First‐line PD‐1/PD‐L1 inhibitors plus chemotherapy versus bevacizumab plus chemotherapy for advanced non‐squamous non‐small cell lung cancer: A Bayesian network meta‐analysis of randomized controlled trials

Abstract Chemotherapy in combination with immune checkpoint inhibitor (ICI) or bevacizumab has demonstrated a superior effect for non‐squamous non‐small cell lung cancer (NS‐NSCLC). There are still few randomized controlled trials (RCTs) investigating the differences between ICI plus chemotherapy (ICI‐chemotherapy) and bevacizumab plus chemotherapy (Bev‐chemotherapy) in first‐line treatment of NS‐NSCLC. We identified RCTs in databases and conference abstracts presented at international conferences by Sep 1, 2021. Bayesian network meta‐analysis was performed using randomized effect consistency model to estimate hazard ratio (HR) and odds ratio (OR). The outcomes included overall survival (OS), progression‐free survival (PFS), overall response rate (ORR), and grade ≥ 3 treatment‐related adverse events (TRAEs). Fifteen RCTs (17 articles) of 6561 advanced NS‐NSCLC patients receiving ICI‐chemotherapy, Bev‐chemotherapy, or chemotherapy at first‐line were eligible for analysis. NMA results showed that first‐line ICI‐chemotherapy prolonged OS (HR 0.79, 0.66–0.94) in patients with advanced NS‐NSCLC compared with Bev‐chemotherapy, while no differences were in PFS, ORR, and grade ≥ 3 TRAEs (p > 0.05). Ranking plots suggested that ICI‐chemotherapy had the most probability to offer the best OS (probability 0.993), PFS (probability 0.658), and ORR (probability 0.565), and Bev‐chemotherapy had the most risks of grade ≥ 3 TRAEs (probability 0.833). Therefore, our findings showed that first‐line ICI‐chemotherapy was associated with better OS than Bev‐chemotherapy in patients with advanced NS‐NSCLC, and more clinical trials are warranted to confirm these results.


| INTRODUCTION
Lung cancer remains the most common cancer and has been the main cause of cancer-related death worldwide. 1,2 Non-small cell lung cancer (NSCLC) accounts for more than 80% of all lung cancers, with most having nonsquamous histology. 3 Despite significant improvements have been made in the treatment of advanced NSCLC, The prognosis remains poor with median overall survival (OS) of 8-10 months and 5-year progression-free survival (PFS) of 4%. 4,5 Currently, the mainstream treatment for advanced non-squamous NSCLC (NS-NSCLC) is to target driver gene mutation such as epidermal growth factor receptor. [6][7][8] However, customarily a small fraction of patients were identified as having driver mutations that can benefit from targeted agents. 9,10 As for patients without sensitive gene mutations, they are unable to benefit from targeted therapy, making the choices of treatment full of passivity for them. 9 Traditional chemotherapy remains the mainstream treatment with a response rate of only 15% ~ 30%.

| Inclusion and exclusion criteria
Studies were included as follows: (1) patients had histologically confirmed with advanced NS-NSCLC (stage IIIB-IV); (2) studies involved the comparisons among ICIchemotherapy, Bev-chemotherapy, and chemotherapy in first-line treatment; (3) study outcomes included objective response rate (ORR), PFS, OS, or grade ≥ 3 treatmentrelated adverse events (TRAEs); (4) study designs were controlled clinical trials (RCTs). Studies were excluded as follows: (1) studies with duplicate publications; (2) study protocols, reviews, meta-analyses, letters, or case reports; (3) studies with data unavailable. If the trial compared one drug of two different dosages, we chose the usual drug dosage. This NMA was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA). 37

K E Y W O R D S
bevacizumab, chemotherapy, immune checkpoint inhibitor, network meta-analysis, non-small cell lung cancer

| Outcome measures
The following outcomes were used for analysis: ORR, PFS, OS, and grade ≥ 3 TRAEs. ORR was calculated by the proportion of complete and partial recession. PFS was defined as the time between randomization and disease progression or death from any cause. OS was calculated as the time in between randomization and death from any cause. Grade ≥ 3 TRAEs were evaluated by the common terminology criteria for adverse events. 38 Hazard ratio (HR) with 95% confidence interval (CI) was calculated for continuous variable analysis (OS and PFS), and odds ratio (OR) with 95%CI were estimated for binary variable analysis (ORR and grade ≥ 3 TRAEs).

| Study selection and data extraction
Two investigators (Z.B. Zhang and Y. Wang) independently screened the studies by title, abstract, and full text based on inclusion and exclusion criteria; any discrepancies were discussed by a third reviewer (J.L. Wang). The extracted data included the first author, publication year, trial name, trial phase, sex, age, smoking history, treatment strategy, drug dose, number of patients, Eastern Cooperative Oncology Group Performance Status (ECOG PS), and the outcomes of OS, PFS, ORR, and grade ≥ 3 TRAEs.

| Quality assessment
The quality of each included study was assessed by another two reviewers (J.Y. Lu and X.W. Ge) independently with the Cochrane Risk of Bias Tools, 39 including six domains: (1) randomization sequence generation and allocation concealment; (2) blinding of patients and personnel; (3) blinding of outcome assessment; (4) incomplete outcome data; (5) selective outcome reporting; (6) other bias. Discrepancies were mediated by a third investigator (J.L. Wang) and resolved by consensus. Quality assessments for each trial are available in Supplement 2.

| Statistical analysis
2.6.1 | Pairwise meta-analysis STATA software (version 15.0) was used for performing pairwise meta-analysis (PMA) in a random-effect model. The I-square was calculated for assessing the extent of variability attributable to heterogeneity between two studies. 40 PMA between ICI-chemotherapy and chemotherapy was conducted stratified by PD-L1 expression, liver metastases, and brain metastases.

| Network meta-analysis
R software (version 4.1.0) with gemtc package (version 1.0) was applied for conducting Bayesian network metaanalysis (NMA) in a random-effect model. Four chains were generated and 100,000 iterations with 10,000 burnins for each chain (the interval of 10). R software was also used for identifying the probability of each treatment to be ranked the best for four endpoints and the surface under the cumulative ranking curves (SUCRAs) were presented in ranking plots. Network plots showed the connection between eligible trails based on the number of trails and sample size. The comparison-adjusted funnel plots tested publication bias. 41,42 To ensure the reliability of NMA, sensitive analysis was conducted by excluding 1 trial with phase II and 3 trials with a sample size of each group less than 100. Inconsistency was assessed by comparing the synthesized HRs of PMA and NMA results. Node splitting analysis was used for assessing the consistency of direct and indirect evidence. 43 The contribution plot was used to measure the percent contribution of direct comparison to the indirect, mixed, and the entire network estimates (Supplementary materials). All tests were two-sided, and a p-value below 0.05 was considered statistically significant.

| Eligible studies and characteristics
All 2284 records were identified from the databases and international conferences. After screening titles and abstracts to exclude duplicate articles and articles that did not meet the inclusion criteria, 15 eligible RCTs (17 articles) with 6541 advanced NS-NSCLC patients receiving three treatments (ICI-chemotherapy, Bev-chemotherapy, and chemotherapy) were included in this NMA. Seven trials [15][16][17][18][19][20][21] compared Bev-chemotherapy with chemotherapy, seven trials 23-30 compared ICI-chemotherapy with chemotherapy, and one trial 35,36 compared ICI-chemotherapy with Bevchemotherapy ( Figure 1). Overall, ICI-chemotherapy, Bevchemotherapy, and chemotherapy were administered to 2328, 1720, and 2513 patients, respectively. Network plots of all eligible trials were shown in Figure 2. Of all 15 trials, 15 reported ORR and PFS, 14 reported OS, and 11 reported grade ≥ 3 TRAEs. Thirteen of 15 trials (86.7%) were phase III RCTs; 12 of 15 RCTs (80%) included more than 100 patients in each arm. Eligible studies and patients' characteristics (Table 1 and Table 2) included the first author, publication year, trial name, trial phase, treatment drug, number of patients, sex, age, stage, smoking status, ECOG PS, brain metastases, and liver metastases. Supplement 2 showed that 20.0% (3/15) were high-quality studies and 46.7% (7/15) were accompanied by a high risk of bias for lacking personnel blinding according to the Cochrane Collaboration tool.

| NMA of outcomes
As listed in Table 3

| Treatment ranking
The treatment ranking plots for four outcomes were shown in Figure 3, which suggested that ICI-chemotherapy had the most probability to offer the best OS (probability 0.993), PFS (probability 0.658), ORR (probability 0.565), and Bev-chemotherapy had the most risks of grade ≥ 3 TRAEs (probability 0.833).

| Heterogeneity and publication bias
Heterogeneity was evaluated in direct, indirect, and network comparations. The results showed that there was little heterogeneity among included trials with I 2 > 50% (Supplement 5). Comparison-adjusted funnel plots were F I G U R E 4 Comparing ICI-chemotherapy with chemotherapy stratified by PD-L1 expression, liver metastases, and brain metastases presented in Supplement 6, which indicated little publication bias among included trials.

| Inconsistency assessment
The inconsistency assessment was performed using nodesplitting analysis for OS, PFS, ORR, and grade ≥ 3 TRAEs, the results showed the consistency among direct, indirect, and network comparations with all p > 0.05. Forest plots of direct, indirect, and network comparations were generated for OS, PFS, ORR, and grade ≥ 3 TRAEs ( Figure 5).

| Sensitivity analysis and contribution of direct comparison
As shown in Supplement 7, we excluded one trial 17 with phase II and three trials 18,19,25 with a sample size of less than 100 in each group. A total of 11 trials, involving 6023 previous untreated NS-NSCLC patients, were enrolled in the sensitivity analysis. The results were similar to the main analysis. Therefore, we included all trials for the robustness of the final NMA results. The contribution plots of direct comparison are presented in Supplement 8.

| DISCUSSION
Immune checkpoint inhibitors (ICIs) have been noted to show promise for cancer treatment due to their advances in cancer treatments and have been clinically approved in a variety of malignancies. [44][45][46] PD-1, an essential regulator of adaptive immune responses, is primarily involved in immune inhibitory signaling and is ectopically expressed on antitumor T cells, while expression of its ligand PD-L1 can be upregulated by tumor cells and block antitumor effects. 47 48 Conventional chemotherapy is the main treatment option for NS-NSCLC patients. 49 The combination of ICIs and chemotherapy have higher efficacy compared to chemotherapy and is therefore recommended by NCCN guidelines as first-line treatment of advanced NS-NSCLC. 50,51 Currently, ICI-chemotherapy has emerged as a new treatment option on advanced NS-NSCLC patients without the driver gene mutation. Bevacizumab, a monoclonal antibody against VEGFR, is beneficial in combination with chemotherapy for advanced NS-NSCLC, and Bev-chemotherapy has been the standard treatment for advanced NS-NSCLC. IMpower-150 trial was the first RCT comparing ICI-chemotherapy with Bevchemotherapy in previous untreated NS-NSCLC patients with advanced-stage, the results showed that patients with ICI-chemotherapy had prolonged survival compared with those with Bev-chemotherapy. 35 The comparison between first-line ICI-chemotherapy and Bev-chemotherapy in advanced NS-NSCLC remains unclear due to the lack of direct comparisons between ICI-chemotherapy and Bev-chemotherapy. Therefore, we performed this NMA to compare the potential of these two combination strategies.
We initially performed NMA among ICI-chemotherapy, Bev-chemotherapy, and chemotherapy for previously untreated patients with advanced NS-NSCLC. Based on 15 RCTs, the results indicated an OS benefit with ICIchemotherapy compared to Bev-chemotherapy, while no differences were observed on ORR, PFS, and grade ≥ 3 TRAEs. Treatment ranking plots revealed that ICI-chemotherapy was most likely to deliver the best OS, PFS, and highest risk of TRAEs ≥3, while Bev-chemotherapy had the highest ORR. Furthermore, we performed subgroup NMA in patients with liver metastases, and no differences in PFS and OS were detected between ICI-chemotherapy and Bevchemotherapy. Our findings demonstrated that first-line ICI-chemotherapy was associated with better OS than Bevchemotherapy in advanced EC patients except for those with liver metastases. A comparison between ICI-chemotherapy and Bev-chemotherapy stratified by biomarkers such as PD-L1 expression was not performed due to the unavailability of data, which requires further investigation.
There were several limitations in this study. First, different PD-1/PD-L1 inhibitors (pembrolizumab, camrelizumab, tislelizumab, and sintilimab) and chemotherapy regimens were used in different RCTs, which may have affected the final results. Second, the genetic status in Bev-chemotherapy trials was unknown, whereas immunotherapy trials involved patients without the driver gene mutation. Third, some trials allowed patients to cross over from chemotherapy to ICI-chemotherapy after progressions such as the KEYNOTE-021 trial and CameL trial, which may decrease the OS benefit of ICI-chemotherapy. Fourth, heterogeneity and publication bias existed, several trials are still ongoing with incomplete data, we thus conducted sensitive analysis and the results were consistent with the main analysis. Lastly, due to the unavailability of data, we only performed a subgroup NMA stratified by liver metastases. Despite the above limitations, our study still confirmed the favorable survival of ICI-chemotherapy versus Bevchemotherapy without more severe TRAEs.

| CONCLUSION
This study elucidates that ICI-chemotherapy is superior to Bev-chemotherapy for improved OS in first-line treatment of advanced NS-NSCLC. More clinical trials are warranted to confirm these results.