Comparative Safety of Immune Checkpoint Inhibitors and Chemotherapy in Advanced Non-Small Cell Lung Cancer: A Systematic Review and Network Meta-Analysis

Backgrounds: Immune checkpoint inhibitors (ICIs) alone or in combination with chemotherapy (CT) are the standard of care for rst-line therapy in metastatic non-small cell lung cancer (NSCLC) patients without actionable mutations. The safety ranking of different ICI and CT combination regimens has not been investigated. This study was aimed to provide a toxicity prole and safety ranking of different ICI and CT combination regimens. Methods: We performed comprehensive searches of phase 2 and 3 randomized clinical trials (RCTs) comparing different ICI regimens (alone or combination) or CT for the rst-line treatment of advanced NSCLC. Outcomes of interest were the cumulative incidence of any treatment-related adverse events (TRAEs), grade 3-5 TRAEs (grade 3-5), any immune-related adverse events (irAEs), and grade 3-5 irAEs (grade 3-5). Odds ratios and 95% credible intervals were calculated as summary statistics to quantify the effect of different ICI combination regimens. Results: We included 23 RCTs from 2016 to 2021 with a total of 14,378 patients. The incidence of any TRAEs and grade 3-5 TRAEs ranked from high to low were ICI-CT (probability: 74.88%), ICI-ICI-CT (50.60%), CT alone (74.79%), ICI-ICI (98.37%), and ICI monotherapy (99.37%). Adding CT to ICI regimens resulted in a higher incidence of any grade or grade 3-5 TRAEs compared to ICI-ICI combinations or ICI monotherapy. However, ICI-ICI-CT combinations did not result in a higher incidence of TRAEs than ICI-CT combinations. For any irAEs and grade 3-5 irAEs, the ranking was ICI-ICI (probability: 97.38), ICI monotherapy (96.98%), ICI-CT (99.44%), and CT alone (99.98%). Notably, the incidence of any grade and grade 3-5 irAEs was lower when adding CT to ICI monotherapy. Conclusion: Lack of head-to-head comparisons, these ndings provide evidence for clinical decision-making when considering different ICI combination regimens for advanced NSCLC patients. monotherapy in non-squamous NSCLC patients with a high PD-L1 level. heterogeneity enrolled in study rst-line NSCLC treatment, 4 second-line treatment trials, 1 maintenance treatment trial, and 2 small cell lung cancer treatment trials. In addition, the recruited number of patients smaller in the present study. The development of irAEs can affect various organs with different clinical symptoms such as endocrine (hyperthyroidism, hypothyroidism, and adrenal insuciency), dermatological (rash and pruritus), gastrointestinal (diarrhea and colitis), hepatic (hepatitis), respiratory (pneumonitis), and kidney (renal insuciency) symptoms. System and organ-specic analyses in the current study indicated that the incidence of irAEs was highest with ICI-ICI-CT and ICI-ICI combinations, followed by ICI monotherapy, ICI-CT, and CT alone in endocrine, dermatologic, hepatic, and respiratory-related irAEs. Pneumonitis is a clinically relevant and potentially life-threatening irAE, and a previous meta-analysis reported that its incidence may be higher with PD-1 inhibitors than PD-L1 inhibitors, and that the incidence was highest with ICI-ICI combinations.[43] However, detailed comparisons between different ICI treatment-related risk proles for organ-specic irAEs are lacking. Several studies have indicated that the development of irAEs after ICI treatment was associated with a better response and survival.[44, 45] However, this association may be biased by differences in histological subtype, disease burden, and lines of treatment. We focused our analyses on TRAEs and irAEs with ICIs, providing detailed data for clinicians which will be helpful when choosing different treatment regimens with regards to both ecacy and


Introduction
Lung cancer remains the leading cause of cancer deaths worldwide, including in Taiwan. [1] Non-small cell lung cancer (NSCLC) accounts for 85% of all lung cancer cases, and more than 70% of NSCLC patients present with locally advanced or metastatic disease (stage III or IV) at the initial diagnosis. [2] Immune checkpoint inhibitors (ICIs) including monoclonal antibodies targeting cytotoxic Tlymphocyte protein-4 (CTLA-4), programmed cell death protein-1 (PD-1), and programmed death ligand-1 (PD-L1), have been introduced to various cancer treatments, and they have become one of the most important therapies for patients with cancer. In addition, they have been shown to provide a higher and more durable treatment response with relatively tolerable toxicity compared to conventional chemotherapy (CT). [3,4] Treatment with ICIs alone or in combination with CT is the standard of care for the rst-line therapy of patients with metastatic NSCLC without driver gene mutations. Clinically, the selection of these different regimens is mainly based on the histological type, PD-L1 expression and outcomes from different trials. [5,6] The toxicity of chemotherapeutic drugs has been extensively researched and can be adequately managed based on clinical experience. However, the side effects of ICIs, especially immune-related adverse events (irAEs), are unique and can involve multiple organs with different onset patterns and severity, which are less familiar to clinicians. The incidence of grade 3-5 irAEs varies according to the ICIs used, dose, and combination regimen, ranging from 6% to 55%. These irAEs can lead to life-threatening conditions and treatment discontinuation, further compromising the outcomes. [7] Currently, both ICI monotherapy and in combination with CT are treatment options for metastatic NSCLC patients with a high PD-L1 expression. A meta-analysis found that the combination of ICIs (pembrolizumab or atezolizumab) and CT resulted in signi cantly better progression-free survival (PFS) and overall survival (OS) compared to pembrolizumab monotherapy in non-squamous NSCLC patients with a high PD-L1 level.
[8] However, retrospective studies have demonstrated a comparable bene t in OS with these two treatment modalities. [9,10] The combination of nivolumab plus ipilimumab with or without two cycles of platinum-doublet CT has also been approved as rst-line treatment for metastatic NSCLC patients without genomic tumor mutations. [5,6] In addition, a recent meta-analysis suggested that anti-PD-1 combinations may be associated with higher survival outcomes than PD-L1 combinations for patients with advanced NSCLC. [11] Given the multiple treatment options and inconsistent e cacy results among different combination strategies, it is important to understand the safety pro les of these regimens. Therefore, the aim of this network meta-analysis (NMA) was to investigate the adverse events associated with ICI monotherapy or in combination with different ICIs or CT as rst-line therapy for advanced NSCLC, and provide a safety ranking of these different modalities.

Data Sources and Search Strategy
This systematic review was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses for Protocols (PRISMA-P) guidelines. [12] We performed a literature search in PubMed, Embase, Cochrane Library, and CENTRAL records for relevant studies published up to July 1, 2021. Abstracts or presentations of international scienti c conferences including the American Society of Clinical Oncology (ASCO), American Association for Cancer Research (AACR), European Society for Medical Oncology (ESMO), and World Conference on Lung Cancer (WCLC) were also checked to ensure that the most recent updated data were included. In addition, reference lists of the included studies were inspected, and experts and physicians in this eld were also consulted. The project was prospectively registered in the PROSPERO database of systematic reviews (CRD42021268464).

Study Selection and Data Extraction
Studies that met the following criteria were included: (1) those that enrolled patients with treatment-naïve advanced (stage IIIB/IIIC/IV) NSCLC; (2) randomized clinical trials (RCTs) using ICI monotherapy or in combination with other ICIs or CT; and (3) phase 2 and 3 trials reporting grade 3-5 or any grade of TRAEs or irAEs according to the de nitions and criteria of the recruited clinical trials.

Quality and Risk of Bias Assessment
The abstract of each study was independently assessed by three authors to con rm the eligibility for analysis. Data from the included studies were independently extracted and checked. Two reviewers independently assessed the risk of bias of the included studies according to the recommendations in the Cochrane Handbook for Systematic Reviews of Interventions 5.1. Each criterion was marked as unclear, low, moderate, or high risk of bias. Disagreements were resolved by consensus or assessed by other authors.

Outcome Measurements
The outcomes of interest were grade 3-5 and any grade of TRAEs and irAEs according to the de nitions used in the trials. We ranked the probability of a treatment regimen being associated with the risk of TRAEs and irAEs from highest to lowest by estimating the risk of each treatment regimen.

Data Synthesis and Statistical Analysis
Studies were pooled to determine the risk of adverse events with different treatment regimens in NMA, and quanti ed with odds ratios (ORs) and 95% credibility intervals (CrIs). Random effect models generating posterior distributions of model parameters were chosen to calculate ORs and 95% CrIs in a Bayesian framework, which were considered to be conservative estimations managing betweenstudy heterogeneity. [13] On the basis of ORs, the posterior probability of being the best to the worst for each treatment regimen was provided and ranked by calculating the surface under the cumulative ranking curves.Subgroup analyses were performed based on system or organ-speci c irAEs, including infusion reactions, endocrine, gastrointestinal, hepatic, respiratory, renal, and dermatologic irAEs. A two-sided p value of <0.05 was considered to be signi cant.The NMA was conducted using the interactive web-based software, MetaInsight version 3.14, powered by the shiny and netmeta packages for R. [14] All Bayesian statistical calculations were performed using R package netmeta [15] and gemtc.
[16] Plots depicting the network geometry and forest plots were generated using MetaInsight.
To facilitate clinical translation, the number needed to bene t (NNTB) and number needed to harm (NNTH) were calculated from the overall ORs and control event rate. [17] The NNTB and NNTH calculations and Cates plots were generated at http://www.nntonline.net/visualrx/.
To ensure the stability and consistency of the ndings, we conducted several sensitivity analyses. First, we repeated the analysis using datasets excluding each single trial one by one. Second, we used the node-splitting method to assess the consistency between direct and indirect comparisons in the NMA. [18]

Study Characteristics
The search strategy is shown in Supplementary Figure S1. A total of 14,378 patients were included from 23 clinical trials conducted between 2016 and 2021.  The characteristics of the included studies are summarized in Table 1. Among these studies, 4 trials included squamous patients only, 7 studies included non-squamous patients only, and 12 included patients of both histologies. The investigated drugs were pembrolizumab in 5 trials, nivolumab in 4, atezolizumab in 5, durvalumab in 2, tislelizumab in 2, sintilimab in 2, cemiplimab in 1, camrelizumab in 1, and pembrolizumab plus nivolumab in 1. Comparative network plots are illustrated in Figure 1.

Risk-of-Bias Assessments
As illustrated in Supplementary Figure S2, all of the studies had a low risk of reporting bias. An unclear risk of performance bias was detected in 17 studies due to the open-label study design, allocation bias was detected in 4 studies due to an unequal ratio of assigned patients, and selection bias was detected in 5 studies as the study populations were from Asia or China only. A high risk of detection bias was detected in the RATIONALE 307 study due to the possibility of investigator bias, and attrition bias was detected in 2 studies (RATIONALE 304, ORIENT-12) due to a high percentage of missing data.

Outcome Assessments
Any grade and grade 3-5 TRAEs The forest plots and results of the NMA for TRAEs are shown in Figure 2. Compared to CT alone, a signi cantly higher risk of any TRAEs was observed with ICI-CT (OR 1.63, 95% CrI 1.12 to 2.39). However, the incidence of any TRAEs was comparable between ICI-ICI-CT combinations and CT alone (OR 1.26, 95% CrI 0.62 to 2.57). Compared to CT alone, a lower incidence of TRAEs was found with ICI monotherapy (OR 0.2, 95% CrI 0.2 to 0.39) and ICI-ICI combinations (OR 0.51, 95% CrI 0.34 to 0.79) (Figure 2A).  Figure 2C shows the ORs of any grade and grade 3-5 TRAEs with different treatment regimens. Compared to ICI monotherapy, CT alone and all ICI-based combinations were associated with a greater risk of any grade and grade 3-5 TRAEs. Of note, ICI-ICI-CT combinations were not associated with higher rates of all grade and grade 3-5 TRAEs compared to ICI-CT combinations or CT alone.

Number needed to bene t (NNTB) and number needed to harm (NNTH)
The frequency and proportion of the NNTB between group comparisons and the Cates plots are illustrated in Supplementary Figure   S3. In summary, the NNTBs were 7 (95% con dence interval [CI] 5 to 11) and 17 (95% CI 9 to 56) for ICI monotherapy and ICI-ICI to prevent a TRAE compared to the CT group, respectively. The NNTH was 37 (95% CI 25 to 136) for ICI-CT to cause a TRAE compared to CT alone. There was no signi cant difference in TRAE rate between ICI-ICI-CT and CT alone. For any irAEs, the NNTHs were 3 (95% CI 2 to 4) for ICI monotherapy, 5 (95% CI 4 to 9) for ICI-CT, and 2 (95% CI 2 to 3) for ICI-ICI to cause an irAE compared to CT alone (Supplementary Figure S4).

Subgroup analyses of system or organ-speci c irAEs
The ranking probability of a treatment regimen being associated with any irAEs varied by organ or system. For all included system or organ-speci c irAEs, CT alone had the lowest ranking (Supplementary Table S1). ICI-ICI-CT was associated with the highest risk of irAEs in endocrine, hepatic, dermatologic irAEs, and infusion reactions, whereas ICI-ICI was associated with the greatest risk of irAEs in gastrointestinal, hepatic irAEs and pneumonitis. ICI-CT was associated with the highest risk of renal irAEs.

Sensitivity analyses
The results of the analyses in which each single trial was sequentially excluded were qualitatively consistent with those without excluding any trial. The OR estimates for any TRAEs, grade 3-5 TRAEs, any irAEs, and grade 3-5 irAEs varied minimally with no change in the ranking orders of probability. In addition, according to the node-splitting model, similarity was observed across direct, indirect, and network comparisons, which implied no signi cant inconsistency (Supplementary Table S2).

Discussion
The results of this NMA demonstrated that ICI-CT combinations were associated with a higher incidence of any grade or grade 3-5 TRAEs compared to ICI-ICI combinations or ICI monotherapy. However, ICI-ICI-CT regimens were not associated with a higher incidence of any grade or grade 3-5 TRAEs than ICI-CT combinations. Surprisingly, the incidence of any grade and grade 3-5 irAEs was lower when CT was added to ICI monotherapy. Subgroup analysis of system or organ-speci c irAEs demonstrated that the incidence of irAEs was highest for ICI-ICI-CT and ICI-ICI combinations and lowest for CT monotherapy.
Several studies have focused on comparisons and meta-analysis of rst-line ICI combination regimens among patients with advanced NSCLC.[8, 11] Dafni et al reported a meta-analysis based on 12 studies with 9,236 metastatic NSCLC patients, and concluded that the addition of CT to ICIs enhanced their treatment e cacy as rst-line treatment for advanced NSCLC.
[8] In terms of toxicity, no signi cant difference in the incidence of any grade ≥3 adverse event was observed between pembrolizumab/CT and CT alone, while a higher incidence of TRAEs was observed with atezolizumab/CT and ipilimumab/CT. In addition, the incidence of TRAEs with ICI-ICI combinations was lower compared to CT alone, and all ICI monotherapies were associated with a signi cantly lower incidence of TRAEs compared to CT alone. These results are in concordance with the current study. However, the incidence of irAEs was not speci ed in their study. In the current study, we found that the incidence of any grade and grade 3-5 irAEs was highest with ICI-ICI combinations followed by ICI monotherapy, and that adding CT to ICI monotherapy reduced the incidence of ICI-related irAEs. Another meta-analysis conducted by Liu et al analyzed 16 studies with 10 ICI combinations and 8,278 advanced NSCLC patients. [11] The safety results indicated that the addition of CT to ICIs increased treatment-related toxicities compared to standard CT, and the authors concluded that toxicity was lowest for the ICI-ICI combinations across all combination treatments. Similarly, their study did not specify the irAEs with different combination treatments. Their subgroup analysis suggested that nivolumab plus ipilimumab plus CT was associated with the best OS in patients with a PD-L1 expression <1%, and that pembrolizumab plus chemotherapy was associated with the best OS in patients with a PD-L1 expression ³1%. However, the incidence of irAEs with different treatment regimens should be taken into consideration, because most clinicians are more familiar with CT-related toxicities rather than irAEs. Although ICI-containing regimens increase the incidence of irAEs, adding CT to ICI monotherapy may decrease the incidence of irAEs. Recently, Berti et al included 16 ICI trials in a meta-analysis and reported rates of any grade irAEs and severe irAEs with ICIs of 37.1% and 18.5%, respectively. [42] Pooled analysis on any, severe and organ-speci c irAEs showed that ICI monotherapy was associated with a signi cantly lower risk of irAEs compared to immuno-chemotherapy. [42] These results are in contrast to those of the present study. A possible explanation is the heterogeneity of the included studies. The enrolled trials in the current study all investigated rst-line NSCLC treatment, whereas Berti's report enrolled 4 second-line treatment trials, 1 maintenance treatment trial, and 2 small cell lung cancer treatment trials. In addition, the recruited number of patients was smaller than in the present study.
The development of irAEs can affect various organs with different clinical symptoms such as endocrine (hyperthyroidism, hypothyroidism, and adrenal insu ciency), dermatological (rash and pruritus), gastrointestinal (diarrhea and colitis), hepatic (hepatitis), respiratory (pneumonitis), and kidney (renal insu ciency) symptoms. System and organ-speci c analyses in the current study indicated that the incidence of irAEs was highest with ICI-ICI-CT and ICI-ICI combinations, followed by ICI monotherapy, ICI-CT, and CT alone in endocrine, dermatologic, hepatic, and respiratory-related irAEs. Pneumonitis is a clinically relevant and potentially lifethreatening irAE, and a previous meta-analysis reported that its incidence may be higher with PD-1 inhibitors than PD-L1 inhibitors, and that the incidence was highest with ICI-ICI combinations. [43] However, detailed comparisons between different ICI treatmentrelated risk pro les for organ-speci c irAEs are lacking. Several studies have indicated that the development of irAEs after ICI treatment was associated with a better response and survival. [44,45] However, this association may be biased by differences in histological subtype, disease burden, and lines of treatment. We focused our analyses on TRAEs and irAEs with ICIs, providing detailed data for clinicians which will be helpful when choosing different treatment regimens with regards to both e cacy and safety.
The strength of this study is that it is the rst NMA to thoroughly examine the risk of irAEs with different ICI combination regimens in currently available trials as rst-line treatment for advanced NSCLC. In addition, we provided risk pro les including ranking, NNTB and NNTH of different ICI combination regimens to improve clinical decision making for treatment strategies. There are several limitations to this study. First, the reported criteria for TRAEs and irAEs in the included RCTs were not exactly the same. Therefore, we analyzed system or organ-speci c irAEs to minimize heterogeneity. Second, we did not conduct subgroup analyses on the basis of histologic types, PD-L1 expression, ICI types, or individual ICI and CT drugs to identify possible differences in the toxicities of these speci c populations or medications. Finally, the median follow-up time was different for all included studies, ranging from 8.0 to 59.9 months.
Awareness and reporting of TRAEs and irAEs may have increased over time, and thus reporting bias may have existed.
In conclusion, this NMA demonstrated that ICI-CT combinations were associated with a higher incidence of any grade or grade 3-5 TRAEs compared to ICI-ICI combinations or ICI monotherapy. However, ICI-ICI-CT combinations were not associated with a higher incidence of any grade or grade 3-5 TRAEs than ICI-CT combinations. Importantly, the incidence of any grade and grade 3-5 irAEs was lower when CT was added to ICI monotherapy. Subgroup analysis of system or organ-speci c irAEs demonstrated that the incidence of irAEs ranked from high to low for ICI-ICI, ICI monotherapy, ICI-CT, and CT alone in terms of endocrine, hepatic, dermatologic, and respiratory-related irAEs. These ndings provide evidence for clinicians on the e cacy and safety of different ICI combination regimens for advanced NSCLC patients.   <p>Risk of any TRAEs and grade 3-5 TRAEs associated with each treatment combination regimen.<sup> </sup>Forest plots of any TRAEs and grade 3-5 TRAEs for different treatment regimens compared with CT alone are shown in (A) and (B), respectively. Figure   2C shows the risk of any TRAEs (blue background) and grade 3-5 TRAEs (pink background) of each comparison among all treatment regimens.<sup> </sup>The pooled odds ratios (95% credible intervals) were the results of comparing the right column treatment regimens with the left column treatment regimens (the reference group).<sup> </sup>Bold text indicates statistical signi cance at p&lt;0.05.</p> Figure 3 <p>Risk of any grade and grade 3-5 irAEs associated with different treatment regimens. Forest plots of any irAEs and grade 3-5 irAEs for different treatment regimens compared with CT alone are shown in (A) and (B), respectively. Figure 3C shows the risk of any irAEs (blue background) and grade 3-5 irAEs (pink background) of each comparison among all treatment regimens. The pooled odds ratios (95% credible intervals) were the results of comparing the right column treatment regimens with the left column treatment regimens (the reference group). Bold text indicates statistical signi cance at p&lt;0.05.</p> Figure 4 <p>Ranking profiles for the different treatment regimens on any TRAEs and grade 3-5 TRAEs (A) and on any irAEs and grade 3-5 irAEs (B). Ranking plots indicate the probability of the highest to lowest risk of different adverse events.</p>