Treatments for brain metastases from EGFR/ALK-negative/unselected NSCLC: A network meta-analysis

Abstract More clinical evidence is needed regarding the relative priority of treatments for brain metastases (BMs) from EGFR/ALK-negative/unselected non-small cell lung cancer (NSCLC). PubMed, EMBASE, Web of Science, Cochrane Library, and ClinicalTrials.gov databases were searched. Overall survival (OS), central nervous system progression-free survival (CNS-PFS), and objective response rate (ORR) were selected for Bayesian network meta-analyses. We included 25 eligible randomized control trials (RCTs) involving 3,054 patients, investigating nine kinds of treatments for newly diagnosed BMs and seven kinds of treatments for previously treated BMs. For newly diagnosed BMs, adding chemotherapy, EGFR-TKIs, and other innovative systemic agents (temozolomide, nitroglycerin, endostar, enzastaurin, and veliparib) to radiotherapy did not significantly prolong OS than radiotherapy alone; whereas radiotherapy + nitroglycerin showed significantly better CNS-PFS and ORR. Surgery could significantly prolong OS (hazard ratios [HR]: 0.52, 95% credible intervals: 0.41–0.67) and CNS-PFS (HR: 0.32, 95% confidence interval: 0.18–0.59) compared with radiotherapy alone. For previously treated BMs, pembrolizumab + chemotherapy, nivolumab + ipilimumab, and cemiplimab significantly prolonged OS than chemotherapy alone. Pembrolizumab + chemotherapy also showed better CNS-PFS and ORR than chemotherapy. In summary, immune checkpoint inhibitor (ICI)-based therapies, especially ICI-combined therapies, showed promising efficacies for previously treated BMs from EGFR/ALK-negative/unselected NSCLC. The value of surgery should also be emphasized. The result should be further confirmed by RCTs.


Introduction
Lung cancer is the leading cause of cancer-related death worldwide, and non-small cell lung cancer (NSCLC) represents approximately 85% of lung cancer cases [1]. Around 25-30% of NSCLC patients develop brain metastases (BMs) [2]; moreover, NSCLC is the most common primary cancer that metastasizes to the brain [2]. The prognosis of NSCLC patients with BMs is dismal, with a median survival time of only approximately 1 month in the absence of treatment [3,4]. Several strategies have been employed to treat BMs from NSCLC, including surgery, targeted therapy, immune checkpoint therapy, chemotherapy, radiotherapy, and their combination [3,5]. Each of these treatments has both advantages and drawbacks, and their relative efficacies are not fully understood.
In recent years, targeted therapies for NSCLC and BMs have rapidly developed. For instance, the epidermal growth factor receptor (EGFR) plays an essential role in lung cancer and depends on its expression status among the population. The mutations in EGFR and its polymorphisms are associated with the onset of carcinogenesis, the prediction of the metastases, and the response to tyrosine kinase inhibitors (TKIs) [6,7]. EGFR mutations are detected in 15-35% of NSCLC, with a higher percentage observed in the Asian population than in Europeans [8][9][10]. For patients with EGFR mutations and BMs, previous studies have shown that third-generation EGFR-TKIs and EGFR-TKIs combined with chemotherapy or radiotherapy have favorable efficacy [11,12]. Anaplastic lymphoma kinase (ALK) rearrangement, which occurs in 2-7% NSCLC, is also a classic target [13]. ALK inhibitors (especially the second and third-generation inhibitors) have shown promising efficacy for NSCLC with BMs [14].
However, there are still a significant number of patients with negative EGFR/ALK NSCLC BMs. In addition, limited to socioeconomic factors, genomic tests cannot cover all the patients, which means the genomic status of many patients remains unknown. Therefore, treatments with broader indications (including surgery, radiotherapy, immune checkpoint therapy, chemotherapy, and other innovative therapies) are more suitable for such patients [15]. Surgery is recommended for BMs that are large, have significant perilesional edema, and result in neurological deficits. It can provide immediate relief from symptomatic mass effects and help to confirm the diagnosis [10]. Radiotherapy, which mainly consists of whole-brain radiation therapy (WBRT) and stereotactic radiosurgery (SRS), is considered the cornerstone of the treatment for BMs [10,16]. WBRT has previously been the standard treatment for BMs. However, considering the neurocognitive toxicity, the value of WBRT has been challenged, and SRS has gradually become popular [10,17,18]. As an alternative to surgical resection, SRS is a high precise localized irradiation given in one fraction. It can achieve a dose to the tumor with a low risk of damage to the surrounding normal brain [16]. SRS is recommended for BMs of a limited number (up to 4) and limited size (up to 3 cm) [18][19][20]. Immune checkpoint inhibitors (ICIs) represent a major breakthrough for treating metastatic NSCLC and have shown preliminarily promising outcomes in patients with BMs from NSCLC [10]. Nevertheless, previous studies generally analyzed single-arm treatments or compared pairwise treatments and could not generate clear hierarchies of treatment approaches. Consequently, we performed a systematic literature review and Bayesian network meta-analysis (NMA) to analyze the comparative efficacy of all types of therapy available for these patients.

Data sources and search strategy
This research was performed following the guidelines provided by the PRISMA (preferred reporting items for systematic reviews and meta-analyses) report [21] (PRISMA Checklist). We searched the PubMed, EMBASE, Web of Science, Cochrane Library, and ClinicalTrials.gov databases from inception until April 10, 2022, without language restrictions, for randomized control trials (RCTs). The search took into account both medical subject headings and text words, using the main search terms "NSCLC," "brain metastasis," and terms specific to the different treatments. The detailed search strategy for the databases is presented in Table S1. The reference lists of the relevant articles were checked for additional articles. The protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO, CRD42021227078).
Ethical approval: The conducted research is not related to either human or animal use.

Study selection
Two investigators, Zhang C.K. and Zhou W.J.L., independently assessed the eligibility of studies based on the title, abstracts, and full texts, resolving disagreements by obtaining a consensus with Guan X.D. We included published and unpublished trials that met the following criteria:

Data extraction and quality assessment
Two authors, Zhang C.K. and Zhou W.J.L., independently extracted data from the eligible studies and assessed the risk of bias in the individual studies. Disagreements were resolved by consensus or referral to a third reviewer, Guan X.D. The extracted items included study details (name of the first author, country, registration number, and phase of the study), participant details (number of participants, age, and gender), intervention and comparison in each arm, and survival outcomes (hazard ratios [HRs] and 95% confidence interval [CI], including the OS rate, PFS rate, and ORR). The quality and risk of bias were assessed for each trial using the Cochrane Collaboration risk of bias tool [22], random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective outcome reporting, and other sources of bias were examined. The quality of each study was categorized as high, low, or unclear.

Statistical analyses
The risk of bias in the RCTs was assessed by Review Manager (RevMan, 5.3, The Cochrane Collaboration, London, UK). The Bayesian NMA was performed using the JAGS program and GEMTC package in R software (version 4.0.2, R Foundation, Vienna, Austria). HRs of CNS-PFS and OS rates were analyzed on a natural log scale and pooled as HRs and 95% credible intervals (CrIs). For studies that did not directly provide HRs, we extracted and estimated the HRs and corresponding standard error from a high-quality Kaplan-Meier curve with the methods described by Tierney [23]. The ORR was pooled with the risk ratio (RR) and corresponding 95% CrI. The simulation was performed using the Markov chain Monte Carlo technique with three different chains, and each of them produced 10,000 interactions with 100,000 burn-in samples and ten thinning rates. Fixed-effect models were used, since in most cases, the treatment of interest was evaluated in only one trial. We assessed statistical inconsistencies by the edge-splitting method to compare direct and indirect evidence. Statistical significance was considered when P < 0.05. Statistical heterogeneity was estimated by the I2 statistic, which indicates what proportion of variability in outcomes was due to heterogeneity rather than chance. An I2 > 50% was regarded as significant heterogeneity, while I2 < 25% indicated a small level of heterogeneity. To assess the robustness and reliability of the results, we also performed a sensitivity analysis in the absence of low-quality trials.

Study selection and characteristics
The procedures of the screening and the reasons for exclusion are shown in Figure 1. A total of 2,099 studies met the search criteria. After title and abstract screening, 47 trials were retrieved, and the full text was reviewed. Ultimately, 25 trials  were included in this NMA. Another five trials were included in a traditional metaanalysis comparing surgery or not. As shown in Table 1, a total of 3,054 participants were enrolled in the selected RCTs. Most participants were male and over 55 years of age. The demographic and clinical characteristics kept a balance between the intervention and control groups in each RCT. The trial publication dates ranged from 2005 to 2022. The bias assessment is presented in Figure S1, with two trials assessed as having a high risk of bias [26,27] and 23 trials assessed as having a low risk of bias [24,25,.
Considering the heterogeneity of the study design and treatment history, we divided the analyses into three parts: (1) NMA about radiotherapy or systemic therapy for previously untreated patients; (2) NMA about ICIs or chemotherapies for previously treated patients; and (3) traditional meta-analysis comparing surgery with radiotherapy alone for patients who had opportunities of surgery.
Three trials about surgery reported ORR (Figure 4e). Surgery showed a favorable trend of ORR for BMs from NSCLC but did not reach a statistical difference (RR: 1.04, 95% CrI: 0.89-1.21, common effect model).

Heterogeneity, consistency, and sensitivity analysis
There was low global heterogeneity for the comparison of radiotherapy-associated regiments (I2 = 0% for OS, I2 = 0% for CNS-PFS, and I2 = 8% for ORR) and low to moderate heterogeneity for the comparison of ICIs (I2 = 22% for OS, I2 = 33% for CNS-PFS, and I2 = 25% for ORR). Furthermore, local heterogeneities were also acceptable between paired treatments (Table S3). In terms of inconsistency, there was no significant difference between direct and indirect comparisons of the OS, CNS-PFS, and ORR ( Figure S3). During the sensitivity analysis, one study with unclear random sequence generation (d)  [27] and another study with imbalanced patients' initial baseline [26] were excluded. The results showed the same ranks compared with those of the original NMA ( Figure  S4). The sensitivity analyses showed that the overall results remained robust.

Discussion
Currently, there is a wide range of alternative treatments available for brain-metastatic NSCLC with negative or unselected EGFR/ALK status. Nevertheless, direct comparisons of such treatments are limited. Our study analyzed the relative efficacy of each treatment for previously treated and untreated BMs. It showed that several ICIs were associated with longer OS and CNS-PFS than chemotherapy in patients with previously treated BMs. Except for nitroglycerin, the addition of EGFR-TKIs, chemotherapy, and other non-ICI systemic innovative medicines to RT did not improve OS, CNS-PFS, and OS. Surgery of BMs was associated with better OS, CNS-PFS, rather than ORR. The reasons for these findings are presented as follows.
Currently, ICIs have been the standard first-line treatments for metastatic NSCLC without sensitizing EGFR or ALK or other druggable mutations [54]. However, the intracranial efficacies of ICIs remained uncertain. The exact mechanism of ICIs for brain tumors was also unclear. First, it may be related to modified immune cell activity rather than direct action in the brain. By immune cell trafficking and T-cell priming in the extracranial immune system, ICIs could produce an effective immune response in the CNS [55,56]. Moreover, due to the infiltration of lymphocytes in BMs [57] and the relatively stable PD-L1 expression level between primary tumors and BMs [58], it can be hypothesized that PD-(L)1 inhibitors might provide similar effects inside and outside the brain [59].
In addition, the choice of monotherapy or combined therapy also affects the therapeutic effect of ICIs for BMs. In our NMA, several combined therapies, including pembrolizumab + chemotherapy and nivolumab + ipilimumab derived relatively better effects. Current NCCN guideline recommends ICIs monotherapy for patients with PD-L1expression more than 50%; whereas, ICIs in combination with chemotherapy is recommended regardless of PD-L1 expression [54]. Consistently, comparisons between different immunotherapy strategies for BMs also showed that combined ICIs with chemotherapy or dual ICIs had favorable efficacies for advanced NSCLC [57,68,69]. Similarly, an RCT also showed that nivolumab + ipilimumab derived better CNS-PFS and ORR than nivolumab monotherapy for BMs from melanoma [70].
The expression level of PD-L1 could also affect the response to ICIs. Studies have demonstrated that ICIs have better efficacy in NSCLC (with or without BM) patients whose PD-L1 expression is ≥1% [59,65,73], while responses can still occur in those with PD-L1 expression <1% or PD-L1negative tumors [44,67,[73][74][75]. Of the eight trials included in this study, six trials did not select patients according to PD-L1 expression [41][42][43][44][45]47], and ICIs still showed promising efficacies. One trial about pembrolizumab monotherapy recruited patients with PD-L1 expression of no less than 1%, but its effect was still inferior to pembrolizumab + chemotherapy [46]. Another trial about cemiplimab included patients with PD-L1 expression at least 50%, and showed relatively superior OS and CNS-PFS [48]. Therefore, we speculated that the therapeutic effect was determined by both PD-L1 expression and the properties of ICIs.
Nowadays, radiotherapy (SRS or WBRT) remains the mainstay of initial therapy for BMs [16]. Previous studies have shown the addition of WBRT to SRS or surgery alone could increase CNS-PFS and local control rate; however, the OS time did not prolong, and the neurocognitive toxicity also increased [18,76,77]. Therefore, local treatment (SRS or surgical resection) without WBRT is recommended for patients with up to four BMs and good physical performance [18]. With the development of the SRS technique, SRS was tried to treat selected patients with multiple (more than four) BMs. Several multicenter studies have found that patients treated with SRS for 5-10 BMs, or even 5-15 BMs derived comparable OS to those with 2-4 BMs [78,79]. WBRT is often considered for patients who are not suitable for SRS or surgery (e.g., innumerable metastases, innumerable metastases, poor physical performance, or other contraindications) [10], and was believed to prolong CNS-PFS [18]. Nevertheless, an RCT found that the WBRT showed no difference with optimal supportive care in terms of OS, quality of life, and dexamethasone for patients unsuitable for resection or SRS [80].
In current analyses, nitroglycerin + WBRT showed favorable effects for BMs. Nitroglycerin has just been used to assist tumor radiotherapy in recent years. It could reduce the radiation resistance by alleviating tumor hypoxia [39]. Nevertheless, the synergistic effect of nitroglycerin with chemoradiotherapy was only tested in several phase II trials of primary NSCLC [81][82][83] and only one trial about BM [39], and the results of primary NSCLC were controversial [81][82][83]. Therefore, the efficacy of nitroglycerin on BMs needs to be further evaluated.
On the other hand, there was no significant advantage of adding chemotherapy, EGFR-TKI, or other non-ICI innovative systemic agents to radiotherapy. Systematic reviews have revealed that radiotherapy + chemotherapy might improve response rates compared with radiotherapy alone; however, this approach does not improve survival outcomes and increases the incidence of adverse reactions in patients with BMs arising from lung cancer [84,85]. Meanwhile, WBRT plus systemic therapy was associated with increased risks for vomiting compared to WBRT alone [84,[86][87][88]. For the same reasons, the CNS (Congress of Neurological Surgeons) and EANO (European Association of Neuro-Oncology) guidelines did not suggest routine use of cytotoxic chemotherapy either alone or following WBRT [16,89].
Although TMZ is recommended to be used with WBRT for patients with BMs arising from triple-negative breast cancer [89], its efficacy on BM arising from NSCLC is controversial. Several trials have yielded mixed results [90][91][92][93], while the current systematic review and metaanalysis determined that adding TMZ to radiotherapy can increase the ORR [87,94,95]. However, it is generally believed that adding TMZ cannot induce a better OS outcome [86,87,96,97]. Therefore, there is insufficient evidence to conclude that there is value in adding TMZ for the treatment of NSCLC with BM [98].
Although EGFR-TKIs were used to treat BMs from EGFR-unselected NSCLC, the effect was not ideal. Currently, with the widespread use of genetic testing technology, it is recommended to screen for EGFR-mutations in NSCLC patients with BMs and treat them with third-generation EGFR-TKIs, which have better blood-brain barrier penetrability and better efficacy [99].
Adding several other innovative systemic treatments to radiotherapy did not show survival benefits, including Veli (polyadenosine-diphosphate-ribose polymerase inhibitor), Enza (serine/threonine kinase inhibitor), and Endo (an antiangiogenic drug). This was not surprising because RCTs evaluating treatments for NSCLC (with or without metastases) have demonstrated that although Veli [100] or Endo [101] demonstrated a favorable trend in PFS and OS outcomes versus chemotherapy alone, the differences were not statistically significant; however, adding Enza to chemotherapy may induce shorter median survival times [102].
Surgical resection of BMs remains one of the mainstays of therapies for patients with BMs from NSCLC [103]. Since these metastases show radioresistance compared to SCLC, surgical resection to relieve the spaceoccupying effect is often the first step in treatments for these patients [10]. In the current analyses, surgery could derive better OS and CNS-PFS than radiotherapy alone for BMs from NSCLC. Consistently, previous studies also found surgery improved survival outcomes of patients with a single brain-metastatic lesion, a good karnofsky performance scale (KPS), and a limited number of extracranial metastases (primary malignancies were not filtered) [104,105].

Limitations
There were several limitations in this study. First, the exact techniques of RT were not compared separately.
On the one hand, the indications and efficacies of WBRT and SRS have been proven by high-quality studies. On the other hand, network comparisons could not form if radiotherapy techniques were discussed separately. Therefore, our analyses mainly focused on the effect of adjuvant systemic therapy on radiotherapy. Discuss RT as a whole is feasible and consistent with previous meta-analyses on BMs [106,107], because the intervention (with adjuvant systemic therapy) and control (without adjuvant systemic therapy) groups in each trial have received radiotherapy of the same technique and dose.
There was still potential selection and publication bias. First, as mentioned above, two trials had unqualified patient inclusion processes; therefore, a sensitivity analysis was conducted, and relatively robust results were ensured. Second, there were a limited number of trials in each comparison. For this reason, we did not use funnel plots to assess publication bias or small-study effects. Third, we assumed that the patients from different trials were similar; however, patients may have had different baseline levels. For example, an unbalanced baseline can arise from the presence or absence of symptoms and the exact number and volume of BMs. All of these factors could give rise to a bias.
Moreover, limited information also restricted our analyses. We could not analyze the adverse effects and quality of life (such as KPS score and other parameters), because such information was not available or not complete. Trials on systemic therapy or radiotherapy usually did not report how many patients underwent surgery and their corresponding outcome. Therefore, we could not analyze the synergy between surgery and other treatment. Included trials about surgery did not perform subgroup analysis according to the surgery details (such as location, size, and the number of BMs), which makes it difficult for us to obtain the corresponding summary results. In addition, most trials about ICIs included in our analyses only reported OS without reporting CNS-PFS and ORR. Such data of interest need to be further explored.

Conclusion
ICI-based therapies, especially ICI-combined therapies, showed promising efficacies for previously treated BMs from EGFR/ALK-negative/unselected NSCLC. Adding chemotherapy, EGFR-TKIs, and some innovative agents (TMZ, Nitro, Endo, Enza, and Veli) to radiotherapy showed limited effects than radiotherapy alone for newly diagnosed BMs from EGFR/ ALK-negative/unselected NSCLC. Surgery could significantly prolong OS and CNS-PFS compared with radiotherapy alone. Limited to heterogeneity and available information, the result of the current network meta-analyses should be further confirmed by RCTs.