First-line PD-1/PD-L1 inhibitors plus chemotherapy vs. chemotherapy alone in stage IIIB-IV non-squamous NSCLC: an updated meta-analysis of phase 3 RCTs.

Introduction
Non-small-cell lung cancer (NSCLC) represents nearly 85% of lung malignancies, predominantly exhibiting a non-squamous phenotype [1]. Despite advancements in early detection and treatment, many individuals still present with stage IIIB-IV disease, showing poor clinical outcomes [2]. Immune checkpoint inhibitors (ICIs), especially agents against PD-1 and PD-L1, have revolutionized the therapeutic landscape of NSCLC [3]. These agents, through reactivating T-cell-mediated immune responses, have demonstrated substantial survival benefits in various clinical settings [4].
Several pivotal phase 3 randomized controlled trials (RCTs) have evaluated the efficacy of PD-1/PD-L1 inhibitors combined with chemotherapy (PIC) in treatment-naïve patients with advanced non-squamous NSCLC (nsqNSCLC) [5–16]. As shown in KEYNOTE-189, adding pembrolizumab to platinum–pemetrexed resulted in marked overall survival (OS) and progression-free survival (PFS) gains over chemotherapy, regardless of PD-L1 status [13]. Similarly, the IMpower132 study showed that atezolizumab combined with chemotherapy also enhanced survival outcomes in patients with metastatic nsqNSCLC [12]. Previous meta-analyses have examined the role of PIC in patients with nsqNSCLC [17, 18]. However, these investigations either pooled all histological subtypes of NSCLC or not entirely based on high-quality RCTs, leaving important gaps for this population.
Despite these promising results, the integration of ICIs into first-line treatment regimens raises several clinical questions. The heterogeneity among trials with respect to patient populations, PD-L1 expression thresholds, and chemotherapy backbones complicates direct comparisons. Moreover, the added toxicity burden of combining ICIs with chemotherapy necessitates careful evaluation of the risk–benefit balance. Although some studies suggest that the combination therapy offers superior efficacy, others indicate that the benefits may be limited to specific subgroups, such as patients with high PD-L1 expression or those without certain comorbidities [5, 14]. Further controversies exist regarding outcomes in patients with brain or liver metastases, those with low PD-L1 expression, and differences observed between Asian and Western populations. These uncertainties underscore the absence of a consensus on the optimal role of PIC in clinical practice [19].
Several pooled analyses have demonstrated that PIC significantly improves outcomes in patients with advanced squamous NSCLC [20]. However, in nsqNSCLC, the long-term durability of survival benefits from immunotherapy, particularly when combined with chemotherapy, remains uncertain. Against this background, an updated meta-analysis was conducted of phase 3 RCTs to evaluate the long-term efficacy and safety of PIC versus chemotherapy in stage IIIB–IV nsqNSCLC. Unlike previous systematic reviews, this study focuses exclusively on non-squamous patients and incorporates the most recent trials with mature follow-up data. This approach enables a more precise assessment of survival, tumor response, and toxicity, thereby addressing an important gap in the literature.

Materials and methods
This meta-analysis was conducted according to the PRISMA guidelines and registered in PROSPERO (ID: CRD420251066166, available from https://www.crd.york.ac.uk/PROSPERO/view/CRD420251066166).

Search strategy
A predefined search framework utilizing keywords such as “PD-1/PD-L1 blockade”, “Lung cancer”, and “Randomized trials” was applied (refer to TableS1). Relevant literature was retrieved from PubMed, EMBASE, Scopus, ScienceDirect, Cochrane Library, and Web of Science, encompassing studies published up to May 20, 2025.

Selection criteria
Studies were eligible if they met the following criteria: (1) phase 3 RCTs published in English; (2) enrolled individuals diagnosed with stage IIIB–IV nsqNSCLC; (3) either reported subgroup data for non-squamous histology or exclusively involved advanced nsqNSCLC populations; (4) evaluated PIC versus standard chemotherapy as initial systemic intervention; and (5) reported at least one clinical endpoint, including OS, PFS, response rates, or adverse events (AEs).
Studies were excluded if they met any of the following: (1) phase 1 or 2 trials; (2) lack of essential survival or safety outcomes; (3) reporting only combined data without separate subgroup results for nsqNSCLC; (4) duplicate publications or conference abstracts lacking complete information.
After deduplication, two reviewers independently screened titles and abstracts of the remaining records against the eligibility criteria. Full-text articles of potentially eligible records were retrieved and assessed independently by both reviewers. Any discrepancies were resolved through discussion, and, if necessary, adjudicated by a third reviewer. The detailed screening process with numbers at each stage and specific reasons for exclusion are presented in the PRISMA 2020 flow diagram (Fig. 1).

Data extraction
Data extraction was independently performed by two reviewers using a standardized template. Discrepancies were resolved through discussion or, if necessary, by a third reviewer. Collected details comprised the first author, publication year, study title, trial phase, size of the nsqNSCLC subgroup, treatment arms, patient characteristics, survival data, response rates, and AEs.

Outcome assessments
Subgroup analyses of OS and PFS were conducted, stratifying by factors including age, gender, ethnicity, ECOG PS, smoking status, disease stage, presence of brain or liver metastases, PD-L1 expression, PD-1/PD-L1 inhibitor type, and platinum chemotherapy type. It should be noted that the OS and PFS subgroup analyses were not pre-specified in the PROSPERO-registered protocol; these analyses were performed post-hoc based on subgroup data reported in the included RCTs. Response rates were subgroup analyzed as objective response rate (ORR), disease control rate (DCR), complete response (CR), partial response (PR), stable disease (SD), and progressive disease (PD). AEs were subgroup analyzed as treatment-emergent AEs (TEAEs), treatment-related AEs (TRAEs), and immune-related AEs (irAEs).

Quality assessment
Risk of bias and methodological quality of included studies were evaluated using the Cochrane risk-of-bias tool and the Jadad scale. Trials scoring between 4 and 7 were classified as high quality studies [21, 22]. Additionally, the GRADE framework was applied to assess evidence certainty, ranging from very low to high [23].

Statistical analysis
Statistical analyses were performed using Review Manager (RevMan) version 5.4 (Cochrane Collaboration, London, UK) and Stata version 17.0 (StataCorp LLC, College Station, TX, USA). Hazard ratios (HRs) and risk ratios (RRs) combined survival (OS and PFS) and dichotomous outcomes (survival rates, response rates, and AEs) from included studies. For AEs, absolute risk differences (ARD) and numbers needed to harm (NNH = 1/ARD) were calculated to complement relative risk estimates and provide clinically interpretable toxicity measures. Heterogeneity was assessed using the I² statistic and Cochrane’s Q test; values of I² >50% or p < 0.10 indicated notable heterogeneity. For high heterogeneity, random-effects models were applied; otherwise, fixed-effects models were used. Publication bias was evaluated with funnel plots and Egger’s and Begg’s statistical tests. Sensitivity analyses were conducted to assess the robustness of the results [24, 25]. To explore potential sources of between-study heterogeneity, univariable meta-regression analyses were performed for OS and PFS using study-level covariates [26]. Statistical significance was defined as p < 0.05.

Results

Search results
This meta-analysis included 34 articles from 12 phase 3 RCTs, encompassing a total of 5,054 patients with advanced nsqNSCLC [5–16, 27–48]. Figure 1 presents the screening process, in line with the PRISMA 2020 framework. Baseline study characteristics are summarized in Table 1. Most selected studies showed robust design and minimal bias (Figure S1, Table S2). Under GRADE methodology, evidence certainty varied between moderate and high across outcomes (Table 2). Five trials originated from China [5, 8, 10, 14, 16], whereas seven were multinational, multicenter studies [6, 7, 9, 11–13, 15]. Six trials exclusively enrolled patients with nsqNSCLC [5, 11–14, 16], with the remaining six providing subgroup analyses for this subtype [6–10, 15].

Survival
PIC was associated with a statistically significant improvement in OS (HR: 0.73 [0.68, 0.79], P < 0.00001) (Fig. 2). Across the 6–60 month follow-up period, OS rates consistently favored the PIC arm (Figure S2). With extended observation time, the advantage of PIC on survival became more evident (Fig. 3).

PIC also was associated with a statistically significant improvement in PFS (HR: 0.59 [0.55, 0.63], P < 0.00001) (Fig. 4). Across the 6–48 month follow-up period, PFS rates remained higher in patients receiving PIC (FigureS3). The PFS advantage similarly grew over extended follow-up (Fig. 5).

Subgroup analysis
Subgroup analyses of OS and PFS demonstrated that PIC improved outcomes across various baseline subpopulations (see outcome evaluations). Notably, patients with brain metastases and those with PD-L1 expression exceeding 50% showed particularly favorable survival trends with PIC treatment in both OS and PFS (Table 3).

Regression analysis
To explore potential sources of heterogeneity (OS I² = 47%; PFS I² = 35%), univariable meta-regression analyses were conducted based on region (China vs. global), inhibitor type (PD-1 vs. PD-L1), and platinum backbone (carboplatin vs. cisplatin/mixed). None of these factors were significantly associated with between-study heterogeneity for either OS or PFS (all P > 0.05), indicating that the pooled results were robust across study-level characteristics (FigureS4).

Responses
Compared with chemotherapy alone, PIC significantly improved tumor response, including ORR (RR: 1.69 [1.55, 1.84], P < 0.00001), DCR (RR: 1.15 [1.10, 1.19], P < 0.00001), CR (RR: 2.13 [1.19, 3.79], P = 0.01), and PR (RR: 1.68 [1.53, 1.84], P < 0.00001). Conversely, the PIC regimen resulted in fewer cases of SD (RR: 0.76 [0.70, 0.83], P < 0.00001) and PD (RR: 0.56 [0.47, 0.68], P < 0.00001) (Table 4, Figure S5).

PIC group was also associated with a prolonged duration of response (DOR) (HR: 0.57 [0.50, 0.65], P < 0.00001) (Fig. 6). DOR rates (DORR) remained higher in the PIC group from 6 to 48 months (Figure S6), and the magnitude of benefit increased with longer follow-up (Figure S7).

Safety
Patients receiving PIC experienced more grade 3–5 (RR: 1.18 [1.09, 1.28], P < 0.0001)/serious TEAEs (RR: 1.75 [1.35, 2.27], P < 0.0001), grade 3–5 (RR: 1.33 [1.23, 1.44], P < 0.00001)/serious TRAEs (RR: 2.17 [1.78, 2.66], P < 0.00001), and total (RR: 1.95 [1.32, 2.87], P = 0.0007)/grade 3–5 irAEs (RR: 2.28 [1.64, 3.18], P < 0.00001). However, overall frequencies of TEAEs and TRAEs did not significantly differ between treatment groups (Table 5, Figure S8).

For TEAEs, the PIC group showed increased rates of any-grade neutrophil count decreased, white blood cell decreased, diarrhea, leukopenia, platelet count decreased, decreased appetite, vomiting, pain in extremity, pyrexia, edema peripheral, rash, headache, hypoalbuminaemia, dysgeusia, pruritus, myelosuppression, hypothyroidism, myalgia, pneumonia, blood creatinine increased, gamma-glutamyltransferase increased, and hyperthyroidism (Table S3). Furthermore, grade 3–5 neutropenia, neutrophil count decreased, anaemia, leukopenia, and myelosuppression occurred more often in this group (Table S4).
In terms of irAEs, patients receiving PIC experienced more any-grade hypothyroidism, pneumonitis, severe skin reactions, hepatitis, hyperthyroidism, colitis, and nephritis (Table S5). Additionally, grade 3–5 nephritis, and colitis were more common in the PIC arm (Table S6).

Sensitivity analysis
Sensitivity analyses demonstrated stability of the combined risk ratios for 18-month OS rates, 24-month PFS rates, total TEAEs, and subgroup analysis of survival (OS and PFS), supporting the consistency of findings. No individual study exerted undue influence on heterogeneity or overall outcome measures, suggesting robustness across all included trials (Figure S9). Additionally, although follow-up durations varied markedly among studies (14.8–65.2 months), an RMST-based sensitivity analysis yielded results consistent with the primary survival outcomes, further supporting the reliability of the pooled estimates. Given the small sample and high heterogeneity in the stage III subgroup (n = 97, I² = 73%), a sensitivity analysis was performed excluding these patients. The pooled results (OS and PFS) remained unchanged, confirming that stage III cases had minimal impact on the overall outcomes.

Publication bias
Funnel plots for OS, PFS, ORR, and grade 3–5 TEAEs demonstrated overall symmetry (Fig. 7). Egger’s test yielded P-values of OS (P = 0.761), PFS (P = 0.095), ORR (P = 0.950), and grade 3–5 TEAEs (P = 0.079) (Figure S10). The Begg’s funnel plots showed near-symmetrical scatter around the pooled estimate, and Kendall’s tau correlation coefficients were non-significant across all endpoints (all P > 0.05) (Figure S10). These findings suggested no significant small-study effects or selective reporting bias.

Discussion
Immunotherapy has significantly transformed the therapeutic landscape for advanced nsqNSCLC [6, 7]. Despite promising evidence from pivotal trials such as KEYNOTE-189 and IMpower132, several clinical uncertainties remain regarding optimal patient selection and treatment stratification [12, 13]. In this updated meta-analysis, we included 12 phase 3 RCTs comprising 5,054 patients with stage IIIB–IV nsqNSCLC. Our pooled results demonstrate consistent improvements in OS and PFS with PIC therapy, with notable benefits observed in subgroups such as patients with brain metastases or PD-L1 expression >50%. PIC was also associated with higher ORR and prolonged DOR but accompanied by increased rates of grade 3–5 irAEs. These findings align with current first-line treatment guidelines and support more individualized decision-making, particularly for patients with high PD-L1 expression or brain metastases.
The meta-analysis demonstrated that PIC significantly prolonged OS (HR: 0.73) and PFS (HR: 0.59) compared with chemotherapy alone. The long-term survival improvement observed from 6 to 60 months suggests that adding ICIs to chemotherapy may induce sustained anti-tumor immunity, potentially through chemotherapy-mediated modulation of the tumor microenvironment, enhanced antigen release, and PD-L1 upregulation [49]. Subgroup analyses revealed meaningful patterns: patients with PD-L1 expression >50% showed the greatest treatment advantage, and exploratory findings indicated relatively stronger efficacy in Asian populations, possibly reflecting underlying genomic or immunologic differences. The observed benefit in patients with brain metastases further suggests that systemic immunomodulation may exert effects within the CNS, potentially aided by tumor-related blood–brain barrier alterations [50]. Nevertheless, given the multiple subgroup comparisons and limited sample sizes in certain categories—such as the OS analysis in patients with brain metastases (HR = 0.47)—these findings should be interpreted cautiously. The increasing magnitude of benefit over longer follow-up supports the characteristic tail plateau seen in ICI therapies, indicating durable disease control in a subset of patients. Clinically, these results reinforce that patients with high PD-L1 expression, brain metastases, and good performance status are most likely to benefit from PIC, whereas those with poor functional status or significant comorbidities may require a more individualized risk–benefit assessment.
The analysis of tumor response profiles in this meta-analysis further supports the superiority of PIC over chemotherapy alone. Significantly increased ORR, DCR, and CR were observed in patients receiving PIC. The improved ORR (RR: 1.69) and prolonged DOR (HR: 0.57) reflect the enhanced immune activation achieved when ICIs are combined with cytotoxic chemotherapy [51]. As shown in the KEYNOTE-042 and IMpower150 trials, patients in the combination therapy arms achieved earlier and deeper responses, potentially contributing to improved survival outcomes [52, 53]. Mo et al. further demonstrated that early tumor shrinkage within the first 8–12 weeks predicts long-term survival in patients treated with ICI-based regimens, particularly in non-squamous histologies [54]. While chemotherapy often induces rapid but transient responses, ICIs can promote sustained immune surveillance, which was evident in the DORR analysis from 6 to 48 months. Long-term follow-ups from KEYNOTE-021 and CheckMate 227 similarly demonstrate sustained tumor control in a subset of patients [7, 55]. Taken together, these refined response patterns suggest that PIC enhances both the depth and persistence of tumor control, contributing to the broader survival gains observed in this meta-analysis. Notably, the analysis also revealed a significantly higher CR rate (RR: 2.13) in the PIC group. Although complete responses remain uncommon in advanced NSCLC, their occurrence is clinically meaningful, as they often correlate with prolonged survival or treatment-free intervals. Moreover, the reduction in PD rates among PIC-treated patients suggests a decrease in both primary and adaptive resistance. In a real-world study by Kaneko et al., adding pembrolizumab to chemotherapy substantially lowered early progression rates in advanced nsqNSCLC [56]. Nonetheless, heterogeneity persists, and future studies should investigate predictive biomarkers—such as PD-L1 expression, TMB, STK11/KEAP1 mutations, and gut microbiome composition—to further refine individualized treatment strategies [57].
While the therapeutic benefits of PIC are substantial, this analysis also demonstrates a higher incidence of treatment-related toxicities, particularly hematologic AEs and irAEs. This pattern is consistent with pivotal trials such as IMpower132, CheckMate 9LA, and KEYNOTE-189, which similarly reported increased hematologic, gastrointestinal, and endocrine toxicities in combination therapy arms [12, 13, 58]. In our analysis, the most common any-grade AEs included anemia, nausea, and neutropenia—events largely driven by chemotherapy and generally reversible with supportive care. In contrast, irAEs such as hypothyroidism, pneumonitis, colitis, hepatitis, and nephritis reflect the unique immune-mediated toxicity profile associated with PD-1/PD-L1 blockade. ARD and NNH analyses further quantified these risks: hypothyroidism (ARD 8.7%, NNH ≈ 12), pneumonitis (ARD 4.7%, NNH ≈ 22), colitis (ARD 1.8%, NNH ≈ 57), and nephritis (ARD 1.3%, NNH ≈ 75) accounted for the most notable increases, with grade 3–5 colitis and nephritis corresponding to NNH ≈ 92 and ≈ 84, respectively. Importantly, while hematologic toxicities tend to be acute and manageable, irAEs may persist long after therapy discontinuation and impose chronic health burdens, particularly endocrine dysfunctions that often require lifelong hormone replacement [59]. Organ-specific irAEs such as pneumonitis or nephritis can be severe or life-threatening if not promptly recognized, emphasizing the need for early monitoring—especially during the first 12–16 weeks of treatment, when most irAEs emerge in clinical practice [60]. Despite these risks, treatment discontinuation rates remained low across trials, suggesting that timely use of corticosteroids, hormone replacement, or immunosuppressants can effectively mitigate toxicity without compromising therapeutic benefit. Notably, accumulating evidence indicates that the occurrence of irAEs may correlate with improved survival, reflecting more active antitumor immunity [61]. Therefore, irAEs should be viewed not only as safety events but also as potential biomarkers of therapeutic responsiveness. Future efforts should focus on developing predictive tools—such as baseline autoantibody profiles, cytokine signatures, or HLA typing—and integrating proactive supportive care strategies to optimize outcomes in patients at higher risk, including older individuals and those with significant comorbidities [61–63].
Several potential sources of bias should be acknowledged when interpreting these findings. First, a considerable proportion of the included trials were conducted in Asian populations, which may lead to an overestimation of benefit relative to Western cohorts, given known differences in tumor genomics, environmental exposures, and immunotherapy responsiveness. Second, although funnel plots showed no major asymmetry, publication bias cannot be fully excluded—particularly for subgroup analyses, where negative or neutral outcomes are less likely to be reported. Third, variations in chemotherapy regimens, PD-L1 testing platforms, and trial design characteristics may have contributed to outcome heterogeneity. These factors underscore the need for cautious interpretation of the pooled estimates and highlight the importance of future trials with harmonized methodologies and more globally representative patient cohorts.
Despite the robustness of the findings, this meta-analysis has several limitations. First, heterogeneity across the included RCTs—such as differences in patient populations, PD-L1 assessment platforms, chemotherapy backbones, and follow-up durations—may have contributed to variability in the pooled outcomes. Second, the analysis relied on aggregated study-level data rather than individual patient data (IPD), limiting the ability to perform refined subgroup analyses, fully evaluate covariate effects, or generate comprehensive absolute risk estimates such as NNT and NNH. Third, inconsistencies in toxicity reporting and methodological differences across trials may influence the precision of safety evaluations. Fourth, the long-term safety profile of PIC remains incompletely defined, highlighting the need for future studies with extended follow-up. Fifth, cross-trial comparability of PD-L1–stratified results is reduced due to non-uniform testing assays and cutoff definitions. Sixth, evidence for stage III disease remains limited owing to small sample size (n = 97) and substantial heterogeneity. Lastly, most trials did not stratify outcomes by clinically relevant subgroups—such as patients with significant comorbidities or varying performance statuses—restricting the generalizability of the findings to these populations.

Conclusion
The addition of PD-1/PD-L1 inhibitors to chemotherapy significantly improves survival and tumor response in patients with stage IIIB–IV non-squamous NSCLC. The most pronounced benefits are observed in specific patient subgroups, including those with PD-L1 expression ≥ 50%, brain metastases, or good performance status (ECOG 0–1). Although immune-related adverse events are more frequent with combination therapy, careful monitoring and timely management can mitigate risks, allowing many patients to continue treatment safely. Clinicians should weigh the potential benefits against uncertainties in toxicity data when selecting first-line therapy. Variability in toxicity reporting and methodological heterogeneity across trials, however, limit the precision of safety and subgroup analyses. Future studies should focus on identifying predictive biomarkers and refining patient selection to maximize therapeutic benefit and ensure safer application across diverse patient populations.

Supplementary Information