEs. For gastrointestinal irAEs ( Figures 4A, B), ICI regimens were associated with higher risk of colitis than chemotherapy, with dual ICIs showing the highest risk (incidence: 3.66% for grade 1-5 and 2.53% for grade 3-5; ranking probability: 91.6% for grade 1-5 and 92.3% for grade 3-5). Nevertheless, a different trend was detected in the case of diarrhea. ICI regimens, especially ICI monotherapy, had a lower risk of causing diarrhea than treatment strategies that included chemotherapy.
For endocrine irAEs, patients administering dual ICIs with or without chemotherapy seemed to be associated with a high risk of hyperthyroidism, hypothyroidism, hypophysitis, thyroiditis, and diabetes. Owing to the low incidence of serious endocrine irAEs, no positive RRs were detected among the individual ICI strategies (Figures 4D-G).
For skin irAEs, the use of ICI monotherapy + chemotherapy, dual ICIs + chemotherapy, and dual ICIs presented the highest risk of pruritus, rash, and severe skin reaction, respectively ( Figures 5A-C). For other irAEs, no significant difference was observed among the ICI regimens ( Figures 5D-H).

Sensitivity Analysis in Network Meta-Analysis
Thirty phase III RCTs, and 29 studies that enrolled patients with NSCLC were separately included in the sensitivity analyses. The observed ranking orders were consistent with the original NMA, irrespective of the overall AEs and irAEs (Table S15).

Major Findings and Interpretation
The current NMA highlights the toxicity profile of ICI-based treatments among patients with advanced lung cancer based on 38 RCTs involving 22,178 patients. Our results indicated that the use of mono-or dual-ICI therapy may reduce the risk of treatment-related AEs at the expense of an increased risk of developing irAEs compared with chemotherapy. During therapy with ICI regimens, patients administering dual ICIs with chemotherapy may experience most AEs (either grade 1-5 or grade 3-5) and irAEs of any grade, and those receiving dual ICIs may have the highest risk of developing serious irAEs. There were differences observed in the toxicity spectra among the five treatment therapies. However, these results should be carefully interpreted because of the limited number of studies on groups receiving dual ICIs with and without chemotherapy.

Comparison With Previous Studies
Recently, developments in lung cancer patients with advanced stages of the disease have shown immunotherapy as a promising treatment option. However, the expanded use of ICIs has resulted in noticeable growth in adverse events, particularly irAEs (6,68). With more treatment options with ICIs now approved for advanced lung cancer, a robust analysis is urgently required to compare the risk of safety profiles among all of the different treatment regimens.
To date, several systematic reviews and meta-analyses have been conducted to assess the safety profiles of ICIs in patients with cancer, yet few studies have focused on patients with lung cancer. The earliest meta-analysis, which pooled 22 RCTs for evaluating rare but severe irAEs resulting from the use of PD-1 and PD-L1 inhibitors in patients with NSCLC, indicated an increased all-grade pneumonitis risk from ICIs than chemotherapy (OR = 2.35, 95% CI, 1.32-4.20, P = 0.004) (15). However, this study had limited value because of the inclusion of minimal trials with a control group, which inevitably led to insufficient evidence for a conclusion; in addition, only ICI monotherapy was investigated. In 2020, an updated NMA of 25 RCTs was conducted for chemotherapy, ICI monotherapy, dual ICIs combination, and ICIs + chemotherapy, and reported a significantly reduced risk of immune-related pneumonitis in patients with lung cancer following ICIs + chemotherapy when compared with dual ICIs combination and ICI monotherapy (14). Furthermore, three other meta-analyses focused on gastrointestinal irAEs, including diarrhea and/or colitis, and consistently found that PD-1/PD-L1 inhibitors might lead to a higher risk of immune-mediated colitis but might result in a The value in each parenthesis represents the probability of risk to rank (%). ICI, immune checkpoint inhibitor; CT, chemotherapy; AE, adverse event; irAE, immune-related adverse event.   The value in each parenthesis represents the probability of risk to rank (%). ICI, immune checkpoint inhibitor; CT, chemotherapy; irAE, immune-related adverse event; G, grade.
reduction in diarrhea when compared with chemotherapy (16)(17)(18). Notably, the above-mentioned studies focused on one specific irAE, making it difficult to unveil the panorama of toxicity from ICIs. Recently, Berti et al. compared overall and organ-specific irAEs between immunotherapy and immunechemotherapy in lung cancer based on 16 phase III clinical trials and found that immunotherapy alone showed a significantly lower risk of irAEs than immunochemotherapy (69). The study investigators excluded phase II RCTs and ignored the discrepancy between ICI monotherapy and dual ICI combination therapy. Given these limitations, the current NMA comprehensively estimated the overall and organ-specific toxicity spectrum among all up to date ICI regimens by pooling all currently available phase II and III clinical trials involving patients with NSCLC and SCLC.

Safety Profile of ICI Regimens in Patients With Lung Cancer
It is well known that the incidence of overall adverse events during ICI monotherapy is lower than that for conventional chemotherapy, while exposure to immunotherapy increases the risk of irAEs, as seen in the present study. Interestingly, further analyses of ICI-based regimens found that ICI monotherapy + chemotherapy decreased the risk of grade 1-5 and grade 3-5 irAEs compared with ICI monotherapy and dual ICIs therapy. This trend was also observed in pneumonitis and myocarditis (grade1-5 and grade 3-5), as well as grade 1-5 hyperthyroidism, hypothyroidism, thyroiditis, severe skin reaction, hepatitis, myositis based on ranking. Consistent with our study, Chen et al. reported that the use of an ICI with chemotherapy led to less pneumonitis than use of ICI monotherapy or dual ICIs combination (14). One possible reason for the decreased risk in irAEs when chemotherapy is added to ICI regimens may be due to the fact that conventional chemotherapy consists of cytotoxic agents that are believed to cause chemotherapy-induced immunosuppression, augmenting stress on the entire immune system and resulting in a reduced immune function (70,71). Another contributing factor might be the use of corticosteroids. In chemotherapy regimens containing cytotoxic agents such as platinum, pemetrexed and taxanes, which are the standard treatments for lung cancer, corticosteroids are commonly prescribed as binding pre-treatment for antiemetic and antiallergy purposes. The baseline or early use of corticosteroids at the time of initiating ICI therapy could blunt a proliferative burst of CD8-positive T cells, which are otherwise needed for the ICI therapeutic response, thus affecting the efficacy and toxicity (72,73). In addition, corticosteroids are recommended immunosuppressive agents for various mild-to-severe irAEs such as pneumonitis, colitis, hepatitis, thyroiditis, rash, etc. (74). Accordingly, the risk of experiencing irAEs may be underestimated in circumstances with corticosteroids. Some studies have pointed out that the incidence of irAEs in patients on a combination of two ICIs was higher than that observed with ICI monotherapy. In the present study, we did not observe any statistical differences between these two groups in terms of overall grade 1-5 irAEs, let alone for rare irAEs, such as thyroiditis, diabetes, myocarditis, nephritis, hepatitis, and myositis. Even between dual ICIs + chemotherapy and ICI monotherapy (both grade 1-5 and grade 3-5 irAEs), significant difference was not detected. However, this trend can be seen from the ranking either in grade 1-5 irAEs or grade 3-5 irAEs. Given the fact that a limited number of studies directly compared dual ICIs (with or without chemotherapy) with ICIs monotherapy, these results should be interpreted with caution. Therefore, more high-quality RCTs are needed to investigate the incidence of irAEs among different ICI-based regimens.

Clinical Implication
Our results provide possible safety speculations for clinical decision-making to tailor the best immunotherapy strategy for each patient with lung cancer. For instance, administration of an individual ICI or dual ICIs plus chemotherapy was reported to have a significantly lower risk of pneumonitis (grade 1-5 and grade 3-5) than ICI monotherapy and dual ICIs therapy and could perhaps be preferred in selected cases of lung fibrosis or severe chronic obstructive pulmonary disease. In addition, ICI monotherapy was associated with the lowest risk for both diarrhea and colitis among ICI regimens and could perhaps be preferred in selected patients in whom gastrointestinal irAEs could be a concern. Of course, these results should be proven in prospective registries or cohorts to better understand the safety of novel ICI-based options in this subset of patients.

Study Strengths and Limitations
The major strength of this study was to depict a full view of the safety profile of ICIs at different levels (risk of overall AEs of any grade and grade 3-5, any irAEs and severe irAEs by individual organs/system). Second, owing to the different toxicity spectrum based on cancer types, the study focused on a specific population of patients (i.e., with lung cancer). Third, except for combination of ICIs and targeted agents, we included all available ICI-based regimens to aid clinicians to tailor the ICI strategy for individual patient with lung cancer.
However, several limitations of this study need to be acknowledged. First, the included RCTs used different terms to describe irAEs. In clinical settings, AEs or irAEs are usually recognised and reported depending on the evaluation of the physician and are diagnosed based on their experience. Therefore, the identification of irAEs might not be completely accurate and might lead to bias in the assessment. In addition, few studies reported irAEs during the entire ICI monotherapy maintenance as separate outcome, making it difficult to investigate irAEs during this period. Second, the included studies showed heterogeneity in terms of subtype of cancer, pharmacological strategy, follow-up time, and other factors. As we focused on outlining the entire safety profile of ICI agents, subgroup analyses based on patients' histology, specified ICIs and kind of chemotherapy were not performed. However, we performed sensitivity analyses and meta-regression in pairwise meta-analysis, as well as sensitivity analyses in NMA, to control for these possible confounders. Third, inconsistent results between direct and indirect comparisons were observed. Unlike direct comparison, network meta-analysis included both direct evidence and indirect effects from the other studies. Because of above-mentioned heterogeneity among RCTs, integrated results possibly underestimated or overestimated the actual results. Fourth, we did not obtain access to comorbidity data, which might be high-risk factors for certain irAEs. Lastly, we did not have the resources to review non-English publications. However, we enrolled studies identified following a comprehensive search of broad databases and are thus confident that this study covered the majority of trials in these special patients. Given aforementioned limitations, further studies are needed to confirm our findings.

CONCLUSIONS
In summary, this network meta-analysis contributes to clarifying the frequency and characteristics of adverse events during ICI treatment in patients with advanced lung cancer. We found that ICI monotherapy + chemotherapy had the best immune-related safety profile, followed by ICI monotherapy, dual ICIs therapy, dual ICIs + chemotherapy for grade 1-5 irAEs, and ICI monotherapy, dual ICIs + chemotherapy, dual ICIs therapy for grade 3-5 irAEs. The safety ranking of ICI-based choices is modulated by specific irAEs and severity.

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 authors.

AUTHOR CONTRIBUTIONS
Z-CG, X-YC, and H-WL are the guarantors of the entire manuscript. Z-CG and Y-DY contributed to the study conception and design, critical revision of the manuscript for important intellectual content, and final approval of the version to be published. J-JC, JF, and Y-JS contributed to the data acquisition, analysis, and interpretation. All authors contributed to the article and approved the submitted version.