Introduction

Brain metastasis (BM) is one of the most common types of metastases in patients with non-small cell lung cancer (NSCLC) and poses a great threat to survival and quality of life for patients.[1,2] Approximately 20% of NSCLC patients are initially diagnosed with BM, and BM may occur in 20% to 50% of NSCLC patients.[3,4] Epidermal growth factor receptor (EGFR) gene mutations are common among NSCLC patients, with the frequency of approximately 50% in Asian patients and 10% to 20% in Caucasian patients;[5–7] and patients with EGFR-sensitive mutations have a higher propensity to develop BM than those who are negative.[8,9] Compared with chemotherapy, EGFR tyrosine kinase inhibitor (TKI) therapy significantly prolongs overall survival (OS) in metastatic EGFR-mutated NSCLC patients.[9] For symptomatic and uncontrolled BM, local therapy is recommended as the first-line treatment.[10] For stable BM, which refers to clinically asymptomatic and controlled BM, systemic therapy is recommended.

Several EGFR-TKIs are recommended for the first-line therapy in metastatic EGFR-mutated NSCLC, such as first-generation EGFR-TKIs (gefitinib, erlotinib, and icotinib),[10] second-generation EGFR-TKIs (dacomitinib and afatinib), and third-generation EGFR-TKIs (osimertinib). Combined therapy of EGFR-TKIs and other drugs (antiangiogenic drugs and chemotherapy) has also been established as a first-line treatment to enhance efficacy and overcome drug resistance.[11] Because of the blood–brain barrier, patients with BM have a poor intracranial response to chemotherapy and first-/second-generation EGFR-TKIs.[12] The penetration (cerebral spinal fluid [CSF]/plasma or CSF/blood) of first-generation and second-generation EGFR-TKIs was only 1.1% to 3.3% and 1.7%, respectively.[13] The BRAIN study indicated that icotinib might be better for patients with EGFR-mutated NSCLC accompanied by multiple BM, with a significantly longer intracranial progression-free survival (PFS) than whole-brain radiation therapy plus chemotherapy.[14] However, no difference in OS was observed in this study. A preclinical study indicated that compared with other EGFR-TKIs, osimertinib provided markedly greater central nervous system activity in EGFR-mutated NSCLC BM models.[15] The penetration (CSF/plasma or CSF/blood) of osimertinib was 2.5–16.0%,[13] and the objective response rate (ORR) was 76%.[16] However, in the subgroup analysis of FLAURA, osimertinib only showed a favorable trend in OS.[16,17] The combination of first-generation TKIs with antiangiogenic agents and chemotherapy improved PFS in advanced, EGFR-mutated NSCLC patients, even in those with BM.[18,19] However, the existing data on treatments for BM are mostly derived from subgroup analyses, lacking phase III studies and direct comparisons. Therefore, the optimal first-line therapy in NSCLC patients carrying EGFR-sensitive mutations with stable BM remains controversial. In addition, the prognosis of patients with BM is poorer than that of patients without BM. Hence, regimens should be selected according to the BM status. To address these questions, we conducted a network meta-analysis (NMA) of randomized controlled trials (RCTs) by comparing a number of interventions with diverse contrasts simultaneously, which is widely used in the absence of head-to-head trial data,[20] to gain insight into the relative efficacy of first-line regimens in advanced EGFR-mutated NSCLC patients with and without BM.

Methods Database and search strategy

RCTs focusing on EGFR-TKIs (alone or in combination) in advanced and EGFR-mutant NSCLC patients, who have not received systematic treatment, were systematically searched from the PubMed, Cochrane Library, and EMBASE with the terms NSCLC, EGFR, TKI, first-line, PFS, OS, and RCT up to December 31, 2021. Detailed search strategies are available in Supplementary Table 1, https://links.lww.com/CM9/B524. To acquire data from unpublished trials, abstracts presented at the conferences of the American Society of Clinical Oncology and the European Society for Medical Oncology in 2021 were further searched. This protocol was registered with the International Platform of Registered Systematic Review and Meta-Analysis Protocols (INPLASY) on October 06, 2020, and it was last updated on October 06, 2020 (registration number INPLASY2020100018).

Outcome definition

The primary and secondary outcomes in this study were PFS and OS, respectively.

Screening studies

To guarantee transitivity, trials with strict criteria for patient grouping were identified and included. Studies that met all of the following inclusion criteria were included in the final analysis: (1) study patients: advanced or metastatic NSCLC with EGFR-sensitive mutations; (2) interventions: EGFR-TKIs with or without anti-vascular endothelial growth factor receptor or chemotherapy; (3) contrasts: chemotherapy or EGFR-TKIs; (4) outcomes: PFS and/or OS; and (5) study design: RCT.

We excluded studies according to the exclusion criteria: (1) without EGFR-mutant patients; (2) without the intervention of EGFR-TKIs; (3) without survival outcome; (4) non-first-line therapy; (5) comment, review, protocol, or editor opinion; and (6) duplicated studies.

Data extraction and risk assessment for bias

Based on eligibility criteria initially predefined by our research working group, the titles and abstracts of identified records were screened, and then the full texts of potentially eligible studies were further assessed. For the final included studies, our research group extracted the data and conducted a risk assessment. A form was pre-designed by a review working group for data extraction, including the following information: (1) basic information: first author, year of publication, and country; (2) trial design: design types, population characteristics, sample size, and treatments in the control and intervention groups, and (3) outcomes: data on PFS and OS.

We used the Cochrane risk-of-bias tool[21] to evaluate allocation concealment (selection bias), random sequence generation (selection bias), blinding of outcome assessment (detection bias), blinding of participants and personnel (performance bias), selective reporting (reporting bias), incomplete outcome data (attrition bias), and other biases.

Statistical analysis

The random-effect model was adopted for the analysis. The hazard ratios (HRs) and the associated standard errors of OS and PFS were used for NMA. The network plots were produced using Stata software (version 15.0, StataCorp LLC, College Station, TX, USA) to suggest interactions among the evaluated regimens from the included trials.[21] Heterogeneity across the included studies was assessed by the Q test and I2 statistics. I2 values <25%, 25% to 50%, and >50% indicated low, moderate, or high heterogeneity, respectively.[22]

The Bayesian framework was used with a Markov chain Monte Carlo simulation technique with R software (version 3.6.3, https://www.r-project.org/). In addition, we applied the HRs and 95% CI to compare the intervention measures.[23] Owing to the limited number of included studies and the lack of direct comparisons for most evaluated interventions, a random-effect model was used using the “gemtc” and “rjags” packages. To fit the model, we set four different series of initial values for five kinds of parameters, including the number of chains, tuning iterations, simulation iterations, and thinning intervals. For both PFS and OS, our analysis generated 10,000 sample iterations with 5000 burn-ins and a thinning interval of five. Using visual inspection of the four chains, we estimated the convergence of iterations to establish homogenous parameter estimation through the density plot.[24] Under the Bayesian framework, overall rankings of therapies were estimated. For the ranking, higher cumulative probability means better regimen. Transitivity and consistency are two key assumptions that had to be met to conduct the NMA. To guarantee transitivity, trials with strict criteria for patient allocation were identified and included, and the same condition for evaluated treatments was optimized. Model fit was assessed by examining the posterior deviance information criterion (DIC). We performed a weighted integration of the evidence through fitting random- effects NMA models. Inconsistency was assessed by comparing the DICs of our primary analyses (based on NMA models that assume consistency between direct and indirect evidence) and the DICs yielded by inconsistency models (which provide effect estimates based on direct evidence only) by using the node-splitting procedure for all loops.

Results Characteristics

We identified 4841 unique records from the dataset, 4534 articles were excluded by initial title and abstract screening, and 307 articles were retrieved and reviewed. Finally, a total of 17 RCTs with 5077 patients were included in this study [Figure 1], among which 12 studies included 825 patients with BM, while 17 studies included 4252 patients without BM [Table 1].[16–19,25–40] A total of 12 intervention treatments, including osimertinib + bevacizumab, aumolertinib, osimertinib, afatinib, dacomitinib, standards of care (SoC, including gefitinib, erlotinib, or icotinib), SoC + apatinib, SoC + bevacizumab, SoC + ramucirumab, SoC + pemetrexed based chemotherapy (PbCT), PbCT, and pemetrexed free chemotherapy (PfCT), were finally included.

Figure 1:

Flow chart of the screening, exclusion and inclusion of this study. ASCO: American Society of Clinical Oncology; ESMO: European Society for Medical Oncology; TKIs: Tyrosine kinase inhibitors.

Table 1 - Baseline characteristics of studies included in the NMA. BM status (n/n) Outcomes Study Region Year (patients registration) Phase Clinical trial number Sample size (n/n) Median age (years) Female (%) Adenocarcinoma (%) Non-smoking (%) BM Non-BM Intervention Control PFS OS FLAURA[16,17] Global (multi-center) 2014–2016 III (double-blind) NCT02296125 279/277 64/64 63.8/62.1 99/98 65.2/63.2 53/63 226/214 Osimertinib SoC Yes Yes CTONG-1509[18] China (multi-center) 2016–2017 III (open-label) NCT02759614 157/154 57/59 61.8/62.3 100/100 NA 44/47 113/107 SoC + bevacizumab SoC Yes Yes NEJ009[19] Japan (multi-center) 2011 III (open-label) UMIN000006340 170/172 64.8/64.0∗ 67.1/62.8 99/99 56.5/56.4 50/38 120/134 SoC + PbCT SoC Yes Yes ENSURE[25] China (multi-center) 2011–2012 III (open-label) NCT01342965 110/107 57.5/56.0 61.8/60.7 95/94 71.8/69.2 None 110/107 SoC PfCT Yes Yes Noronha et al[26] India (mono-center) 2016–2018 III (open-label) CTRI/2016/08/007149 174/176 54/56 49.4/47.2 98/97 83.3/85.2 30/34 144/142 SoC + PbCT SoC Yes No Han et al[27] China (mono-center) 2011 II (open-label) NCT02148380 40/41 NA 62.5/56.1 100/100 67.5/65.9 8/7 32/34 SoC + PbCT SoC No Yes ARCHER1050[28] Global (multi-center) 2013 III (open-label) NCT01774721 227/225 62/61 64.3/55.6 NA 64.8/64.0 None 227/225 Dacomitinib SoC Yes Yes LUX-LUNG-3[29] Global (multi-center) 2009 III (open-label) NCT00949650 186/97 62/61† 63.9/67.0† 100/100 67.4/70.4 20/15 166/82 Afatinib PbCT Yes Yes LUX-LUNG-6[29] East-Asia (multi-center) 2010 III (open-label) NCT01121393 213/104 58/58†[30] 64.0/68.0† 100/100 74.8/81.1 28/18 185/86 Afatinib PfCT Yes Yes LUX-LUNG-7[31] Global (multi-center) 2011 IIb (open-label) NCT01466660 160/159 63/63 56.9/66.7 99/99 66.3/66.7 26/24 134/135 Afatinib SoC Yes Yes JO25567[32,33] Japan (multi-center) 2011–2012 II (open-label) JapicCTI-111390 75/77 67/67 60.0/66.2 99/99 56.0/58.4 None 75/77 SoC + bevacizumab SoC Yes Yes NEJ026[34–35] Japan (multi-center) 2015–2016 III (open-label) UMIN000017069 112/112 67/68 63.4/65.2 98/100 58.0/57.1 36/36 76/76 SoC + bevacizumab SoC Yes Yes RELAY[36] Global (multi-center) 2016–2018 III (double-blind) NCT02411448 224/225 65/64 62.9/63.1 96/97 59.8/61.8 None 224/225 SoC + ramucirumab SoC Yes No ACTIVE[37] China (multi-center) NA III (double-blind) NCT02824458 157/156 57/60 58.0/60.3 NA 73.2/77.6 51/41 106/115 SoC + apatinib SoC Yes No WJOG9717L[38] Japan (multi-center) 2018 II (open-label) UMIN000030206 61/61 67/66∗ 60.7/62.3 100.0//98.4 62.3/49.2 18/23 48/33 Osimertinib + bevacizumab Osimertinib Yes No AENEAS[39] China (multi-center) 2018-2019 III (open-label) NCT03849768 214/215 59/62∗ 62.6/62.8 98.1/98.1 72.9/67.0 56/59 158/156 Aumolertinib SoC Yes No BEVERLY[40] Italy (multi-center) 2016-2019 III (open-label) NCT02633189 80/80 65.9/67.7 65.0/62.5 NA 57.5/46.3 None 80/80 SoC + bevacizumab SoC Yes Yes

Data are expressed as intervention/control unless indicated otherwise. ∗Mean age. †The Median of age and the proportion of sex of different treatments came from all patients.BM: Brain metastases; EGFR: Epidermal growth factor receptor; NA: Not available; NMA: Network meta-analysis; OS: Overall survival; PbCT: Pemetrexed based chemotherapy; PfCT: Pemetrexed free chemotherapy; PFS: Progression-free survival; SoC: Standards of care, including gefitinib, erlotinib, and icotinib.

Risk of bias

The risk of bias assessment for included studies was mainly performed according to the Cochrane handbook. There is a low to medium risk of bias [Figure 2]. All included studies described the method used to generate the allocation sequence, concealed the allocation sequence, and reported the related outcomes based on the trial protocol and other potential biases [Figure 2]. No significant bias was found in most studies, except trial NCT01466660 (Lung 7), which had a high risk of bias in terms of completeness because of the inconsistent number of patients. In this study, there were 50 patients with BM in the baseline demographics, but there were 51 patients in the subgroup analyses.[31] Furthermore, most studies did not clearly state the method for blinding in terms of the intervention.

Figure 2:

Risk of bias assessment for inclusion literatures by Cochrane handbook. (A) Global assessment for the risk of different kinds of bias for all trials. (B) Independent assessment of risk of bias for each trial.

Network meta-analysis of the efficacy of different treatments

This NMA identified 16 studies for PFS and 12 studies for OS. In patients with BM, there were 11 and eight studies with PFS [Figure 3A] and OS [Figure 3B] outcomes, respectively. For patients without BM, 16 studies reported PFS [Figure 3C], while 12 studies included OS outcomes [Figure 3D].

Figure 3:

Network diagrams of comparisons on different outcomes of treatments in advanced, EGFR-mutant NSCLC patients according to the BM statuses. (A) Comparisons for PFS in patients with BM. (B) Comparisons for OS in patients with BM. (C) Comparisons for PFS in patients without BM. (D) Comparisons for OS in patients without BM. The size of the nodes relates to the number of participants in that intervention type, and the thickness of lines between the interventions relates to the number of studies for that comparison. Red lines indicate the original comparisons are available and blue lines indicate the original comparisons are not available. BM: Brain metastases; EGFR: Epidermal growth factor receptor; NSCLC: Non-small cell lung cancer; OS: Overall survival; PbCT: Pemetrexed based chemotherapy; PfCT: Pemetrexed free chemotherapy; PFS: Progression-free survival; SoC: Standards of care, including gefitinib, erlotinib, and icotinib.

PFS for all advanced, EGFR-mutant NSCLC patients

Compared with afatinib, SoC, PbCT, and PfCT, osimertinib was associated with better PFS for all patients with advanced and EGFR-mutant NSCLC (HR = 0.62, 95% CI: 0.43–0.89; HR = 0.47, 95% CI: 0.36–0.59; HR = 0.31, 95% CI: 0.17–0.51; HR = 0.16, 95% CI: 0.10–0.24; Figure 4A). Compared with SoC + apatinib, afatinib, SoC, PbCT, and PfCT, SoC + PbCT improved the PFS of all patients with advanced and EGFR-mutant NSCLC (HR = 0.66, 95% CI: 0.46–0.99; HR = 0.63, 95% CI: 0.44–0.87; HR = 0.47, 95% CI: 0.38–0.57; HR = 0.31, 95% CI: 0.18–0.49; HR = 0.16, 95% CI: 0.10–0.23; Figure 4A). Osimertinib + bevacizumab, aumolertinib, SoC + PbCT, osimertinib, SoC + apatinib, dacomitinib, SoC + ramucirumab, SoC + bevacizumab, and afatinib all showed superior efficacy in PFS for all advanced, EGFR-mutant NSCLC patients compared with SoC (HR = 0.42, 95% CI: 0.23–0.77; HR = 0.47, 95% CI: 0.35–0.62; HR = 0.47, 95% CI: 0.38–0.57; HR = 0.47, 95% CI: 0.36–0.59; HR = 0.71, 95% CI: 0.50–0.98; HR = 0.62, 95% CI: 0.47–0.82; HR = 0.64, 95% CI: 0.48–0.85; HR = 0.60, 95% CI: 0.50–0.73; HR = 0.75, 95% CI: 0.58–0.99), PbCT (HR = 0.27, 95% CI: 0.12–0.60; HR = 0.31, 95% CI: 0.17–0.52; HR = 0.31, 95% CI: 0.18–0.49; HR = 0.31, 95% CI: 0.17–0.51; HR = 0.47, 95% CI: 0.26–0.80; HR = 0.42, 95% CI: 0.24–0.68; HR = 0.43, 95% CI: 0.25–0.71; HR = 0.40, 95% CI: 0.24–0.62; HR = 0.59, 95% CI: 0.35–0.70), and PfCT (HR = 0.14, 95% CI: 0.07–0.27; HR = 0.16, 95% CI: 0.10–0.24; HR = 0.16, 95% CI: 0.10–0.23; HR = 0.16, 95% CI: 0.10–0.24; HR = 0.24, 95% CI: 0.15–0.38; HR = 0.21, 95% CI: 0.14–0.35; HR = 0.21, 95% CI: 0.14–0.33; HR = 0.20, 95% CI: 0.14–0.29; HR = 0.25, 95% CI: 0.18–0.35). Aumolertinib, SoC + PbCT, osimertinib, and SoC + ramucirumab showed superior efficacy in PFS for all advanced, EGFR-mutant NSCLC patients compared with afatinib (HR = 0.62, 95% CI: 0.41–0.90; HR = 0.63, 95% CI: 0.44–0.87; HR = 0.62, 95% CI: 0.43–0.89; HR = 0.49, 95% CI: 0.35–0.70), and SoC + PbCT showed superior efficacy than SoC + apatinib (HR = 0.66, 95% CI: 0.46–0.99). No statistical difference was observed in PFS between PbCT and SoC (HR = 1.50, 95% CI: 0.99–2.46), but both regimens showed favorable PFS compared with PfCT (SoC vs. PfCT, HR = 0.33, 95% CI: 0.24–0.47; PbCT vs. PfCT, HR = 0.50, 95% CI: 0.32–0.85). Additionally, there were no significant differences in PFS in all advanced, EGFR-mutant NSCLC patients treated with osimertinib + bevacizumab, aumolertinib, SoC + PbCT, or osimertinib; there were also no significant differences in SoC + apatinib, dacomitinib, SoC + ramucirumab, and SoC + bevacizumab [Figure 4A].

Figure 4:

HRs and its 95% confidence intervals from network meta–analysis of different therapeutic regimens in advanced, EGFR mutation NSCLC patients. Data in each cell are HRs (95% confidence intervals) for the comparison of row-defining treatment vs. column-defining treatment. (A) Pooled HRs for PFS in all patients. (B) Pooled HRs for PFS in patients with BM. (C) Pooled HRs for PFS in patients without BM. (D) Pooled HRs for OS in all patients. (E) Pooled HRs for OS in patients with BM. (F) Pooled HRs for OS in patients without BM. Significant results are in bold. BM: Brain metastases; EGFR: Epidermal growth factor receptor; HR: Hazard ratio; NMA: Network meta–analysis; NSCLC: Non-small cell lung cancer; OS: Overall survival; PbCT: Pemetrexed based chemotherapy; PfCT: Pemetrexed free chemotherapy; PFS: Progression-free survival; SoC: Standards of care, including gefitinib, erlotinib, and icotinib.

PFS for advanced and EGFR-mutant NSCLC patients with or without BM

SoC + PbCT improved PFS in patients with brain metastases compared with SoC (HR = 0.40, 95% CI: 0.17–0.95; Figure 4B). No significant differences were found among osimertinib + bevacizumab, aumolertinib, and SoC + PbCT [Figure 4B]. For patients without BM, the results were similar to the findings in all patients [Figure 4C]. Osimertinib + bevacizumab, osimertinib, aumolertinib, SoC + PbCT, dacomitinib, SoC + ramucirumab, SoC + bevacizumab, and afatinib showed superior efficacy compared with SoC (HR = 0.43, 95% CI: 0.20–0.90; HR = 0.46, 95% CI: 0.31–0.68; HR = 0.51, 95% CI: 0.34–0.77; HR = 0.50, 95% CI: 0.38–0.66; HR = 0.62, 95% CI: 0.43–0.89; HR = 0.64, 95% CI: 0.44–0.94; HR = 0.61, 95% CI: 0.48–0.76; HR = 0.71, 95% CI: 0.50–1.00), PbCT (HR = 0.29, 95% CI: 0.11–0.74; HR = 0.31, 95% CI: 0.15–0.62; HR = 0.34, 95% CI: 0.17–0.69; HR = 0.34, 95% CI: 0.18–0.64; HR = 0.42, 95% CI: 0.21–0.82; HR = 0.43, 95% CI: 0.22–0.87; HR = 0.41, 95% CI: 0.22–0.74; HR = 0.48, 95% CI: 0.31–0.75), and PfCT (HR = 0.14, 95% CI: 0.06–0.32; HR = 0.15, 95% CI: 0.09–0.26; HR = 0.17, 95% CI: 0.09–0.29; HR = 0.16, 95% CI: 0.10–0.26; HR = 0.21, 95% CI: 0.12–0.35; HR = 0.21, 95% CI: 0.12–0.39; HR = 0.20, 95% CI: 0.12–0.31; HR = 0.23, 95% CI: 0.16–0.34) in terms of PFS in patients without BM. SoC + apatinib showed relatively superior PFS in patients without BM compared with PbCT (HR = 0.44, 95% CI: 0.22–0.92) and PfCT (HR = 0.21, 95% CI: 0.12–0.39), but presented a similar effect with SoC (HR = 0.65, 95% CI: 0.42–1.03). There was no observed statistical difference in PFS in patients without BM treated with PbCT or SoC (HR = 1.49, 95% CI: 0.84–2.64), but both showed favorable PFS compared with PfCT (PfCT vs. SoC, HR = 3.09, 95% CI: 2.06–4.55; PbCT vs. PfCT, HR = 0.14, 95% CI: 0.06–0.32). In addition, no significant differences were observed among patients without BM treated with osimertinib + bevacizumab, osimertinib, aumolertinib, SoC + PbCT, SoC + apatinib, dacomitinib, SoC + ramucirumab, SoC + bevacizumab, or afatinib [Figure 4C].

OS for all advanced, EGFR-mutant NSCLC patients

SoC + PbCT had superior OS compared with SoC (HR = 0.61, 95% CI: 0.43–0.84) and PfCT (HR = 0.57, 95% CI: 0.33–0.91), and SoC + bevacizumab had superior OS compared with SoC (HR = 0.69, 95% CI: 0.55–0.89) [Figure 4D]. There were not significant differences in others treatments.

OS for advanced and EGFR-mutant NSCLC patients with or without BM

For patients with BM, no significant differences in OS were found among all treatments [Figure 4E]. For patients without BM, SoC + PbCT (HR = 0.64, 95% CI: 0.36–1.14) and SoC + bevacizumab (HR = 0.70, 95% CI: 0.47–1.09) seemed to have a trend of benefit for OS compared with SoC, but no significant differences were found for other treatments [Figure 4F].

Ranking of treatment regimens

Bayesian ranking profiles of all evaluated treatment regimens in different populations are shown in Figure 5. In terms of PFS, both in all patients and patients without BM, osimertinib + bevacizumab was most likely to rank the first, with cumulative probabilities of 51.2% and 47.1%, respectively. For those with BM, osimertinib + bevacizumab ranked the first for PFS, with a cumulative probability of 34.5%, followed by aumolertinib, with a cumulative probability of 28.3%. For OS, in all patients and patients with or without BM, SoC + PbCT was most likely to rank the first, with cumulative probabilities of 58.4%, 46.8%, and 37.3%, respectively.

Figure 5:

Bayesian ranking profiles of comparable treatments on efficacy for EGFR-mutated NSCLC patients with or without brain metastases. Profiles indicate the probability of each comparable treatment being ranked from first to last on PFS and OS in all, brain metastases and non-brain metastases patients. (A) The ranking of afatinib for PFS and OS in all, brain metastases and non-brain metastases patients. (B) The ranking of PbCT for PFS and OS in all, brain metastases and non-brain metastases patients. (C) The ranking for dacomitinib on PFS and OS in all and non-brain metastases patients. (D) The ranking of osimertinib for PFS and OS in all, brain metastases and non-brain metastases patients. (E) The ranking of SoC for PFS and OS in all, brain metastases and non-brain metastases patients. (F) The ranking of SoC+apatinib for PFS in all, brain metastases and non-brain metastases patients. (G) The ranking of SoC+bevacizumab for PFS and OS in all, brain metastases and non-brain metastases patients. (H) The ranking of SoC+PbCT for PFS and OS in all, brain metastases and non-brain metastases patients. (I) The ranking of SoC+ramucirumab for PFS in all and non-brain metastases patients. (J) The ranking of PfCT for PFS and OS in all, brain metastases and non-brain metastases patients. (K) The ranking of osimertinib+bevacizumab for PFS in all, brain metastases, and non-brain metastases patients. (L) The ranking of aumolertinib for PFS in all, brain metastases, and non-brain metastases patients. EGFR: Epidermal growth factor receptor; NSCLC: Non-small cell lung cancer; OS: Overall survival; PbCT: pemetrexed based chemotherapy; PfCT: pemetrexed free chemotherapy; PFS: Progression-free survival; SoC: Standards of care, including gefitinib, erlotinib, and icotinib.

Heterogeneity and inconsistency assessment

Most comparisons from one or two studies, minor heterogeneities could be found. However, high heterogeneities could be found from the comparisons between PfCT and afatinib (51.90%) for PFS in all patients, and SoC + bevacizumab and SoC for PFS in patients with brain metastases (60.60%). The fit of the consistency model was used in the analysis due to the superiority of lower DIC than that from the inconsistency model. Differences between the consistency model and the inconsistency model were accepted (less than 5 [Supplementary Table 2, https://links.lww.com/CM9/B524]). No significant differences were found between direct and indirect estimates for any outcome (P >0.05) [Supplementary Table 3, https://links.lww.com/CM9/B524].

Discussion

The brain is a common metastatic site in patients with NSCLC, resulting in a relatively poor prognosis.[41] Despite the rapid development of treatment agents—EGFR-TKIs for NSCLC, their efficacy is poorer in advanced, EGFR-mutant, BM NSCLC patients than in those without BM.[42] Patients with EGFR-mutated NSCLC are prone to develop BM, and the medical demands for patients with BM are largely unmet. Therefore, the status of BM should be considered in clinical practice while selecting the optimal treatment. A few NMAs have drawn conclusions regarding the optimal therapy in EGFR-mutated NSCLC, but no data have indicated the preferable approach for BM patients.[43] Additionally, some combined regimens of osimertinib + bevacizumab, first-generation EGFR-TKIs with antiangiogenic drugs or chemotherapy were not included into analysis,[44] and OS data from the FLAURA, JO25567, and NEJ026 studies and updated results from the CTONG 1509, RELAY, ACTIVE, AENEAS, and WJOG9717L studies were not included in previously NMAs.[43,44] Thus, we performed a NMA to investigate the survival outcomes of advanced EGFR-mutated NSCLC patients with different BM statuses and therapeutic regimens to identify the optimal therapeutic strategy. To our knowledge, this meta-analysis included the largest sample size to date and involved mostly current regimens to evaluate the optimal therapy for EGFR-mutated NSCLC patients with BM. Furthermore, this NMA identified the optimal EGFR-TKI regimen for non-BM, NSCLC patients.

A previous study showed that osimertinib has superior efficacy and decreases the risk of intracranial progression compared with standard EGFR-TKIs in NSCLC patients with BM.[45] Our study suggests that osimertinib + bevacizumab ranked the first for PFS, and SoC + PbCT ranks the first for OS in them. Due to immaturity of OS data, the OS comparison among osimertinib + bevacizumab, aumolertinib, and osimertinib was lacking. Previous studies demonstrated that bevacizumab was beneficial for patients with BM,[46–48] our results suggested that osimertinib + bevacizumab was related to the most favorable PFS for those with BM compared with others treatments. However, previous studies observed a shorter PFS with osimertinib plus bevacizumab than osimertinib alone in patients with EGFR T790M-mutated NSCLC,[49] and failed to show