A systematic review and network meta-analysis of first-line EGFR-TKI-based combination strategies for EGFR-mutant non-small cell lung cancer with brain metastases

Abstract

Brain metastases are common in epidermal growth factor receptor (EGFR)-mutated non-small cell lung cancer (NSCLC) and remain clinically challenging despite the improved central nervous system activity of third-generation EGFR tyrosine kinase inhibitors (EGFR-TKIs). Because direct comparisons between active first-line combination strategies are limited, this systematic review and network meta-analysis (NMA) evaluated the relative efficacy and safety of EGFR-TKI-based combinations in this population. PubMed, Embase, Web of Science, and the Cochrane Library were searched from inception to December 31, 2025. Randomized controlled trials (RCTs) and non-randomized comparative studies were eligible if they enrolled patients with EGFR-mutated NSCLC and brain metastases and compared EGFR-TKI monotherapy with EGFR-TKI-based combinations incorporating chemotherapy, anti-angiogenic therapy, radiotherapy, or bispecific antibody therapy. Bayesian NMAs were performed for overall survival (OS), progression-free survival (PFS), and intracranial PFS (iPFS). RCT-only analyses were conducted as robustness assessments. Grade ≥ 3 treatment-related adverse events (TRAEs) were summarized descriptively. Forty-eight studies involving 5,432 patients were included, of which 17 were RCTs. In the overall network, EGFR-TKI plus anti-angiogenic therapy showed the most favorable OS estimate versus monotherapy (HR, 0.63; 95% CI, 0.49–0.80), whereas EGFR-TKI plus chemotherapy showed the most favorable estimates for PFS (HR, 0.52; 95% CI, 0.44–0.61) and iPFS (HR, 0.36; 95% CI, 0.25–0.52). RCT-only analyses most consistently supported the PFS benefit of EGFR-TKI plus chemotherapy; OS rankings were less definitive, and randomized iPFS evidence remained sparse. Radiotherapy-related results were exploratory because the radiotherapy node pooled WBRT, SRS/SRT, mixed approaches, and variable dose-fractionation schedules. EGFR-TKI plus chemotherapy showed the most consistent evidence for delaying progression. The apparent OS advantage of EGFR-TKI plus anti-angiogenic therapy, and findings for radiotherapy- and bispecific antibody-based combinations, should be interpreted cautiously. Treatment selection should be individualized according to treatment goals, EGFR-TKI generation, intracranial disease burden, expected toxicity, and patient tolerability.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1038/s41598-026-64728-w.

Introduction

Lung cancer is the second most commonly diagnosed malignancy and the leading cause of cancer-related mortality worldwide¹. Non-small cell lung cancer (NSCLC) is the predominant histological subtype, accounting for approximately 85% of all lung cancer cases². With the development of precision medicine, epidermal growth factor receptor (EGFR) mutations have been identified as key oncogenic drivers in NSCLC. The prevalence of EGFR mutations varies substantially by ethnicity, occurring in approximately 10–15% of Western patients and up to 40–50% of Asian patients^3,4. Despite the transformative impact of EGFR tyrosine kinase inhibitors (EGFR-TKIs) on the treatment of advanced NSCLC, central nervous system (CNS) involvement, particularly brain metastases, remains a major clinical challenge. Epidemiological studies have reported that 20–40% of patients with NSCLC develop brain metastases. This incidence is even higher among patients with EGFR-mutant disease, with 3-year cumulative rates ranging from 29.4% to 60.3%^5,6. The high frequency of CNS involvement is partly attributable to the blood-brain barrier (BBB), which restricts the intracranial penetration of many systemic therapies and allows the brain to function as a pharmacological sanctuary site^7,8. Brain metastases are frequently associated with debilitating neurological symptoms, including headache, nausea, limb paresis, and cognitive impairment, and can substantially impair patients’ quality of life⁹. Moreover, the presence of brain metastases is an independent predictor of poor prognosis and reduced overall survival (OS).

Local therapies, including whole-brain radiotherapy (WBRT) and stereotactic radiosurgery (SRS), can provide intracranial disease control, but their use may be limited by the risk of neurocognitive toxicity, particularly after WBRT¹⁰. First- and second-generation EGFR-TKIs, such as gefitinib, erlotinib, and afatinib, have demonstrated substantial efficacy in systemic disease control; however, their intracranial activity is limited by relatively poor CNS penetration. Consequently, intracranial progression has been a major pattern of treatment failure with these agents^11,12. The development of the third-generation EGFR-TKI osimertinib, which has improved BBB penetration and enhanced CNS activity, represented an important therapeutic advance. In the pivotal FLAURA trial, osimertinib achieved a higher intracranial objective response rate and a longer median duration of intracranial response than first-generation EGFR-TKIs. Accordingly, osimertinib is now widely regarded as an established first-line standard of care for treatment-naïve patients with EGFR-mutant NSCLC and CNS metastases^13,14.

However, acquired resistance to osimertinib eventually develops in most patients. In addition, certain high-risk subgroups, including patients with a high intracranial tumor burden or leptomeningeal metastases, may experience suboptimal outcomes with EGFR-TKI monotherapy alone. To overcome this therapeutic limitation, several EGFR-TKI-based combination strategies have been investigated. These include combinations with chemotherapy, anti-angiogenic agents, radiotherapy, and, more recently, bispecific antibody-based regimens¹⁵. From a mechanistic perspective, anti-angiogenic agents such as bevacizumab and ramucirumab may enhance antitumor activity by inhibiting VEGF-mediated angiogenesis and modulating abnormal tumor vasculature. Vascular normalization may improve drug delivery to intracranial lesions by reducing abnormal vascular permeability, lowering interstitial pressure, and modifying the blood-tumor barrier microenvironment¹⁶. However, this rationale is most clearly supported for first-generation EGFR-TKI-based combinations, whereas its incremental value in the context of osimertinib, which already has substantial CNS penetration, remains less certain. Several phase III randomized controlled trials (RCTs), including NEJ026¹⁷,CTONG1509/ARTEMIS¹⁸,and RELAY¹⁹, have reported that EGFR-TKI plus anti-angiogenic therapy prolongs progression-free survival (PFS) compared with EGFR-TKI monotherapy. Chemotherapy-based combinations are supported by a different biological rationale. Because EGFR-mutant NSCLC may contain heterogeneous tumor cell populations, chemotherapy may help eliminate EGFR-TKI-insensitive subclones and delay the emergence of acquired resistance. This strategy was supported by the recent FLAURA2 trial, which showed that first-line osimertinib plus platinum-based chemotherapy significantly prolonged PFS compared with osimertinib monotherapy in patients with advanced EGFR-mutant NSCLC. Importantly, the PFS benefit was also observed in patients with baseline CNS metastases²⁰. Other emerging strategies involve distinct mechanisms of action. For example, the MARIPOSA trial evaluated first-line amivantamab, an EGFR-MET bispecific antibody, plus lazertinib versus osimertinib monotherapy in patients with advanced EGFR-mutant NSCLC²¹. This regimen demonstrated improved PFS compared with osimertinib alone, with emerging survival data suggesting potential additional benefit. Nevertheless, because evidence for bispecific antibody-based combinations remains limited, particularly in patients with brain metastases, these findings should be interpreted cautiously.

Although the aforementioned combination strategies have shown clinical activity in individual trials, the optimal first-line approach for patients with EGFR-mutant NSCLC and brain metastases remains uncertain. This uncertainty is largely attributable to the lack of direct head-to-head comparisons between active combination regimens, as most available studies have compared a single EGFR-TKI-based combination strategy with EGFR-TKI monotherapy. In addition, potential efficacy gains must be balanced against increased toxicity, including a higher risk of treatment-related adverse events (TRAEs). This balance is particularly important in patients with brain metastases, who may have neurological symptoms, impaired performance status, or a limited tolerance for intensive combination therapy²².

Previous network meta-analyses have evaluated EGFR-TKI-based treatment strategies in advanced EGFR-mutant NSCLC. However, many did not specifically focus on patients with brain metastases or incorporate recent pivotal trials, such as FLAURA2 and MARIPOSA. Moreover, prior analyses have often emphasized treatment ranking, whereas the present study aims to characterize how different first-line combination strategies may vary across clinically relevant endpoints, including OS, PFS, iPFS, and safety. By synthesizing the available direct and indirect evidence, this study provides an updated, endpoint-specific framework to inform individualized treatment selection, while acknowledging the limitations imposed by indirect comparisons, clinical heterogeneity, and differences in EGFR-TKI generation.

Methods

Study design and registration

This systematic review and NMA was conducted and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement. The study protocol was prospectively registered in the International Prospective Register of Systematic Reviews under the registration number CRD420261396304.

Search strategy and inclusion criteria

A systematic literature search was performed in PubMed, Embase, Web of Science, and the Cochrane Library from database inception to December 31, 2025. The search strategy was developed according to the Population, Intervention, Comparator, Outcomes, and Study Design framework. Search terms included disease-related terms, such as “non-small cell lung cancer,” “brain metastases,” and “EGFR mutation,” as well as intervention-related terms, such as “tyrosine kinase inhibitor,” “chemotherapy,” “radiotherapy,” and “anti-angiogenic agents.” Only studies published in English were eligible for inclusion. To ensure comprehensive retrieval, the reference lists of all included studies and relevant reviews were also manually screened. The full search strategy for each database is provided in Supplementary Material 1.

Studies were eligible for inclusion if they met all of the following criteria: (1) enrolled patients with pathologically confirmed NSCLC harboring EGFR mutations and brain metastases; (2) were comparative studies evaluating at least two protocol-defined treatment strategies; and (3) reported at least one efficacy outcome of interest, including OS, PFS, or intracranial progression-free survival (iPFS), with a corresponding hazard ratio (HR) and 95% confidence interval (CI), or provided sufficient information for HR estimation. Studies were excluded if they met any of the following criteria: (1) enrolled patients without documented EGFR mutations or included mixed populations from which data for the EGFR-mutant subgroup could not be extracted; (2) were non-original research articles, including reviews, case reports, case series, clinical guidelines, editorials, letters, or conference abstracts; or (3) lacked extractable time-to-event data for the key efficacy outcomes.

Data extraction and quality assessment

Two reviewers independently performed full-text screening and data extraction using a standardized data collection form. Discrepancies were resolved through discussion, and, when necessary, adjudicated by a third senior reviewer to reach consensus.

The extracted data included study-level characteristics, such as first author, publication year, study design, and trial phase; baseline patient characteristics, including sample size, median age, sex distribution, EGFR mutation status, and, where available, performance status; and intervention details, including the specific EGFR-TKI used, EGFR-TKI generation, and type of combination strategy. Efficacy outcomes were extracted as HRs with corresponding 95% CIs for OS, PFS, and intracranial iPFS. Safety outcomes, including the incidence of grade ≥ 3 TRAEs, were extracted where available.

Because treatment-specific survival estimates for patients with leptomeningeal metastases were rarely reported, leptomeningeal metastasis-related data were not included in the quantitative NMA. These data were recorded descriptively when available.

The risk of bias was assessed independently by two reviewers using the Cochrane Risk of Bias tool for RCTs and the Risk Of Bias In Non-randomized Studies of Interventions tool for non-randomized studies. Disagreements were resolved by consensus or third-reviewer adjudication. RCT assessments were visualized using Review Manager version 5.3.

Study endpoints

The efficacy endpoints were OS, PFS, and iPFS. OS was defined as the time from randomization or treatment initiation to death from any cause. PFS was defined as the time from randomization or treatment initiation to documented disease progression or death from any cause. iPFS was defined as the time from randomization or treatment initiation to documented intracranial disease progression or death from any cause. Disease progression for PFS and iPFS was assessed according to the response criteria used in each included study, most commonly the Response Evaluation Criteria in Solid Tumors version 1.1.

The safety endpoint was the incidence of grade ≥ 3 TRAEs. Because adverse-event definitions, reporting thresholds, and follow-up durations varied substantially across studies, safety outcomes were summarized descriptively rather than synthesized by NMA.

Statistical analysis

For quantitative synthesis, studies were included if they reported HRs with corresponding 95% CIs for OS, PFS, or iPFS, or if they provided Kaplan-Meier curves from which HRs could be estimated. When HRs were not directly reported, Kaplan-Meier curves were digitized using GetData Graph Digitizer, version 2.26. HRs were used as the primary effect measure because OS, PFS, and iPFS are time-to-event outcomes.

The primary efficacy analyses for OS, PFS, and iPFS were performed within a Bayesian NMA framework using GeMTC, version 0.14.3. Treatment effects were reported as HRs with 95% CIs. An effect estimate was considered statistically significant when the 95% CI did not include 1. Model convergence was assessed using the potential scale reduction factor (PSRF), with values approaching 1 indicating acceptable convergence.

Treatment nodes were defined according to therapeutic strategy rather than individual EGFR-TKI agent. The predefined treatment nodes included EGFR-TKI monotherapy, EGFR-TKI plus chemotherapy, EGFR-TKI plus radiotherapy, EGFR-TKI plus anti-angiogenic therapy, and EGFR-TKI plus bispecific antibody therapy. This strategy-level classification was used because the available evidence was not sufficiently connected to support a stable network based exclusively on individual EGFR-TKI agents or third-generation EGFR-TKI-based regimens across all treatment categories. Given the potential influence of EGFR-TKI generation on CNS penetration and intracranial efficacy, generation-stratified subgroup analyses were performed for OS, PFS, and iPFS where data permitted.

WBRT, SRS/SRT, mixed radiotherapy approaches, and different dose-fractionation schedules were retained within a broad EGFR-TKI plus radiotherapy node because modality-specific HRs for OS, PFS, and iPFS were not consistently available. Therefore, radiotherapy estimates were interpreted as class-level rather than modality-specific effects.

Outcome-specific networks were constructed separately for OS, PFS, and iPFS. All three networks were anchored by EGFR-TKI monotherapy and lacked direct comparisons between active combination strategies; therefore, formal node-splitting inconsistency assessment was not applicable, and active-combination comparisons relied on indirect evidence and the transitivity assumption. Ranking probabilities were interpreted as exploratory.

To assess the potential impact of non-randomized evidence, RCT-only NMAs were performed for OS, PFS, and iPFS after excluding retrospective and non-randomized studies. Fixed-effect Bayesian models were used for the main RCT-only analyses because randomized comparisons were limited in some nodes, and random-effects models were fitted as sensitivity analyses.

Radiotherapy-sequence analysis was conducted separately as a conventional pairwise subgroup meta-analysis because radiotherapy timing was considered a within-class modifier rather than an independent treatment node. Pooled HRs were calculated using an inverse-variance approach, heterogeneity was assessed with I², and subgroup differences were evaluated using conventional P values.

Results

Literature search and study selection

The literature search and study selection process is summarized in the PRISMA flow diagram (Fig. 1). The initial database search identified 16,941 records. After removal of duplicates, 13,752 unique records were screened by title and abstract. Following full-text assessment according to the eligibility criteria, 48 studies, including 17 randomized controlled trials, were included in the quantitative synthesis.

Fig. 1.

Flow diagram of the meta-analysis for inclusion and exclusion of studies.

Study characteristics

The 48 included studies enrolled 5,432 patients across five EGFR-TKI-based treatment strategies. 10 studies evaluated EGFR-TKI plus chemotherapy^20,23–31, 11 evaluated EGFR-TKI plus anti-angiogenic therapy^{17,18,32–40}, 26 evaluated EGFR-TKI plus radiotherapy^41–66,a and one evaluated EGFR-TKI plus bispecific antibody therapy^67,68. Because the bispecific antibody node was informed by a single RCT, its estimates and rankings were considered exploratory. Leptomeningeal metastases were inconsistently reported and treatment-specific HRs for this subgroup were generally unavailable. Detailed characteristics and quality assessments are summarized in Table 1.

Table 1.

Characteristics of studies included in this network meta-analysis.

study	Research type	Histological type	Patient no.(T/C)	Intervention	Control treatment	OS HR (95% CI)	PFS HR (95% CI)	iPFS HR (95% CI)
Jänne, P. A.2024	Ⅲ	SQ and No sq	116/110	Osimertinib plus platinum-pemetrexed	osimertinib	0.72(0.52–0.99)	NA	NA
Li, J.2025	Prospective cohort	adenocarcinoma	14/16	Osimertinib plus chemotherapy	Osimertinib	NA	0.76(0.22–2.60)	NA
Kanda, S.2025	Ⅲ	No sq	68/74	Osimertinib plus intercalated chemotherapy	Osimertinib	1.21(0.80–1.84)	0.77(0.53–1.11)	NA
Planchard, D.2023	Ⅲ	SQ and No sq	116/110	Osimertinib plus chemotherapy	Osimertinib	NA	0.47(0.33–0.66)	NA
Hou, X.2023	Ⅲ	SQ and No sq	80/81	gefitinib plus chemotherapy	Osimertinib	0.65(0.43–0.99)	0.39(0.27–0.58)	0.36(0.25–0.53)
Shang, K.2022	retrospective studies	SQ and No sq	12/18	Gefitinib/erlotinib/icotinib/osimertinib plus chemotherapy	Gefitinib/erlotinib/icotinib/osimertinib	1.80(0.79–4.10)	0.46(0.20–1.03)	NA
Wang, Q.2021	retrospective studies	SQ and No sq	9/17	Gefitinib Plus chemotherapy	Gefitinib	NA	1.20(0.30–1.80)	NA
Noronha, V.2020	Ⅲ	SQ and No sq	30/34	Gefitinib Plus chemotherapy	Gefitinib	NA	0.53 (0.29–0.98	NA
Hosomi, Y.2019	Ⅲ	SQ and No sq	50/38	Gefitinib Plus chemotherapy	Gefitinib	0.66(0.40–1.07)	0.32(0.19–0.53)	NA
Lu, S. 2025	Ⅲ	SQ and No sq	310/314	Aumolertinib plus plati- num-pemetrexed	Aumolertinib	NA	0.56(0.37–0.83)	NA
Lee, Y.2022	Ⅱ	SQ and No sq	29/30	Erlotinib plus bevacizumab	Erlotinib	1.27(0.58–2.79)	0.54(0.31–0.95)	0.18(0.05–0.67)
Shi, H.2025	retrospective studies	No sq	19/27	Afatinib plus bevacizumab	Afatinib	NA	1.34(0.60–3.02)-	NA
Zhou, H. Q.2024	Ⅲ	SQ and No sq	49/50	Gefitinib plus anlotinib	Gefitinib plus placebo	NA	0.47(0.29–0.77)	NA
Le, X.2025	Ⅱ	SQ and No sq	40/24	Osimertinib plus ramucirumab	Osimertinib	NA	0.65(0.32–1.31)	NA
Kenmotsu, H.2022	Ⅱ	No sq	18/23	Osimertinib plus bevacizumab	Osimertinib	NA	0.83(0.36–1.94)	NA
Chiu, T. H.2022	Real-world research	SQ and No sq	43/89	Gefitinib/erlotinib plus bevacizumab	Gefitinib/erlotinib	0.28(0.10–0.78)	NA	NA
Zhao, H.2021	Ⅲ	No sq	51/41	Gefitinib plus apatinib	Gefitinib plus Placebo	NA	0.87(0.50–1.52)	NA
Zhou, Q.2021	Ⅲ	SQ and No sq	44/47	Erlotinib plus bevacizumab	Eerlotinib	0.62(0.38–1.01)	0.48(0.27–0.84)	NA
Saito, H.2019	Ⅲ	No sq	36/36	Erlotinib plus bevacizumab	Erlotinib	0.84(0.43–1.63)	0·78(0·42 − 1·43)	NA
Stinchcombe, T. E.2019	Ⅱ	No sq	11/14	Erlotinib plus bevacizumab	Erlotinib	1.14(0.46–2.84)	1.72(0.92–3.23)	NA
Jiang, T.2019	retrospective studies	No sq	59/149	Gefitinib/erlotinib/icotinib plus bevacizumab	Gefitinib/erlotinib/icotinib	0.47(0.31–0.70)	0.46(0.33–0.64)	0.55(0.39–0.76)
Zhou, J.2023	retrospective studies	adenocarcinoma	96/117	Osimertinib plus RT	Osimertinib	0.87(0.61–1.24)	NA	0.42(0.31–0.58)
Chiou, G.Y.2022	retrospective studies	adenocarcinoma	190/90	Gefitinib/erlotinib/ Afatinib/osimertinib plus SRS	Gefitinib/erlotinib/ Afatinib/osimertinib	0.82(0.55–1.22)	NA	0.25(0.18–0.34)
Liu, Y.2021	retrospective studies	adenocarcinoma	77/57	Gefitinib/erlotinib/icotinib plus RT	Gefitinib/erlotinib/icotinib	0.56(0.33–0.94)	0.62(0.43–0.91)	0.51(0.33–0.79)
Gu, Y.2021	retrospective studies	adenocarcinoma	60/186	Gefitinib/erlotinib/icotinib/afatinib/dacomitinib/osimertinib/almonertinib plus RT	Gefitinib/erlotinib/icotinib/afatinib/dacomitinib/osimertinib/almonertin	0.92(0.72–1.18)	NA	0.75(0.59–0.96)
Hyun, D. G.2020	retrospective studies	SQ and No sq	66/107	Gefitinib/erlotinib/afatinib plus RT	Gefitinib/erlotinib/afatinib	1.34(0.98–1.83)	NA	0.72(0.42–1.23)
Saruwatari, K.2019	retrospective studies	adenocarcinoma	30/51	Gefitinib/erlotinib plus RT	Gefitinib/erlotinib	0.42 (0.20–0.88)	0.38 (0.19–0.77)	NA
He, Z. Y.2019	retrospective studies	adenocarcinoma	56/48	Gefitinib/erlotinib/icotinib plus RT	Gefitinib/erlotinib/icotinib	0.83(0.54–1.27)	NA	0.58(0.38–0.89)
An, N.2019	retrospective studies	adenocarcinoma	35/29	Gefitinib/erlotinib/icotinib plus RT	Gefitinib/erlotinib/icotinib	0.40(0.17–0.93)	NA	0.43(0.20–0.91)
Saida, Y.2019	retrospective studies	SQ and No sq	39/65	Gefitinib/erlotinib/icotinib plus RT	Gefitinib/erlotinib/icotinib	0.86(0.54–1.36)	NA	0.82(0.53–1.27)
Wang, C.2019	retrospective studies	SQ and No sq	53/40	Gefitinib/erlotinib/icotinib/afatinib plus RT	Gefitinib/erlotinib/icotinib/afatinib	0.75(0.39–1.44)	NA	0.38(0.19–0.75)
Lee, J. H.2019	retrospective studies	SQ and No sq	96/75	Gefitinib/erlotinib/afatinib plus RT	Gefitinib/erlotinib/afatinib	0.52(0.20–1.32)	NA	NA
Chen, Y.2019	retrospective studies	adenocarcinoma	49/29	Gefitinib/erlotinib/icotinib plus RT	Gefitinib/erlotinib/icotinib	0.51(0.20–1.28)	NA	0.33(0.16–0.69)
Wang, W.2018	retrospective studies	SQ and No sq	119/62	Gefitinib/erlotinib/icotinib plus RT	Gefitinib/erlotinib/icotinib	0.74(0.55–0.99)	NA	0.73(0.54–0.98)
Sung, S.2018	retrospective studies	adenocarcinoma	40/41	Gefitinib/erlotinib plus RT	Gefitinib/erlotinib	0.78(0.42–1.45)	NA	0.39(0.21–0.72)
Li, S. H.2018	retrospective studies	adenocarcinoma	17/11	Afatinib plus WBRT	Afatinib	0.99(0.20–4.82)	NA	NA
Chen, H.2018	retrospective studies	adenocarcinoma	66/39	Gefitinib/erlotinib/icotinib plus RT	Gefitinib/erlotinib/icotinib	0.44(0.31–0.63)	NA	0.18(0.07- 0.43)
Ke, S.B.2018	retrospective studies	adenocarcinoma	60/79	Gefitinib/erlotinib plus WBRT	Gefitinib/erlotinib	1.15(0.69–1.91)	NA	0.54(0.37–0.78)
Yu, M.2018	retrospective studies	SQ and No sq	24/18	Gefitinib/erlotinib/icotinib plus RT	Gefitinib/erlotinib/icotinib	0.33(0.19–0.58)	NA	0.47(0.25–0.90)
Magnuson, W. J.2017	retrospective studies	SQ and No sq	220/131	Erlotinib plus RT	Erlotinib	0.53(0.30–0.94)	NA	0.82(0.65–1.04)
Fan, Y. 2017	retrospective studies	SQ and No sq	56/41	Icotinib plus RT	Icotinib	0.78 (0.52–1.17)	NA	0.66 (0.44–0.99)
Liu, Y.2017	retrospective studies	SQ and No sq	49/64	Gefitinib/erlotinib/icotinib plus WBRT	Gefitinib/erlotinib/icotinib	0.33 (0.13–0.87)	NA	0.34 (0.19–0.61)
Zhu, Q.2017	retrospective studies	SQ and No sq	67/66	Gefitinib/erlotinib plus WBRT	Gefitinib/erlotinib	0.55(0.34–0.90)	NA	0.62 (0.41–0.93)
Byeon, S.2016	retrospective studies	SQ and No sq	59/62	Gefitinib/erlotinib plus WBRT	Gefitinib/erlotinib	1.16(0.49–2.73)	NA	1.20 (0.81–1.77)
Jiang, T.2016	retrospective studies	SQ and No sq	51/116	Gefitinib/erlotinib/icotinib plus WBRT	Gefitinib/erlotinib/icotinib	1.56(0.75–3.63)	1.09(0.68–1.76)	1.32(0.83–2.23)
Chen, Y.2016	retrospective studies	adenocarcinoma	53/79	Gefitinib/erlotinib plus WBRT	Gefitinib/erlotinib	1.03 (0.59–1.80)	NA	0.59 (0.41–0.84)
Lee, C. K.2026	Ⅱ	adenocarcinoma	39/40	Osimertinib plus SRS	Osimertinib	0.66(0.35–1.24)	0.85(0.50–1.43)	0.80(0.46–1.37)
Cho, B. C.2024	III	SQ and No sq	178/173	Amivantamab plus lazertinib	Osimertinib	0.67(0.50–0.9)	0.69(0.53–0.92)	0.79(0.61–1.02)

Risk of bias assessment

RCT risk-of-bias results are presented in Supplementary Fig. 1. Several open-label trials had high risk of performance or detection bias, and some domains were unclear because methodological details were insufficiently reported. Overall, the randomized evidence varied in methodological quality, although most trials were judged to have a low risk of bias across several key domains. The 31 retrospective or non-randomized comparative studies were assessed using the Risk Of Bias In Non-randomized Studies of Interventions tool. The results of this assessment are presented in Supplementary Table 1.

Meta-analysis results

The contribution of each study to the OS, PFS, and iPFS networks was determined by the availability of extractable endpoint-specific HRs. A study-by-endpoint contribution matrix, including HR availability and data source, is provided in Supplementary Table 2.

The outcome-specific treatment networks are shown in Fig. 2A-C. A total of 38, 24, and 27 studies contributed to the quantitative network meta-analyses of OS, PFS, and iPFS, respectively. Although the overall evidence base included 48 studies across five treatment categories, the network geometry was assessed separately for each outcome.All three efficacy networks had a star-shaped structure centered on EGFR-TKI monotherapy. No direct head-to-head comparisons were available between active combination regimens. Therefore, comparisons among combination strategies were based on indirect evidence through the common comparator, EGFR-TKI monotherapy. Given the potential influence of EGFR-TKI generation on CNS penetration and intracranial efficacy, generation-stratified subgroup analyses were performed where data permitted.

Fig. 2.

Network plots for overall survival (OS) (A), progression-free survival (PFS) (B), and intracranial progression-free survival (iPFS) (C). Node size is proportional to the number of participants receiving each treatment, and edge width is proportional to the number of studies contributing to each direct comparison.

OS

The OS NMA included 38 studies. Compared with EGFR-TKI monotherapy, the most favorable OS estimate was observed for EGFR-TKI plus anti-angiogenic therapy (HR, 0.63; 95% CI, 0.49–0.80), followed by bispecific antibody therapy (HR, 0.67; 95% CI, 0.50–0.90), radiotherapy (HR, 0.77; 95% CI, 0.70–0.84), and chemotherapy (HR, 0.82; 95% CI, 0.67–0.99) (Fig. 3A). No indirect comparison between active combinations showed a statistically significant OS difference (Supplementary Table 3 A). Ranking probabilities favored anti-angiogenic therapy but were interpreted as exploratory because of the star-shaped common-comparator network (Supplementary Table 4).

Fig. 3.

Forest plot of comparison of OS (A), PFS (B), and iPFS (C).

In the EGFR-TKI generation subgroup analysis (Fig. 4A), OS benefits were most apparent in first-generation EGFR-TKI-based studies, with HRs of 0.63, 0.65, and 0.67 for anti-angiogenic therapy, chemotherapy, and radiotherapy, respectively. For third-generation EGFR-TKI-based regimens, a statistically significant OS benefit was observed only for bispecific antibody therapy (HR, 0.67), whereas chemotherapy (HR, 0.87) and radiotherapy (HR, 0.81) were not significant.

Fig. 4.

Forest plot of subgroup analyses stratified by EGFR-TKI generation: overall survival (A), progression-free survival (B), and intracranial progression-free survival (C).

No extractable OS HR was available for third-generation EGFR-TKI plus anti-angiogenic therapy. Therefore, although EGFR-TKI plus anti-angiogenic therapy ranked most favorably for OS in the overall network, this finding should not be interpreted as evidence of a confirmed OS benefit when anti-angiogenic therapy is added to osimertinib. Finally, limited data for second-generation EGFR-TKIs showed no significant OS benefit from the addition of radiotherapy (HR, 0.99).

PFS

The PFS NMA included 24 studies. In the overall treatment-strategy-level network, EGFR-TKI-based combination strategies generally showed favorable PFS estimates compared with EGFR-TKI monotherapy(Fig. 3B). EGFR-TKI plus chemotherapy had the most favorable PFS estimate versus EGFR-TKI monotherapy (HR, 0.52; 95% CI, 0.44–0.61). Significant PFS benefits were also observed for EGFR-TKI plus anti-angiogenic therapy and EGFR-TKI plus radiotherapy, whereas EGFR-TKI plus bispecific antibody therapy showed a favorable PFS estimate based on a single RCT (HR, 0.69).In the indirect comparison between active combination strategies, EGFR-TKI plus chemotherapy was associated with a significantly greater PFS benefit than EGFR-TKI plus radiotherapy (HR, 0.72; 95% CI, 0.53–0.96) (Supplementary Table 3B). Ranking probabilities suggested that EGFR-TKI plus chemotherapy had the highest probability of being the best-ranked strategy for PFS, with a ranking probability of 91.1%, whereas EGFR-TKI monotherapy had the highest probability of being ranked last, with a probability of 99.1% (Supplementary Table 4). However, as these comparisons were based on an indirect common-comparator network, the ranking results should be interpreted cautiously.

The subgroup analysis stratified by EGFR-TKI generation showed distinct patterns of efficacy (Fig. 4B). The addition of chemotherapy was associated with a consistent and statistically significant PFS benefit across first-, second-, and third-generation EGFR-TKI-based regimens, with HRs ranging from 0.43 to 0.61.In contrast, the PFS benefits of EGFR-TKI plus anti-angiogenic therapy and EGFR-TKI plus radiotherapy were mainly observed in first-generation EGFR-TKI-based combinations. For third-generation EGFR-TKI plus anti-angiogenic therapy, only two studies provided PFS data, and the results were numerically favorable but not statistically significant; no extractable OS or iPFS HRs were available for this subgroup. Finally, third-generation EGFR-TKI plus bispecific antibody therapy showed a significant PFS benefit (HR, 0.69), although this estimate was based on a single RCT and should be considered preliminary.

iPFS

The iPFS NMA included 27 studies. Compared with EGFR-TKI monotherapy, EGFR-TKI plus chemotherapy had the most favorable iPFS estimate (HR, 0.36; 95% CI, 0.25–0.52), followed by EGFR-TKI plus anti-angiogenic therapy (HR, 0.51; 95% CI, 0.37–0.71), EGFR-TKI plus radiotherapy (HR, 0.61; 95% CI, 0.56–0.66), and EGFR-TKI plus bispecific antibody therapy (HR, 0.79; 95% CI, 0.61–1.00). The estimate for EGFR-TKI plus bispecific antibody therapy was not statistically significant and was supported by only one RCT, precluding a definitive conclusion regarding its intracranial efficacy.

In indirect comparisons, EGFR-TKI plus chemotherapy showed significantly more favorable iPFS estimates than radiotherapy and bispecific antibody therapy, with a non-significant numerical advantage over anti-angiogenic therapy (Supplementary Table 3 C). Anti-angiogenic therapy also showed a significantly more favorable iPFS estimate than bispecific antibody therapy. Ranking probabilities favored chemotherapy but were again interpreted cautiously because all comparisons were indirect (Supplementary Table 4).

The subgroup analysis stratified by EGFR-TKI generation provided further insight into intracranial efficacy (Fig. 4C). Among first-generation EGFR-TKI-based regimens, the addition of chemotherapy was associated with the most favorable iPFS estimate (HR, 0.43; 95% CI, 0.33–0.56). Radiotherapy was associated with significant iPFS benefits when added to both first-generation and third-generation EGFR-TKIs, with HRs of 0.69 and 0.85, respectively. Anti-angiogenic therapy was also associated with a significant iPFS benefit when combined with first-generation EGFR-TKIs (HR, 0.60). In contrast, third-generation EGFR-TKI plus bispecific antibody therapy did not show a statistically significant improvement in iPFS.

Subgroup analysis results

RCT-only NMA as a primary robustness check

Because the main treatment-strategy network included comparative evidence from both randomized and non-randomized studies, RCT-only networks were constructed for OS, PFS, and iPFS. These analyses were used to assess whether the comparative hierarchy among EGFR-TKI-based combination strategies was influenced by study design. In accordance with the prespecified analytical strategy, analyses were first restricted to RCTs. Sensitivity analyses were then conducted using an alternative model specification. These analyses were not intended to replace the full evidence synthesis. Rather, they were performed to evaluate the consistency and precision of the network estimates within a more restricted randomized evidence base.

For OS, the RCT-only analysis generally retained the favorable direction of effect observed in the main network. However, differences among combination strategies became less distinct. In the full primary network, EGFR-TKI plus anti-angiogenic therapy had the most favorable OS ranking. When the analysis was restricted to RCTs, all combination nodes had favorable point estimates compared with EGFR-TKI monotherapy. These included EGFR-TKI plus radiotherapy (HR, 0.66; 95% CI, 0.35–1.20), EGFR-TKI plus bispecific antibody therapy (HR, 0.67; 95% CI, 0.50–0.90), EGFR-TKI plus chemotherapy (HR, 0.78; 95% CI, 0.64–0.95), and EGFR-TKI plus anti-angiogenic therapy (HR, 0.82; 95% CI, 0.59–1.10). After the alternative model was applied, the point estimates remained directionally favorable, but the CIs became wider: EGFR-TKI plus radiotherapy, 0.66 (95% CI, 0.30–1.50); EGFR-TKI plus bispecific antibody therapy, 0.67 (95% CI, 0.37–1.20); EGFR-TKI plus chemotherapy, 0.78 (95% CI, 0.57–1.10); and EGFR-TKI plus anti-angiogenic therapy, 0.85 (95% CI, 0.56–1.30) (Supplementary Fig. 2). These findings indicate that the OS results were directionally consistent with the primary analysis. However, when the evidence base was restricted to RCTs and between-study heterogeneity was considered, the relative ordering of active combination strategies became less precise. Therefore, the OS ranking derived from the full network should be interpreted as hypothesis-generating rather than definitive.

For PFS, the RCT-only analysis and the model-based sensitivity analysis provided the most consistent support for the findings of the primary network. In the main network, EGFR-TKI plus chemotherapy had the most favorable estimate for delaying disease progression. This pattern was largely retained in the RCT-only network. EGFR-TKI plus chemotherapy had the most favorable PFS estimate compared with EGFR-TKI monotherapy (HR, 0.50; 95% CI, 0.42–0.60), followed by EGFR-TKI plus anti-angiogenic therapy (HR, 0.69; 95% CI, 0.56–0.85) and EGFR-TKI plus bispecific antibody therapy (HR, 0.69; 95% CI, 0.52–0.91). EGFR-TKI plus radiotherapy also had a favorable but imprecise estimate (HR, 0.85; 95% CI, 0.50–1.40). After the alternative model was applied, EGFR-TKI plus chemotherapy remained the most consistently supported strategy for PFS (HR, 0.49; 95% CI, 0.36–0.68). EGFR-TKI plus anti-angiogenic therapy also maintained a significant but more moderate PFS benefit (HR, 0.71; 95% CI, 0.52–0.97). EGFR-TKI plus bispecific antibody therapy (HR, 0.69; 95% CI, 0.34–1.40) and EGFR-TKI plus radiotherapy (HR, 0.85; 95% CI, 0.37–1.90) continued to show favorable trends. However, their CIs crossed unity (Supplementary Fig. 3). These findings suggest that the PFS hierarchy was relatively stable across the primary network, the RCT-only fixed-effect analysis, and the RCT-only random-effects sensitivity analysis. Among the available combination strategies, EGFR-TKI plus chemotherapy had the most consistent randomized evidence for PFS benefit. EGFR-TKI plus anti-angiogenic therapy also provided supportive randomized evidence.

For iPFS(Supplementary Fig. 4), the RCT-only analysis supported the overall trend observed in the primary network. However, the strength of inference was limited by the scarcity of randomized intracranial evidence. EGFR-TKI plus anti-angiogenic therapy had the most favorable iPFS estimate compared with EGFR-TKI monotherapy (HR, 0.18; 95% CI, 0.05–0.66), followed by EGFR-TKI plus chemotherapy (HR, 0.36; 95% CI, 0.25–0.52). These findings were directionally consistent with the primary network. In both analyses, EGFR-TKI plus chemotherapy and EGFR-TKI plus anti-angiogenic therapy appeared to be the main strategies associated with improved intracranial disease control.

Overall, the RCT-only sensitivity analysis did not materially overturn the main direction of the primary network. The PFS findings were the most robust, particularly for TKI plus chemotherapy. The iPFS findings also remained directionally supportive of TKI plus chemotherapy, but were limited by sparse randomized evidence. In contrast, the OS treatment hierarchy became less definitive after restriction to RCTs, suggesting that OS-based rankings in the primary network should be regarded as hypothesis-generating rather than definitive.

Mutation type

Subgroup analyses by EGFR mutation subtype suggested broadly consistent intracranial benefits across exon 19 deletion and L858R, although OS patterns varied by endpoint and treatment strategy.

For iPFS, EGFR-TKI plus chemotherapy ranked most favorably in both exon 19 deletion (HR, 0.29; 95% CI, 0.17–0.50) and L858R (HR, 0.34; 95% CI, 0.19–0.62) subgroups (Supplementary Fig. 5A-B and Supplementary Table 4). For OS, chemotherapy and anti-angiogenic therapy were significant in the exon 19 deletion subgroup, whereas anti-angiogenic therapy and radiotherapy were significant in the L858R subgroup (Supplementary Fig. 6A-B). Because these analyses used aggregate study-level data, they should be interpreted as exploratory.

Age

Age-based subgroup analyses showed generally similar iPFS patterns in patients aged ≥ 65 and < 65 years, but they were based on aggregate data and did not formally test treatment-by-age interactions (Supplementary Figs. 7 and 8).

For iPFS, chemotherapy and anti-angiogenic therapy were significant in both age subgroups, whereas radiotherapy was not. For OS, anti-angiogenic therapy was significant in both age subgroups; chemotherapy and radiotherapy showed numerically favorable but non-significant estimates in patients aged < 65 years, and chemotherapy was not significant among patients aged ≥ 65 years.

Overall, EGFR-TKI plus chemotherapy and EGFR-TKI plus anti-angiogenic therapy showed favorable iPFS estimates in both age subgroups, whereas the iPFS benefit of EGFR-TKI plus radiotherapy was not statistically significant. These findings suggest that age alone may not fully explain differences in intracranial treatment benefit, although further analyses using individual patient-level data are needed to determine whether age modifies the effect of specific combination strategies.

Sex

Sex-based subgroup analyses suggested that the efficacy patterns of different EGFR-TKI-based combination strategies varied between male and female patients for iPFS and OS. However, because these analyses were based on aggregate study-level data and were not designed to formally test treatment-by-sex interactions, the findings should be interpreted as exploratory.

For iPFS (Supplementary Fig. 9 A and 9B), EGFR-TKI plus chemotherapy was associated with significantly improved iPFS compared with EGFR-TKI monotherapy in both male and female patients. The HR was 0.43 (95% CI, 0.25–0.73) in male patients and 0.28 (95% CI, 0.16–0.50) in female patients. EGFR-TKI plus anti-angiogenic therapy was associated with a significant iPFS benefit in female patients (HR, 0.56; 95% CI, 0.40–0.78), whereas the estimate in male patients was numerically favorable but not statistically significant (HR, 0.57; 95% CI, 0.32–1.00). EGFR-TKI plus radiotherapy was also associated with a significant iPFS benefit in female patients (HR, 0.59; 95% CI, 0.44–0.80), but not in male patients (HR, 0.73; 95% CI, 0.53–1.00).

For OS (Supplementary Fig. 10 A and 10B), EGFR-TKI plus anti-angiogenic therapy was associated with significantly improved survival compared with EGFR-TKI monotherapy in both male patients (HR, 0.46; 95% CI, 0.25–0.82) and female patients (HR, 0.53; 95% CI, 0.37–0.76). EGFR-TKI plus chemotherapy showed numerically favorable but statistically non-significant OS estimates in both sexes. EGFR-TKI plus radiotherapy did not show a clear OS benefit in either male or female patients.

Overall, EGFR-TKI plus chemotherapy showed favorable iPFS estimates in both sexes, while EGFR-TKI plus anti-angiogenic therapy showed significant OS benefits in both male and female patients. Some sex-specific differences were observed for iPFS, particularly for anti-angiogenic therapy and radiotherapy; however, these findings should be considered hypothesis-generating and require validation using individual patient-level data.

Exploratory pairwise meta-analysis of radiotherapy sequence

To further explore whether the timing of radiotherapy modified the efficacy of EGFR-TKI plus radiotherapy, we conducted a conventional pairwise subgroup meta-analysis stratified by radiotherapy sequence. The EGFR-TKI plus radiotherapy node included WBRT, SRS/SRT, mixed radiotherapy approaches, and variable dose-fractionation schedules. Because modality-specific HRs were unavailable in many studies, a reliable WBRT-versus-SRS/SRT comparison could not be performed. Therefore, the pooled estimates should be interpreted as overall radiotherapy-class effects rather than evidence supporting any specific radiotherapy modality, dose schedule, or sequencing strategy.

For OS (Supplementary Fig. 11 A), concurrent EGFR-TKI plus radiotherapy was associated with a favorable estimate compared with EGFR-TKI monotherapy (HR, 0.81; 95% CI, 0.71–0.92; P = 0.001), whereas sequential radiotherapy followed by EGFR-TKI did not show a statistically significant OS benefit (HR, 0.95; 95% CI, 0.79–1.16; P = 0.63). The overall pooled estimate suggested a modest OS benefit with the addition of radiotherapy (HR, 0.85; 95% CI, 0.76–0.95; P = 0.003). However, the test for subgroup differences was not statistically significant (P = 0.17), indicating that the apparent difference between concurrent and sequential approaches should be interpreted cautiously.

For PFS (Supplementary Fig. 11B), concurrent EGFR-TKI plus radiotherapy showed a numerically favorable but statistically non-significant estimate (HR, 0.79; 95% CI, 0.61–1.02; P = 0.07). The sequential subgroup was informed by limited evidence. The overall pooled estimate suggested a PFS benefit with the addition of radiotherapy (HR, 0.72; 95% CI, 0.57–0.92; P = 0.008), although notable inter-study heterogeneity was observed.

For iPFS (Supplementary Fig. 11 C), concurrent EGFR-TKI plus radiotherapy was associated with a favorable intracranial disease-control estimate (HR, 0.53; 95% CI, 0.48–0.59). Sequential radiotherapy followed by EGFR-TKI also showed a statistically significant but smaller iPFS benefit (HR, 0.79; 95% CI, 0.64–0.97; P = 0.02), with moderate heterogeneity (I² = 59%; P = 0.05). The overall pooled estimate suggested an iPFS benefit with the addition of radiotherapy (HR, 0.58; 95% CI, 0.53–0.64; P < 0.00001), but substantial overall heterogeneity was present (I² = 76%; P < 0.00001).

Overall, concurrent EGFR-TKI plus radiotherapy showed numerically more favorable OS and iPFS estimates than sequential radiotherapy followed by EGFR-TKI. However, these findings should be regarded as exploratory and hypothesis-generating because much of the available evidence was derived from retrospective studies, and because radiotherapy modality, dose-fractionation schedule, intracranial tumor burden, and patient selection varied substantially across studies. These findings are broadly consistent with the hypothesis raised by Deng et al.⁶⁹ that radiotherapy timing may influence outcomes in selected patients. The ongoing STARLET trial⁷⁰ may help clarify the optimal timing of SRS combined with osimertinib using prospective individual patient data. Therefore, the present results should not be interpreted as establishing an optimal radiotherapy sequence, but rather as supporting the need for further prospective evaluation.

Grade ≥ 3 treatment-related adverse events

Sixteen studies reported grade ≥ 3 TRAEs. Owing to heterogeneity in adverse-event definitions, reporting thresholds, and follow-up duration, safety outcomes were summarized descriptively by treatment class rather than included in a safety NMA (Supplementary Table 5).

EGFR-TKI plus chemotherapy was mainly associated with hematologic and systemic toxicities, including anemia, neutropenia, leukopenia, thrombocytopenia, hepatic dysfunction, fatigue, and gastrointestinal events. These toxicities should be considered in relation to patient fitness, bone marrow reserve, and hepatic function.

EGFR-TKI plus anti-angiogenic therapy was mainly associated with vascular or anti-VEGF-related toxicities, including hypertension, proteinuria, bleeding-related events, diarrhea, and rash; therefore, cardiovascular risk, bleeding risk, renal function, and blood pressure control are important in treatment selection.

Radiotherapy-related grade ≥ 3 TRAEs were less uniformly reported and included fatigue, radiation-related neurotoxicity, and radiation-induced cerebral necrosis. Interpretation is limited by differences in radiotherapy modality, dose, fractionation, intracranial tumor burden, and follow-up.

Safety evidence for EGFR-TKI plus bispecific antibody therapy was limited to a single RCT and mainly reflected known EGFR/MET antibody-related toxicities, including dermatologic toxicity, infusion-related reactions, and systemic adverse events.

Overall, safety profiles differed across treatment classes. EGFR-TKI plus chemotherapy was mainly associated with hematologic and systemic toxicities, EGFR-TKI plus anti-angiogenic therapy with vascular and renal toxicities, and EGFR-TKI plus radiotherapy with radiation-related neurologic toxicities. Therefore, treatment selection should balance endpoint-specific efficacy benefits against treatment-class-specific toxicity patterns and individual patient tolerability.

Publication bias, heterogeneity, model convergence, and network-structure assessment

Funnel plots were visually inspected to explore small-study effects and potential publication bias (Supplementary Fig. 12). Model convergence was assessed using the PSRF, convergence plots, and density plots (Supplementary Figs. 13–15). The PSRF values were close to 1, indicating acceptable convergence of the Bayesian models.

Because all outcome-specific networks had a star-shaped common-comparator structure centered on EGFR-TKI monotherapy and contained no closed loops, formal node-splitting inconsistency assessment was not statistically applicable. Therefore, inconsistency could not be quantitatively evaluated and was instead considered descriptively according to the network geometry. This network structure was also taken into account when interpreting indirect comparisons and treatment-ranking probabilities.

Discussion

This Bayesian common-comparator NMA summarized first-line EGFR-TKI-based combinations for EGFR-mutant NSCLC with brain metastases. In the overall network, EGFR-TKI plus anti-angiogenic therapy showed the most favorable OS estimate, whereas EGFR-TKI plus chemotherapy showed the most favorable PFS and iPFS estimates. However, all efficacy networks were star-shaped and lacked direct head-to-head comparisons between active combinations. Therefore, endpoint-specific rankings should be interpreted as hypothesis-generating rather than definitive.

EGFR-TKI generation represents an important source of clinical heterogeneity because CNS penetration and intracranial efficacy differ across agents. Therefore, the pooled all-generation network should be interpreted as a strategy-level comparison rather than as evidence that different EGFR-TKIs are interchangeable. Generation-stratified analyses suggested that some benefits were more apparent in studies using first-generation EGFR-TKIs, whereas evidence for third-generation EGFR-TKI-based combinations remained limited and endpoint-specific. These findings underscore the need for caution when extrapolating results derived predominantly from first-generation EGFR-TKI-based studies to current osimertinib-based treatment practice.

The RCT-only analysis supported the overall direction of the full-network findings but also showed that the robustness of the evidence differed across endpoints. The PFS benefit of EGFR-TKI plus chemotherapy remained the most consistent finding, whereas OS-based rankings became less stable after restriction to randomized evidence, and iPFS estimates were limited by sparse randomized intracranial data. These findings suggest that the results should be interpreted as endpoint-specific evidence rather than as indicating uniform superiority of any combination strategy across all efficacy outcomes.

Compared with previous network meta-analyses, the present study provides updated evidence focused specifically on patients with EGFR-mutant NSCLC and brain metastases. Ma et al.⁷¹ conducted a NMA but did not provide detailed subgroup analyses for this population and did not include the most recent FLAURA2 data. Taslimi et al.⁷² had a literature search cutoff in 2020 and therefore did not capture key evidence on third-generation EGFR-TKI-based combinations. By incorporating more recent studies and evaluating OS, PFS, and iPFS simultaneously, our analysis may help characterize how different combination strategies vary across clinically relevant endpoints. Nevertheless, these findings should not be interpreted as establishing a definitive treatment hierarchy, particularly for treatment nodes supported by limited evidence, such as EGFR-TKI plus bispecific antibody therapy. Treatment selection should remain individualized according to EGFR-TKI generation, intracranial disease burden, treatment goals, performance status, expected toxicity, and patient tolerability.

The favorable OS estimate for anti-angiogenic therapy is biologically plausible because blood-tumor barrier disruption may produce heterogeneous drug distribution and elevated interstitial pressure⁷³, and VEGF signaling has been implicated in angiogenesis and acquired resistance to EGFR-TKIs⁷⁴. Nevertheless, this evidence was derived mainly from first-generation EGFR-TKI studies, and the incremental benefit of adding anti-angiogenic therapy to osimertinib remains uncertain. Because the OS advantage was attenuated in RCT-only analyses, it should be regarded as hypothesis-generating; the more consistent randomized signal for this strategy is PFS benefit.

The most consistent finding across the full network and the RCT-only analyses was the PFS benefit of EGFR-TKI plus chemotherapy. This strategy remained associated with improved PFS compared with EGFR-TKI monotherapy in both fixed-effect and random-effects RCT-only models. The systemic benefit of adding chemotherapy may be explained, at least in part, by the clonal heterogeneity hypothesis. EGFR-mutant NSCLC may contain EGFR-TKI-insensitive subclones even at initial diagnosis, and early cytotoxic elimination of these resistant populations may delay disease progression⁷⁵. For iPFS, EGFR-TKI plus chemotherapy also showed favorable estimates. Local disruption of the BBB at metastatic sites may allow cytotoxic agents to reach therapeutic concentrations in intracranial lesions⁷⁶. However, because the RCT-only iPFS network was sparse and the random-effects model produced wide CIs, the intracranial benefit of EGFR-TKI plus chemotherapy should be considered supportive rather than definitive. Therefore, EGFR-TKI plus chemotherapy appears to provide the most consistent evidence for delaying disease progression, whereas its potential intracranial disease-control benefit requires further prospective confirmation.

The present study adds to previous evidence in several ways. Earlier meta-analyses by Chen et al.⁷⁷ and Jiang et al.⁴² primarily compared first-generation EGFR-TKIs with WBRT and therefore provide limited guidance for current osimertinib-based practice. A more recent study by Dai et al.¹⁵ compared EGFR-TKI plus anti-angiogenic therapy with EGFR-TKI plus chemotherapy, but did not specifically focus on patients with brain metastases and did not include more recent pivotal evidence, including FLAURA2 and MARIPOSA. By incorporating recent trials and evaluating OS, PFS, iPFS, and safety together, the present analysis provides an updated overview of first-line EGFR-TKI-based combination strategies in this clinically important subgroup.

Several aspects of this study may improve its clinical interpretability. First, the inclusion of recent trials, including FLAURA2 and MARIPOSA, allowed the analysis to reflect more contemporary treatment strategies, including osimertinib plus chemotherapy and bispecific antibody-based therapy. Second, the NMA framework enabled the synthesis of direct comparisons against EGFR-TKI monotherapy and indirect comparisons between active combination strategies through a common comparator, although no direct head-to-head comparisons between active combinations were available. Third, we explored radiotherapy sequence using a conventional pairwise subgroup meta-analysis. However, because the radiotherapy node included heterogeneous modalities, including WBRT, SRS/SRT, mixed approaches, and variable dose-fractionation schedules, this analysis should be regarded as exploratory and cannot determine the optimal radiotherapy subtype, dose, or timing strategy.

The interpretation of radiotherapy-based combination strategies should be cautious because the EGFR-TKI plus radiotherapy node pooled WBRT, SRS/SRT, and mixed radiotherapy approaches. Given the clinical differences among these modalities and the substantial heterogeneity observed across studies, the pooled estimate should be regarded as hypothesis-generating rather than as evidence supporting any specific radiotherapy modality, dose-fractionation schedule, or sequencing strategy.

Although concurrent EGFR-TKI plus radiotherapy showed favorable estimates in some analyses, the RCT-only analyses did not confirm consistent PFS or iPFS benefits. The potential benefit of early radiotherapy remains biologically and clinically plausible, as EGFR-TKIs may impair DNA damage repair after radiotherapy and thereby enhance radiosensitivity⁷⁸, whereas delaying radiotherapy until intracranial progression may cause some patients to miss the optimal window for local treatment⁴⁷. However, radiotherapy allocation in retrospective studies is strongly influenced by intracranial tumor burden, neurological symptoms, performance status, expected survival, treatment timing, and physician preference. Therefore, the role and optimal timing of radiotherapy combined with EGFR-TKIs should be regarded as suggestive rather than definitive and require further prospective randomized evaluation.

Evidence for EGFR-TKI plus bispecific antibody therapy was also preliminary. The node was supported by a single randomized trial of amivantamab plus lazertinib, so class-level conclusions for patients with brain metastases are not yet justified⁶⁸. Further randomized trials with longer follow-up and CNS-specific endpoints are warranted.

Several limitations should be acknowledged. First, studies using first- to third-generation EGFR-TKIs were pooled, which introduced clinical heterogeneity because these agents differ in systemic efficacy, CNS penetration, and intracranial activity. Generation-stratified analyses were performed where data allowed, but pooled all-generation estimates should remain strategy-level comparisons. Second, the primary network included both randomized and non-randomized comparative studies. Retrospective and non-randomized evidence may be affected by confounding by indication and treatment-selection bias, particularly for radiotherapy-based strategies, in which treatment allocation is often influenced by intracranial tumor burden, neurological symptoms, performance status, expected survival, and physician preference. Although RCT-only analyses were performed as robustness checks, sparse randomized evidence, particularly for iPFS, limits the certainty of OS- and iPFS-based treatment rankings. Third, the radiotherapy node pooled WBRT, SRS/SRT, mixed approaches, and different dose-fractionation schedules because modality-specific HRs were not consistently extractable. Thus, the radiotherapy estimates should not be interpreted as evidence favoring a specific modality, dose schedule, or sequence. Fourth, the outcome-specific networks had a star-shaped common-comparator structure centered on EGFR-TKI monotherapy and lacked direct head-to-head comparisons between active combination strategies. As no closed loops were present, formal inconsistency testing using node-splitting was not statistically applicable. Consequently, indirect comparisons and treatment-ranking probabilities should be interpreted cautiously in light of this network structure and the underlying transitivity assumption. Fifth, the EGFR-TKI plus bispecific antibody node was supported by only one RCT. Therefore, its effect estimates, indirect comparisons, and ranking probabilities remain exploratory and require validation in further randomized trials with mature survival data and CNS-specific outcomes.Finally, data limitations precluded a separate analysis of patients with leptomeningeal metastases. Leptomeningeal metastases were inconsistently reported across the included studies, and treatment-specific HRs for OS, PFS, or iPFS in this subgroup were generally unavailable. Therefore, the present findings should not be directly extrapolated to patients with leptomeningeal metastases.

Conclusion

First-line treatment for patients with EGFR-mutant NSCLC and brain metastases should not rely on a universal treatment hierarchy. EGFR-TKI plus chemotherapy showed the most consistent RCT-supported evidence, particularly for delaying disease progression, whereas EGFR-TKI plus anti-angiogenic therapy showed favorable PFS and iPFS estimates but less definitive OS evidence. Because CNS penetration and intracranial activity differ across EGFR-TKI generations, pooled findings across EGFR-TKI generations should be interpreted as treatment-strategy-level evidence rather than as evidence that different EGFR-TKIs are interchangeable. Findings for radiotherapy-based and bispecific antibody-based combinations remain exploratory because of radiotherapy heterogeneity and limited randomized data. Therefore, treatment selection should be individualized according to therapeutic goals, EGFR-TKI generation, intracranial disease burden, molecular characteristics, expected toxicity, and patient tolerability.

Supplementary Information

Below is the link to the electronic supplementary material.

Supplementary Material 1 ^{(15.9KB, docx)}

Supplementary Material 2 ^{(15.9KB, docx)}

Supplementary Material 3 ^{(1.4MB, docx)}

Supplementary Material 4 ^{(18.5KB, docx)}

Supplementary Material 5 ^{(216.8KB, docx)}

Author contributions

**Jiayun Ma: ** Conceptualization; Formal analysis; Methodology; Validation; Writing – original draft; Writing – review & editing; **Weixing Zhao: ** Conceptualization; Methodology; Software; Visualization; Writing – original draft; Writing – review & editing; **Wanjing Guo: ** Formal analysis;; Supervision; Writing – original draft; Writing – review & editing; **Zirui Li: ** Software; Supervision; Writing – original draft; Writing – review & editing; **Xinxin Lu: ** Validation; Visualization; Writing – original draft; Writing – review & editing; **Xiujin Yang: ** Formal analysis; Supervision; Writing – original draft; Writing – review & editing; **Chenyu Jia: ** Software; Supervision; Writing – original draft; Writing – review & editing; **Qi Zhou: ** Formal analysis; Investigation; Writing – original draft; Writing – review & editing; **Jun Jiang: ** Conceptualization; Funding acquisition; Methodology; Writing – original draft; Writing – review & editing.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This paper is funded by the Science and Technology Support Qinghai Project (2025-QY-253) from the Science and Technology Department of Qinghai Province.

Data availability

All data generated or analysed during this study are included in this article. Further enquiries can be directed to the corresponding author.

Declarations

Competing interests

The authors declare no competing interests.