Exceeding Radiation Dose to Volume Parameters for the Proximal Airways with Stereotactic Body Radiation Therapy Is More Likely for Ultracentral Lung Tumors and Associated with Worse Outcome

Simple Summary The optimal way to treat central (CLT) and ultracentral (UCLT) lung tumors with curative radiation is unclear. We evaluated 83 patients with CLT and UCLT who underwent a curative radiotherapy technique called stereotactic body radiation therapy (SBRT). On statistical analysis, patients with UCLT had worse overall survival. Using a cohort of patients matched for relevant variables such as gender and performance status, we evaluated radiation doses to critical central structures such as the airway and heart. In this group, patients with UCLT were more likely to exceed dose constraints as compared CLT, particularly constraints regarding the airway. Additionally, patients had worse non-cancer associated survival when radiation doses were higher than 18 Gy to 4cc’s of either the trachea or proximal bronchial tree. Based on these findings, patients with UCLT have worse outcomes which could be secondary to higher radiation doses to the trachea and proximal bronchial tree. Abstract The preferred radiotherapeutic approach for central (CLT) and ultracentral (UCLT) lung tumors is unclear. We assessed the toxicity and outcomes of patients with CLT and UCLT who underwent definitive five-fraction stereotactic body radiation therapy (SBRT). We reviewed the charts of patients with either CLT or UCLT managed with SBRT from June 2010–April 2019. CLT were defined as gross tumor volume (GTV) within 2 cm of either the proximal bronchial tree, trachea, mediastinum, aorta, or spinal cord. UCLT were defined as GTV abutting any of these structures. Propensity score matching was performed for gender, performance status, and history of prior lung cancer. Within this cohort of 83 patients, 43 (51.8%) patients had UCLT. The median patient age was 73.1 years with a median follow up of 29.9 months. The two most common dose fractionation schemes were 5000 cGy (44.6%) and 5500 cGy (42.2%) in five fractions. Multivariate analysis revealed UCLT to be associated with worse overall survival (OS) (HR = 1.9, p = 0.02) but not time to progression (TTP). Using propensity score match pairing, UCLT correlated with reduced non-cancer associated survival (p = 0.049) and OS (p = 0.03), but not TTP. Within the matched cohort, dosimetric study found exceeding a D4cc of 18 Gy to either the proximal bronchus (HR = 3.9, p = 0.007) or trachea (HR = 4.0, p = 0.02) was correlated with worse non-cancer associated survival. In patients undergoing five fraction SBRT, UCLT location was associated with worse non-cancer associated survival and OS, which could be secondary to excessive D4cc dose to the proximal airways.


Introduction
Lung cancer remains the deadliest malignancy in the world, with an estimated 2.2 million new cases and 1.8 million deaths in 2020 alone [1]. Of new diagnoses, approximately 23% will present with early stage non-small cell lung cancer (NSCLC) [2]. While surgery remains the standard of care for these patients, stereotactic body radiation therapy (SBRT) is a viable therapeutic option for non-operative candidates or those who wish to avoid surgery [3].
Ultracentral tumors (UCLT) represent an extreme presentation of this clinical scenario, where the gross tumor volume (GTV) abuts or planning target volume (PTV) overlaps these central structures and are thought to be even higher risk for treatment [13]. While several groups have found UCLT to be associated with worse outcomes, others have reported no difference in survival when compared to CLT [7,11,[14][15][16][17][18][19][20][21][22][23][24]. Most data for this clinical scenario are derived from retrospective studies with a limited number of patients, where the definition of CLT and UCLT can vary [13]. Complicating matters, there is no consensus regarding treatment regimen and planning for CLT and UCLT [13,16]. RTOG 0813 utilized a five fraction SBRT regimen for CLT, whereas other groups have attempted hypofractionated regimens of up to 15 fractions [10,19,[23][24][25][26][27]. Taken together, interpretation of the current literature is challenging, particularly regarding the preferred dose-fractionation scheme and relevant dose constraints for this clinical scenario. The SBRT for Ultra-central NSCLC-a Safety and Efficacy Trial (SUNSET) is an ongoing, prospective, multi-institution phase 1 dose escalation study which may provide future guidance for UCLT [10].
In this current study, we evaluated the outcomes of CLT and UCLT using a fivefraction SBRT regimen, correlating survival-based endpoints with dosimetric parameters from both the RTOG 0813 trial and the active protocol SUNSET.

Patients Population
Between February 2007 to April 2019, the charts of 563 patients who underwent definitive SBRT for thoracic tumors at our institution were reviewed. After excluding those treated for non-NSCLC tumors, 438 patients remained and were characterized in a previous investigation [28]. The current analysis was limited to those who underwent five fraction SBRT for a total of 83 patients. Data were collected under approval from the institutional review board at Roswell Park Comprehensive Cancer Center (EDR-171710).

Patient Evaluation and Follow-Up
Within the cohort, all patients underwent positron emission tomography with diagnostic computed tomography (PET/CT) imaging, had pathologic confirmation of NSCLC, and nodal evaluation was at the discretion of the treating physician. Patients with tumors ≤5 cm and who were node-negative via PET/CT with or without nodal evaluation were eligible for SBRT. Staging was completed via the American Joint Commission on Cancer 8th edition and while no patients had T3 disease, two were characterized as T4 due to invasion into the mediastinum [29]. All patients were evaluated by both a thoracic surgeon and radiation oncologist with multidisciplinary committee input as needed. Typical indications for SBRT included poor surgical candidacy due to medical comorbidities, patient unwilling to have surgical resection, or avoidance of a higher risk surgical procedure such as pneumonectomy. Follow-up was performed as previously described [4,28]. Briefly, patients underwent a CT scan of the chest approximately 3 months after treatment, then repeated every 3-6 months after up to a year, then yearly. Areas of suspicion are further characterized by PET/CT and biopsied if necessary.

Clinical Data
Relevant clinicopathologic data were collected for each patient via chart review. Staging was completed via the American Joint Commission on Cancer 8th edition [29]. History of diabetes was defined as those receiving active treatment for diabetes mellitus and included insulin-dependent and non-insulin dependent patients. Criterion for history of heart disease and prior lung cancer were previously defined [28]. Relapse was defined as a progression on imaging or pathologic confirmation of recurrence. Local failure was defined as progression within the PTV, regional failure was defined as progression within thoracic lymph nodes or lung parenchyma excluding the treated lesion, and distant failure was defined as progression outside the thorax. All recorded instances of local failure were confirmed by biopsy. Acute toxicity was determined via chart review of follow up visits within 6 months of treatment using the Common Terminology Criteria for Adverse Events (CTCAE) Version 5.0 to grade adverse events.

Central and Ultracentral Lung Tumors
To classify CLT and UCLT, imaging for all patients included in the study was reviewed. CLT were defined as 2 cm within the trachea or proximal bronchial tree, mediastinum, great vessels, or spinal cord [7,13]. UCLT were defined as GTV directly abutting any of the above structures [7,13].

SBRT
Patients underwent CT simulation in the supine position with arms above their head using a thoracic Medical Intelligence BodyFIX®immobilization system (Elekta, Stockholm, Sweden). Tumor motion management included either abdominal compression or respiratory gating, as previously described [4,30]. Dose delivery techniques employed included non-coplanar 3-dimensional conformal fields (3DCRT) or volumetric modulated arc therapy (VMAT). Treatments were planned using Eclipse (Varian Medical Systems, Palo Alto, CA, USA). Heterogeneity correction was employed for all patients. Tumors were prescribed such that 95% of the PTV volume was covered by 100% of the prescribed dose [25]. Treatment was typically delivered twice weekly with a minimal interval of a day between treatments.

Dosimetric Analysis
Organs at risk (OARs) were contoured at the time of treatment per RTOG 0813 [25]. The aorta was separated into ascending and descending aorta using the inferior aspect of the branching of the right pulmonary artery as the inferior border between the two structures. Radiation plans were reviewed in Eclipse (Varian Medical Systems, Palo Alto, CA, USA) and evaluated using the dose constraints per RTOG 0813 and SUNSET [10,25].

Statistics
Freedom-from progression (FFP) was defined as the date of treatment to the date of documented recurrence. Patients who died without a history of relapse were censored. Non-cancer associated survival was defined as the date of treatment to date of death without a history of relapse. Patients who died with a history of relapse were censored. Overall survival (OS) was defined as the date of treatment to date of death due to any cause. For all relevant endpoints, patients who were lost to follow-up prior to an event were censored. To compare differences between groups, Pearson χ 2 was used for ordinal and categorical variables whereas Wilcoxon test was used for continuous variables. To assess for associations between variables and relevant outcomes, univariate Cox regression was employed. Variables with p-values <0.1 on univariate analysis were incorporated into a multivariate Cox regression model. Dose per fraction was treated as an ordinal variable of increasing intervals of 0.5 Gy per fraction starting at 10 Gy per fraction and ending at 12 Gy per fraction. Propensity score matching for variables significant for either non-cancer associated survival or OS on univariate study (gender, Karnofsky Performance Status (KPS)) (80-100, <80), and history of prior lung cancer was performed using package MatchIt version 3.0.2 using a 1:1 ratio and nearest neighbor method, caliper length of 0.1. Kaplan-Meier survival estimation with log-rank testing was used to assess the relationship between tumor location and survival within the matched and unmatched cohorts [31]. To evaluate the impact of competing risks for FFP and non-cancer associated survival, competing risk regression and cumulative incidence analysis were performed. All p-values were two-sided. Variables with p < 0.05 were considered significant. Statistical analyses were performed using IBM SPSS Version 26 and R version 4.0.2.
Cumulative incidence analysis revealed similar findings for FFP and a trend for worse non-cancer associated survival (p = 0.06) (Table S1).

Dosimetric Differences between Central and Ultracentral Lung Tumors
To explore whether the decrease in non-cancer associated survival and OS of UCLT could be explained by dosimetric parameters, we analyzed differences in maximum and volume doses to relevant OARs as defined by RTOG 0813 and SUNSET within the matched cohort (Table 3). UCLT had significantly higher point and volume doses to the trachea and esophagus (Table 3). Maximum dose percentage and D0.33cc to the proximal bronchus was significantly higher with UCLT, however only a statistical trend was observed regarding D4cc and maximum point doses (Table 3). While significant differences in maximum doses were observed for the ascending aorta, doses to the descending aorta were not statistically different between CLT and UCLT (Table 3).

Dosimetric Differences between Central and Ultracentral Lung Tumors
To explore whether the decrease in non-cancer associated survival and OS of UCLT could be explained by dosimetric parameters, we analyzed differences in maximum and volume doses to relevant OARs as defined by RTOG 0813 and SUNSET within the matched cohort (Table 3). UCLT had significantly higher point and volume doses to the trachea and esophagus (Table 3). Maximum dose percentage and D0.33cc to the proximal bronchus was significantly higher with UCLT, however only a statistical trend was observed regarding D4cc and maximum point doses (Table 3). While significant differences in maximum doses were observed for the ascending aorta, doses to the descending aorta were not statistically different between CLT and UCLT (Table 3).
When examining by dose constraint, UCLT only had more esophagus maximum dose (4000 cGy) failures per SUNSET (Table 4). In contrast, UCLT had significantly more failures of the trachea and bronchus dose to volume constraints, as well as bronchus and ascending aorta maximum dose percentage by RTOG 0813 (Table 4).

Discussion
This retrospective cohort comparing CLT and UCLT treated with 5000 cGy or 5500 cGy in 5 fractions, on multivariate analysis, found that: (1) the D4cc constraints for both the proximal bronchus and trachea were significantly associated with non-cancer associated survival and (2) UCLT was associated with worse OS. In a propensity score matched cohort, UCLT had worse non-cancer associated survival and OS. Within that matched cohort, UCLT had higher doses to the proximal airway, esophagus, and ascending aorta which corresponded to a significant increase in an inability to meet RTOG 0813 dose constraints.
The D4cc constraint for proximal bronchus and trachea was more likely to be exceeded in UCLT versus CLT (p < 0.05). Except tumor laterality, there were no significant differences between CLT and UCLT across numerous other clinicopathologic variables.
The definition of CLT and UCLT can be heterogenous in the literature [13]. While the "no-fly zone" definition or 2 cm within only the proximal bronchial tree is commonly used such as in RTOG 0813, whether this margin refers to gross tumor or PTV can differ [13]. Others have included the great vessels, mediastinum, and esophagus to CLT and UCLT definitions as well [7,10].
In the current study, CLT were defined at 2 cm within the proximal airway, mediastinum, great vessels, or spinal cord while UCLT directly abutted these structures. Gross tumor volume was used to characterize location as it is less sensitive to differences in tumor motion control or radiation technique. While this approach is similar to RTOG 0813, we acknowledge the importance of proximity of other central structures such as the esophagus. SUNSET expanded this definition to include the esophagus and pulmonary arteries, yet was based on PTV. As such, our definition represents a hybrid of the two studies. Despite the differences in definitions for tumor location, we feel the recommended dose constraints from RTOG 0813 and SUNSET are still applicable in this setting, however acknowledge that this may impact comparing our findings to those in RTOG 0813 and SUNSET.
Factors including tumor volume, history of diabetes, and prior history of lung cancer impacted outcome, which is consistent with previous reports [17,23,28,33]. UCLT location was associated with worse non-cancer association and OS within a matched pair cohort; however, only OS on multivariate Cox regression. The different observations between the two models may be driven by increased stringency of the multivariate model and the lower number of non-cancer death events.
Interestingly, there was a female majority within the cohort which is uncommon for lung cancer patients. Similar findings were observed in the entire dataset, which contained peripheral tumors as well, suggesting this is not an artifact of focusing on CLT and UCLT. The reason for this observation is not clear [28]. Additionally, the use of VMAT was associated with worse FFP. Consistent with RTOG 0813, our institution historically utilized 3DCRT for this clinical scenario [25]. Particularly for patients treated earlier in the study period, VMAT was used if 3DCRT could not meet dose constraints. This selection bias may account for an unfavorable outcome observed with VMAT.
Although UCLT are associated with worse outcomes, the cause is not well understood. The use of concurrent anti-angiogenic agents has been linked to an increased risk of severe toxicity, but was not frequent within our primary NSCLC patient cohort [36,37].
Unless clinically suspected or discovered during diagnostic work-up, endobronchial invasion was not explicitly investigated for all patients in this cohort. Patients with endobronchial invasion have an increased risk of death from either SBRT or hypofractionated regimens [21,36,38], or even when using a conventionally fractionated radiation [39,40]. As such, modification of dose-fractionation schemes alone may be insufficient to account for this high-risk factor.
A biologically effective dose (BED) of ≥180 Gy 3 to the proximal bronchial tree has been correlated with treatment related mortality [14,37]. A BED of 180 Gy 3 corresponds to approximate 45 Gy in 5 fractions when using an α/β ratio of 3, which is less than the lowest prescription dose of this cohort [13]. Similarly, a recent prospective trial evaluating 56 Gy in 8 fractions for UCLT revealed an increased risk of fatal pulmonary hemorrhage with D0.2cc exceeding an EQD2 of 80 Gy to the proximal airway [24]. Furthermore, while Manyam et al. found doses exceeding 47.5 Gy to D0.33cc of the proximal bronchial tree to be associated with increased non-pneumonitis toxicity, it was not correlated with non-cancer associated or OS in the current study [32]. While the maximum dose percentage to the proximal bronchus was exceeded in 13.8% and 48.3% of CLT and UCLT respectively, there was no correlation between failure of this constraint and any endpoint. On the other hand, D4cc >18 Gy to either the trachea or bronchus correlated with non-cancer associated survival in a multivariate model and may represent a more relevant constraint than point doses. To the best of our knowledge, we are the first to report this significant relationship between D4cc proximal airway dose and non-cancer associated survival. While the several dose to volume parameters for the proximal airway were evaluated prospectively for 56 Gy in 8 fractions, D4cc was not included [24]. Interestingly, exceeding the D5cc to the esophagus constraint was no longer significantly correlated with non-cancer associated survival on multivariate study, likely due to the covariance between esophageal and tracheal dose.
Consistent with our findings, prior studies found that failure to meet the RTOG 0813 dose constraints can be common when treating UCLT [15]. Despite this fact, the rate of acute grade 3 toxicity was low within this cohort (7.5% for CLT, 9.3% for UCLT) and similar to previous reports [7,11,14,18,35]. When evaluating a dose of 56 Gy in 8 fractions for UCLT, the HILUS study observed grade 3+ adverse events in 33.8% of patients, 15.4% of which were grade 5 [24]. While no grade 4+ toxicity was observed, the cause of death of patients who die outside of our institution cannot be determined given the retrospective nature of this investigation. As exceeding the D4cc trachea or bronchial dose was associated with increased non-cancer associated death, it is possible these patients suffered life-threatening or fatal toxicities which could not be characterized due to this limitation.
The mechanism in which exceeding D4cc constraints for either the trachea or proximal bronchi increases the risk of death is not clear. Presumably, increased dose to these structures could result in pulmonary hemorrhage secondary to damage to the pulmonary vasculature. However, it is worth noting that with palliative regimens such a 20 Gy in 5 fractions these constraints are likely routinely exceeded and felt to be well tolerated. It is possible that those treated with palliative intent are at risk of treatment related death, yet this is not captured to the limited life span inherent to that patient population, considering that after only a year UCLT location began to stratify OS. Alternatively, the respective dose to volume parameters may be surrogates for other structures (such as pulmonary vessels) yet to be characterized.
Although there are numerous studies evaluating different dose fractionation schemes for UCLT, given the heterogeneity in inclusion criteria, differences in prescription plans, and small population size of retrospective studies, it is difficult to compare these options [13,20]. SUNSET, a phase 1, multi-institutional dose escalation study, will evaluate the maximally tolerated SBRT dose associated with ≤30% rate of grade ≥ 3 toxicity at 2 years [10]. SUNSET will begin with 60 Gy in 8 fractions and escalate to 5 fractions or de-escalate to 60 Gy in 15 fractions if the initial treatment is too toxic.
Several limitations apply to our study. The relatively small retrospective patient cohort reduces statistical power and generalizability, however unfortunately this is common in reports dealing with UCLT [12]. Furthermore, occult nodal spread to levels 5 and 6 may be common in UCLT and difficult to assess via imaging and endoscopic evaluation. Nevertheless, relapse risk was similar between both groups. Additionally, most patients were treated with a 3DCRT technique and may no longer be representative of contemporary patients treated with VMAT [23]. Lastly, due to the small sample size, not all significant variables could be included in propensity score matching, yet all relevant variables remained well balanced between the two cohorts.

Conclusions
Compared to CLT, UCLT were found to have worse OS on multivariate Cox regression and reduced non-cancer associated survival, and OS in a propensity score matched pair cohort. Within the matched cohort, UCLT demonstrated more frequent inability to meet dose constraints to structures such as the proximal airway, esophagus, and ascending aorta. Exceeding the D5cc esophageal constraint was associated with worse outcomes on univariate, but not multivariate study. In contrast, failure of the D4cc trachea or bronchus constraints were associated with reduced non-cancer associated survival within a matched cohort. Lastly, failure of the tracheal D4cc constraint was correlated with reduced noncancer association and OS within the whole, unmatched patient cohort. Based on these findings, we recommend inclusion of the D4cc dose constraints for the proximal airway, and are considering prospective investigation on whether dose to these structures can be safely limited at the expense of the PTV.