Immune Checkpoint Inhibitors +/− Chemotherapy for Patients With NSCLC and Brain Metastases: A Systematic Review and Network Meta‐Analysis

ABSTRACT

Background

Multiple studies have demonstrated the intracranial efficacy of immune checkpoint inhibitors (ICI) +/− chemotherapy. The efficacy of chemoimmunotherapy compared to ICI alone in patients with metastatic NSCLC and brain metastases (BM) remains unknown.

Methods

A systematic review and network meta‐analysis were performed to evaluate ICI efficacy and the influence of additional chemotherapy on survival outcomes in treatment‐naïve metastatic NSCLC with BM. Randomized phase II/III studies with at least one treatment arm with an ICI were eligible. Overall survival (OS) and progression‐free survival (PFS) in patients with and without BM were assessed.

Results

Ten studies were included, totaling 6560 patients, 770 with BM. Pairwise meta‐analysis revealed that patients with BM treated with ICI +/− chemotherapy had improved PFS (hazard ratio [HR] 0.49; 95% CI 0.40–0.60) and OS (HR 0.55; 95% CI 0.44–0.68) versus chemotherapy alone. Patients without BM treated with ICI +/− chemotherapy also had improved PFS and OS compared to chemotherapy alone. In the network meta‐analysis of patients with BM, chemoimmunotherapy demonstrated improved PFS compared to ICI alone (HR 0.64; 95% CI 0.43–0.96; p = 0.03). No significant difference was observed in OS. In the population of patients without BM, no significant differences in PFS or OS were observed between chemoimmunotherapy versus ICI alone.

Conclusion

This meta‐analysis confirms that ICIs with or without chemotherapy are superior to chemotherapy alone for the first‐line management of metastatic NSCLC with and without BM. This network meta‐analysis suggests combination chemoimmunotherapy offers PFS benefit over ICI monotherapy in BM patients, warranting direct comparisons in clinical trials.

Trial Registration: PROSPERO: CRD42024501345

A network meta‐analysis was performed to compare the efficacy of immune checkpoint inhibitors (ICIs) with and without chemotherapy, in the first‐line management of metastatic non‐small cell lung cancer (NSCLC) with brain metastases (BM). ICIs with or without chemotherapy were superior to chemotherapy alone in the management of patients with metastatic NSCLC with and without BM. In the network meta‐analysis, chemoimmunotherapy demonstrated a progression‐free survival benefit over ICIs alone in patients with NSCLC and BM, supporting further clinical trial investigation.

1. Introduction

Lung cancer is the leading cause of cancer mortality globally with non‐small cell lung cancer (NSCLC) accounting for 80%–90% of cases [1, 2]. Most patients present with advanced or metastatic disease [3], and 10%–30% of patients will have brain metastases (BM) at diagnosis or over the course of their disease [4, 5]. Despite improvement in diagnostic imaging and treatment advances, the presence of BM often indicates a poor prognosis [6].

Immune checkpoint inhibitors (ICIs) have revolutionized the treatment of metastatic lung cancer [7]. First‐line chemoimmunotherapy [8, 9, 10, 11] and immunotherapy [12, 13, 14] for tumors with PD‐L1 expression of ≥ 50% are standard of care for patients with metastatic NSCLC without targetable mutations. However, clinical trials often exclude patients with active, symptomatic, or untreated BM. Accordingly, data for this patient subgroup are limited to subgroup analyses in phase III studies, small phase II trials, or observational cohort studies.

Pembrolizumab monotherapy has demonstrated an intracranial response of 29.7% in a phase II trial by Goldberg et al. [15]. High intracranial objective response rates (iORR) have been observed in chemoimmunotherapy combination trials for metastatic NSCLC with BM. In the Atezo‐Brain phase II trial, atezolizumab combined with chemotherapy, in metastatic NSCLC with untreated BM, demonstrated an iORR of 42.7% and a median intracranial progression‐free survival (PFS) of 6.9 months [16]. The CAP‐Brain study, using camrelizumab combined with chemotherapy, reported an iORR of 52.5%, with a median intracranial PFS and overall survival (OS) of 7.6 months and 21 months respectively [17].

It remains uncertain whether combination chemoimmunotherapy is more effective than ICI monotherapy in treating metastatic NSCLC with BM. There are no prospective studies directly comparing chemoimmunotherapy combinations to ICIs without chemotherapy. As such, we performed a systematic review and network meta‐analysis (NMA), which facilitates an indirect comparison of the established trials that individually investigated either chemoimmunotherapy or immunotherapy versus chemotherapy, to investigate the differences in efficacy between chemoimmunotherapy and ICIs in patients with metastatic NSCLC with BM.

2. Methods

2.1. Search Strategy and Selection Criteria

The systematic review and NMA were performed according to the Preferred Reporting Items for Systematic Review and Meta‐Analyses (PRISMA) Statement. The inclusion criteria and prespecified analysis were registered with PROSPERO.

A comprehensive literature search using PubMed (MEDLINE), EMBASE, and Cochrane Library to identify all relevant articles from the inception of each database up to January 21, 2024. Our search strategy included the following search terms: “randomized,” “non‐small cell lung cancer”, (“PD‐1” OR “PD‐L1” OR “CTLA‐4”) AND (“brain” OR “central nervous system”). To ensure comprehensive capture, an additional manual reference check of pertinent literature to identify additional studies. The complete search strategy is presented in Table S1.

Eligibility was determined by independent screening of titles and abstracts by LJB and NY, with full texts reviewed if eligibility could not be determined. Disagreements regarding eligibility were resolved via a third author BYK.

To be eligible for inclusion, studies need to meet the following criteria:

1. Phase II or III randomized controlled trials involving treatment‐naïve, unresectable, stage IV NSCLC

2. Intervention arm including either (a) anti‐PD‐(L)1 monoclonal antibodies, (b) anti‐PD‐(L)1 and anti‐CTLA‐4 monoclonal antibodies, (c) anti‐PD‐(L)1 monoclonal antibodies with platinum‐based chemotherapy or (d) anti‐PD‐(L)1 and anti‐CTLA‐4 monoclonal antibodies with platinum‐based chemotherapy

3. Control arm involving platinum‐based chemotherapy

4. Trials reporting data on OS, PFS, objective response rate (ORR) or iORR

5. Subgroup or posthoc analysis reporting on the outcomes of patients with baseline BM

Studies were excluded if they were not performed in the first‐line setting or included treatment with anti‐vascular endothelial growth factor (VEGF) inhibitors, tyrosine kinase inhibitors, antibody drug‐conjugates, or other novel therapies.

2.2. Statistical Analysis

The outcomes of interest were OS, PFS, ORR, and iORR. The effect sizes were assessed using hazard ratio (HR) with 95% confidence interval (CI) for median OS and PFS, and odds ratio (OR) with 95% CI for ORR and iORR. Data were collected from studies and analyzed using Covidence systematic review software (Veritas Health Innovation, Melbourne, Australia), Cochrane Revman Web version 7.7.2 and R, version 4.2.3 (R Foundation for Statistical Computing, Vienna, Austria).

Using a random effects model, direct pairwise meta‐analyses were performed for the following groups: Group 1: trials investigating ICIs with or without chemotherapy versus chemotherapy; Group 2: ICIs without chemotherapy (anti‐PD‐(L)1 +/− anti‐CTLA‐4) versus chemotherapy; and Group 3: ICIs with chemotherapy (anti‐PD‐(L)1 +/− anti‐CTLA‐4 plus chemotherapy) versus chemotherapy (Table S2). The intervention effect was analyzed using the logarithmic transformation of the hazard ratio (log[HR]) for median OS and PFS, as is standard for time‐to‐event data. Data unable to be included in the meta‐analyses were presented qualitatively. The presence of potential publication bias was assessed graphically using a funnel plot.

Indirect comparisons of the outcomes described above were performed between Group 2 (ICIs) versus Group 3 (ICIs + chemotherapy) via an NMA using frequentist methods [18] and the netmeta package in R (Table S2). NMA is only considered valid if the interstudy heterogeneity is low (I ² < 50%) with acceptable transitivity (the assumption that the outcome of indirect comparisons between treatments can be reliably inferred from their direct comparisons with a common third treatment [chemotherapy]). Thus, we only used meta‐analyses with a low‐to‐moderate level of heterogeneity for the NMA. A p value of ≤ 0.05 was considered statistically significant within the NMA.

Heterogeneity was evaluated using the Q‐test and I ² [19]. A Q‐test p value of > 0.1 or I ² < 50% was considered to have a low‐to‐moderate level of heterogeneity; a Q‐test p value of < 0.05 or I ² ≥ 50% suggested high heterogeneity. When heterogeneity was detected, sensitivity analyses were performed to evaluate the effects of excluding the study most likely responsible for the heterogeneity and to determine if this exclusion altered the pooled analysis results.

2.3. Assessment of Risk of Bias in Included Studies

All studies were assessed for risk of bias using the Cochrane Risk of Bias 2 tool for assessing risk of bias in randomized controlled trials [20]. The summary of the risk of bias assessment for the included studies is detailed in Table S3.

2.4. Certainty Assessment

Each article was evaluated for evidence quality using the GRADE system, which was used to guide the assessment of the strength of recommendations pertaining to each outcome in the pairwise meta‐analysis [21]. The summary of findings and certainty of the evidence by the GRADE system are detailed in Table S4.

3. Results

3.1. Study Characteristics

We identified 1786 records in total, 1762 in the databases, and 4 by citation searching. After deduplication and screening, 61 full‐text studies were retrieved and assessed for eligibility (Figure 1). Ten publications across 12 trials (including a pooled analysis of 3 studies) met the eligibility criteria and were included in the final analysis [11, 13, 22, 23, 24, 25, 26, 27, 28, 30]. This included 6560 patients, 770 with BM, and 5790 without BM. The studies included in the NMA satisfied the technical requirement that each treatment must be represented by at least one study and were connected to the network (Figure 2).

FIGURE 1.

PRISMA flowchart.

FIGURE 2.

NMA nodes: Frequentist framework was used to generate comparisons between the median overall survival and the median progression‐free survival. The size of the light blue circle and the thickness of the dark blue lines are proportional to the number of studies included. Solid lines represent meta‐analyses (direct comparisons), and dotted lines represent network meta‐analyses (indirect comparisons) performed. (A) NMA plot by each type of therapy. (B) Network meta‐analysis further delineated by type of immunotherapy.

Of the included studies (Table 1), two assessed Anti‐PD‐1 monotherapy [13, 32], one assessed Anti‐PD‐1 and Anti‐CTLA‐4 immunotherapy [29], one assessed Anti‐PD‐1, Anti‐CTLA‐4, and chemotherapy [11], and six used combination Anti‐PD‐(L)1 and chemotherapy [22, 23, 24, 25, 26, 27]. Of the included studies, two [13, 27] did not analyze OS data for patients with BM and one [24] did not analyze PFS data for patients with BM. Only three studies assessed ORR in the BM population: CheckMate‐9LA [11], EMPOWER‐Lung 3 [22], and the Pooled KEYNOTE analysis [26]. Given all three of these studies were examining chemoimmunotherapy, an NMA could not be performed for ORR. No studies reported on iORR and thus iORR was not evaluated in the meta‐analysis or NMA.

TABLE 1.

Characteristics of studies included in network meta‐analysis.

Study	Primary endpoint	Experimental arm (n)	Control arm (n)	PFS HR (95% CI)			OS HR (95% CI)			ORR (%)
				Overall	BM	No BM	Overall	BM	No BM	Overall	BM	No BM
CameL [25, 29]	PFS	205	207	0.60 (0.45–0.79)	0.14 (0.01–0.88)	0.60 (0.45–0.80)	0.73 (0.53–1.02)	0.45 (0.10–1.73)	0.72 (0.56–0.92)	60.5 vs. 38.6	—	—
Checkmate 227 Part 1 [29]	PFS and OS	582	583	0.50 (0.39–0.65)	0.77 (0.51–1.15)	0.79 (0.68–0.91)	0.62 (0.48–0.81)	0.63 (0.43–0.92)	0.76 (0.66–0.87)	—	—	—
Checkmate‐227 Part 2 [24]	OS	377	378	0.67 (0.55–0.82)	—	—	0.86 (0.69–1.08)	0.92 (0.53–1.59)	0.80 (0.66–0.97)	48.0 vs. 29.0	—	—
Checkmate‐9LA [11, 31]	OS	361	358	0.70 (0.59–0.83)	0.44 (0.28–0.69)	0.77 (0.64–0.92)	0.74 (0.62–0.87)	0.45 (0.29–0.70)	0.80 (0.67–0.96)	—	39.2 vs. 20.0	—
Pooled analysis of KEYNOTE‐021, 189 and 407 [26]	OS and PFS (KEYNOTE‐189, KEYNOTE‐407) ORR (KEYNOTE‐021)	748	550	—	0.44 (0.31–0.62)	0.55 (0.48–0.63)	—	0.48 (0.32–0.70)	0.63 (0.53–0.75)	—	39.0 vs. 19.7	54.6 vs. 31.8
EMPOWER‐Lung 1 [28, 32]	OS and PFS	283	280	0.54 (0.43–0.68)	0.45 (0.22–0.92)	0.56 (0.44–0.71)	0.57 (0.42–0.77)	0.17 (0.04–0.76)	0.60 (0.44–0.83)	39.0 vs. 20.0	—	—
EMPOWER‐Lung 3 [22]	OS	312	154	0.56 (0.44–0.70)	0.53 (0.22–1.31)	0.54 (0.43–0.69)	0.71 (0.53–0.93)	0.42 (0.14–1.26)	0.68 (0.51–0.90)	43.3 vs. 22.7	20.8 vs. 28.6	45.1 vs. 22.4
GEMSTONE‐302 [27]	PFS	320	129	0.48 (0.39–0.60)	0.29 (0.15–0.56)	0.54 (0.43–0.68)	0.67 (0.50–0.90)	—	—	63.4 vs. 40.3	—	—
KEYNOTE‐024 [12, 13]	PFS	154	151	0.50 (0.39–0.65)	0.55 (0.20–1.56)	0.50 (0.36–0.68)	0.62 (0.48–0.81)	—	—	46.1 vs. 31.1	—	—
ORIENT‐11 [23, 33]	PFS	266	131	0.60 (0.45–0.79)	0.57 (0.28–1.16)	0.60 (0.45–0.82)	0.49 (0.36–0.63)	0.50 (0.26–0.92)	0.49 (0.37–0.64)	—	—	—

Abbreviations: BM: brain metastases; CI: confidence interval; HR: hazard ratio; ORR: objective response rate; OS: overall survival; PFS: progression‐free survival.

The pooled analysis of KEYNOTE‐021, ‐189, and ‐407 [26] was included as it described subanalyses of the BM population from each of the studies which were not independently reported in KEYNOTE‐407 and ‐021. It incorporated an early phase trial of KEYNOTE‐021 Cohort G, where all patients were negative for EGFR and ALK mutations and had not received prior systemic therapy. To prevent data overlap, the article reporting results of KEYNOTE‐189 [34], which independently reported outcome data specifically for patients with BM, was excluded. No RCTs were identified that directly compared chemoimmunotherapy versus ICIs.

All studies utilized 4–6 cycles of platinum doublet chemotherapy as the control arm, except for the KEYNOTE‐021 (cohort G) study, included as part of the pooled analysis [26]. Despite being a phase 1 study, KEYNOTE‐021 was pooled with the KEYNOTE‐407 and KEYNOTE‐189 at a patient level [26]. The decision to retain this study in the analysis was made to enhance the robustness of the comparative data and ensure a comprehensive evaluation of patient outcomes across varying therapeutic regimens. Carboplatin or cisplatin were used as the platinum agents; pemetrexed was used in patients with non‐squamous histopathology; paclitaxel, nab‐paclitaxel, or gemcitabine was used in patients with squamous histopathology.

Patient and trial characteristics are reported in Table 2. Patients with BM made up a small proportion of the population in all studies ranging from 3.6% to 17.0%. The majority of the population in all studies were current or former smokers and male. However, there were some notable differences between the studies: eight enrolled patients with both non‐squamous and squamous NSCLC. However, two studies enrolled patients with non‐squamous disease only [25, 33]. PD‐L1 expression differed throughout the studies with some only enrolling patients with PD‐L1 ≥ 1% or ≥ 50%. Additionally, three different assays were utilized for PD‐L1 assessment across the studies including 28‐8 pharmDX, 22C3 pharmDX, and SP263.

TABLE 2.

Characteristics of included patients in the network meta‐analysis.

Study	Experimental arm	Total (n)	Histology, n (%)	PD‐L1 expression, n (%)	Assay used for PD‐L1 assessment	BM, n (%)	Treatment of BM	Current/former smokers, n (%)	Males, n (%)	Median age	ECOG PS 0/1, n (%)
CameL [25, 29]	Camrelizumab and chemotherapy	412	Nonsquamous: 412/412 (100.0)	PD‐L1 ≥ 50%: 50/412 (12.1) PD‐L1 1%–49%: 205/412 (49.8) PD‐L1 < 1%: 118/412 (28.6) Unquantifiable: 39/419 (8.7)	22C3 pharmDX	15 (3.6)	Prior local therapy and subsequent stability required	—	295 (71.6)	60	412 (100.0)
Checkmate 227 Part 1 [29]	Ipilimumab and Nivolumab	1166	Nonsquamous: 838/1166 (71.9) Squamous: 328/1166 (28.1)	PD‐L1 ≥ 1%: 793/1166 (68.0) PD‐L1 < 1%: 373/1166 (32.0)	28‐8 pharmDX	134 (11.5)	Prior local therapy and subsequent stability required	996 (85.4)	778 (66.7)	62.5	1158 (99.3)
Checkmate‐227 Part 2 [24]	Nivolumab and chemotherapy	755	Nonsquamous: 543/755 (71.9) Squamous: 212/755 (28.1)	PD‐L1 ≥ 50%: 171/755 (22.6) PD‐L1 1%–49%: 234/755 (31.0) PD‐L1 < 1%: 304/755 (40.3) Unquantifiable: 46/755 (6.1)	28‐8 pharmDX	76 (10.1)	Prior local therapy and subsequent stability required	615 (81.5)	530 (70.2)	63.5	749 (99.2)
Checkmate‐9LA [11, 31]	Ipilimumab, nivolumab and chemotherapy	719	Nonsquamous: 495/719 (68.8) Squamous: 224/719 (31.2)	PD‐L1 ≥ 50%: 174/719 (24.2) PD‐L1 1%–49%: 233/719 (32.4) PD‐L1 < 1%: 264/719 (36.7) Unquantifiable: 48/719 (6.7)	28‐8 pharmDX	122 (17.0)	Prior local therapy and subsequent stability required	621 (86.4) 36 872 862	504 (70.1)	65	717 (99.7)
Pooled analysis of KEYNOTE‐021189 and 407 [26]	Pembrolizumab and chemotherapy	1298	Nonsquamous: 739/1298 (56.9) Squamous: 559/1298 (43.1)	PD‐L1 ≥ 50%: 385/1298 (29.7) PD‐L1 1%–49%: 435/1298 (33.5) PD‐L1 < 1%: 428/1298 (33.0) Unquantifiable: 50/1298 (3.9)	22C3 pharmDX	171 (13.2)	Asymptomatic untreated BM allowed in KN‐189 and ‐407	1160 (89.4)	866 (66.7)	64.25	1292 (99.5)
EMPOWER‐Lung 1 [28, 32]	Cemiplimab	563	Nonsquamous: 320/563 (56.8) Squamous: 243/563 (43.2)	PD‐L1 ≥ 50%: 563/563 (100.0)	22C3 pharmDX	68 (12.1)	Prior local therapy and subsequent stability required	563 (100.0)	479 (85.1)	64	563 (100.0)
EMPOWER‐Lung 3 [22]	Cemiplimab and chemotherapy	466	Nonsquamous: 266/466 (57.1) Squamous: 200/466 (42.9)	PD‐L1 ≥ 50%: 152/466 (32.6) PD‐L1 1%–49%: 175/466 (37.6) PD‐L1 < 1%: 139/466 (29.8)	SP263	31 (6.7)	Prior local therapy and subsequent stability required	399 (85.6)	391 (83.9)	63	462 (99.1)
GEMSTONE‐302 [27]	Sugemalimab and chemotherapy	479	Nonsquamous: 287/479 (59.9) Squamous: 192/479 (40.1)	PD‐L1 ≥ 1%: 291/479 (60.8) PD‐L1 < 1%: 188/479 (39.2)	SP263	67 (14.0)	Untreated asymptomatic BM allowed	351 (73.3)	383 (80.0)	63	479 (100.0)
KEYNOTE‐024 [12, 13]	Pembrolizumab	305	Nonsquamous: 249/305 (81.6) Squamous: 56/305 (18.4)	PD‐L1 ≥ 50%: 305/305 (100.0)	22C3 pharmDX	28 (9.2)	Prior local therapy and subsequent stability required	281 (92.1)	187 (61.3)	65	304 (99.7)
ORIENT‐11 [23, 33]	Sintilimab and chemotherapy	397	Nonsquamous: 397/397 (100.0)	PD‐L1 ≥ 50%: 168/397 (42.3) PD‐L1 1%–49%: 100/397 (25.2) PD‐L1 < 1%: 129/397 (32.5)	22C3 pharmDX	58 (14.6)	Untreated asymptomatic BM allowed	258 (65.0)	303 (76.3)	61	397 (100.0)

Abbreviations: BM = brain metastases; ECOG PS = Eastern Cooperative Oncology Group performance status; PD‐L1 = programmed death‐ligand 1.

3.2. Pairwise Meta‐Analyses for PFS

In the overall analysis for PFS, nine studies comprising 694 patients with BM and 5805 patients without BM were included. In Group 1, ICIs with or without chemotherapy versus chemotherapy alone for metastatic NSCLC improved median PFS significantly in patients with BM (HR 0.49; 95% CI 0.40–0.60; p < 0.001; I ² = 11%; Figure 3A) and in patients without BM (HR 0.60; 95% CI 0.52–0.68; p < 0.001; I ² = 72%; Figure 3B).

FIGURE 3.

Forest plot for PFS: The use of immunotherapy with or without chemotherapy was associated with superior PFS in patients with and without BM. (A) Forest plot and pooled HRs for PFS comparing ICIs with or without chemotherapy (Group 1) versus chemotherapy alone in patients with BM. (B) Forest plot and pooled HRs for PFS comparing with or without chemotherapy (Group 1) versus chemotherapy alone in patients without BM. CI: confidence interval; ICI: immune checkpoint inhibitor; Log[HR]: logarithm of the hazard ratio, SE: standard error.

The types of experimental arms were assessed independently. In Group 2, ICI without chemotherapy (anti‐PD‐1 or anti‐PD‐1 plus anti‐CTLA‐4) compared to chemotherapy improved median PFS significantly in patients with BM (HR 0.66; 95% CI 0.47–0.92; p = 0.01; I ² = 0%; Figure S1A) and in patients without BM (HR 0.62; 95% CI 0.46–0.83; p = 0.001; I ² = 81%; Figure S1B). In Group 3, ICI with chemotherapy (anti‐PD‐1 or anti‐PD‐1 plus anti‐CTLA‐4 plus chemotherapy) was compared to chemotherapy alone improved median PFS significantly in patients with BM (HR 0.43; 95% CI 0.34–0.54; p < 0.001; I ² = 0%; Figure S2A) and patients without BM (HR 0.58; 95% CI 0.50–0.67; p < 0.001; I ² = 61%; Figure S2B).

3.3. Pairwise Meta‐Analyses for OS

In the overall analysis for OS, eight studies comprising 675 patients with BM and 5776 patients without BM were included. In Group 1, ICI with or without chemotherapy compared to chemotherapy alone for metastatic NSCLC, improved median OS significantly in patients with BM (HR 0.55; 95% CI 0.44–0.68; p < 0.001; I ² = 12%; Figure 4A) and in patients without BM (HR 0.72; 95% CI 0.66–0.78; p < 0.001; I ² = 17%; Figure 4B).

FIGURE 4.

Forest plot for OS: The use of immunotherapy with or without chemotherapy was associated with superior OS in patients with and without BM. (A) Forest plot and pooled HRs for OS comparing ICIs with or without chemotherapy (Group 1) versus chemotherapy alone in patients with BM. (B) Forest plot and pooled HRs for OS comparing ICIs with or without chemotherapy (Group 1) versus chemotherapy alone in patients without BM. CI: confidence interval; ICI: immune checkpoint inhibitor; Log[HR]: logarithm of the hazard ratio, SE: standard error.

In Group 2, ICI without chemotherapy compared to chemotherapy alone demonstrated a numeric trend toward improved median OS in patients with BM (HR 0.40; 95% CI 0.12–1.36; p = 0.14; I ² = 65%; Figure S3A). In patients without BM, median OS was significantly improved with ICI compared to chemotherapy (HR 0.71; 95% 0.57–0.88; p = 0.002; I ² = 47%; Figure S3B). In Group 3, chemoimmunotherapy compared to chemotherapy alone improved median OS significantly in patients with BM (HR 0.54; 95% CI 0.42–0.68; p < 0.001; I ² = 0%; Figure S4A) and patients without BM (HR 0.71; 95% CI 0.64–0.79; p < 0.001; I ² = 22%; Figure S4B).

3.4. Pairwise Meta‐Analyses for ORR

In the studies that reported the ORR according to the presence or absence of BM, most reported on the ORR in the BM population and did not report on the population without BM. In patients with BM, treatment with ICIs with or without chemotherapy, improved the likelihood of a response compared to chemotherapy alone (OR 2.34; 95% CI 1.37–4.00; p = 0.002; I ² = 0%; Figure S5).

The GEMSTONE‐302 trial [27] was the only included trial to assess intracranial PFS. The addition of sugemalimab to chemotherapy improved intracranial PFS when compared with chemotherapy alone (HR 0.31; 95% CI 0.17–0.58).

3.5. Heterogeneity and Transitivity

The publication bias was minimal between the studies for both PFS and OS for the population without brain metastases (Figures S6 and S7). However, slight asymmetry was observed in the plots for the population with BM. This likely reflects variations in the sample size and variable reporting of the population with BM. The pairwise meta‐analyses had low I ² except for the analysis evaluating PFS in patients without BM. A sensitivity analysis revealed that Checkmate 227 Part 1 [29] was the study increasing heterogeneity in the group, likely due to the difference in the HR compared to the remaining groups. Removing this study from the assessment of median PFS in patients without BM reduced the heterogeneity (HR 0.57; 95% CI 0.51–0.64; p < 0.001; I ² = 49%; Figure S8).

3.6. Network meta‐analysis for PFS and OS

Using the pairwise meta‐analyses, without significant heterogeneity, indirect comparisons were made between groups treated with ICIs versus chemoimmunotherapy. Thus, the sensitivity analysis (excluding Checkmate 227 Part 1) was used for the assessment of PFS in the cohort without BM. Receipt of chemotherapy in addition to immunotherapy was associated with significantly improved median PFS compared to immunotherapy alone for patients with BM (HR 0.64; 95% CI 0.43–0.96; p = 0.03; Figure 5A). There was no difference in median PFS for patients who received chemoimmunotherapy versus immunotherapy without BM (HR 0.92; 95% CI 0.70–1.21; p = 0.56; Figure 5A).

FIGURE 5.

Forest plot for NMA: The use of immunotherapy with chemotherapy was associated with superior PFS in patients with BM compared with immunotherapy alone. (A) Forest plot for network meta‐analyses of median PFS in patients with and without BM. (B) Forest plot for network meta‐analyses of median OS in patients with and without BM. BM: brain metastases; Chemo: chemotherapy; CI: confidence interval; HR: hazard ratio, ICI: immune checkpoint inhibitor; NMA: network meta‐analysis; OS: overall survival; PFS: progression‐free survival.

There was no difference in median OS for patients who received chemoimmunotherapy versus immunotherapy in patients with BM (HR 0.99; 95% CI 0.56–1.74; p = 0.97; Figure 5B) versus no BM (HR 0.99; 95% 0.81–1.20; p = 0.92; Figure 5B).

An NMA was performed and categorized by type of immunotherapy (Figure 2B). There was high heterogeneity in some of the delineated pairwise analyses, and with a small number of studies, sensitivity analyses were not able to be performed. Thus this further analysis is exploratory and hypothesis‐generating only.

All ICI therapies, with or without chemotherapy were superior to chemotherapy alone in PFS and OS for patients without BM and OS for patients with BM.

For patients with BM, treatment with Anti‐PD‐(L)1 and chemotherapy showed superior PFS compared to Anti‐PD‐1 and Anti‐CTLA‐4 (HR 0.55; 95% CI 0.34–0.89; p = 0.01; Figure S9).

For patients without BM, treatment with either Anti‐PD‐1 alone (HR 0.70; 95% CI 0.54–0.91; p < 0.01) or Anti‐PD‐(L)1 with chemotherapy (HR 0.71; 95% CI 0.58–0.87; p < 0.001) showed superior PFS compared to treatment with both Anti‐PD‐1 and Anti‐CTLA‐4 plus chemotherapy (Figure S10). For patients with or without BM, there was no significant difference between treatments for OS when compared indirectly (Figures S11 and S12).

4. Discussion

This is the first NMA to analyze first‐line ICIs compared with chemoimmunotherapy in patients with metastatic NSCLC and BM. Our findings confirm that ICIs with or without chemotherapy improve ORR, PFS, and OS compared with chemotherapy alone for patients with metastatic NSCLC with or without BM, in the first‐line setting. Our NMA also demonstrated a significantly improved PFS with chemoimmunotherapy compared to immunotherapy in patients with metastatic NSCLC with BM.

Multiple phase II trials have demonstrated the efficacy of chemoimmunotherapy without prior local therapy in patients with metastatic NSCLC and BM [16, 17]. The Atezo‐Brain study reported an iORR of 42.7% and median intracranial PFS of 6.9 months using atezolizumab combined with chemotherapy in patients with metastatic non‐squamous NSCLC with untreated BM [16]. In the CAP‐BRAIN study, camrelizumab combined with chemotherapy similarly demonstrated an iORR of 52.5%, with a median intracranial PFS of 7.6 months [17].

In contrast, single‐agent ICIs have varying efficacy in NSCLC with BM. Both pembrolizumab and atezolizumab monotherapy have demonstrated activity in NSCLC with BM in Phase II and III trials, reporting an iORR of between 30% and 32% [15, 35, 36, 37]. While, a pooled analysis of the KEYNOTE studies comparing pembrolizumab to chemotherapy demonstrated pembrolizumab improved PFS and OS in patients without BM, however, in patients with BM, no significant OS or PFS difference was seen [38]. Multiple retrospective studies have also highlighted the benefit of anti‐PD‐1 therapy without prior local therapy [39, 40, 41]. In contrast, Tozuka et al., in a small retrospective analysis, reported limited efficacy of anti‐PD‐(L)1 monotherapy for the management of untreated BM in NSCLC [42].

This NMA demonstrates improvement in PFS with chemoimmunotherapy for patients with BM. However, prospective studies have not yet evaluated the comparative efficacy of the addition of chemotherapy to ICIs in NSCLC for BM. An FDA pooled analysis of first‐line chemoimmunotherapy in the treatment of PD‐L1 ≥ 50% advanced NSCLC demonstrated improved response rates and PFS compared to immunotherapy alone [43]. However, there was no difference observed in OS. While this study evaluated subgroups of smokers, age, and ECOG, no analysis was performed on the population with BM. However, this was explored in a registry study by Brown et al. which reported an improved iORR and OS with chemoimmunotherapy compared with immunotherapy alone, in patients with metastatic NSCLC with BM [41]. Specifically, patients receiving chemoimmunotherapy had an iORR of 58% versus 31% with immunotherapy alone (p = 0.01).

Recently, the FLAURA‐2 trial also demonstrated the addition of platinum‐pemetrexed chemotherapy to Osimertinib leads to better CNS efficacy (progression or death due to CNS involvement) than Osimertinib monotherapy in patients with EGFR‐mutant NSCLC [44, 45]. While Osimertinib alone [46] and platinum‐pemetrexed chemotherapy [47, 48] each exhibit intracranial activity independently, it has been suggested that CNS metastases disrupt the blood–brain barrier, enhancing the penetration of chemotherapeutic agents [45]. This may explain the improved CNS outcomes in patients treated with both chemotherapy and Osimertinib. Considering the known efficacy of anti‐PD‐(L)1 monotherapy on BM in patients with NSCLC [15, 36, 37], it is reasonable to hypothesize that the addition of chemotherapy is also synergistic and thus improves survival outcomes in this population, explaining the findings in our analysis.

While there was an improvement in PFS with chemoimmunotherapy compared to immunotherapy in patients with BM, there was no improvement seen in OS. This may be explained by multiple factors. First, some of the studies allowed for crossover, obscuring some of the differences in OS [13, 23, 25, 27, 28]. Secondly, these studies were not specifically powered to assess the OS differences in the subpopulation of patients with BM. Thirdly, treatment of progression likely varied amongst the study populations and may have included local therapies or subsequent systemic therapies, mitigating the mortality risk post‐progression, thus diluting the impact of treatments on OS.

The tumor immune microenvironment of BM from NSCLC has also been shown to be more immunosuppressed compared to that of the corresponding primary lung tumors [49]. Thus immunotherapy alone may not be sufficient to achieve optimal outcomes. However, treatment with Anti‐PD‐1 therapy has been shown to activate CD4+ T cells in the brain leading to local IFN‐gamma production that weakens the blood–brain barrier, allowing improved drug penetration [50]. ICIs also increase the migration of immune cells to the brain [51]. The addition of chemotherapy is thought to then further augment the modulation of the immune response by ICIs [52, 53, 54].

Bevacizumab has also been shown to reduce BM incidence in patients with NSCLC [55] by inhibiting angiogenesis and formation of CNS micro‐metastases [56, 57]. Although this meta‐analysis did not explore the benefit of VEGF‐inhibitors, the IMpower 150 study noted a reduced rate of new BM development with the addition of atezolizumab to bevacizumab and chemotherapy, though not statistically significant (HR 0.68; 95% CI 0.39–1.19) [58]. The TASUKI‐52 trial also demonstrated a trend toward improved PFS with the addition of nivolumab to bevacizumab and chemotherapy (HR 0.65; 95% CI 0.36–1.18) [59]. No trials have to date evaluated chemoimmunotherapy with or without bevacizumab for the management of BM in NSCLC.

In our meta‐analysis, 11.7% of patients had BM, which is congruent with the reported incidence of BM in patients with NSCLC of 10.4% [5]. However, in two of the trials, only 3.6%–6.7% of patients had BM [22, 25]. Historically, patients with metastatic NSCLC with BM are often excluded or under‐represented in clinical trials, which highlights the need for increased representation and outcome reporting for this subgroup in trial populations [60].

Our study supports the activity of chemoimmunotherapy in patients with BM, suggesting it may be more effective than immunotherapy alone in improving PFS. However, it is important to note that most patients included required local therapy prior to clinical trial enrolment (Table 2). For select patients, there may be a role for upfront systemic therapy with local treatments reserved as salvage treatment [16, 17, 41], deferring the risk of radiation necrosis, cognitive dysfunction and other long‐term neurological complications [61, 62, 63]. However, current guidelines have not determined a consensus recommendation regarding deferring local therapy in favor of systemic therapy [64]. With an increase in diagnosis of patients with asymptomatic BM [65], secondary to improvements in screening and imaging, the optimal management, and sequencing for patients with metastatic non‐oncogene addicted NSCLC must be elucidated in prospective clinical trials.

4.1. Limitations

While this NMA demonstrates improved PFS with chemoimmunotherapy compared to immunotherapy alone in patients with BM, there are several limitations to our study. Firstly, only three included studies [23, 26, 27], as well as KEYNOTE‐189 and ‐407 (but not ‐021) from the pooled analysis [26], allowed recruitment of patients without prior local therapy who had asymptomatic, stable BM. Thus, assessment of the true effect of the addition of chemotherapy to immunotherapy in patients with untreated intracranial disease was unable to be assessed. Secondly, we assessed the subgroup or posthoc analyses of these phase II and III clinical trials which increases the risk of type 1 errors due to small sample size or underpowered subgroup analyses. Thirdly, there was variability in PD‐L1 expression levels and PD‐L1 assays across trials, with studies such as KEYNOTE‐024 [12, 13] and EMPOWER‐Lung 1 [28, 32] only enrolling patients exhibiting PD‐L1 expression of at least 50%. Typically, the use of Anti‐PD‐(L)1 agents alone is not advised for patients with PD‐L1 expression below 50%. However, our NMA indicates that chemoimmunotherapy enhanced PFS in patients with BM, suggesting a significant benefit from adding chemotherapy across a broad range of PD‐L1 expressions. Lastly, the indirect comparisons performed in an NMA do not replace a prospective randomized trial.

5. Conclusions

In this NMA, patients with BM and NSCLC had improved median PFS with first‐line chemoimmunotherapy compared to immunotherapy alone. This was not observed in the cohort of patients without BM. Thus, the addition of chemotherapy to ICIs may enhance the treatment efficacy of patients with NSCLC and BM. In the absence of direct randomized trial comparison, this offers the best contemporary evidence regarding the most appropriate first‐line systemic therapy for patients with NSCLC and BM without a targetable oncogene mutation. Thus, chemoimmunotherapy should be considered in suitable patients with BM, regardless of PD‐L1 status. Prospective randomized trials are required to confirm our findings.

Author Contributions

Lauren Julia Brown: conceptualization, methodology, formal analysis, investigation, data curation, writing – original draft preparation, visualization, project administration. Nicholas Yeo: formal analysis, investigation, data curation, writing – original draft preparation. Harriet Gee: writing – reviewing and editing. Benjamin Y. Kong: conceptualization, supervision, writing – reviewing and editing. Eric Hau: conceptualization, supervision, writing – reviewing and editing. Inês Pires da Silva: supervision, writing – reviewing and editing. Adnan Nagrial: supervision, writing – reviewing and editing.

Conflicts of Interest

L.J.B. and N.Y. declare no conflicts of interest. H.G. received honoraria from Astra Zeneca. E.H. received honoraria and research funding from Astra Zeneca and honoraria from Novartis. I.P.S. is a Consultant Advisor and speaker for MSD and received honoraria from Roche, Novartis, and Bristol Myers Squibb. B.Y.K. received honoraria from Astra Zeneca. A.N. is an Advisory board member for MSD, BMS, Roche, Astra Zeneca, Pfizer Merck, Serono.

Supporting information

Data S1.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.