First-line therapies with maintenance regimens for driver gene-negative advanced non-small cell lung cancer (NSCLC): a systematic review and network meta-analysis of 61 randomized trials

Introduction

Lung cancer remains the leading cause of cancer-related mortality worldwide (1). Non-small cell lung cancer (NSCLC) accounts for >85% of lung cancer cases (2). Over 50% of patients with NSCLC are diagnosed at an advanced stage (stage IIIB–IV), with poor clinical outcomes (3). The 5-year survival rate is approximately 26% for stage IIIB disease and nearly 0% for stage IVB disease (4). Treatment options for advanced NSCLC remain limited. However, recent advancements, including the development of molecularly targeted agents and immune checkpoint inhibitors (ICIs), have expanded the therapeutic landscape (5).

Driver genes play critical roles in tumor growth, survival, and metastasis, and their abnormalities are directly and indirectly implicated in oncogenesis. Targeted therapies directed at driver alterations can improve outcomes and reduce toxicity in selected patients with advanced NSCLC (6). For patients with epidermal growth factor receptor (EGFR) or anaplastic lymphoma kinase (ALK) alterations, National Comprehensive Cancer Network (NCCN) guidelines recommend tyrosine kinase inhibitors as preferred first-line treatment (7,8). However, approximately one-third of advanced NSCLC cases lack targetable genetic aberrations and are therefore ineligible for appropriate targeted therapy (TT) (9). Additionally, the driver mutations commonly found in lung adenocarcinoma are rarely observed in squamous (SQ) NSCLC, resulting in a lack of specific treatment strategies for SQ NSCLC (10).

For patients with wild-type driver genes, platinum-based chemotherapy (CT) remains the mainstay (5). Median overall survival (OS) with platinum-based CT is approximately 8–12 months, and long-term survival remains poor for most patients without targetable mutations (11,12). Because platinum-based treatments have reached a plateau of effectiveness, there is an urgent need for more effective and tolerable therapeutic regimens (13). Attempts to extend induction CT beyond 4–6 cycles have not improved survival and may increase severe toxicity, as shown in prior studies and meta-analyses (14-17). Consequently, maintenance therapy (MT) has received increasing attention (18). MT refers to continued treatment after a planned induction phase in patients without progression, aiming to sustain disease control. MT may continue an induction agent (continuation maintenance) or switch to a non-cross-resistant agent (switch maintenance) (19).

Several randomized controlled trials (RCTs) have reported improved outcomes with MT in advanced NSCLC (20-23). Accordingly, NCCN guidelines recommend MT for patients without progression after initial therapy as a standard approach in driver-gene-negative disease (8). However, multiple induction regimens and MT approaches are available, creating uncertainty about the optimal first-line-to-maintenance strategy. Many RCTs randomized patients only after induction (i.e., among those without progression) and measured outcomes from MT initiation (20-22). Prior meta-analyses largely evaluated MT in isolation rather than the complete induction-plus-maintenance strategy (24-26).

The optimal integration of induction and MT within a unified first-line strategy remains unclear. Valuable insights may be gained from studies in which randomization occurs before first-line initiation and outcomes are measured throughout the full treatment course (27). Therefore, there is a need for comprehensive synthesis and assessment of consistency across such trials (26). Conventional pairwise meta-analyses provide high-quality evidence but are restricted to head-to-head comparisons (28,29). Network meta-analysis (NMA) allows simultaneous comparison of multiple interventions, including those without direct head-to-head trials (30). A prior NMA evaluated MT-containing strategies in selected molecular/clinical subgroups (27). However, that analysis focused mainly on EGFR and did not comprehensively address other driver-negative populations; moreover, newer trials warrant an updated synthesis.

In this study, we systematically identified RCTs evaluating first-line strategies that incorporate MT in patients with advanced driver-gene wild-type NSCLC. We conducted a comprehensive meta-analysis assessing the efficacy and safety of these regimens and rated certainty of evidence using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach (31). Using a Bayesian NMA framework, we aimed (I) to identify the most favorable induction-plus-maintenance strategies and (II) to compare outcomes across CT backbones when combined with MT. We present this article in accordance with the PRISMA-NMA reporting checklist (32) (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-aw-2087/rc).

Methods

Inclusion criteria

Types of studies

We included RCTs that randomized patients before first-line treatment initiation and incorporated MT in at least one treatment arm. Trials enrolling molecularly unselected populations were eligible only if extractable data were available for patients with driver-gene wild-type NSCLC. Given the low prevalence of targetable mutations in SQ NSCLC, SQ-only trials were included even when molecular testing was not reported. Conference abstracts were included when they provided sufficient extractable data for analysis. This review was registered in PROSPERO (CRD42021215862).

Types of participants

Eligible participants were adults with histologically or cytologically confirmed stage IIIB–IV (ineligible for curative surgery or radiotherapy) or recurrent disease and confirmed driver-gene wild-type status. Participants had not received prior systemic therapy before first-line treatment.

Types of interventions

We included any first-line regimen with a prespecified MT component. Regimens without MT were eligible only when they were comparators within the same RCT against an MT-containing strategy.

Outcome measurements

Primary outcomes were OS and progression-free survival (PFS). Secondary outcomes included objective response rate [ORR = complete response (CR) rate + partial response (PR) rate] and severe adverse effects (SAEs; grade ≥3 treatment-related adverse events) (33,34).

Exclusion criteria

We excluded RCTs that (I) incorporated surgery or radiotherapy, (II) enrolled fewer than five patients with driver-gene wild-type NSCLC, or (III) offered MT only as an optional (non-protocolized) treatment.

Electronic search

We systematically searched the Cochrane Library, PubMed and Embase with no restrictions on language, publication year, or publication status. The full PubMed search strategy is provided in available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf. We also hand-searched reference lists of included studies and relevant reviews to identify additional trials.

Study selection

Records were managed in Endnote X9 (35) and duplicates were removed. Two blinded reviewers (H.M.F. and Y.Z.) screened titles and abstracts independently and assessed full texts. Studies that met all of the selection criteria were finally reviewed. Disagreements were resolved through discussion.

Data extraction and management

Using a standardized Excel form (Microsoft Excel 2013) (36), we extracted publication details (first author, year of publication, country of study), study design features (method of randomization, allocation concealment, blinding, follow-up), participant characteristics (sample size, allocation per arm, mean age, gender, Eastern Cooperative Oncology Group (ECOG) performance status, NSCLC histological subtype, AJCC/UICC stage, type and proportion of wild-type driver genes), intervention details (first-line treatment, MT regimen), outcomes, and type of data analyses [e.g., intention-to-treat (ITT), modified ITT].

For dichotomous data (ORR and SAEs), the total number of participants and those experiencing the event in each intervention group were extracted. For survival outcomes (OS and PFS), we extracted hazard ratios (HRs) with corresponding 95% credible intervals (CIs) as reported. Survival data from independent committee assessments was preferred if it was available, or investigator assessment data was selected. When HRs and 95% CIs were not available, we estimated them from Kaplan-Meier survival curves using the method of Tierney et al. (37). When outcomes were reported only graphically, data were digitized using Engauge Digitizer.

Bias risk

Two reviewers (H.M.F. and Y.Z.) independently assessed risk of bias using the Cochrane ‘Risk of bias (RoB)’ tool (38) across seven domains (random sequence generation (per study), allocation concealment (per study), blinding of participants and personnel (per outcome), blinding of outcome assessment (per outcome), incomplete outcome data (per outcome), selective outcome reporting (per study) and other bias). Each domain was rated as ‘low’, ‘high’ or ‘unclear’ risk of bias for each included study. Disagreements were resolved by discussion or consulting the third reviewer (J.H.T.).

Quality of evidence

Two authors (H.M.F. and Y.Z.) independently assessed certainty of evidence for primary outcomes using the GRADE framework, considering risk of bias, imprecision, inconsistency, indirectness, and publication bias (31). Certainty was graded as high, moderate, low, or very-low. We followed the approach suggested by Brignardello and colleagues to evaluate confidence in evidence from a NMA (39-42). In summary, for each comparison of the evidence network, the certainty of direct and indirect evidence was rated separately. Direct evidence was assessed using the standard GRADE approach. For the indirect evidence, we focused on the first order loop. In the absence of a first order loop, we used a higher order loop. The direct comparisons forming the indirect comparison were also assessed by GRADE approach, and the lower confidence rating was the initial confidence rating of the indirect comparison. In the presence of intransitivity, the initial confidence rating of the indirect comparison was rated down further. We evaluated the certainty of both the direct and indirect evidence without considering imprecision if network evidence existed. For the network evidence, if the network estimate was dominated by either the direct or the indirect estimate, we based the network quality rating on the dominant estimate and would not rate it down because of incoherence. But if both sources of evidence contributed to a similar degree, we used the higher of the two quality ratings as the network quality rating, and rated it down further for incoherence. Then, imprecision was considered (43).

Data synthesis

We visualized network geometry in STATA 14.2 (44), with nodes representing treatment regimens and edges representing direct comparisons. Node and edge sizes were proportional to sample size and number of trials (45). Trials not connected to the network were summarized descriptively. Transitivity was assessed by comparing distributions of key effect modifiers across treatment nodes.

Statistical analysis

Bayesian NMAs were performed using the Graphical Mixed Treatments Comparisons (GEMTC) package in R software (46,47). A hierarchical Bayesian model with three different initial values and 50,000 burn-in simulations for each chain was used. Posterior summaries were based on 100,000 subsequent simulations. Convergence was evaluated via Brooks-Gelman-Rubin plots (48). Vague or flat priors for trials baselines and treatment effect priors were used.

We summarized dichotomous outcomes using pooled risk ratios (RRs) and time-to-event outcomes using pooled HRs, each with 95% CIs. We estimated treatment ranking probabilities and calculated surface under the cumulative ranking curve (SUCRA) values (49). Random- and fixed-effects models were compared using the deviance information criterion (DIC). The model with the lowest DIC was preferred (a difference >5 is considered significant) and was used to explain our results. When DICs were similar between random effects and fixed effect models, we chose fixed effect model to explain our results. Because the assumption of equal between-study heterogeneity across comparisons in random effects model might result in an implausibly wide CI in sparse networks (42). Global and local inconsistency were examined via inconsistency models and node-splitting methods.

Assessment of heterogeneity

Statistical heterogeneity was assessed using the I² statistics and P values; substantial statistical heterogeneity was defined as P value <0.10 and/or the value of I² statistic >50% (50).

Dealing with missing data

When data were missing, authors were contacted. ITT data were prioritized. Otherwise, we used the data available to us, but the potential impact was addressed in the assessment of risk of bias.

Subgroup analyses and sensitivity analyses

Subgroup analyses and network meta-regression were performed to explore statistical heterogeneity and inconsistency if at least 10 studies were available, focusing on histology and programmed cell death ligand 1 (PD-L1) expression. Sensitivity analyses were conducted to test result robustness.

Measures for publication bias

Comparison-adjusted funnel plot was drawn to assess small sample effects when ≥10 studies were available.

Results

Search results and characteristics of included studies

All databases were systematically searched up to 11 December 2024, and 35,235 records were retrieved. After removing duplicates and screening titles and abstracts, 177 articles were retained for full-text review. Ultimately, 61 RCTs were included; see available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx for the full list (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf).

The 61 included RCTs were published between 2005 and 2024 and enrolled 30,488 participants. Almost all studies included patients with an ECOG performance score of 0 to 2, with the exception of six conference-abstract-only studies [(available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) (A12,A19,A37,A39,A49-50)]. Across 60 studies, SQ NSCLC accounted for 32% of participants; one abstract-only study (A49) did not report histology details. Among patients with non-squamous (NSQ) NSCLC, 89% had wild-type driver genes. The median age ranged from 57 to 78 years. Among studies reporting sex, 70% of participants were male; seven [(available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) (A12,A16,A18,A39,A49-50,A63)] studies did not report sex distribution. Most patients were EGFR-negative, with other negative driver genes including ALK, ROS1, KRAS, and RET. Thirty (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) studies required tumor tissue for PD-L1 assessment. We categorized interventions into seven groups: (I) monotherapy; (II) immunotherapy (I); (III) CT + anti vascular endothelial growth factor (VEGF) monoclonal antibody (mAb); (IV) CT + I; (V) CT + single immunotherapy (SI) + other agents; (VI) CT + other agents (e.g., Hsp27 inhibitor or JAK inhibitor); and (VII) CT + TT. For analytical consistency, we assigned labels to each regimen; detailed definitions are provided in Figure 1. Study characteristics and label assignments are summarized in available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx.

Figure 1 Treatment regimens. Beva, bevacizumab; CI, credible interval; CT, chemotherapy (cisplatin or carboplatin based); DI, dual immunotherapy; EGFR, epidermal growth factor receptor; Gem, gemcitabine; HR, hazard ratio; m, maintenance; MET, mesenchymal-epithelial transition factor; OR, objective response; PD-L1, programmed cell death ligand 1; Pem, pemetrexed; SD, stable disease; SI, single immunotherapy; TIGIT, T cell immunoreceptor with Ig and ITIM domains; TKI, tyrosine kinase inhibitor; VEGF, vascular endothelial growth factor.

Risk of bias in included studies

Overall risk of bias judgments are summarized in Figure 2, and study-level assessments are shown in available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf. Overall, 15 trials [(available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) (A6,A26,A30-A33,A36,A38,A42,A47-48,A51-52,A54,A58,A60)] were judged to have low risk of bias across all seven domains of “RoB” tool. Nineteen trials [(available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) (A1,A4-5,A12,A15-20,A24,A27,A37,A39,A41,A50,A53,A62-63)] reported random assignment but did not describe the randomization procedure in detail; therefore, selection bias was rated as unclear. The remaining trials adequately described the randomization procedures in detail. Twenty-six trials (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) had high risk of performance bias, primarily because of open-label designs. Thirteen trials [(available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) (A1,A4,A10,A12,A16,A18-20,A37,A41,A50,A62-63)] were judged to be at unclear risk of performance bias, as they were designed as double-blind trials but did not provide a detailed explanation of blinding of participants and personnel. Seven trials [(available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) (A9,A15,A17,A28,A50,A55-56)] had high risk of detection bias because PFS or tumor response was assessed by investigators rather than an independent review committee. With the exception of eight studies [(available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) (A12,A16,A18-19,A40-41,A50,A63)] with insufficient information, the remaining trials were judged to have low risk of attrition bias because primary survival analyses were conducted using ITT populations. Nine conference-abstract-only studies [(available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) (A12,A16,A18-19,A37,A39,A41,A50,A63)] were rated as having unclear reporting bias because protocols were unavailable and reporting was limited.

Figure 2 Risk of bias graph. Review authors’ judgements about each risk of bias item presented as percentages across all included studies.

Transitivity

Clinical and methodological characteristics were broadly comparable across studies, supporting the transitivity assumption. Most interventions were administered intravenously with broadly similar dosing schedules. All trials enrolled participants with comparable disease stages and age ranges, and included both sexes.

Effects of interventions

The DIC values for all outcome models in the NMA are presented in available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf. GRADE evaluations of direct evidence are summarized in Tables S3,S4, while assessments of indirect evidence are provided in the league tables for OS and PFS outcomes. Different colors are used to indicate the certainty of evidence levels.

OS and PFS

For OS, 53 RCTs (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) involving 22 treatments were included in this network (Figure 3A). For PFS, 55 RCTs (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) involving 23 treatments were included (Figure 3B). The number of studies contributing to each pairwise comparison is presented in available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf. The largest subset of studies (n=20) focused on comparing the effectiveness of CT + SI versus CT alone [(available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) (A3,A27-28,A30-32,A34-36,A38,A43-45,A47-48,A51-52,A57-58,A60-61)]. Most comparisons were evaluated in only a single trial. The primary outcomes of all interventions compared with CT are illustrated in the forest plot (Figure 4). Complete network results, including HRs, 95% CIs, and SUCRA values, are presented in the league tables (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf). A ranking plot comparing OS and PFS outcomes is shown in Figure 5, with the X-axis and Y-axis representing SUCRA values for OS and PFS, respectively. Based on the SUCRA rankings, the most effective therapies for OS were DI (SUCRA =0.91), CT + DI m:SI (SUCRA =0.90), CT + SI + anti-IL-1β mAb (SUCRA =0.90), CT + SI + Bevacizumab (SUCRA =0.85), and CT + DI (SUCRA =0.79). For PFS, the most effective therapies were CT + SI + Bevacizumab (SUCRA =0.96), CT + SI + anti-IL-1β mAb (SUCRA =0.94), CT + SI (SUCRA =0.88), CT + DI (SUCRA =0.73), DI (SUCRA =0.73). All six treatments demonstrated improved OS and PFS compared to CT, with the exception of CT + DI m:SI, which significantly improved OS (HR 0.61, 95% CI: 0.44 to 0.86, moderate-certainty evidence) but not PFS (HR 0.81, 95% CI: 0.52 to 1.28, low-certainty evidence). We have moderate confidence in the evidence suggesting that DI, CT + DI and CT + SI improve both OS and PFS compared with CT. Additionally, CT + SI + Bevacizumab appears to improve both OS (moderate-certainty evidence) and PFS in comparison to CT; however, the PFS benefit is supported only by low-certainty evidence. CT + SI + anti-IL-1β mAb may be a promising therapy, but the conclusion is currently based on limited evidence of low-quality [one study (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx, A37)]. Given the clinical importance of OS, we considered OS, PFS, and evidence certainty together and identified DI, CT + DI m:SI, CT + SI + Bevacizumab, and CT + DI as the most favorable options overall.

Figure 3 Network graph for primary outcomes. (A) Network graph for OS. (B) Network graph for PFS. CT, chemotherapy; DI, dual immunotherapy; EGFR, epidermal growth factor receptor; MET, mesenchymal-epithelial transition factor; OS, overall survival; PFS, progression-free survival; SI, single immunotherapy; TIGIT, T cell immunoreceptor with Ig and ITIM domains; TKI, tyrosine kinase inhibitor; VEGF, vascular endothelial growth factor.

Figure 4 Forest plot for the overall population (all treatments compared with CT). CI, credible interval; CT, chemotherapy; DI, dual immunotherapy; EGFR, epidermal growth factor receptor; HR, hazard ratio; MET, mesenchymal-epithelial transition factor; OS, overall survival; PFS, progression-free survival; SI, single immunotherapy; TIGIT, T cell immunoreceptor with Ig and ITIM domains; TKI, tyrosine kinase inhibitor; VEGF, vascular endothelial growth factor.

Figure 5 Ranking plot of SUCRA values. CT, chemotherapy; DI, dual immunotherapy; EGFR, epidermal growth factor receptor; MET, mesenchymal-epithelial transition factor; OS, overall survival; PFS, progression-free survival; SI, single immunotherapy; SUCRA, surface under the cumulative ranking curve; TIGIT, T cell immunoreceptor with Ig and ITIM domains; TKI, tyrosine kinase inhibitor; VEGF, vascular endothelial growth factor.

Tumor response

In the tumor response network, 48 RCTs (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) involving 20 treatments were included (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf). The league table for this network is presented in available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf. Based on SUCRA rankings, CT + SI + Bevacizumab (SUCRA =0.91), CT + SI (SUCRA =0.89), CT + SI + anti-IL-1β mAb (SUCRA =0.87), and CT + DI (SUCRA =0.82) were the most effective in improving ORR (Figure 6), consistent with the findings for PFS. Compared with CT, RRs and 95% CIs for all the four regimens exceeded 1, indicating improved objective response.

Figure 6 SUCRA values for ORR and safety. CT, chemotherapy; DI, dual immunotherapy; EGFR, epidermal growth factor receptor; MET, mesenchymal-epithelial transition factor; ORR, objective response rate; SI, single immunotherapy; SUCRA, surface under the cumulative ranking curve; TIGIT, T cell immunoreceptor with Ig and ITIM domains; TKI, tyrosine kinase inhibitor; VEGF, vascular endothelial growth factor.

Safety

All studies reported adverse effects, but 49 RCTs (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) involving 20 treatments specifically reported SAEs (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf). Based on SUCRA values, SI (SUCRA =0.96), epidermal growth factor receptor tyrosine kinase inhibitor (EGFR-TKI) (SUCRA =0.94), Bevacizumab + EGFR-TKI (SUCRA =0.87), DI m:SI (SUCRA =0.86), DI (SUCRA =0.77) ranked as the safest regimens (Figure 6). The incidences of grade 3 or higher adverse events associated with SI, EGFR-TKI and DI m:SI were lower than those with CT (RRs and 95% CIs <1) (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf). Bevacizumab + EGFR-TKI and DI were as safe as CT (RRs and 95% CIs included 1).

Subgroup analysis and meta regression

Subgroup analyses and meta-regressions were conducted based on tumor histology and PD-L1 expression levels (<1%, ≥1%, 1–49%, and ≥50%). In the histology-based analysis, CT regimens were stratified to evaluate whether the efficacy of various CT regimens combined with MTs differed.

For patients with NSQ NSCLC, 33 studies (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) provided OS data and 33 studies (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) provided PFS data, forming the corresponding networks (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf). Studies that could not be connected to others within the network are listed in available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf. Among patients with advanced NSQ NSCLC, regimen A (Pemetrexed + Carboplatin or Cisplatin with Pemetrexed maintenance) was the most frequently used CT, followed by regimen B (Paclitaxel + Carboplatin or Cisplatin). The outcomes of all interventions compared to regimen A are illustrated in a forest plot (Figure 7), and full OS and PFS results are presented in the league tables (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf). According to SUCRA rankings (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf), A + SI + anti-IL-1β mAb, B + SI + Bevacizumab, and A + SI + Bevacizumab were the most effective therapies for improving OS and PFS, but the confidence in these findings is low (compared to A). After excluding low/very-low certainty evidence, A + DI (OS SUCRA =0.76, PFS SUCRA =0.66), DI (OS SUCRA =0.75), and A + SI (OS SUCRA =0.74, PFS SUCRA =0.80) were the most effective therapies, with moderate confidence for OS benefits (compared to A).

Figure 7 Forest plot for NSQ NSCLC (all treatments compared with regimen A). CI, credible interval; CT, chemotherapy; DI, dual immunotherapy; HR, hazard ratio; NSCLC, non-small cell lung cancer; NSQ, non-squamous; OS, overall survival; PFS, progression-free survival; SI, single immunotherapy; TIGIT, T cell immunoreceptor with Ig and ITIM domains.

For patients with SQ NSCLC, 23 studies [(available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) (A12-15,A17-18,A25,A27,A29,A32,A36-38,A44-48,A55,A57-58,A60-61)] provided OS data (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf) and 20 studies [(available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) (A12-15,A17-18,A25,A27,A32,A36-38,A45-48,A57-58,A60-61)] provided PFS data (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf), forming two sub-networks due to incomplete connectivity. Studies that could not be connected are listed in available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf, and league tables are provided in available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf and available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf. Among SQ patients, regimen B was the most frequently used CT, followed by C (Gemcitabine + Carboplatin or Cisplatin), and D (Nabpaclitaxel + Carboplatin) was least used. The outcomes of interventions compared to B or C are shown in the forest plot (Figure 8). After comprehensive evaluation of SUCRA rankings, HRs, and 95% CI, and excluding interventions with low or very-low GRADE evidence, the most effective regimens for OS were DI, B + SI, C + SI, and B + DI, while B/C/D + SI were the most effective for PFS (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf). We then performed a meta-regression analysis on all studies that could be connected within the network. For both OS and PFS, the interaction parameter β included 0, indicating no significant interaction between treatment outcomes and tumor histology (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf).

Figure 8 Forest plot for SQ NSCLC (all treatments compared with regimen B and C). CI, credible interval; CT, chemotherapy; DI, dual immunotherapy; EGFR, epidermal growth factor receptor; HR, hazard ratio; MET, mesenchymal-epithelial transition factor; NSCLC, non-small cell lung cancer; OS, overall survival; PFS, progression-free survival; SI, single immunotherapy; SQ, squamous; TIGIT, T cell immunoreceptor with Ig and ITIM domains; VEGF, vascular endothelial growth factor.

A total of 22 studies [(available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) (A3,A21-22,A27-31,A34,A36-38,A43-A48,A51-A53,A55,A57-A58)] provided OS data, and 23 studies [(available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-1.docx) (A3,A21-22,A27-28,A30-32,A34-38,A43-44,A46-A48,A51-A53,A57-A58,A60-61)] provided PFS data stratified by PD-L1 expression levels. Figure 9 is a forest plot demonstrating all treatments compared to CT across the PD-L1 subgroups, with complete results and SUCRA values presented in available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf. Among patients with PD-L1 expression <1%, DI (SUCRA =0.86) showed the greatest OS benefit, followed by CT + DI (SUCRA =0.77), while CT + SI + Bevacizumab (SUCRA =0.97) showed the greatest PFS benefit. For patients with PD-L1 ≥1%, CT + SI (OS SUCRA =0.86; PFS SUCRA =0.82) and CT + SI + Bevacizumab (OS SUCRA =0.78; PFS SUCRA =0.99) were the most effective therapies. In the 1%-49% subgroup, aside from CT + SI + anti-IL-1β mAb (highest PFS SUCRA =0.91), CT + SI provided the best OS (SUCRA =0.74) and PFS (SUCRA =0.82). For patients with PD-L1 ≥50%, CT + SI + Bevacizumab (OS SUCRA =0.77; PFS SUCRA =0.87) and CT + SI (OS SUCRA =0.76; PFS SUCRA =0.70) were the most effective therapies, excluding CT + SI + anti-IL-1β mAb (the highest PFS SUCRA value =0.95). A final meta-regression was conducted across PD-L1 subgroups (<1%, 1–49%, and ≥50%). For both OS and PFS, the interaction parameter β does not include 0, indicating a significant interaction between PD-L1 expression and treatment effect (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf).

Figure 9 Forest plot across the PD-L1 subgroups (all treatments compared with CT). CI, credible interval; CT, chemotherapy; DI, dual immunotherapy; HR, hazard ratio; OS, overall survival; PD-L1, programmed cell death ligand 1; PFS, progression-free survival; SI, single immunotherapy.

Measures for publication bias

The comparison-adjusted funnel plot suggested no substantial small-study effects (Egger test P=0.5) (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf).

Sensitivity analysis

Both the random-effects model and the fixed-effects model were applied. The results were not substantially different, and the choice of model did not meaningfully affect the interpretation of our results.

Inconsistency

We conducted an assessment of global inconsistency and compared the fit between the consistency and inconsistency models. The consistency model of this NMA demonstrated a fit equal to or better than the inconsistency model (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf). We further assessed local inconsistency using node-splitting (available online: https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf, and https://cdn.amegroups.cn/static/public/jtd-2025-aw-2087-2.pdf). Local inconsistency was considered present if P<0.05, and we considered downgrading the quality of evidence according to GRADE criteria.

Discussion

This NMA included 61 RCTs, one-third of which compared CT + SI [CT plus anti-PD-(L)1 with anti-PD-(L)1 maintenance] with CT alone. CT + SI was associated with statistically significant improvements in OS and PFS relative to CT alone, supported by moderate-certainty evidence. Based on the NMA, DI [dual anti-PD-(L)1 plus anti-CTLA-4, administered until disease progression, unacceptable toxicity, or up to 2 years], CT + DI m:SI (CT + DI with SI maintenance), CT + SI + Bevacizumab [CT + anti-PD-(L)1 plus Bevacizumab, MT: anti-PD-(L)1 plus Bevacizumab], and CT + DI (CT + DI with DI maintenance) emerged as the most effective options in the overall population. Compared with CT alone, all four treatments significantly improved OS with moderate-certainty evidence and PFS, except for CT + DI m:SI, which improved OS (HR 0.61, 95% CI: 0.44 to 0.86) but did not improve PFS (HR 0.81, 95% CI: 0.52 to 1.28, low-certainty evidence). CT + SI + Bevacizumab ranked highest for PFS, although this finding was supported by low-certainty evidence. In contrast, DI (with or without CT) was associated with improved PFS versus CT, supported by moderate-certainty evidence. Based on SUCRA values for safety, DI ranked most favorable (SUCRA =0.77), followed by CT + DI m:SI (SUCRA =0.53), CT + DI (SUCRA =0.25), and CT + SI + Bevacizumab (SUCRA =0.23). Taken together, DI appears to provide the most favorable balance of efficacy and safety in the overall population. This conclusion is consistent with Alifu et al. 2023 (51), who reported durable OS and PFS benefits for DI compared with CT. By contrast, Chai et al. 2022 (52) conducted an NMA of 19 trials evaluating the efficacy and safety of first-line therapies in patients with advanced NSQ NSCLC lacking driver-gene mutations. In that analysis, DI was estimated to be statistically comparable to CT for both OS and PFS. The discrepancy arises because their analysis did not include key studies, such as the updated results of CheckMate 227 (53), which demonstrated that DI improved both OS and PFS compared with CT, irrespective of PD-L1 expression levels.

We additionally performed subgroup analyses stratified by histology and PD-L1 expression. In NSQ NSCLC, A + DI [Pemetrexed + Carboplatin or Cisplatin + dual anti-PD-(L)1 plus anti-CTLA-4, MT: Pemetrexed + dual anti-PD-(L)1 plus anti-CTLA-4] (OS SUCRA =0.76; PFS SUCRA =0.66), DI (OS SUCRA =0.75), and A + SI [Pemetrexed + Carboplatin or Cisplatin + anti-PD-(L)1, MT: Pemetrexed + anti-PD-(L)1] (OS SUCRA =0.74; PFS SUCRA =0.80) were identified as the most effective therapies. All three regimens significantly improved OS relative to regimen A (Pemetrexed + Carboplatin or Cisplatin, MT: Pemetrexed), with moderate certainty in the effect estimates. In addition, A + DI and A + SI improved PFS versus A, supported by moderate-certainty evidence. Direct head-to-head evidence comparing DI and regimen A for PFS is currently lacking. When jointly considering efficacy and safety, DI remains a preferred option for NSQ NSCLC, despite the absence of direct PFS comparisons against regimen A. In SQ NSCLC, DI, B/C + SI [B:Paclitaxel + Carboplatin or Cisplatin or C:Gemcitabine + Carboplatin or Cisplatin + anti-PD-(L)1, MT: anti-PD-(L)1], and B + DI [Paclitaxel + Carboplatin or Cisplatin + dual anti-PD-(L)1 plus anti-CTLA-4, MT: dual anti-PD-(L)1 plus anti-CTLA-4] were the most effective regimens for improving OS, while B/C/D + SI (D:Nabpaclitaxel + Carboplatin) were the most effective for improving PFS. Considering both efficacy and safety, DI remained the preferred regimen and showed an OS benefit versus regimen C with moderate-certainty evidence. However, evidence for PFS with DI in SQ NSCLC—and for direct DI versus regimen B comparisons—remains limited. B + SI is also a favorable option, supported by high-certainty evidence for both OS and PFS compared with B alone.

In the PD-L1 <1% subgroup, DI ranked highest for OS (SUCRA =0.86), followed by CT + DI (SUCRA =0.77). CT + SI + Bevacizumab (SUCRA =0.97) ranked highest for PFS. In patients with PD-L1 ≥1%, CT + SI (OS SUCRA =0.86) ranked highest for OS, whereas CT + SI + Bevacizumab (SUCRA =0.97) ranked highest for PFS. In the PD-L1 1–49% subgroup, CT + SI ranked highest for both OS (SUCRA =0.74) and PFS (SUCRA =0.82). For PD-L1 ≥50%, CT + SI + Bevacizumab (OS SUCRA =0.77; PFS SUCRA =0.87) and CT + SI (OS SUCRA =0.76; PFS SUCRA =0.70) were the most effective regimens. However, the wide 95% CI for OS in the comparison of CT + SI + Bevacizumab versus CT (HR =0.6, 95% CI: 0.40 to 0.93) suggests greater precision and confidence in the OS benefit of CT + SI (compared to CT, HR =0.63, 95% CI: 0.56 to 0.72). In summary, for PD-L1 <1%, DI is recommended due to its superior OS benefit. However, its PFS benefit is not ideal (compared to CT, HR =0.7, 95% CI: 0.50 to 0.99), and CT + DI may be a suitable alternative. This finding is consistent with Peters et al. 2025 (54), who reported long-term benefit with DI (with or without CT) in metastatic NSCLC with PD-L1 <1%. For PD-L1 1%, CT + SI is the recommended treatment. Collectively, these results suggest that anti-PD-(L)1 monotherapy has limited efficacy in PD-L1<1%, whereas combining anti-PD-(L)1 with anti-CTLA-4 can confer clinical meaningful benefit. Adding CT to DI may improve PFS, but it does not appear to materially increase OS. With increasing PD-L1 expression, anti-PD-(L)1 therapies can act independently of anti-CTLA-4 to achieve therapeutic efficacy.

In summary, DI has substantial potential and demonstrates clinical benefits in patients with advanced driver-gene wild-type NSCLC, regardless of histologic subtype, particularly among those with PD-L1 expression <1%. A key limitation is the unsatisfactory PFS benefit. In NSQ NSCLC with PD-L1<1%, adding regimen A to DI-based therapy may improve PFS. By contrast, in SQ NSCLC with PD-L1 <1%, adding regimen B or C did not appear to improve PFS. In NSQ NSCLC with PD-L1 ≥1%, A + SI may represent an optimal option. Similarly, in SQ NSCLC with PD-L1 ≥1%, B + SI or C + SI may be the most suitable strategies.

The advent of immunotherapy has transformed the therapeutic landscape of cancer. There is a growing interest in using immunotherapies in (neo)adjuvant, maintenance, and even monotherapy settings (55,56). Over recent years, mounting evidence suggests that cytotoxic CT agents have various immunostimulatory effects, leading to enhanced therapeutic responses when combined with immunotherapy. CT is known to augment antitumor immunogenicity through several complementary mechanisms, including increased tumor infiltration by T cells, enhanced activity and maturation of APCs, downregulation of immunosuppressive Tregs, reduction in myeloid-derived suppressor cells (MDSCs) at tumor sites, and induction of immunogenic cell death (ICD). Some agents, including platinum-based CT, have also been shown to increase tumor PD-L1 expression, which has been associated with improved responses to PD-1/PD-L1 blockade. Through these mechanisms, cytotoxic drugs exert synergistic effects on immunotherapies, converting a ‘cold’ immune microenvironment into a ‘hot’ one with an inflammatory profile (57,58). Consistent with this biology, our NMA suggests that CT plus anti-PD-(L)1 therapy is effective in patients with PD-L1 ≥1%. In contrast, in PD-L1 <1%, anti-PD-(L)1 therapy may require co-administration with anti-CTLA-4 mAb to achieve clinical benefit. CTLA-4 (CD152) is a member of the B7/CD28 family that negatively regulates T cell function. Both CTLA-4 and PD-1 act as brakes in immune regulatory pathways. It is thought that CTLA-4 signaling primarily affects T cell priming and PD-1 signaling mainly impacts the effector stage of T cell activity. Dual blockade of CTLA-4 and PD-1 pathways may restore immune priming and enable effective T cell responses (59). Thus, anti-PD-(L)1 and anti-CTLA-4 immunotherapies have distinct but complementary mechanisms that together help reactivate antitumor immunity (54).

Bevacizumab is a humanized mAb that inhibits angiogenesis and tumor vascular growth (60). Our results suggest that adding bevacizumab to CT + SI may improve PFS versus CT alone, however, the certainty of evidence is low. We also identified other candidate strategies, including the addition of an anti-IL-1β mAb to CT + SI. Nevertheless, further evidence is needed to confirm their efficacy.

Limitations

Several limitations of this review should be acknowledged. Firstly, the included studies were conducted over an extended time period, and many studies did not report PD-L1 expression levels. Consequently, insufficient data were available to perform subgroup analyses based on PD-L1 expression in some cases. Hence, certain results should be interpreted with caution. Secondly, most comparisons in the NMA were informed by a single trial, resulting in a lack of head-to-head evidence for many treatment comparisons. Thirdly, the networks for some outcomes were not fully connected, limiting comparisons to treatments within the same sub-network.

Conclusions

In NSQ NSCLC with PD-L1 <1%, DI ranked highest for OS, and adding CT backbone regimen A may improve PFS. In SQ NSCLC with PD-L1 <1%, DI likewise ranked highest for OS, whereas adding CT backbones B or C did not show a clear PFS benefit. In NSQ NSCLC with PD-L1 ≥1%, A + SI may be the most appropriate regimen; in SQ NSCLC with PD-L1 ≥1%, B + SI or C + SI may be optimal. Overall, these findings provide comparative evidence to support individualized selection of first-line plus maintenance strategies aimed at maximizing survival benefits in specific patient subgroups. However, additional evidence is needed to accurately compare continuous dual immunotherapies with other MT-containing regimens.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the PRISMA-NMA reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-aw-2087/rc

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-aw-2087/prf

Funding: This research was supported by Scientific Research Project of the Health and Wellness Industry in Gansu Province (No. GSWSKY2024-35).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-aw-2087/coif). All authors report that this research was supported by Scientific Research Project of the Health and Wellness Industry in Gansu Province (No. GSWSKY2024-35). The authors have no other conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Jeon H, Wang S, Song J, et al. Update 2025: Management of Non‑Small-Cell Lung Cancer. Lung 2025;203:53. [Crossref] [PubMed]
2. Wang H, Niu X, Jin Z, et al. Immunotherapy resistance in non-small cell lung cancer: from mechanisms to therapeutic opportunities. J Exp Clin Cancer Res 2025;44:250. [Crossref] [PubMed]
3. Jänne PA, Wang X, Socinski MA, et al. Randomized phase II trial of erlotinib alone or with carboplatin and paclitaxel in patients who were never or light former smokers with advanced lung adenocarcinoma: CALGB 30406 trial. J Clin Oncol 2012;30:2063-9.
4. Goldstraw P, Chansky K, Crowley J, et al. The IASLC Lung Cancer Staging Project: Proposals for Revision of the TNM Stage Groupings in the Forthcoming (Eighth) Edition of the TNM Classification for Lung Cancer. J Thorac Oncol 2016;11:39-51. [Crossref] [PubMed]
5. Okamoto I, Nokihara H, Nomura S, et al. Comparison of Carboplatin Plus Pemetrexed Followed by Maintenance Pemetrexed With Docetaxel Monotherapy in Elderly Patients With Advanced Nonsquamous Non-Small Cell Lung Cancer: A Phase 3 Randomized Clinical Trial. JAMA Oncol 2020;6:e196828. [Crossref] [PubMed]
6. De Lucia A, Mazzotti L, Gaimari A, et al. Non-small cell lung cancer and the tumor microenvironment: making headway from targeted therapies to advanced immunotherapy. Front Immunol 2025;16:1515748. [Crossref] [PubMed]
7. West H, McCleod M, Hussein M, et al. Atezolizumab in combination with carboplatin plus nab-paclitaxel chemotherapy compared with chemotherapy alone as first-line treatment for metastatic non-squamous non-small-cell lung cancer (IMpower130): a multicentre, randomised, open-label, phase 3 trial. Lancet Oncol 2019;20:924-37. [Crossref] [PubMed]
8. National Comprehensive Cancer Network (NCCN). NCCN clinical practice guidelines in oncology (NCCN guidelines®): non-small cell lung cancer. Version 3.2025. Available online: https://www.nccn.org/
9. Chang Q, Zhang Y, Xu J, et al. First-line pemetrexed/carboplatin or cisplatin/bevacizumab compared with paclitaxel/carboplatin/bevacizumab in patients with advanced non-squamous non-small cell lung cancer with wild-type driver genes: A real-world study in China. Thorac Cancer 2019;10:1043-50. [Crossref] [PubMed]
10. Lau SCM, Pan Y, Velcheti V, et al. Squamous cell lung cancer: Current landscape and future therapeutic options. Cancer Cell 2022;40:1279-93. [Crossref] [PubMed]
11. Chen R, Manochakian R, James L, et al. Emerging therapeutic agents for advanced non-small cell lung cancer. J Hematol Oncol 2020;13:58. [Crossref] [PubMed]
12. Paz-Ares L, Ciuleanu TE, Cobo M, et al. First-line nivolumab plus ipilimumab combined with two cycles of chemotherapy in patients with non-small-cell lung cancer (CheckMate 9LA): an international, randomised, open-label, phase 3 trial. Lancet Oncol 2021;22:198-211. [Crossref] [PubMed]
13. Reck M. Beyond the TRIBUTE trial: integrating HER1/EGFR tyrosine kinase inhibitors with chemotherapy in advanced NSCLC. Future Oncol 2006;2:47-51. [Crossref] [PubMed]
14. Smith IE, O'Brien ME, Talbot DC, et al. Duration of chemotherapy in advanced non-small-cell lung cancer: a randomized trial of three versus six courses of mitomycin, vinblastine, and cisplatin. J Clin Oncol 2001;19:1336-43. [Crossref] [PubMed]
15. Socinski MA, Schell MJ, Peterman A, et al. Phase III trial comparing a defined duration of therapy versus continuous therapy followed by second-line therapy in advanced-stage IIIB/IV non-small-cell lung cancer. J Clin Oncol 2002;20:1335-43. [Crossref] [PubMed]
16. Rossi A, Chiodini P, Sun JM, et al. Six versus fewer planned cycles of first-line platinum-based chemotherapy for non-small-cell lung cancer: a systematic review and meta-analysis of individual patient data. Lancet Oncol 2014;15:1254-62. [Crossref] [PubMed]
17. Genestreti G, Di Battista M, Cavallo G, et al. Maintenance therapy in non-small cell lung cancer. Expert Rev Anticancer Ther 2015;15:839-46. [Crossref] [PubMed]
18. Joshi M, Jiang Y, Belani CP. Maintenance therapy for advanced non-small-cell lung cancer: switch versus continuation. Expert Opin Pharmacother 2012;13:685-97. [Crossref] [PubMed]
19. Mok TS, Ramalingam SS. Maintenance therapy in nonsmall-cell lung cancer: a new treatment paradigm. Cancer 2009;115:5143-54. [Crossref] [PubMed]
20. Ciuleanu T, Brodowicz T, Zielinski C, et al. Maintenance pemetrexed plus best supportive care versus placebo plus best supportive care for non-small-cell lung cancer: a randomised, double-blind, phase 3 study. Lancet 2009;374:1432-40. [Crossref] [PubMed]
21. Wu YL, Kim JH, Park K, et al. Efficacy and safety of maintenance erlotinib in Asian patients with advanced non-small-cell lung cancer: a subanalysis of the phase III, randomized SATURN study. Lung Cancer 2012;77:339-45. [Crossref] [PubMed]
22. Johnson BE, Kabbinavar F, Fehrenbacher L, et al. ATLAS: randomized, double-blind, placebo-controlled, phase IIIB trial comparing bevacizumab therapy with or without erlotinib, after completion of chemotherapy, with bevacizumab for first-line treatment of advanced non-small-cell lung cancer. J Clin Oncol 2013;31:3926-34. [Crossref] [PubMed]
23. Owonikoko TK, Ramalingam SS, Belani CP. Maintenance therapy for advanced non-small cell lung cancer: current status, controversies, and emerging consensus. Clin Cancer Res 2010;16:2496-504. [Crossref] [PubMed]
24. Zhou F, Jiang T, Ma W, et al. The impact of clinical characteristics on outcomes from maintenance therapy in non-small cell lung cancer: A systematic review with meta-analysis. Lung Cancer 2015;89:203-11. [Crossref] [PubMed]
25. Zhang C, Huang C, Wang J, et al. Maintenance or Consolidation Therapy for Non--Small-Cell Lung Cancer: A Meta-Analysis Involving 5841 Subjects. Clin Lung Cancer 2015;16:e15-23. [Crossref] [PubMed]
26. Tan PS, Lopes G, Acharyya S, et al. Bayesian network meta-comparison of maintenance treatments for stage IIIb/IV non-small-cell lung cancer (NSCLC) patients with good performance status not progressing after first-line induction chemotherapy: results by performance status, EGFR mutation, histology and response to previous induction. Eur J Cancer 2015;51:2330-44. [Crossref] [PubMed]
27. Tan PS, Bilger M, de Lima Lopes G, et al. Meta-analysis of first-line therapies with maintenance regimens for advanced non-small-cell lung cancer (NSCLC) in molecularly and clinically selected populations. Cancer Med 2017;6:1847-60. [Crossref] [PubMed]
28. Xiu-xia L, Ya Z, Yao-long C, et al. The reporting characteristics and methodological quality of Cochrane reviews about health policy research. Health Policy 2015;119:503-10. [Crossref] [PubMed]
29. Ge L, Hou L, Yang Q, et al. A systematic review and network meta-analysis protocol of adjuvant chemotherapy regimens for resected gastric cancer. Medicine (Baltimore) 2019;98:e14478. [Crossref] [PubMed]
30. Jansen JP, Crawford B, Bergman G, et al. Bayesian meta-analysis of multiple treatment comparisons: an introduction to mixed treatment comparisons. Value Health 2008;11:956-64. [Crossref] [PubMed]
31. Guyatt GH, Oxman AD, Kunz R, et al. What is "quality of evidence" and why is it important to clinicians? BMJ 2008;336:995-8. [Crossref] [PubMed]
32. Hutton B, Salanti G, Caldwell DM, et al. The PRISMA extension statement for reporting of systematic reviews incorporating network meta-analyses of health care interventions: checklist and explanations. Ann Intern Med 2015;162:777-84. [Crossref] [PubMed]
33. Eisenhauer EA, Therasse P, Bogaerts J, et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer 2009;45:228-47. [Crossref] [PubMed]
34. National Cancer Institute. National Cancer Institute Common Terminology Criteria for Adverse Events v4.0. NCI, NIH, DHHS 29 May 2009; Vol. NIH publication # 09-7473. Available online: https://evs.nci.nih.gov/ftp1/CTCAE/CTCAE_4.03/Archive/CTCAE_4.0_2009-05-29_QuickReference_8.5x11.pdf
35. EndNote X9 [Software]. Clarivate Analytics. EndNote X9. Boston (MA): Clarivate Analytics, 2018.
36. Microsoft Excel 2013 [Software]. Microsoft Corp, Redmond, WA. Available online: www.microsoft.com.
37. Tierney JF, Stewart LA, Ghersi D, et al. Practical methods for incorporating summary time-to-event data into meta-analysis. Trials 2007;8:16. [Crossref] [PubMed]
38. Higgins JP, Green S, editors. Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0 (updated March 2011). The Cochrane Collaboration, 2011. Available online: handbook.cochrane.org
39. Puhan MA, Schünemann HJ, Murad MH, et al. A GRADE Working Group approach for rating the quality of treatment effect estimates from network meta-analysis. BMJ 2014;349:g5630. [Crossref] [PubMed]
40. Brignardello-Petersen R, Bonner A, Alexander PE, et al. Advances in the GRADE approach to rate the certainty in estimates from a network meta-analysis. J Clin Epidemiol 2018;93:36-44. [Crossref] [PubMed]
41. Brignardello-Petersen R, Mustafa RA, Siemieniuk RAC, et al. GRADE approach to rate the certainty from a network meta-analysis: addressing incoherence. J Clin Epidemiol 2019;108:77-85. [Crossref] [PubMed]
42. Brignardello-Petersen R, Murad MH, Walter SD, et al. GRADE approach to rate the certainty from a network meta-analysis: avoiding spurious judgments of imprecision in sparse networks. J Clin Epidemiol 2019;105:60-7. [Crossref] [PubMed]
43. Zhao Y, Feng HM, Tian J, et al. Stereotactic body radiation therapy versus more fractionated radical radiotherapy for adults with stage I/II non-small cell lung cancer: a systematic review and network meta‐analysis. Cochrane Database Syst Rev 2022;2022:CD014744.
44. Stata [Software]. Stata, Version 14.2. College Station, TX, USA: StataCorp, 2015. Available online: www.stata.com
45. Ge L, Tian JH, Li L, et al. Mesh fixation methods in open inguinal hernia repair: a protocol for network meta-analysis and trial sequential analysis of randomised controlled trials. BMJ Open 2015;5:e009369. [Crossref] [PubMed]
46. Van Valkenhoef. Network Meta-analysis using Bayesian methods. 2016. Available online: https://rdrr.io/cran/gemtc/
47. R 2017 [Software]. R Foundation for Statistical Computing R: a language and environment for statistical computing. Version 3.4.2. Vienna, Austria: R Foundation for Statistical Computing, 2017. Available online: www.R-project.org.
48. Brooks SP, Gelman A. General methods for monitoring convergence of iterative simulations J Comput Graph Stat 1998;7:434-55.
49. Salanti G, Ades AE, Ioannidis JP. Graphical methods and numerical summaries for presenting results from multiple-treatment meta-analysis: an overview and tutorial. J Clin Epidemiol 2011;64:163-71. [Crossref] [PubMed]
50. Deeks JJ, Higgins JPT, Altman DG, editors. Chapter 10: Analysing data and undertaking meta-analyses. In: Higgins JPT, Thomas J, Chandler J, et al., editors. Cochrane Handbook for Systematic Reviews of Interventions version 6.0 (updated July 2019). Cochrane, 2019. In: Cochrane Handbook for Systematic Reviews of Interventions version 6.0 (updated July 2019). Available online: www.training.cochrane.org/handbook, 2019.
51. Alifu M, Tao M, Chen X, et al. Checkpoint inhibitors as dual immunotherapy in advanced non-small cell lung cancer: a meta-analysis. Front Oncol 2023;13:1146905. [Crossref] [PubMed]
52. Chai Y, Wu X, Bai H, et al. Combined Immunotherapy with Chemotherapy versus Bevacizumab with Chemotherapy in First-Line Treatment of Driver-Gene-Negative Non-Squamous Non-Small Cell Lung Cancer: An Updated Systematic Review and Network Meta-Analysis. J Clin Med 2022;11:1655. [Crossref] [PubMed]
53. Brahmer JR, Lee JS, Ciuleanu TE, et al. Five-Year Survival Outcomes With Nivolumab Plus Ipilimumab Versus Chemotherapy as First-Line Treatment for Metastatic Non-Small-Cell Lung Cancer in CheckMate 227. J Clin Oncol 2023;41:1200-12. [Crossref] [PubMed]
54. Peters S, Paz-Ares LG, Reck M, et al. Long-Term Survival Outcomes With First-Line Nivolumab Plus Ipilimumab-Based Treatment in Patients With Metastatic NSCLC and Tumor Programmed Death-Ligand 1 Lower Than 1%: A Pooled Analysis. J Thorac Oncol 2025;20:94-108. [Crossref] [PubMed]
55. Johnson DB, Nebhan CA, Moslehi JJ, et al. Immune-checkpoint inhibitors: long-term implications of toxicity. Nat Rev Clin Oncol 2022;19:254-67. [Crossref] [PubMed]
56. Grivas P, Monk BJ, Petrylak D, et al. Immune Checkpoint Inhibitors as Switch or Continuation Maintenance Therapy in Solid Tumors: Rationale and Current State. Target Oncol 2019;14:505-25. [Crossref] [PubMed]
57. Bailly C, Thuru X, Quesnel B. Combined cytotoxic chemotherapy and immunotherapy of cancer: modern times. NAR Cancer 2020;2:zcaa002. [Crossref] [PubMed]
58. Principe DR, Kamath SD, Korc M, et al. The immune modifying effects of chemotherapy and advances in chemo-immunotherapy. Pharmacol Ther 2022;236:108111. [Crossref] [PubMed]
59. Seidel JA, Otsuka A, Kabashima K. Anti-PD-1 and Anti-CTLA-4 Therapies in Cancer: Mechanisms of Action, Efficacy, and Limitations. Front Oncol 2018;8:86. [Crossref] [PubMed]
60. Garcia J, Hurwitz HI, Sandler AB, et al. Bevacizumab (Avastin®) in cancer treatment: A review of 15 years of clinical experience and future outlook. Cancer Treat Rev 2020;86:102017. [Crossref] [PubMed]