The Promise of Chemotherapy‐Free Strategies in Advanced Driver‐Negative NSCLC: A Systematic Review and Network Meta‐Analysis of Antiangiogenic Combination Therapies

ABSTRACT

Background

Antiangiogenic combination therapy—antiangiogenic agents combined with immune checkpoint inhibitors and/or chemotherapy—has become an important treatment strategy for advanced driver‐negative non‐small cell lung cancer (NSCLC). We conducted a network meta‐analysis to compare efficacy and safety and identify optimal antiangiogenic combinations.

Methods

We searched PubMed, Embase, Web of Science, and Cochrane Library for randomized controlled trials (RCTs) that evaluated antiangiogenic combination therapies. Primary outcomes were progression‐free survival (PFS), overall survival (OS), objective response rate (ORR), and incidence of grade ≥ 3 treatment‐related adverse events (TRAEs).

Results

Nine treatment regimens comprising 5954 patients were included. The network meta‐analysis indicated that the chemotherapy‐free regimen of sintilimab + anlotinib achieved the greatest progression‐free survival (PFS) benefit, compared with chemotherapy (HR = 0.39, 95% CI 0.23–0.67), and the lowest incidence of grade ≥ 3 treatment‐related adverse events (TRAEs) (RR = 0.57, 95% CI 0.32–0.95). Recombinant human endostatin (Endostar) + chemotherapy provided the largest overall survival (OS) benefit vs. chemotherapy (HR = 0.46, 95% CI 0.36–0.58). The triplet regimen of atezolizumab, bevacizumab, and chemotherapy yielded the largest improvement in objective response rate (ORR) vs. bevacizumab + chemotherapy (OR = 1.90, 95% CI 1.40–2.60). Across most subgroup analyses, regimens combining immunotherapy, an antiangiogenic agent, and chemotherapy conferred the greatest PFS and OS benefits.

Conclusions

This network meta‐analysis demonstrates that antiangiogenic combinations improve outcomes in driver‐negative advanced NSCLC. Endostar + chemotherapy offers the greatest OS benefit, atezolizumab–bevacizumab–chemotherapy improves ORR, and sintilimab–anlotinib provides superior PFS with lower toxicity. Treatment should be tailored based on clinical factors, with further validation in multiethnic trials.

1. Introduction

Non‐small cell lung cancer (NSCLC) accounts for more than 85% of lung cancers and remains a leading cause of cancer‐related mortality worldwide. In 2022, approximately 2.48 million new lung cancer cases occurred globally, with more than 1.8 million deaths. China accounted for approximately 1.06 million cases and over 700,000 deaths, representing the largest national disease burden [1, 2, 3]. Among patients with advanced NSCLC, those without actionable driver mutations (e.g., EGFR, ALK, ROS1) have historically received platinum‐based doublet chemotherapy as first‐line treatment. However, the median overall survival (OS) remains 12–15 months [4, 5], and chemotherapy‐related adverse events (e.g., myelosuppression, gastrointestinal toxicities) significantly impair quality of life [6]. In recent years, immune checkpoint inhibitors (ICIs) combined with chemotherapy have become a standard first‐line treatment. Nonetheless, clinical benefit remains limited in PD‐L1–negative and never‐smoker populations, highlighting the need to optimize treatment strategies [7].

Antiangiogenic agents modulate the tumor microenvironment by inhibiting VEGF‐mediated signaling, alleviating immunosuppression, and enhancing T‐cell infiltration [8]. Bevacizumab, an anti‐VEGF monoclonal antibody, combined with chemotherapy, is an established regimen (e.g., ECOG 4599) [9]. However, the phase III IMpower150 trial first demonstrated that a quadruple regimen—atezolizumab, bevacizumab, carboplatin, and paclitaxel—significantly improved overall survival (OS) (median 19.5 months), with notable benefits in patients with liver metastases and EGFR‐resistant disease [10, 11]. A study presented at the 2023 World Conference on Lung Cancer (WCLC) further suggested that recombinant human endostatin (Endostar), combined with a PD‐1 inhibitor and chemotherapy, achieved a median progression‐free survival (PFS) of 16.7 months and an objective response rate (ORR) of 57.7% [12], supporting the value of antiangiogenic agents in combination therapy.

An ongoing debate concerns the incremental value of adding antiangiogenic agents to immunotherapy‐based chemotherapy regimens. Meta‐analyses suggest that chemo‐immunotherapy provides a greater overall survival (OS) benefit (HR = 0.73) compared to bevacizumab + chemotherapy (HR = 0.89); however, the difference appears attenuated in PD‐L1–negative and never‐smoker populations [13]. Notably, the IMpower150 trial showed a significantly higher objective response rate (ORR) with the quadruple regimen than with the triple regimen [10, 11]. Mechanistic studies provide a rationale for synergy: Low‐dose antiangiogenic therapy promotes vascular normalization and augments CD8+ T‐cell activity, often at doses lower than those used in monotherapy to balance efficacy and toxicity [14, 15]. Consequently, selected subgroups (e.g., never‐smokers, PD‐L1–negative patients) may derive greater benefit from triple‐ or quadruple‐agent regimens that include an antiangiogenic agent.

For patients unable to tolerate or unwilling to receive chemotherapy, a chemotherapy‐free strategy—immunotherapy combined with an antiangiogenic agent—has emerged as an area of active investigation. In 2024, a case report in Frontiers in Oncology described a PD‐L1–low advanced NSCLC patient with marked tumor shrinkage and no severe toxicity after receiving a PD‐L1 inhibitor + recombinant human endostatin [16]. Another case involving malignant pleural effusion (MPE) reported reduced effusion and durable clinical benefit with the same chemotherapy‐free regimen [17]. A single‐center experience from Shanghai Chest Hospital, evaluating a domestic dual‐agent combination (immunotherapy + an antiangiogenic targeted therapy), reported improved quality of life and a median overall survival (OS) exceeding 20 months [18]. However, the generalizability of chemotherapy‐free regimens (e.g., to PD‐L1–high patients), their long‐term safety, and their comparative efficacy vs. standard chemo‐immunotherapy must be validated in large, controlled studies.

Recombinant human endostatin (Endostar) is a domestically developed, multi‐target antiangiogenic agent in China with pleiotropic mechanisms of action. A study referred to as “ENPOWER” reportedly showed that endostatin + a PD‐1 inhibitor and chemotherapy achieved an objective response rate (ORR) of 57.7%, comparable to the ORR with the bevacizumab‐containing quadruple regimen in IMpower150 (63.5%), and a longer median progression‐free survival (PFS) (16.7 vs. 8.3 months) [10, 12]. Regarding safety, Endostar has been reported to show a lower incidence of certain adverse events than bevacizumab and to lack specific bleeding‐risk contraindications, with no substantial additive toxicity when combined with chemotherapy [19, 20]. Currently, Endostar is approved for use in non–small‐cell lung cancer in China. However, evidence in non‐Asian populations remains limited, which constrains broader international adoption. Furthermore, the optimal dosing when combined with PD‐1 inhibitors warrants investigation (e.g., Low‐dose enhancement strategies) [14].

A schematic overview of the mechanisms of action of the PD‐1/PD‐L1 inhibitors and antiangiogenic agents evaluated in this study, as well as their interactions, is depicted in Figure 1. Briefly, antiangiogenic therapy can normalize tumor vasculature, alleviate immunosuppression within the tumor microenvironment, and thereby enhance the delivery and efficacy of immune checkpoint inhibitors and cytotoxic chemotherapy. This mechanistic framework underpins the clinical development of antiangiogenic combination regimens in driver‐negative advanced NSCLC.

FIGURE 1.

Mechanistic rationale for combining antiangiogenic therapy with PD‐1/PD‐L1 blockade in driver‐negative advanced NSCLC. (A) Engagement of PD‐1 on CD8⁺ T cells by PD‐L1 on tumor cells induces T‐cell exhaustion, whereas anti–PD‐1/PD‐L1 antibodies restore cytotoxic activity and tumor cell killing. (B) Antiangiogenic agents targeting VEGF signaling induce regression of tumor blood vessels, reduce perfusion, and promote cancer cell death mediated by ischemia and hypoxia. (C) When appropriately dosed, antiangiogenic therapy normalizes the tumor vasculature and enhances antitumor immune responses, resulting in synergistic tumor killing.

This study employs a network meta‐analysis (NMA) to integrate direct and indirect evidence and address three questions: (1) How do chemotherapy‐free regimens (immunotherapy + an antiangiogenic agent) compare with chemotherapy‐containing regimens? (2) What is the incremental value of adding an antiangiogenic agent to chemo‐immunotherapy? (3) How do Endostar‐ and bevacizumab‐containing combinations compare in efficacy and safety? The findings aim to inform evidence‐based, personalized treatment for driver‐negative advanced NSCLC and provide a rationale for the clinical application of domestically developed antiangiogenic agents.

2. Materials and Methods

This study adhered to the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses Protocols (PRISMA‐P) guidelines [21]. After comparing frequentist and Bayesian frameworks for network meta‐analysis, we selected a Bayesian NMA to address the complexity of multiple‐treatment comparisons [22]. The study protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO); ID: CRD420251088297.

3. Search Strategy and Inclusion Criteria

We searched PubMed, Embase, the Cochrane Library, and Web of Science from inception to January 31, 2025, for phase II/III randomized controlled trials (RCTs). All included studies were restricted to English‐language publications. The search employed MeSH/Emtree terms and keywords related to non–small‐cell lung carcinoma, vascular endothelial growth factor (VEGF/VEGFA), angiogenesis inhibitors, and combination therapy; full strategies are provided in Supporting Information 1. Eligible studies enrolled adults with advanced NSCLC and compared antiangiogenic combination regimens either against each other or against non–antiangiogenic regimens. Trials identified through registries were screened, but only studies reporting outcome data were included. Studies were included if they reported at least one primary outcome (progression‐free survival [PFS] or overall survival [OS]); secondary outcomes included objective response rate (ORR) and grade ≥ 3 treatment‐related adverse events (TRAEs). We excluded review articles, case reports, guidelines, animal studies, single‐arm trials, cohort studies, incomplete reports, duplicate publications, and meta‐analyses. Additionally, we hand‐searched the reference lists of included trials to ensure comprehensive capture of eligible studies. When multiple articles reported the same trial, we used the most recent or most complete report; disagreements regarding study selection were adjudicated by a third reviewer.

3.1. Data Extraction

From each eligible study, we extracted (i) study characteristics (author/title, publication year, study phase, design), (ii) patient characteristics, (iii) treatment regimens and comparators, and (iv) outcomes (progression‐free survival [PFS], overall survival [OS], objective response rate [ORR], and grade ≥ 3 treatment‐related adverse events [TRAEs]) (Table 1).

TABLE 1.

Baseline characteristics of the included studies.

Study	Year	Phase	Stage	Histology	N	E/C	Network comparator (s)	Medians of OS months	OS, HR (95% CI)	Medians of PFS months	PFS, HR (95% CI)	ORR (%)	≥ 3 TRAEs (%)
APPLE	2023	III	IV	Non‐squamous	287	143	Atezo + Bev + CT	28	0.99 (0.71–1.38)	9.3	0.97 (0.75–1.25)	67.80%	56.86%
						144	Atezo + CT	26.9	—	9.5	—	54.30%	55.67%
SUNRISE	2023	II	IV	ALL NSCLC	99	49	Sinti + Anlotinib	NA	0.81 (0.44–1.50)	12.6	0.39 (0.23–0.67)	44.90%	28.00%
						50	CT	NA	—	5.7	—	18.00%	48.98%
IMpower150	2021	III	IV	Non‐squamous	1,047	350	Atezo + Bev + CT	19.5	0.80 (0.67–0.95)	8.3	0.57 (0.48–0.67)	63.50%	60.31%
						359	Atezo + CT	19	0.84 (0.71–1.00)	6.9	0.82 (0.70–0.97)	NA	44.00%
						338	Bev + CT	14.7	—	6.8	—	48.00%	51.52%
BEYOND	2015	III	IV	Non‐squamous	276	138	Bev + CT	24.3	0.68 (0.50–0.93)	9.2	0.40 (0.29–0.54)	53.62%	67.14%
						138	CT	17.7	—	6.5	—	25.40%	61.94%
JO19907	2012	II	IIIB‐IV	Non‐squamous	175	117	Bev + CT	> 22	0.99 (0.65–1.50)	6.9	0.61 (0.42–0.89)	60.70%	NA
						58	CT	> 22	—	5.9	—	31.00%	NA
AVAiL	2010	III	IIIB‐IV	Non‐squamous	698	351	Bev + CT	NA	1.03 (0.86–1.23)	6.5	0.82 (0.68–0.98)	30.40%	80.55%
						347	CT	NA	—	6.1	—	20.10%	75.23%
ECOG4599	2006	III	IIIB‐IV	ALL NSCLC	850	417	Bev + CT	12.3	0.79 (0.67–0.92)	6.2	0.66 (0.57–0.77)	35.00%	58.78%
						433	CT	10.3	—	4.5	—	15.00%	23.64%
Robert C et al.	2015	II	IV	Non‐squamous	140	69	Ram + CT	13.9	1.03 90% (0.74–1.42)	7.2	0.75 (0.55–1.03)	49.30%	80.60%
						71	CT	10.4	—	5.6	—	38.00%	76.81%
Thomas, S.	2017	II	IV	Non‐squamous	140	71	Ram + CT	10.4	0.93 (0.68–1.27)	5.6	0.88 (0.64–1.22)	46.50%	NA
						69	CT	11.3	—	5.4	—	24.60%	NA
Yan Sun et al.	2013	II	IIIB‐IV	ALL NSCLC	347	230	Endostar + CT	NA	0.46 (0.37–0.59)	NA	NA	40.00%	NA
						117	CT	NA	—	NA	—	23.93%	NA
TASUKI‐52	2021	III	IIIB‐IV	ALL NSCLC	548	273	Nivo + Bev + CT	NA	0.85 (0.63–1.14)	12.1	0.56 (0.43–0.71)	61.50%	31.14%
						275	Bev + CT	NA	—	8.1	—	50.50%	22.91%
Kubota, K	2017	III	IV	Non‐squamous	318	156	Motesanib + CT	22.8	1.01 (0.74–1.38)	6.1	0.88 (0.68–1.13)	60.10%	54.08%
						162	CT	21.6	—	5.6	—	41.60%	19.12%
Shi, Y. K.	2017	III	IV	ALL NSCLC	546	365	Endostar + CT	NA	NA	5.4	0.692 (0.523–0.915)	NA	NA
						181	CT	NA	NA	4.7	—	NA	NA
PRONOUNCE	2015	III	IV	Non‐squamous	361	182	CT	10.5	1.07 (0.83–1.36)	4.44	1.06 (0.84 to 1.35)	23.60%	NA
						179	Bev + CT	11.7	—	5.49	—	27.40%	NA
Han, B	2011	II	IIIB‐IV	ALL NSCLC	122	61	Endostar + CT	17.6	NA	7.1	0.883 (0.604–1.292)	39.30%	24.59%
						61	CT	15.8	NA	6.3	—	23.00%	36.07%

3.2. Data Analysis

Objective response rate (ORR) was synthesized as odds ratios (ORs) with 95% confidence intervals (CIs). Progression‐free survival (PFS) and overall survival (OS) were summarized as hazard ratios (HRs) with 95% CIs, and grade ≥ 3 treatment‐related adverse events (TRAEs) as relative risks (RRs) with 95% CIs. Between‐study heterogeneity was assessed using the I² statistic; I² < 25% and > 50% indicated low and substantial heterogeneity, respectively [23]. Network meta‐analysis was implemented in R (v4.4.0) using JAGS and the GeMTC package under a Bayesian Markov chain Monte Carlo (MCMC) framework. We used 20,000 adaptation (burn‐in) iterations followed by 50,000 sampling iterations [24, 25]. A consistency model was assumed to estimate relative treatment effects and derive rank probabilities. For ranking, we summarized posterior rank probabilities and calculated SUCRA values; results were visualized as rank‐probability histograms. SUCRA values are used to probabilistically rank treatment regimens, with higher values indicating a greater probability of achieving a better ranking. However, these rankings are inherently relative and should be interpreted in conjunction with the corresponding effect size estimates. Model convergence was assessed using the Brooks–Gelman–Rubin potential scale reduction factor (PSRF), with values approaching 1 indicating adequate convergence. Network plots and comparison‐adjusted funnel plots were generated in Stata/SE 18 to explore small‐study effects and publication bias [26].

3.3. Risk of Bias Assessment

We used the Cochrane Risk of Bias tool (RoB 1.0) to assess the risk of bias for each study. Our assessment addressed seven domains: (1) random sequence generation, (2) allocation concealment, (3) blinding of participants and personnel, (4) blinding of outcome assessment, (5) incomplete outcome data, (6) selective reporting, and (7) other biases. For each domain, risk was classified as low (green), unclear (yellow), or high (red). Two investigators independently assessed the risk of bias; discrepancies were resolved by discussion with a third reviewer.

4. Results

4.1. Baseline Characteristics of Included Studies

The initial screening identified 14,365 records (PubMed: 2988; Embase: 7876; Cochrane Library: 1462; Web of Science: 2039). After sequential screening, 15 randomized controlled trials (RCTs) involving 5954 patients and nine treatment regimens were included (Figure 2) [9, 10, 18, 19, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37]. Nine were phase III trials [9, 10, 27, 28, 30, 34, 35, 36, 37], and six were phase II trials [18, 19, 29, 31, 32, 33]. All participants were adults with pathologically confirmed advanced non‐small cell lung cancer (NSCLC). The active regimens were: Bevacizumab + chemotherapy (Bev + CT; n = 1815) [9, 10, 28, 29, 30]; Endostar + chemotherapy (Endostar+CT; n = 595) [33, 36]; ramucirumab + chemotherapy (Ram+CT; n = 140) [31, 32]; motesanib + chemotherapy (Motesanib+CT; n = 156) [35]; atezolizumab + chemotherapy (Atezo+CT; n = 503) [10, 27]; atezolizumab + bevacizumab + chemotherapy (Atezo+Bev + CT; n = 493) [10, 27]; nivolumab + bevacizumab + chemotherapy (Nivo+Bev + CT; n = 273) [34]; and sintilimab + anlotinib (Sinti+Anlo; n = 49) [18]. The control group received platinum‐based chemotherapy (CT) alone.

FIGURE 2.

Flow diagram for the selection of the studies.

All included studies were published between 2000 and January 2025. Risk of bias was assessed using the Cochrane Risk of Bias tool, with results shown in Figure 3B.

FIGURE 3.

(A) Risk of bias graph. (B) Risk of bias summary. (C) Network map: Each circular node indicates a type of treatment; the node size is proportional to the total number of participants who received that treatment. Each line depicts a type of head‐to‐head comparison, and the linewidth is proportional to the total number of studies comparing the connected treatments.

Seven open‐label trials did not implement participant blinding, one trial did not report blinding procedures, and none reported investigator blinding. Four studies had relatively small sample sizes, whereas most others were rated as high quality. A network diagram (Figure 3C) depicts the nine interventions; node size/edges reflect the number of studies and direct comparisons.

4.2. Network Meta‐Analysis Results

4.2.1. PFS

Fourteen studies were included in the PFS analysis. Direct comparisons showed that, vs. CT, Bev + CT (HR = 0.70, 95% CI 0.64–0.78), Endostar+CT (HR = 0.75, 95% CI 0.60–0.95), and Sinti+Anlo (HR = 0.39, 95% CI 0.23–0.67) were significantly superior. Compared with Bev + CT, Atezo+Bev + CT (HR = 0.59, 95% CI 0.51–0.70), Atezo+CT (HR = 0.79, 95% CI 0.67–0.93), and Nivo+Bev + CT (HR = 0.56, 95% CI 0.44–0.72) were significantly superior (Figure 4A). In the network (indirect) analysis, Sinti+Anlo had the highest probability of being the best PFS regimen (SUCRA = 46.7%). Pooled PFS effects are shown in Figure 5A; rank‐probability histograms are provided in Figure S1A.

FIGURE 4.

Pairwise Meta‐analysis Based on Head‐to‐Head Comparisons. (A) Progression‐Free Survival (PFS). (B) Overall Survival (OS). (C) Objective Response Rate (ORR). (D) Grade 3–5 treatment‐related adverse events (Grade 3–5 TRAE).

FIGURE 5.

Indirect comparisons among the various treatment groups. (A) Progression‐Free Survival (PFS). (B) Overall Survival (OS). (C) Objective Response Rate (ORR). (D) Grade 3–5 treatment‐related adverse events (Grade 3–5 TRAE).

4.2.2. OS

Thirteen studies were included in the OS analysis. Direct comparisons showed that, vs. CT, Bev + CT (HR = 0.88, 95% CI 0.79–0.97) and Endostar+CT (HR = 0.46, 95% CI 0.36–0.58) significantly improved OS. Compared with Bev + CT, Atezo+Bev + CT (HR = 0.80, 95% CI 0.68–0.95) and Atezo+CT (HR = 0.84, 95% CI 0.71–0.99) significantly outperformed Bev + CT (Figure 4B). In the network analysis, Endostar+CT ranked highest for OS (SUCRA = 95%). Pooled OS effects are shown in Figure 5B; rank‐probability histograms are provided in Figure S1B.

4.2.3. ORR

Fourteen studies were included in the ORR analysis. Direct comparisons showed that, vs. CT, Endostar+CT (OR = 2.1, 95% CI 1.4–3.3), Motesanib+CT (OR = 2.1, 95% CI 1.4–3.2), Ram+CT (OR = 2.0, 95% CI 1.2–3.3), and Sinti+Anlo (OR = 3.8, 95% CI 1.5–10.0) achieved higher ORR. Compared with Bev + CT, Nivo+Bev + CT (OR = 1.6, 95% CI 1.1–2.2) and Atezo+Bev + CT (OR = 1.9, 95% CI 1.4–2.6) significantly improved ORR, and Atezo+CT was inferior to Atezo+Bev + CT (OR = 0.56, 95% CI 0.34–0.92) (Figure 4C). In the network analysis, Atezo+Bev + CT ranked highest for ORR (SUCRA = 50.5%). Pooled ORR effects are shown in Figure 5C; rank‐probability histograms are provided in Figure S1C.

4.2.4. Grade ≥ 3 TRAEs

Ten studies were included in the safety analysis. Direct comparisons showed that, vs. CT, only Sinti+Anlo reduced Grade ≥ 3 TRAEs (RR = 0.57, 95% CI 0.32–0.95), whereas other regimens showed higher risks or no significant differences (Figure 4D). In the network analysis, Sinti+Anlo ranked safest (SUCRA = 67%). Pooled TRAE effects are shown in Figure 5D; rank‐probability histograms are provided in Figure S1D.

4.3. Subgroup Analysis Results

4.3.1. Subgroup Analysis Based on Gender

In male patients, both PFS and OS showed significant between‐regimen differences. For PFS, Bev + CT (HR = 0.45, 95% CI 0.33–0.60), Nivo+Bev + CT (HR = 0.24, 95% CI 0.16–0.35), and Sinti+Anlo (HR = 0.41, 95% CI 0.23–0.72) were superior to CT (Figure 6A). Indirect comparisons ranked Nivo+Bev + CT as optimal for male PFS (Figure 7A). For OS, Atezo+CT (HR = 0.80, 95% CI 0.66–0.98) and Atezo+Bev + CT (HR = 0.71, 95% CI 0.58–0.87) outperformed Bev + CT, whereas CT was inferior to Bev + CT (HR = 1.44, 95% CI 1.20–1.73) (Figure 6B). Indirect comparisons ranked Atezo+Bev + CT as optimal for male OS (Figure 7B).

FIGURE 6.

Forest plots of subgroup analyses for Progression‐Free Survival (PFS) and Overall Survival (OS). (A) PFS in the male subgroup. (B) OS in the male subgroup. (C) OS in the female subgroup. (D) PFS in patients with ECOG PS = 1. (E) OS in patients with ECOG PS = 1. (F) OS in patients with ECOG PS = 0. (G) PFS in patients aged < 65 years. (H) PFS in patients aged ≥ 65 years. (I) PFS in ever‐smokers. (J) OS in ever‐smokers. (K) PFS in never smokers. (L) OS in never smokers.

FIGURE 7.

Indirect comparisons of subgroup analyses for Progression‐Free Survival (PFS) and Overall Survival (OS). (A) PFS in the male subgroup. (B) OS in the male subgroup. (C) OS in the female subgroup. (D) PFS in patients with ECOG PS = 1. (E) OS in patients with ECOG PS = 1. (F) OS in patients with ECOG PS = 0. (G) PFS in patients aged < 65 years. (H) PFS in patients aged ≥ 65 years. (I) PFS in ever‐smokers. (J) OS in ever‐smokers. (K) PFS in never smokers. (L) OS in never smokers.

Only OS data were available for female patients; Atezo+Bev + CT (HR = 0.91, 95% CI 0.70–1.20) and Atezo+CT (HR = 0.85, 95% CI 0.65–1.10) did not significantly outperform Bev + CT (Figure 6C), and indirect comparisons showed no significant difference between the two regimens (HR = 0.94, 95% CI 0.70–1.26) (Figure 7C).

4.3.2. Subgroup Analysis Based on ECOG Score

In patients with PS = 1, the greatest PFS benefit was observed with Nivo+Bev + CT vs. Bev + CT (HR = 0.58, 95% CI 0.43–0.79); Bev + CT vs. CT (HR = 0.42, 95% CI 0.31–0.56) and Sinti+Anlo vs. CT (HR = 0.44, 95% CI 0.27–0.72) were also superior (Figures 6D, 7D). For OS, Atezo+Bev + CT outperformed Bev + CT (HR = 0.78, 95% CI 0.64–0.96); Atezo+CT (HR = 0.91, 95% CI 0.74–1.10) showed no significant advantage, and CT was inferior to Bev + CT (HR = 1.40, 95% CI 1.10–1.60) (Figure 6E). Indirect comparisons ranked Atezo+Bev + CT as optimal for OS in PS = 1 (Figure 7E).

For PS = 0, only OS data were available: Atezo+Bev + CT (HR = 0.78, 95% CI 0.60–1.00) and Atezo+CT (HR = 0.75, 95% CI 0.57–0.97) were superior to Bev + CT (Figure 6F), but indirect comparisons showed no significant difference between them (Figure 7F).

4.3.3. Subgroup Analysis Based on Age

In patients < 65 years, Bev + CT demonstrated a PFS advantage over CT (HR = 0.39, 95% CI 0.30–0.51). Sinti+Anlo showed no significant PFS advantage vs. CT (HR = 0.54, 95% CI 0.28–1.05). Nivo+Bev + CT outperformed Bev + CT (HR = 0.50, 95% CI 0.36–0.70) (Figure 6G), and indirect comparisons ranked Nivo+Bev + CT as optimal for PFS (Figure 7G).

In patients ≥ 65 years, for PFS, Bev + CT (HR = 0.59, 95% CI 0.38–0.92) and Sinti+Anlo (HR = 0.40, 95% CI 0.18–0.87) were superior to CT. Nivo+Bev + CT was superior to Bev + CT (HR = 0.65, 95% CI 0.48–0.89), but Sinti+Anlo did not differ significantly from Nivo+Bev + CT (Figures 6H, 7H).

4.3.4. Subgroup Analysis Based on Smoking Status

Among smokers, Bev + CT (vs. CT: HR = 0.51, 95% CI: 0.38–0.69) and Endostar+CT (HR = 0.63, 95% CI: 0.44–0.90) were superior to CT, and Nivo+Bev + CT outperformed Bev + CT (HR = 0.56, 95% CI: 0.44–0.72) (Figure 6I). Indirect comparisons identified Nivo+Bev + CT as the optimal PFS regimen among smokers (Figure 7I). For OS, Atezo+Bev + CT (HR = 0.81, 95% CI: 0.67–0.97) and Atezo+CT (HR = 0.83, 95% CI: 0.69–1.00) outperformed Bev + CT; CT was inferior to Bev + CT (HR = 1.30, 95% CI: 0.86–2.00), with no significant difference between Atezo+Bev + CT and Atezo+CT (Figures 6J, 7J).

Among never‐smokers, Bev + CT (HR = 0.37, 95% CI: 0.26–0.54) and Sinti+Anlo (HR = 0.36, 95% CI: 0.17–0.76) were superior to CT; Nivo+Bev + CT showed no significant advantage over Bev + CT or Sinti+Anlo (Figures 6K, 7K). For OS, Atezo+Bev + CT was superior to Bev + CT (HR = 0.69, 95% CI: 0.48–1.00), but indirect comparisons showed no significant difference between Atezo+Bev + CT and Atezo+CT (Figures 6L, 7L).

4.4. Analysis of Convergence, Inconsistency, Publication Bias, and Heterogeneity

Comparison‐adjusted funnel plots were symmetric, with no evidence of substantial publication bias (Figure 8). The results of Begg's and Egger's tests are shown in Figure 2, and their P‐values are summarized in Table S1. Heterogeneity analysis is presented in Figure 3; overall, network heterogeneity was low. Posterior density plots for comparative treatment effects vs. the common reference for PFS, OS, ORR, and grade 3–5 TRAEs are shown in Figure 4. The potential scale reduction factor (PSRF) was approximately 1, indicating adequate convergence (Figure 5).

FIGURE 8.

Funnel plots to assess potential publication bias for (A) progression‐free survival (PFS), (B) overall survival (OS), and (C) objective response rate (ORR).

5. Discussion

This study systematically evaluated the efficacy and safety of nine antiangiogenic combination regimens in advanced non‐small cell lung cancer (NSCLC) with driver‐negative status using a Bayesian network meta‐analysis integrating data from 15 randomized controlled trials. The principal findings are summarized in three key aspects:

First, adding an antiangiogenic agent to chemo‐immunotherapy confers additional clinical benefit in selected settings. The triple regimen of atezolizumab, bevacizumab, and chemotherapy significantly improved ORR vs. bevacizumab + chemotherapy (OR = 1.90, 95% CI: 1.40–2.60). This finding is mechanistically consistent with the survival advantage of the quadruple regimen in IMpower150, wherein antiangiogenic therapy promotes vascular normalization, improves the immune microenvironment, and enhances cytotoxic drug delivery [10]. Notably, subgroup analyses showed comparable OS advantages in men (HR = 0.71) and in patients with ECOG PS = 1 (HR = 0.78), although potential increases in proteinuria and hypertension warrant consideration. For ECOG PS = 0, chemo‐immunotherapy without bevacizumab (e.g., atezolizumab + chemotherapy) may be a more appropriate choice.

Second, recombinant human endostatin (Endostar) + chemotherapy demonstrated greater survival benefits and a favorable safety profile relative to bevacizumab‐based therapy. Compared with chemotherapy alone, Endostar+CT substantially improved OS (HR = 0.46, 95% CI: 0.36–0.58), corresponding to a 54% relative risk reduction. Its SUCRA of 95% ranked it highest among the evaluated regimens, although SUCRA reflects relative ranking rather than absolute certainty. Plausible mechanisms include the downregulation of VEGF/VEGFR signaling and inhibition of lymphangiogenesis via VEGF‐C pathway suppression. Furthermore, Endostar has been reported to modulate additional pathways, suggesting broad antiangiogenic activity; reported targets include HIF‐1α, matrix metalloproteinases (MMPs), basic FGF (bFGF), and integrin αvβ3 [38, 39]. In network estimates, Endostar‐based regimens were associated with fewer grade ≥ 3 treatment‐related adverse events (TRAEs) than bevacizumab‐based regimens (RR = 0.54, 95% CI: 0.30–0.94). However, evidence in non‐Asian populations remains limited, highlighting the need for multiethnic phase III trials.

Third, the chemotherapy‐free regimen of sintilimab + anlotinib offers an effective, lower‐toxicity option for selected patients. This regimen outperformed chemotherapy in PFS (HR = 0.39) and in safety (Grade ≥ 3 TRAEs: RR = 0.57), with a notable PFS advantage in patients ≥ 65 years (HR = 0.40). Anlotinib inhibits VEGFR/FGFR/PDGFR family kinases, potentially reversing immunosuppressive features of the tumor microenvironment and synergizing with PD‐1 blockade [40, 41]. Subgroup analysis suggested greater PFS benefit in never‐smokers (HR = 0.36), and a single‐center report from Shanghai Chest Hospital described a median OS > 20 months [18]. However, given the small sample size (n = 49), current evidence supports use primarily in older adults or in patients with organ dysfunction who are unable to tolerate chemotherapy.

These mechanisms, illustrated in Figure 1, provide a plausible biological explanation for the superior PFS, OS, and ORR with antiangiogenic combination regimens in our network meta‐analysis.

Subgroup findings offer practical guidance for personalized treatment. Among smokers, nivolumab + bevacizumab + chemotherapy improved PFS vs. bevacizumab + chemotherapy (HR = 0.56), consistent with the hypothesis that tobacco‐related tumor mutational burden may enhance responsiveness to immunotherapy. In never‐smokers, sintilimab + anlotinib showed a favorable benefit–risk profile. ECOG performance status also informs selection: PS = 1 may favor bevacizumab‐containing triple regimens, whereas PS = 0 may be appropriate for de‐escalation to chemo‐immunotherapy without antiangiogenic therapy. By age, patients ≥ 65 years may benefit from the lower toxicity of chemotherapy‐free regimens; overall safety ranking favored Sinti+Anlo (SUCRA = 67%), although subgroup‐specific SUCRA was not estimated.

This study has several limitations: Imbalanced sample sizes across regimens (e.g., Sinti+Anlo, n = 49); predominance of Asian cohorts in trials of Endostar, potentially limiting generalizability; and the absence of subgroup analyses according to PD‐L1 expression, histologic subtype, and sites of metastasis. Additionally, the trials included in this network meta‐analysis span a long time period, covering both the pre‐ICI era and the contemporary ICI era. Temporal changes in diagnostic and therapeutic strategies are an inevitable source of heterogeneity, and this should be considered when interpreting long‐term survival outcomes such as overall survival. Future research should include optimization of antiangiogenic dosing strategies (e.g., low‐dose “normalization” approaches); phase III trials evaluating Endostar in combination with immune checkpoint inhibitors; and identification of predictive biomarkers of benefit.

In summary, our findings support the following: (1) For Chinese patients with advanced NSCLC, Endostar (recombinant human endostatin) + chemotherapy may be a preferred first‐line option; (2) triple therapy combining immunotherapy, antiangiogenic therapy, and chemotherapy appears suitable for male patients and those with ECOG PS = 1; and (3) chemotherapy‐free regimens are reasonable alternatives for patients unable to tolerate chemotherapy. Treatment selection should integrate smoking status, ECOG performance status, and age to inform personalized decision‐making.

Author Contributions

Zirui Li: conceptualization, writing – original draft. Weixing Zhao: methodology. Wanjing Guo and Xinxin Lu: formal analysis. Chenyu Jia and Qi Zhou: supervision. Jiayun Ma and Xiujin Yang: resources. Jun Jiang: writing – review and editing, funding acquisition.

Funding

This work was supported by the Science and Technology Support Qinghai Project from the Science and Technology Department of Qinghai Province (2025‐QY‐253). This paper is funded by the Science and Technology Support Qinghai Project (2025‐QY‐253) from theScience and Technology Department of Qinghai Province.

Ethics Statement

The authors have nothing to report.

Consent

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure 1 Ranking of treatments based on network meta‐analysis for (A) Progression‐Free Survival (PFS), (B) Overall Survival (OS), (C) Objective Response Rate (ORR), and (D) Grade 3–5 treatment‐related adverse events (Grade 3–5 TRAE).

Figure 2 Assessment of Publication Bias using Egger's and Begg's Tests. Funnel plots from (A, C, E) Egger's test and (B, D, F) Begg's test for the outcomes of (A, B) Progression‐Free Survival (PFS), (C, D) Overall Survival (OS), (E, F) Objective Response Rate (ORR). Asymmetry in the plots may indicate the presence of small‐study effects or publication bias.

Figure 3 Assessment of Heterogeneity in the Network Meta‐Analysis for Progression‐Free Survival (PFS) and Overall Survival (OS). Forest plots display the extent of heterogeneity for each treatment comparison within the network. (A) Results for Progression‐Free Survival (PFS). (B) Results for Overall Survival (OS).

Figure 4 Posterior Density Plots from Network Meta‐Analysis. Density plots for the comparative effects of various treatments relative to a common reference are presented for (A) Progression‐Free Survival (PFS), (B) Overall Survival (OS), (C) Objective Response Rate (ORR), and (D) Grade 3–5 treatment‐related adverse events (Grade 3–5 TRAE).

Figure 5 Convergence Diagnostics for the Bayesian Network Meta‐Analysis Models. (A) Progression‐Free Survival (PFS), (B) Overall Survival (OS), (C) Objective Response Rate (ORR), and (D) Grade 3–5 treatment‐related adverse events (Grade 3–5 TRAE).

Table 1 P‐value of Begg's test and Egger's test.

Data S1: Search strategy.

Data Availability Statement

The data that supports the findings of this study are available in the Supporting Information of this article.