Stereotactic body radiotherapy for Ultra-Central lung Tumors: A systematic review and Meta-Analysis and International Stereotactic Radiosurgery Society practice guidelines

Background: Stereotactic body radiotherapy (SBRT) is an effective and safe modality for early-stage lung cancer and lung metastases. However, tumors in an ultra-central location pose unique safety considerations. We performed a systematic review and meta -analysis to summarize the current safety and efficacy data and provide practice recommendations on behalf of the International Stereotactic Radiosurgery Society (ISRS)


Introduction
Stereotactic body radiotherapy (SBRT) involves highly conformal radiotherapy delivered with high doses per fraction over a small number of fractions. It is the standard treatment option for peripherally located medically inoperable early-stage non-small cell lung cancer (NSCLC) patients and may be an option for operable stage I small cell lung cancer patients [1][2][3]. Its role has also been well established for metastasisdirected therapy (MDT) in the setting of oligometastatic disease [4][5][6], a treatment paradigm that is rapidly gaining traction across multiple disease sites within oncology.
Generally, lung SBRT is safe and effective, normally without grade 4 or 5 toxicities, for targets in the periphery of the lung parenchyma [7]. In contrast, targets in proximity to the central airway and mediastinal structures may pose increased risks due to the dose tolerances of the nearby organs-at-risk (OARs). The remarkably high toxicity rates observed in the early phase II trial from Indiana University using 60-66 Gy in 3 fractions set the precedence for the RTOG-defined "no-fly-zone", encompassing the 2 cm radius of the trachea and proximal bronchial tree as well as mediastinal structures [8]. With careful risk-adapted fractionation and total dose reduction however, subsequent studies have shown that SBRT can be safely delivered to central lesions with alternative dose/fractionation schedules [9]. However, a subset of central tumors that abut the central airway, esophagus, or other mediastinal structures are termed, "ultra-central", and associated with a substantial risk of high-grade toxicity from SBRT [10][11][12]. A 2019 systematic review of this population determined a 5 % risk of grade 5 toxicity from SBRT, predominantly from endobronchial hemorrhage [11]. More recently, the phase II HILUS trial reported significant risks of grade ≥3 (34 %) and grade 5 (15 %) toxicity following ultra-central SBRT for early-stage NSCLC [12]. While a relatively low dose of 56 Gy in 8 fractions was prescribed, concerns related to dose inhomogeneity within the target volume have been raised, further highlighting the balance of optimizing tumor control and decreasing OAR toxicity when considering SBRT for this patient population [13].
Given these trial results, and the publication of additional institutional series, the objective of this study was to perform an updated systematic review and meta-analysis to inform clinical decision making and guideline recommendations on behalf of the International Stereotactic Radiosurgery Society (ISRS). We hypothesize that with a more nuanced attention to radiotherapy prescription technical details and dosimetric constraints, SBRT for ultra-central lung tumors can be delivered safely and effectively.

Evidence acquisition
The Preferred Reporting Items for Systematic Reviews and Metaanalyses (PRISMA) and Meta analysis of Observational Studies in Epidemiology (MOOSE) reporting guidelines were used to guide the conduct of this review [14,15]. A comprehensive search was conducted in PubMed/MEDLINE and EMBASE databases for all articles published from January 1,1990, to February 2, 2022. Search terms included "SBRT" or "stereotactic body radiotherapy" or "SABR" or "stereotactic ablative radiotherapy" and "ultra-central" or "ultracentral" and "lung" or "pulmonary". Database searches were supplemented by manual searching of manuscript references. Inclusion criteria consisted of patient cohorts (1) with ultra-central lung tumors, (2) treated with SBRT (defined as ≥5 Gy per fraction using photons), and (3) at least one endpoint of interest was reported in terms of local control (LC) and/or toxicity. The definition of ultra-central varied based on the specific study, but generally included tumors in which the gross tumor volume (GTV) or planning target volume (PTV) abuts or overlaps the PBT or other mediastinal structures. Studies with <5 eligible lesions, reirradiation, non-English language, hilar/mediastinal nodal tumors, or containing a mixed patient cohort with the inability to discern ultracentral outcomes from others, were excluded. In the event of multiple publications of the same clinical cohort, the most recent study was included while others were excluded.

Data extraction
Search results were imported into Covidence (Veritas Health Innovation, Melbourne, Australia) for determination of eligibility. All studies were screened by two authors (MY and AVL), with conflicts settled by consensus. Data collection was performed on standardized extraction forms, in which baseline clinical, radiotherapy-specific, and outcomes data were extracted. Risk of bias for individual studies was assessed using a modified Newcastle Ottawa Scale (mNOS) [16]. When outcome measures and their variances were not stated within included studies, Kaplan Meier curves were digitized using Web Plot Digitizer, version 4.6, to extract pertinent 1-and 2-year values [17]. In the setting of heterogeneous dose fractionation schedules, doses were converted to biologically effective dose (BED) using the following formula: where n is the total number of fractions, d is the dose per fraction, and α/β is the alpha/beta ratio of the tumor (α/β = 10) or organ at risk (OAR) for late toxicity (α/β = 3). OAR doses are also expressed as Equivalent Dose in 2 Gy fractions (EQD2) using the formula: EQD2 α/β = nd( d d+α/β ). In order to standardize reported maximum doses (Dmax) to OAR, and taking into account heterogeneity between various methods of dose calculation, we selected a volumetric threshold of ≤0.5 cc as representative of Dmax for studies where it was not explicitly reported (e.g. D0.1 cc and D0.5 cc would be extracted as Dmax).

Outcomes definitions
The primary outcomes of interest were the 1-and 2-year rates of LC and incidences of grade 3-5 toxicity. Toxicity events were based upon grading defined by the Common Terminology Criteria for Adverse Events (CTCAE) version 4 or 5 [18,19]. Toxicities were stratified into bronchial and non-bronchial toxicities, with the former including hemoptysis, bronchial stenosis, and bronchial fistulisation.

Statistical analysis
Endpoints reported in the included studies were weighted by inverse variance and combined using a random effects model, with the pooled effect estimates depicted as forest plots with corresponding 95 % confidence intervals (95 % CI). The presence of publication bias was assessed visually with the use of funnel plots and quantified by Egger's test [20]. A p-value threshold of 0.05 was used for statistical significance, suggesting the presence of publication bias. The Freeman-Tukey double arcsine transformation was used. Inter-study heterogeneity was quantified by the Cochran Q test and the I 2 statistic. A Cochran Q < 0.1 represented significant heterogeneity. I 2 values exceeding 25 %, 50 %, and 75 % representing low, moderate, and high heterogeneity, respectively [21]. We also determined the value of τ, which represents the standard deviation of the pooled endpoint due to study heterogeneity.
Meta-regression was performed to compare summary effect sizes of LC and bronchial toxicity endpoints in relation to respective predictor variables, that is PTV volume and BED for LC, and PBT Dmax for bronchial toxicity. Meta-regression comprised a univariable linear regression weighted by individual study inverse variance. All data analysis was performed in Stata (Stata Corp) using the metaprop, metabias, and metafunnel packages and R (R Foundation) using the metareg package.

Study descriptions
A total of 27 studies published between 2010 and 2022, consisting of 1183 unique patients, met the criteria for study inclusion ( Figure A1, Table 1). One study was presented as an abstract while the others were full publications [22]. All studies were retrospective with the exception of the HILUS trial, which was a prospective phase 2 observational study [12]. Risk of bias assessment using the mNOS scale determined that most studies (81 %) were of high quality and received 7/8 stars. One study received full scores [12] and another received 5/8 stars, representing the lowest score [22] (Table A1). Median follow up ranged from 7.6 to 44.5 months for studies reporting this value. Studies were heterogeneous for the proportion of primary NSCLC versus metastatic disease, with 9 studies (33 %) consisting entirely of the former. The definition of ultra-central lesion varied (Table A2). PTV overlap with the PBT was included in the definition across all studies (100 %) while GTV overlap was included in only 16 studies (59 %). PTV overlap with other mediastinal structures including great vessels and esophagus were included in 14 studies (52 %), while only 4 studies (15 %) allowed for direct GTV overlap with these structures.
When looking specifically at the subset of patients with primary NSCLC only, the 1-year and 2-year LC were 95 %. Inter-study heterogeneity was high for included studies (I 2 = 96.3 % and 94.2 %).

Discussion
Carefully delivered SBRT, with attention to adjacent OAR dosing and location of "hot spots," is safe and effective in the treatment of ultracentral lung cancers and lung metastases. Local control was high, with pooled estimates of 1-year local of 92 % for both lung cancers and metastatic disease, a 2-year local control of 89 %, and a significant positive correlation with BED 10 . When looking at NSCLC histology alone, the 1-and 2-year pooled rates of local control were 95 %. Severe toxicity was generally low, with a risk of grade 3-4 toxicity of 6 %, and predominantly related to pneumonitis. Pooled grade 5 toxicity risk was 4 %, with most events (58 %) due to hemoptysis.
These results support the observations from a previous systematic review that consisted of only 250 patients [11]. The median 1-year and 2-year local control rates were slightly higher at 96 % and 92 %, respectively. Toxicity rates were higher compared to the current study, with a median treatment-related mortality risk of 5 % and grade ≥3 toxicity risk of 10 %. Similarly, Rim et al. included 291 patients in their meta-analysis and determined an excellent 2-year local control rate of Abbreviations: BEDbiologically effective dose; IDLisodose line; EoDevery other day; nrnot reported; CK -Cyberknife. *Value represents the entire patient cohort, including ultra-central and centrally located lesions. ǂ Median follow up "not reached".  97 %, albeit with a much higher grade ≥3 toxicity rate of 23 % [35]. In their updated meta-analysis, they observed pooled grade ≥3 toxicity rates of 9 % and grade 5 rates of 6 %, more in line with the current study [36]. The lower grade ≥3 toxicity risk that we observe may be attributed to the inclusion of more contemporary studies within our meta-analysis, in which tumor targeting and radiotherapy dose conformality are improved, and more cautious dose schedules are utilized.
The pooled local control rates from our study are excellent, and comparable to those observed from prospective trials of primary lung SBRT. TROG 09.02 was a phase 3 randomized trial of 101 patients comparing 48 Gy/4 fractions with conventional radiotherapy (50 Gy or 66 Gy), of which the former resulted in better local control, with a 2-year rate of 89 % [37]. It should be noted that targets on this trial were noncentral, defined as >2 cm from the bifurcation of the lobar bronchi or 1 cm from mediastinum. Similarly, RTOG 0813 was a phase II single arm prospective trial of 120 patients with central or ultra-central lung cancers [9]. The 2-year local control rate for evaluable patients was also 89.4 %. Consistent with previous series, we observed a dose-response with increasing BED 10 significantly correlated with higher local control probability. In a Japanese multi-institutional analysis, the risk of local recurrence was only 8.4 % with a BED 10 of ≥100 Gy, whereas it was 42.9% for those less than 100 Gy [38]. Our meta-regression suggests ~90 % local control probability at a BED 10 of 100 Gy for tumors of any histology ( Figure A3A). Despite possible differences in traditional radiosensitivity, the local control rates are similar between our mixed histology and NSCLC subset analyses, consistent with the results of a previous multi-institutional investigation [39]. This suggests that SBRT in general is an effective modality in controlling traditionally radioresistant tumors, which is also reflected in other literature series of SBRT for renal cell carcinoma and sarcoma [40,41].
Since the results of the seminal Indiana University trial reporting a 2year freedom from severe toxicity rate of only 54 % in central lung tumors, there has been significant uncertainty in the use of SBRT in central lung cancers [8]. Further work characterized higher toxicity risks for tumors located closer to the PBT and introduced the concept of "ultracentral" lung cancers [42,43]. However, with gentler dose fractionations, while still delivering doses with a BED ≥ 100 Gy 10 , several series have suggested the safety and efficacy of SBRT this setting with gradual increasing uptake in its utilization [44]. Recently, the results of the phase-II HILUS trial of 56 Gy/8 fractions in the treatment of ultra-central lung tumors has again called into question the safety of SBRT in this setting, with a reported grade 5 toxicity rate of 15 % [12]. However, several details of the trial warrant mention. Firstly, the prescription isodose line as 67 %, suggesting that the Dmax in the center of the PTV reached on average as high as 150 % (84 Gy). Secondly, the PTV margin was generous, and could be as large as 15 mm beyond the GTV [13]. A combination of these factors, in addition to the minimum coverage requirement of 80 % in any overlapping region of PTV and ipsilateral PBT resulted in very high doses to this critical OAR, with a median Dmax of 97 Gy and D0.5 cc of 53 Gy in 8 fractions.
The results of our meta-analyses suggest an overall low risk of grade ≥ 3 toxicity, of which the pooled fatal toxicity risk was 4 %. No significant association was determined between PBT Dmax and toxicity risk from meta-regression, however, this is likely due to the small number of studies included that provided pertinent dosimetric details. Mihai [30]. In all cases of fatal hemoptysis in the HILUS trial, the EQD2 3 Dmax to the main bronchus was 100 Gy (BED 3 = 167 Gy) [12]. Tekatli et. al determined a range of 70-90 Gy in 12 fractions (BED 3 205-315 Gy or EQD2 3 123-189 Gy) for all patients experiencing fatal hemoptysis [28]. Taken together, these results suggest an important role of PBT dose and the risk of developing fatal toxicities, with a threshold that may be lower than the 180 Gy BED 3 (EQD2 3 = 108 Gy) reported in a previous meta-analysis [11]. Nevertheless, a dosimetric study determined a threshold of D0.03 cc < 50 Gy in 5 fractions (BED 3 = 217 Gy, EQD2 3 = 130.2 Gy) as being the optimal dosimetric endpoint for predicting grade 2 + non-pneumonitis toxicity, suggesting a potentially higher threshold [45]. The in-progress SUNSET trial allows for a Dmax of 64 Gy in 8 fractions to the PBT (BED 3 = 235 Gy, EQD2 3 = 141 Gy) [46]. Taken together, these results suggest that increasing dose to the PBT is associated with airway toxicity risk. Care should be taken during the planning process to avoid significant "hot spots" in this critical OAR, likely < 133-150 Gy BED 3 (EQD2 3 80-90 Gy) as a conservative constraint. Appropriate prescription dose schedules should be selected to safely achieve these aims, and tolerance further decreased if other concomitant risk factors exist as described below.
We did note several other potential risk factors for fatal toxicity. Endobronchial tumor was noted in 8 cases of fatal toxicity, largely from hemoptysis or fistula development [12,28,29,33]. Antiplatelet and/or anticoagulant use was noted in 17 cases of grade 5 toxicity hemoptysis [12,27,28]. These are well established risk factors previously observed, with intuitive biological reasoning and observed even in the setting of conventional radiotherapy [47,48]. Use of targeted therapies have been associated with increased toxicity risk in the setting of SBRT. In particular, vascular endothelial growth factor (VEGF) inhibitor therapy has been established as a risk factor for significant SBRT-related toxicities, with a proposed mechanism of decreasing vascular density and inhibiting the repair of mucosal damage [49]. Significant rates of severe bowel toxicity have been reported with VEGF administration within 13 months of SBRT to abdominopelvic targets, a common site of treatment for colorectal cancers in which VEGF inhibitors are a cornerstone of therapy [50]. Specific to ultra-central lung tumors, a retrospective analysis of 88 patients determined that receipt of a VEGF inhibitor within 30 days of SBRT was associated with a significantly higher risk of fatal hemoptysis (HR 16.9, p < 0.001) than those who did not receive an agent [51]. Use of VEGF inhibitor was also observed with fatal hemoptysis events in several other series included within the current analysis [23,27,28]. The concomitant use of other targeted agents and immunotherapy is less well characterized, however recent reports suggest certain classes are more toxic than others when combined with radiotherapy, such as BRAF and MEK inhibitors or CTLA-4 antibodies   [52]. The management of concomitant systemic therapies and SBRT can be guided by the recent EORTC-ESTRO consensus recommendations, which were developed through a Delphi consensus process following a systematic review of the topic, and provide specific guidance based on the class of systemic therapy [53]. In our analysis, one case of grade 5 pneumonitis and grade 5 hemoptysis occurred in a patient who received nivolumab and another who received everolimus post SBRT, respectively [12,24]. Lastly, we observed 5 cases of grade 5 toxicity in patients with interstitial lung disease (ILD), of which 2 were pneumonitis and 3 were hemoptysis [24,28,30]. In the two cases of pneumonitis, the bilateral lung V20s were about 10 % [24,30], suggesting that extra precaution should be taken to minimize lung dose or avoid SBRT in patients with ILD. Previous reports have suggested a treatment related mortality rate of 15.6 % in ILD patients, with the risk being as high as 33 % in the subset of patients having idiopathic pulmonary fibrosis (IPF) [54]. The phase II ASPIRE-ILD trial aims to better characterize the safety and efficacy of SBRT in ILD patients, in which a strict constraint of V20 < 10 % must be adhered to amongst other conformality criteria [44].
Limitations of the current analysis warrant mention. Firstly, there are various potential sources of bias that may influence our effect estimates and results. Although most studies included in this systematic review and meta-analysis scored highly on the mNOS, only 1 of the 27 included studies were prospective, and none were randomized. Retrospective studies are inherently prone to selection bias and variability in endpoint definition. Endpoints are known prior to study initiation, thereby allowing for misattribution of toxicities to treatment or non-treatment related factors, or under-reporting of events such as recurrence. Follow up was also limited, with most studies (60 %) having a median follow up <2 years. Long term toxicities may be under-represented, particularly in patients with good long-term prognoses. Furthermore, significant publication bias was detected in our quantitative analyses. However, our review process was extensive with careful review of many studies in the initial phases of screening. We further mitigated this by restricting our inclusion criteria to studies reporting on 5 or more observations, so that highly selected case reports or series, typically reporting extreme outcomes, were excluded. Significant heterogeneity was observed in our pooled endpoint estimates. This is reflective of the heterogeneous nature of the source evidence in terms of disease histology, dose fractionations, differential radiation prescription techniques, and other clinical factors influencing the risk of toxicity and oncologic outcomes. Not all studies reported each endpoint of interest which may introduce another source of bias. We attempted to mitigate this by maximizing endpoint acquisition from each study, including through the recapitulation of digitally reconstructed Kaplan Meier curves. Lastly, there is evidence suggesting that not all structures of the PBT may have the same sensitivity to radiotherapy. In the HILUS study, the incidence of grade 5 hemoptysis was only 4 % in tumors >1 cm from the main bronchi/trachea versus 18 % in tumors that were ≤1 cm [12]. Due to the paucity of reporting PBT substructure dosimetry in other included studies, we were unable to further define constraints beyond those reported in HILUS. Based on Fig. 3 however, we show that doses ≥90EQD2 3 to any part of the PBT poses significant toxicity risk, supporting a constraint lower than this threshold. There appears to be a larger proportion of toxicity events within proximity to larger airway structures such as the trachea and main bronchi.
We anticipate an increasing clinical demand for ultra-central lung SBRT as the evidence supporting its use in early lung cancers and oligometastatic disease amounts. As such, we provide recommendations for the use of SBRT in ultra central lung tumors in Table 3. Generally, the goal of therapy is to maximize target dose and coverage, but without compromising safety and abiding by strict dose constraints for OARs. Patient selection is also critical, in that underlying diseases and other concomitant drug therapies may influence treatment risk. The results of the ongoing SUNSET (NCT03306680) and LUNGTECH trials (NCT01795521) will provide further prospective evidence to guide the use of SBRT in this patient population [46,55]. The introduction of novel radiotherapy technologies such as the MR-linac may allow for safer SBRT delivery through fraction-by-fraction adaptive replanning in scenarios where OAR dosimetry may be affected by anatomic variations. This is also under study specific for ultra-central tumors (LUNG STAAR, NCT04917224) [56]. Particle therapy may be another modality capable of minimizing dose to normal tissues, however clinical evidence is needed [57]. Nevertheless, the results of the current study are convincing that SBRT for ultra-central lung tumors is both safe and effective in carefully selected patient populations.

Conclusion
In conclusion, SBRT as a treatment for ultra-central lung tumors generally yields high rates of local control. Toxicity rates may be acceptable, with the most common severe toxicities being radiation pneumonitis and hemoptysis. Nevertheless, there is a real risk of severe toxicity, and therefore SBRT must be performed with extreme caution. Treating clinicians must be cognizant of patient, tumoral, dosimetric, Abbreviations: NSnot stated, APantiplatelet, ACanticoagulation, VEGFvascular endothelial growth factor, BEDbiologically effective dose, RPradiation pneumonitis, ILDinterstitial lung disease, IPFidiopathic pulmonary fibrosis, GTVgross tumor volume, PTVplanning target volume, V20 -volume (%) receiving 20 Gy, PBTproximal bronchial tree, NSCLCnon small cell lung cancer, D#cc -highest dose (Gy) covering a certain cubic centimeter of volume.
and medication factors that may increase the risk of fatal toxicities.
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.  [53].

Endobronchial tumors
Given the significant risk of hemoptysis, these patients should be treated with extreme caution. All efforts must be taken to mitigate other risk factors for inducing hemoptysis. Non-SBRT regimens using lower dose per fraction (i.e., 60 Gy/15 fractions) should be considered and a more conservative constraint to the PBT such as a Dmax < 100 Gy BED3 or EQD2 3 < 60 Gy (10 % grade 5 bleeding risk as per modelling from HILUS trial [12] from all patients, including those without endobronchial disease) should be adopted. Interstitial Lung Disease Given the high risk of pneumonitis, limiting normal lung volume is critical. A V20 < 10 % is a strict cutoff. All efforts should be made to treat on trial.