Comparison among stereotactic body radiation therapy, lobectomy, segmentectomy, and wedge resection for clinical stage I non-small cell lung cancer: a network meta-analysis.

Introduction
Non-small cell lung cancer (NSCLC), which constitutes approximately 85% of all lung cancer cases, is associated with high prevalence and mortality rates worldwide (1,2). For early-stage NSCLC, there is no doubt that surgery has been the standard treatment method. In 1995, the Lung Cancer Study Group reported a randomized controlled trial (RCT) comparing lobectomy and sublobar resection, and lobectomy became the standard procedure for early-stage patients (3). However, with the development of surgery, an increasing number of studies have found that sublobar resection is not inferior to lobectomy. For individuals with clinical stage IA NSCLC <2 cm, wedge resection and lobectomy had similar outcomes regarding recurrence-free survival (RFS) (4). Compared to lobectomy, segmentectomy resulted in significantly higher 5-year overall survival (OS) for patients with small-peripheral NSCLC (5). There were differences between different lung resections, and not all early-stage NSCLC patients need to undergo lobectomy to achieve better treatment outcomes (4).
In recent years, stereotactic body radiation therapy (SBRT) has shown satisfactory effectiveness in the treatment of early-stage NSCLC (6). Although recommended as the standard care method for patients who were medically inoperable for various reasons, the favorable outcomes achieved with SBRT in inoperable cases suggest its potential applicability in surgical candidates as well (6,7). However, its comparison with surgical results has been controversial. A retrospective study, after conducting propensity score matching (PSM), revealed that there were no significant differences in OS and disease-free survival between SBRT and surgery (8). Even a previous pooled analysis of two randomized trials showed that compared to the surgical group, the SBRT group has a higher 3-year OS and RFS for operable stage I NSCLC (9). However, a meta-analysis summarizing 30 studies showed that for early-stage NSCLC patients, the 3-year OS and cancer-specific survival in the surgical group were better than those in the SBRT group (10). It is worth noting that previous studies viewed lung resections as a whole and compared its effectiveness with SBRT, without considering the heterogeneity between different lung resections. Therefore, the differences in effectiveness between specific lung resections and SBRT are unclear.
To evaluate the differences in effectiveness between specific lung resections and SBRT, this study conducted a network meta-analysis for comparing survival outcomes and treatment-related complication incidence among clinical stage I NSCLC patients who underwent SBRT or lung resections (lobectomy, segmentectomy, and wedge resection). We present this article in accordance with the PRISMA-NMA reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1568/rc) (11).

Methods

Data sources and searches
To identify the relevant studies, two authors (J.Z. and J.H.) independently searched the records in the electronic databases of PubMed, EMBASE, and The Cochrane Library. We used the following keywords: “Carcinoma, Non-Small-Cell Lung”, “stage I”, “lobectomy”, “segmentectomy”, “wedge resection”, and “stereotactic ablative radiotherapy”. All relevant articles were searched before February 8, 2025, and detailed search criteria can be found in the Appendix 1. To guarantee the comprehensiveness of this research, a thorough screening was conducted on the references cited in the retrieved articles. Additionally, unpublished studies listed in the ClinicalTrials.gov registry were meticulously examined. The protocol of this study has been registered in the International Prospective Register of Systematic Reviews (PROSPERO; number CRD42023412394).

Study selection criteria
The studies incorporated into this network meta-analysis were selected based on the following eligibility criteria: (I) patients diagnosed with clinical stage I NSCLC; (II) comparison of two or more treatment modalities among SBRT, lobectomy, segmentectomy, and wedge resection; (III) at least one outcome among OS, RFS, or treatment-related complication incidence was reported; (IV) cohort study or RCT. Studies were excluded from this network meta-analysis for the following reasons: (I) absence of full-text reports; (II) unavailability of valid data; (III) a low Newcastle-Ottawa Scale score for cohort studies (score <7). Initial screening was conducted based on the title and abstract, followed by a thorough examination of the entire study and a comprehensive assessment to exclude literature that did not meet the inclusion criteria. When there were disagreements regarding the selected research, the third author (H.Y.) independently conducted a check to eliminate discrepancies and reached a consensus.

Data extraction and quality assessment
Two authors (J.Z. and J.H.) independently extracted data for each included study into a spreadsheet, and the third author (H.Y.) resolved conflicts when disagreement occurred. The data extracted mainly contained the first author, publication year, country, research period, study design, clinical stage of each study, different treatments, number of patients, age, female proportion, radiologic tumor diameter, forced expiratory volume in one second (FEV1)%, and diffusion capacity of the lung for carbon monoxide (DLCO)% in different intervention groups. Since the included studies spanned a wide time range, some applied the 7th edition and others the 8th edition of the IASLC staging system. Therefore, we accepted the definition of clinical stage I NSCLC as reported by the original study authors and did not impose a unified staging definition. If a study used propensity score matching, the matched data were prioritized. The primary outcome was OS, and the secondary outcomes were RFS and treatment-related complication incidence. Engauge Digitizer 11.3 software was used when survival data could not be directly extracted (12). The quality of the included cohort studies and RCTs was evaluated using the Newcastle-Ottawa Quality Assessment Scale and Cochrane Collaboration’s tool, respectively (13,14). Studies with scores ≥7 and moderate risk or below were considered eligible for final inclusion in the network meta-analysis.

Statistical analysis
Stata software (version 17.0) was used to draw network plots of treatments. The thickness of lines between nodes and the size of the nodes represented the number of studies (15). The survival outcomes we analyzed were HR with 95% confidence intervals (CIs) for OS and RFS. Furthermore, we summarized the effect of treatments on treatment-related complication incidence using the odds ratio (OR) and the corresponding 95% CI. Bayesian hierarchical random effects model was used to perform network meta-analysis in R software (version 4.2.1; Package gemtc) (16). This method is widely used in network meta-analyses for comparing multiple treatments, especially when indirect evidence is involved (17,18). Random effects and consistency model were computed by using Markov chain Monte Carlo (MCMC) methods with Gibbs sampling based on simulations of 100,000 iterations, 50,000 adaptions and a thinning interval of 10 in each of 4 chains (19). Model convergence was assessed through iteration plots and Gelman-Rubin diagnostics (20). Within the bayesian framework, the network meta-analysis estimated the overall rankings of treatments by generating the cumulative ranking curve for each treatment (21). The greater the surface under the cumulative ranking curve (SUCRA), the higher the likelihood of an intervention being considered the most effective measure.
Three critical assumptions underlying the network meta-analysis are similarity (that requires all studies included within a network to be comparable in terms of key factors that can be potential treatment effect modifiers), homogeneity (no significant variation in treatment effects among studies of same comparison), and consistency (the direct and indirect results are consistent). The comparability of baseline characteristics among patient cohorts across different treatment modalities was evaluated using the independent samples T-test in SPSS software (version 25). To visually represent the distribution of patient characteristics, box plots were generated (Figure S1). I2 statistic was used to assess heterogeneity in studies (22). Global inconsistency was assessed by comparing the deviance information criterion (DIC) of consistency and inconsistency models. A difference of less than 5 indicates adherence to the consistency assumption, allowing the use of a consistency model (23). Furthermore, the node-splitting approach was used to perform local inconsistency analysis within studies of closed loops in the network evidence plot (24). This involved generating effect size and CI for the indirect comparison to estimate inconsistency between direct and indirect comparison results. For publication bias analysis, an adjusted network funnel plot was plotted when the number of included studies exceeded 10.

Subgroup analysis
Three subgroup analyses were conducted, stratified by patient clinical stage (clinical IA stage and clinical stage I), regions (Asia and the West), and publication year (2009–2015 and 2016–2025). Subgroup analysis only focused on the primary outcome OS. Additionally, considering various surgical methods collectively, a pairwise meta-analysis was conducted to compare the effectiveness of SBRT and surgery in patients.

Results

Study selection
The literature selection process and reasons for study exclusion were shown in Figure 1. A total of 10,080 records were retrieved by database searching. After removing 4,020 duplications, 6,060 articles were screened for eligibility based on their titles and abstracts and 5,655 irrelevant articles were eliminated following application of the inclusion and exclusion criteria. After assessing the full-text articles, 375 articles were excluded due to the study population (128 records), interventions (103 records), outcomes (46 records), and study design (57 records) not meeting the standards, inability to obtain the full text (32 records), and inadequate research quality (9 records). Ultimately, 30 articles were included in the network meta-analysis.

Study characteristics and baseline characteristics
The basic characteristics of enrolled studies and patients’ baseline characteristics were shown in Table 1. There were 28 retrospective cohort studies and 2 RCTs, respectively, conducted across diverse geographical regions, encompassing 5 studies in Europe (5,25-28), 9 in North America (29-37), and 16 in Asia (38-53). There were 14 studies comparing segmentectomy and lobectomy (25,31,33,36,42,43,45-49,51-53), 7 studies comparing SBRT with lobectomy (26-28,35,39,40,44), 4 studies comparing wedge resection vs. segmentectomy (30,38,41,50), 2 studies were enrolled that compared wedge resection with SBRT (32,34), and one study comparing wedge resection with lobectomy (5). Additionally, two three-arm trials were included (29,37). The period of these studies is 2009–2025. A total of 27 studies reported on OS, with 14 and 10 studies focusing on RFS and treatment-related complication incidence, respectively.

Quality assessment
Quality assessment scores were provided in Table 1. The evaluation results showed that 6 studies scored 7 points, 16 scored 8 points, and 6 scored 9 points. Two RCTs were both at moderate risk of bias.

Network meta-analysis results
The network evidence plots for each outcome were shown in Figure 2. The trajectory plot and convergence diagnosis plot showed that the model had satisfactory stability and convergence (Figures S2,S3).
In terms of the OS, the pooled analysis indicated that lobectomy [hazard ratio (HR) 0.65, 95% CI: 0.53–0.79], segmentectomy (HR 0.64, 95% CI: 0.50–0.82), and wedge resection (HR 0.72, 95% CI: 0.55–0.93) were associated with superior OS in patients with clinical stage I NSCLC compared to SBRT. There were comparable OS among the three lung resections (Figure 3A). The treatments were ranked as follows: segmentectomy (SUCRA, 0.797), lobectomy (SUCRA, 0.747), wedge resection (SUCRA, 0.453), and SBRT (SUCRA, 0.003) (Figure 3B). The comparison-adjusted funnel plot indicated that there might be some publication bias in the included studies (Figure 3C).
In terms of the RFS, the pooled analysis showed that patients who underwent lobectomy (HR 0.65, 95% CI: 0.46–0.83) or segmentectomy (HR 0.68, 95% CI: 0.46–0.91) had superior RFS compared to those treated with SBRT. Patients who underwent wedge resection had similar RFS (HR 0.78, 95% CI: 0.49–1.12) to those treated with SBRT, and there were comparable RFS between different lung resections (Figure 4A). According to the ranking plot, the top-ranked treatment was lobectomy (SUCRA, 0.868), segmentectomy (SUCRA, 0.729), followed by wedge resection (SUCRA, 0.374), and finally SBRT (SUCRA, 0.029) (Figure 4B). Moreover, no publication bias was found in the selected studies (Figure 4C).
Similar treatment-related complication incidence was found in each treatment group (Figure 5A). Wedge resection (SUCRA, 0.906) was associated with the lowest incidence of treatment-related complication, followed by segmentectomy (SUCRA, 0.582), lobectomy (SUCRA, 0.372), and SBRT (SUCRA, 0.140) (Figure 5B). The visual examination of the comparison-adjusted funnel plot did not indicate a significant publication bias in treatment-related complication incidence (Figure 5C).

Subgroup analysis
In clinical stage IA patients, the OS of patients among the four treatments were comparable [lobectomy vs. segmentectomy: HR, 1.01 (0.94, 1.10); lobectomy vs. wedge resection: HR, 0.58 (0.31, 1.08); segmentectomy vs. wedge resection: HR, 0.57 (0.31, 1.07); lobectomy vs. SBRT: HR, 0.99 (0.85, 1.16); segmentectomy vs. SBRT: HR, 0.98 (0.83, 1.16); wedge resection vs. SBRT: HR, 1.71 (0.91, 3.26)] (Figure 6A). The ranking of effectiveness for OS in this subgroup was observed as follows: segmentectomy (SUCRA, 0.737), lobectomy (SUCRA, 0.622), wedge resection (SUCRA, 0.598), and SBRT (SUCRA, 0.043) (Figure 6B). Comparison-adjusted funnel plot showed no publication bias in the selected studies (Figure 6C). In Asian patients, both lobectomy and segmentectomy were associated with better OS compared to SBRT [lobectomy: HR, 0.71 (0.54, 0.95); segmentectomy: HR, 0.70 (0.52, 0.95); wedge resection: HR, 0.83 (0.55, 1.26)], while in Western patients, only lobectomy showed superior outcomes to SBRT [lobectomy: HR, 1.58 (1.11, 2.26); segmentectomy: HR, 1.57 (0.94, 2.63); wedge resection: HR, 1.43 (0.94, 2.18)] (Figure S4A). In Asian patients, the treatments ranked as segmentectomy, lobectomy, wedge resection, and SBRT (Figure S4B), while in Western patients, were lobectomy, segmentectomy, wedge resection, and SBRT (Figure S4C). Among patients from 2009 to 2015, lobectomy and segmentectomy exhibited superiority over SBRT, a trend similarly observed in the subgroup of patients from 2016 to 2025 (Figure S5A). Lobectomy showed better efficacy for the former, while segmentectomy was better for the latter (Figure S5B,S5C). There may be partial publication bias in included studies (Figure S6A-S6D). A total of 20 (8,26-28,32,34,35,37,39,40,44,54-62) studies were included in this pairwise meta-analysis. Pooled analysis revealed that surgery was significantly superior to SBRT in OS for clinical stage I NSCLC (HR 1.25, 95% CI: 0.83–1.89, P<0.01) (Figure S7).

Similarity, homogeneity and consistency
The results of the similarity assessment showed differences in age, tumor diameter, FEV1%, and DLCO% among patients in different treatment groups (Table S1). Heterogeneity was detected in the direct comparisons of SBRT vs. lobectomy (I2=96%) and wedge resection vs. SBRT (I2=90%) in the OS, SBRT vs. lobectomy in the RFS (I2=90%), SBRT vs. lobectomy (I2=99%) and wedge resection vs. segmentectomy (I2=74%) in the treatment-related complication incidence (Table S2). The results of local inconsistency indicated that P values were all >0.05 (Figures S8,S9). Additionally, the results of global inconsistency showed that the difference of DIC values between the consistent and inconsistent models was <5 (Table S3), satisfying the consistency assumption.

Discussion
This systematic review and network meta-analysis were conducted to assess the survival outcomes and treatment-related complication incidence among patients with clinical stage I NSCLC treated with SBRT, lobectomy, segmentectomy, or wedge resection. The study had the following key findings. First, lobectomy and segmentectomy demonstrated superior OS and RFS to SBRT in patients with clinical stage I NSCLC. Second, no significant disparities were observed in treatment-related complication incidence among the patients treated with the four distinct treatments. Third, for clinical stage IA NSCLC patients, SBRT may be comparable to specific lung resections in OS.
SBRT has become the standard treatment for stage I NSCLC patients who are not candidates for surgery due to poor general health or comorbidities because of its high local control rate and low toxicity (63,64). In addition, during the study period from 2012 to 2018 in the United States, about 8.9% of patients treated with SBRT had refused a clinician-recommended surgery (65). Surgical anxiety, the need for postoperative hospitalization, and the convenience of SBRT may explain why more patients are opting for SBRT. Therefore, an increasing number of studies comparing SBRT and surgery to determine which is more effective for early-stage NSCLC patients. Many retrospective studies have reported comparable survival outcomes between SBRT and surgery in operable stage I NSCLC patients (8,54,58-61). However, some studies using PSM have found that patients in the surgical group have significantly higher OS than those in the SBRT group (27,55,65). In contrast, a pooled analysis involving 58 operable NSCLC patients demonstrated a significantly higher 3-year OS in the SBRT group compared to the surgery group (surgery group:79%, SBRT group: 95%; P=0.037) (9). Due to the limited sample size, caution should be exercised in interpreting these findings. Most previous studies compared the differences in the effectiveness of various surgeries as a whole with SBRT, with limited research focusing on comparisons between specific lung resections and SBRT. By conducting a network meta-analysis, our study systematically collated and evaluated these specific comparative findings. We found that lobectomy, segmentectomy, and wedge resection were superior to SBRT in OS of patients. No significant difference in OS was observed among patients who underwent the three lung resections. More importantly, the SUCRA values indicated that patients in the segmentectomy group may have better OS, followed by the lobectomy, wedge resection, and SBRT groups. The finding was consistent with the Japan Clinical Oncology Group (JCOG) 0802 trial, which showed that segmentectomy significantly improved the OS of clinical stage IA NSCLC patients compared with lobectomy (47). In addition, a supplementary analysis of the JCOG 0802 indicated that the non-lung cancer death was associated with the number of resected segments. The more segments resected, the higher the incidence of non-lung cancer mortality (66). Compared to lobectomy, segmentectomy preserves more lung parenchyma, leading to lower mortality from other causes. As for the SBRT, our findings were consistent with most previous studies that the SBRT was associated with inferior OS, whether compared to surgery as a whole or to specific lung resection (27,34,57).
Early-stage NSCLC patients undergoing surgery may have a lower recurrence rate due to the complete resection in primary tumor and the accurate assessment of lymph node statuses. This study revealed that lobectomy and segmentectomy provided better RFS compared to SBRT, while the RFS in patients who underwent wedge resection was comparable to that of SBRT. No significant difference in RFS was observed among patients who underwent the three lung resections. This result was consistent with the post-hoc analysis results of Cancer and Leukemia Group B (CALGB) 140503 (29), which showed that lobectomy, segmentectomy, and wedge resection provided comparable RFS. Notably, based on the SUCRA values, the lobectomy group may have the best RFS, followed by segmentectomy, wedge resection, and SBRT, which differed to that the segmentectomy ranked first in OS. Compared with segmentectomy, the complete removal of a lung lobe significantly reduces the risk of local recurrence and improves the RFS of patients (47). Such discrepancies between OS and RFS in lobectomy and segmentectomy were also found in JCOG0802 trial. As a non-anatomical resection, wedge resection is associated with higher recurrence rate. Therefore, the RFS of patients underwent wedge resection was slightly inferior to that of the lobectomy and segmentectomy groups, and similar to that of the SBRT group.
While differences in effectiveness between SBRT and lung resections were observed, the treatment-related complication incidence were comparable between them for patients with early-stage NSCLC. SBRT delivers a very precise and high dose of radiation to a lung tumor in a small number of fractions and achieves more than 90% of local control rate (67). However, patients were prone to developing radiation pneumonia and radiation dermatitis after SBRT (26,32,57). In lung resection groups, the common treatment-related complication incidence of lobectomy included atrial fibrillation and pneumonia, while air leak was more likely to occur in segmentectomy and wedge resection (5,36,41,48). Although most studies indicated that treatment-related complications were more frequent and severe in surgery group (44,59,61), this study only investigated the incidence of treatment-related complications. Patients treated with SBRT may experience milder symptoms of complications, but the frequency of occurrence could be similar. Our analysis aimed to provide an overall estimate of the incidence of treatment-related complications, rather than a strict type-by-type comparison among lung resections and SBRT.
Although surgical resection remains the standard treatment for patients with stage I NSCLC, many previous studies have shown that for selected patients, SBRT may achieve comparable outcomes under specific conditions (68). Paul and colleagues demonstrated that for patients with tumors ≤2 cm, cancer-specific survival was not significantly different between surgery and SABR (68). Similarly, the revised STARS trial confirmed that there was no significant difference in OS between the surgical group and the SBRT group when only patients with clinical stage IA NSCLC with a tumor diameter <3 cm were included (69). In contrast, when clinical stage IB patients with tumor diameter >3 cm but ≤4 cm were included, the OS of patients in the SBRT group decreased significantly compared to those in the surgical group (44). This may be explained by the better prognosis of stage IA patients, which allows SBRT to demonstrate more pronounced benefits, whereas the poorer prognosis of stage IB patients and the limited tumor control of SBRT contribute to inferior outcomes. Considering the impact of tumor stage and histological features on treatment effectiveness, we performed several subgroup analyses. For clinical stage IA NSCLC patients, the OS of patients treated with SBRT and the three lung resections were similar, consistent with previous studies. Moreover, prior studies have suggested that for early-stage NSCLC patients unwilling to undergo lobectomy or at high operative risk, SBRT is an excellent alternative (70), for which our meta-analysis further provides evidence-based support. However, despite the similar HR for OS between SBRT and surgery, our study found that SBRT had the lowest SUCRA value, indicating a lower overall effectiveness in comparison with the surgical treatments. This suggests that while SBRT may offer comparable OS to surgery, its overall treatment effectiveness, considering factors such as local control and complication rates, does not reach the level of surgery. In addition, the impact of tumor components on effectiveness is also worth paying attention to. For patients with subsolid NSCLC, 5-year all-cause survival for surgical patients and those who underwent SBRT were found to be similar. However, in solid NSCLC patients, SBRT significantly reduces the 5-year all-cause survival compared to surgery (71). At present, there is a lack of sufficient literature to explore the impact of different tumor components on the effectiveness of different treatments, and further related research is needed.
Several inevitable limitations exist in our study. First, because the majority of the included studies were retrospective in nature, systematic differences inevitably existed in treatment selection and baseline characteristics, which may have influenced the comparability between groups. Consequently, this network meta-analysis does not fully satisfy the similarity hypothesis between studies, as patients using different treatment methods are difficult to be completely consistent in age, tumor size, FEV1%, and DLCO%. Second, subgroup analysis stratified by stage only includes clinical stage IA NSCLC patients. Some characteristics, like the proportion of solid component, potentially influence the efficacy in different treatments. Third, some of the data included in the study were extracted through software or converted into mean and standard deviation through statistical calculations (72,73), resulting in potential biases. This method was essential because in some cases, raw data cannot be directly obtained.

Conclusions
For clinical stage I NSCLC patients, lobectomy and segmentectomy are superior to SBRT. Wedge resection is associated with similar RFS but better OS to SBRT. No significant difference in treatment-related complication incidence was observed among the three lung resections and SBRT groups. In clinical stage IA NSCLC, SBRT may provide comparable OS to lung resections.

Supplementary
The article’s supplementary files as