Comparison of Efficacy and Safety of Single and Double Immune Checkpoint Inhibitor-Based First-Line Treatments for Advanced Driver-Gene Wild-Type Non-Small Cell Lung Cancer: A Systematic Review and Network Meta-Analysis

Background Immune checkpoint inhibitors (ICIs) have improved survival for advanced wild-type non-small cell lung cancer (NSCLC) significantly, but few studies compared single ICI (SICI)-based treatments and double ICIs (DICI)-based treatments. We summarized the general efficacy of ICI-related treatments, compared the efficacy and safety of SICI-based [programmed death 1 (PD-1)/programmed death-ligand 1 (PD-L1) or cytotoxic T lymphocyte-associated antigen 4 (CTLA-4) inhibitors ± chemotherapy (CT)] and DICI-based (PD-1/PD-L1 inhibitors+CTLA-4 inhibitors ± chemotherapy) treatments vs. CT in the first-line treatment. Methods We included phase II/III randomized controlled trials (RCTs), including patients with histologically confirmed stage IIIB–IV driver-gene wild-type NSCLC who received first-line ICI-related therapy in at least one arm. PubMed, Embase, and Cochrane Library were searched from January 1, 2005, to December 31, 2020. This network meta-analysis was performed in a Bayesian framework using GEMTC and JAGS package in R.3.6.1. The research was registered with PROSPERO (CRD42020184534). Results Twenty RCTs were involved, including 13,032 patients and 17 treatment regimens. The results showed that ICI-based therapies could provide a pooled median overall survival (mOS) (POS) of 15.79 (95% CI: 14.85–16.73) months, and there were no significant differences in OS, progression-free survival (PFS), objective response rate (ORR), and grade 3 or higher adverse events (≥3AEs) between DICI-based treatments (POS: 14.81, 12.11–17.52 months) and SICI-based treatments (POS: 16.17, 14.59–17.74 months) in overall patients. However, DICI-based treatments had significantly prolonged the OS over SICI-based treatments in squamous and PD-L1 <1% subgroups. The ranking of OS benefit by Bayesian surface under the cumulative ranking curve (SUCRA) spectrum showed that DICI+chemotherapy ranked first for overall population and subgroups including squamous, non-squamous, any level of PD-L1 expression, smoking, male, Eastern Cooperative Oncology Group performance status (ECOG PS) = 0/1, age < 65/≥65 while SICI+CT for low tumor mutation burden (TMB), non-smoking, and female subgroups, and DICI for high TMB subgroups. Conclusions In the first-line therapy for advanced wild-type NSCLC, both SICI- and DICI-based treatments could bring significant overall advantages over chemotherapy, with comparable outcomes of efficacy and ≥3AEs. DICI-based treatments were more effective than SICI-based treatments in squamous and PD-L1 <1% subgroups. For most populations, DICI+chemotherapy could be the best choice with a survival benefit, while SICI+chemotherapy has established its position actually. Systematic Review Registration [PROSPERO], identifier [CRD42020184534].

Both SICI-based and DICI-based treatments have achieved certain success. However, no studies have been conducted to compare the two treatments directly. In theory, DICI-based treatments could target more immune checkpoints and should be more effective but may also produce more side effects. It has become a huge challenge perplexing clinicians whether DICIbased therapies are more effective and whether there exists the best treatment or beneficial populations among SICI, SICI+CT, DICI, and DICI+CT. To address such questions reasonably, we conducted an integrated analysis and network meta-analysis (NMA). Our study summarized the general effects of related treatments and compared the efficacy and safety among SICI, SICI+CT, DICI, DICI+CT, and CT in the first-line treatment of advanced wild-type NSCLC, which will provide valuable evidence for clinical decision-making.

Literature Searching Strategies
This NMA was performed according to the PRISMA extension statement (Supplementary Table 1

Inclusion Criteria
Published phase II/III RCTs reported in English that compared at least two first-line treatments, at least one arm containing ICIs, for histologically confirmed advanced (stage III-IV) driver-gene wild-type NSCLC patients who did not receive prior systemic therapies. The hazard ratio (HR) and 95% confidence interval (CI) of OS and PFS are available.

Exclusion Criteria
Trials involving targeted therapy for driver-gene mutation NSCLC patients or therapies other than ICIs or CT, such as surgery, radiotherapy, antiangiogenesis, immune cells, and cancer vaccines, or currently unavailable drugs such as the anti-TIGIT antibody tiragolumab. Trials that only reported outcomes of maintenance therapy were also excluded.

Data Extraction and Risk of Bias Assessment
We extracted study name, first author, publication year, number and characteristics of patients, OS, PFS, objective response rate (ORR), and the incidence of grade 3 or higher adverse events (≥3AEs) related to treatments. For the same study that reported outcomes of different follow-up times, we extracted the most recent data.
We assessed the bias risk of RCTs using the Cochrane Risk of Bias Tool, including seven items: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective outcome reporting, and other sources of bias (31). RCTs can be evaluated as low, high, or ambiguous risk of bias. Data extraction and risk of bias assessment were conducted by two independent investigators (QX and XZ).

Data Analysis
To judge the median OS (mOS) of each treatment tentatively, we performed pairwise meta-analyses with the frequentist method for head-to-head trials. Heterogeneity between studies was assessed using the Q test and I 2 statistics. The random model was used when I 2 ≥ 50 or p < 0.05, in which heterogeneity was considered statistically significant (32).
For survival variables (OS/PFS) and binary variables (ORR/ ≥3AEs), HR or odds ratio (OR) and corresponding 95% CIs were pooled according to the fixed or random model, which were compared using deviance information criteria (DIC) (33). We used the JAGS and GEMTC package in R.3.6.1 for Bayesian NMA using a Markov Chain Monte Carlo simulation technique. For each outcome, 150,000 sample iterations were generated with 100,000 burn-ins and a thinning interval of 1. To ensure the convergence of the model, visual inspection methods of trace plots and Brooks-Gelman-Rubin diagnostic were adopted (34). We used Stata 16.0 to generate network plots, indicating more directly the relationships between treatments. For network consistency, node splitting analysis was used to evaluate the differences between direct and indirect comparisons in the closed loop of treatments. Transitivity was evaluated using visual graphics for patient characteristics between treatment groups and control groups, respectively. To estimate the probability of each treatment being at each rank, we calculated the surface under the cumulative ranking curve (SUCRA). The higher SUCRA value represents that a treatment is to be ranked on the top more likely (35).
The assumption of transitivity was accepted because no variability of population baselines was identified in the treatment group and control group among studies except for KEYNOTE 021 (11,12), which showed a significant deviation of male proportion (Supplementary Figure 1). The risk of bias assessment was summarized in Supplementary Figure 2

Integrated Analysis of Median Overall Survival
We firstly performed an integrated analysis of mOS in eligible studies to get a pooled OS of current treatment strategies for advanced wild-type NSCLC. The pooled mOS (POS) of ICIbased treatments was 15 Figure 2C).

Network Meta-Analysis of Pathology Subgroup
In the squamous NSCLC subgroup, both SICI-based treatments and DICI-based treatments achieved significant OS advantages compared to CT only, while SICI-based treatments achieved significantly shorter mOS than that in DICI-based treatments (HR = 1.25, 95% CI: 1.01-1.54) ( Figure 3C   SICI+CT was superior to SICI and DICI, while DICI was significantly better than SICI (Supplementary Figure 5B).

Network Meta-Analysis of Tumor Mutation Burden Subgroup
The superiority of SICI-based and DICI-based treatments over CT in OS and PFS was observed in the high TMB subgroup. However, there was no statistical difference between SICI and DICI. In the low TMB subgroup, there was also no statistical difference in mOS and mPFS between SICI-based or DICI-based treatments and CT ( Figure 3F and Supplementary Figure 4D).
In the high TMB populations, SICI, DICI, SICI+CT, and DICI+CT showed significant prolongation of both OS and PFS in contrast to those of CT ( Figure 5D and Supplementary Figure 5D). In the low TMB populations, only SICI+CT showed a significant advantage over CT in mOS (HR = 0.75, 95% CI: 0.56-1.00) and mPFS (HR = 0.60, 95% CI: 0.46-0.77). In addition, the mPFS of SICI and DICI was statistically inferior to that of CT ( Figure 5D and Supplementary Figure 5D).

Network Meta-Analysis of Smoking, Gender, Age, or Eastern Cooperative Oncology Group Subgroup
In smokers, all ICI-based measures significantly prolonged OS compared with CT, and SICI+CT was inferior to DICI+CT (HR = 1.28, 95% CI: 1.05-1.57) (Supplementary Figure 6A). In non-smokers, the four ICI-based strategies achieved equal outcomes on OS with CT (Supplementary Figure 6B). In males, they yielded superior OS than CT, while DICI is the same with DICI+CT (HR = 1.01, 95% CI: 0.82-1.26). DICI was significantly better than SICI; DICI and DICI+CT were also superior to SICI +CT (Supplementary Figure 7A). DICI, SICI+CT, and DICI+CT all showed significant OS benefits compared with CT regardless of age ( Supplementary  Figures 8A, B). In patients <65 years old, the mOS of SICI+CT was significantly shorter than that of DICI+CT (HR = 1.29, 95% CI: 1.00-1.67) (Supplementary Figure 8B). In Eastern Cooperative Oncology Group performance status (ECOG PS) = 0 populations, DICI, SICI+CT, and DICI+CT obtained significantly longer mOS than CT, while DICI+CT dramatically reduced the risk of death by 52% (HR = 0.48, 95% CI: 0.32-0.72). When combined with CT, the efficacy of SICI+CT was significantly worse than that of DICI+CT (HR = 1.70, 95% CI: 1.10-2.63) (Supplementary Figure 9A). In the ECOG PS = 1 subgroup, SICI, DICI, SICI+CT, and DICI+CT all achieved significant OS benefits compared with CT, while there were no statistical differences among the four ICI-based measures (Supplementary Figure 9B).

Rank Probabilities
The Bayesian ranking curves of comparable treatments in different populations are shown in Supplementary Figures  S10A, B (ranking profiles and corresponding SUCRA are shown in Supplementary Figures 11A, B and Supplementary  Figures 12A, B). The result of Bayesian ranking is approximately consistent with NMA. Overall, DICI+CT was most likely to be ranked first for mOS; SICI+CT was ranked first for mPFS and ORR (Supplementary Figure 10A). In subgroup analysis, mOS of DICI+CT ranked first for squamous, non-squamous, any PD-L1 expression, smoking, males, ECOG PS = 0/1, age <65/≥65; SICI+CT for low TMB, non-smoking, and females; DICI for high TMB (Supplementary Figures 11A, B; Supplementary Figures  12A, B).

Inconsistency Assessment and Sensitivity Analyses
The fit of the consistency model in most comparisons was better than that of the inconsistency model, except for mOS (overall, non-squamous, females subgroups), mPFS (overall, squamous, non-squamous, PD-L1 ≥50% subgroups), and ORR, for which the random model was used (Supplementary Table 3). Inconsistency between direct and indirect comparisons using the node-splitting approach did not show significant differences in comparisons except for mOS and mPFS in the low TMB subgroup (Supplementary Table 4).

Network Meta-Analysis of Specific Treatment Regimens
We compared the efficacy and safety of specific treatment regimens (Supplementary Figure 15).  Figures 16A, B).

DISCUSSION
As mentioned above, to compare and evaluate the efficacy of SICI-and DICI-based therapies in advanced wild-type NSCLC, we performed an integrated analysis of survival outcomes and NMA among these first-line treatment strategies. Despite those negative primary endpoints of many ICI-related RCTs, we found that ICI-based therapies could provide a POS of nearly 16 months for overall patients with advanced NSCLC. Furthermore, both SICI-based therapies (POS: 16.17 months) and DICI-based therapies (POS: 14.81 months) had significant OS benefits compared with CT, without significant difference in mOS, mPFS, ORR, and ≥3AEs between the two ICI-based strategies. DICI-based therapies were significantly superior to SICI-based therapies in squamous and PD-L1 <1% subgroups on mOS. DICI was more effective than SICI in PD-L1 <1% and male subgroups. In subgroups such as smoking, male, age <65, ECOG PS = 0, DICI+CT obtained significantly longer OS than SICI+CT. Bayesian ranking spectrum showed that DICI+CT had the best OS advantage in the overall population and squamous, non-squamous, any PD-L1 level, smoking, male, ECOG PS = 0/1, <65/≥65 subgroups; SICI+CT ranked first in subgroups of low TMB, non-smoking, and female subgroups, while DICI ranked first in high TMB subgroups. In our NMA, the overall efficacy of SICI-based and DICIbased therapies was consistent possibly due to the limited number of RCTs on DICI-based therapies with different conclusions. Notably, DICI-based therapies were significantly superior to SICI-based therapies in low immunogenicity subgroups (squamous or PD-L1 <1%), suggesting that dualtarget interventions can improve the immune response by transforming the "cold" tumors to "hot" tumors and thereby lead to better efficacy. Interestingly, in populations with potentially high immune responses (smoking, male, <65, ECOG PS = 0), DICI+CT also brought more OS benefits than SICI+CT. In terms of specific treatment regimens, NIV+IPI, with or without CT, all obtained positive survival results and got Food and Drug Administration (FDA) approval, while DUR+TRE ± CT failed to replicate the success of NIV+IPI ± CT. So how to match the anti-PD-1/L1 and anti-CTLA-4 correctly is the key to get the most considerable benefit of DICI. Interestingly, when comparing anti-CTLA-4 plus anti-PD-1 therapy with anti-PD-1 monotherapy, we found that the OS of NIV+IPI was significantly higher than that of NIV monotherapy or DUR monotherapy, which is consistent with the finding of the previous study (39). However, the OS benefit of NIV+IPI vs. that of PEM monotherapy is comparable, manifesting that PEM may amplify the efficacy of SICI.
Obviously, further explorations are needed. The key to applying DICI-based treatments reasonably focuses on how to reduce the side effects of anti-CTLA-4 and maximize the efficacy and synergy of ICIs combined with CT. Although the current exploration of DICI-based regimens is still insufficient, with the increasing number of related studies and the effective control of drug dose and toxicities, such strategy possesses great potential to improve the survival of patients with advanced NSCLC to a large extent. For example, some novel anti-PD-L1 antibodies, such as M7824 (40) and YM101 (41), exhibited broader ranges of antitumor spectrum compared to the SICI recently. These biologicals simultaneously blocked transforming growth factor (TGF)-b and PD-L1 pathways, or targeted some new immune checkpoints other than PD-1/L1 or CTLA-4, thus having potential to overcome resistance to SICIs or the present DICI treatment in future clinical practices.
We found that SICI-based therapies also obtained satisfactory results. Due to a large number of such studies and participants involved, the integrated results and NMA comparison were more reliable and robust. Based on the current comparative results, SICI-based therapies, especially SICI+CT, were the first-line treatment regimen with definite efficacy and tolerable side effects. In terms of specific treatment regimens, SIN+CT and PEM+CT ranked in the top on OS, with equal ≥3AEs to that of CT alone. Therefore, SICI+CT is currently the most practical treatment for the unscreened population. How to optimize the period and duration of medication to achieve the unity of efficacy improvement and side effect reduction remains a key problem to be resolved.
Our study also has several limitations. First, some studies were classified as moderate or high risk of bias because of inadequate randomization, allocation concealment, and blinding. Second, although all the studies in our analysis included patients with advanced wild-type NSCLC, some studies included a few patients with driver-gene mutated NSCLC. Thirdly, mOS data in some studies were immature and were extracted or calculated from interim analysis or the latest meeting abstracts. Fourth, it is not possible to compare all treatment strategies in each subgroup due to the limited availability of outcomes. For example, the comparison of mPFS in the PD-L1 1%-49% subgroup lacked data on DICI-based therapies. Fifth, the prediction of SUCRA for treatment strategy ranking is not absolute; when SUCRA prediction contradicts NMA results, the HR estimation of NMA should be given priority. Finally, due to the limited number of RCTs and participants involved in DICI-based therapies, the reliability and robustness of related NMA results and conclusions need to be further verified.

CONCLUSIONS
In the first-line therapy for advanced wild-type NSCLC, both SICI-based and DICI-based treatments could bring significant overall advantages vs. CT, with comparable outcomes for mOS and ≥3AEs. DICI-based treatments were more effective than SICI-based treatments in squamous and PD-L1 <1% subgroups, while DICI in combination with CT could be the best first-line choice for most populations. We need more research to further evaluate the efficacy and safety of DICI-based treatments. At the same time, SICI-based therapies have established their position in the current first-line treatment. In addition, NMA and ranking possibilities of specific regimens could provide strong evidence for clinical selection of individualized treatment regimens to maximize survival benefits for related patients.

DATA AVAILABILITY STATEMENT
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

AUTHOR CONTRIBUTIONS
LL is the corresponding author. QX and XZ are joint first authors. LL contributed to the study concept and design. QX and XZ took part in the initial literature search and assessed the eligibilities of feasible studies. QX and XZ interpreted the findings and wrote the first draft of the manuscript. QX, XZ, MH, XD, JG, LS, SL, KH, and JW prepared the figures and tables. LL revised and edited the manuscript. All authors approved the final version of the manuscript. LL is the guarantor of this study and accepts full responsibility for the work, had access to the data, and controlled the decision to publish. The corresponding authors attest that all listed authors meet authorship criteria and that no other person meeting the criteria has been omitted. All authors contributed to the article and approved the submitted version.