Significant Correlation Between Overall Survival and Mean Lung Dose in Lung Stereotactic Body Radiation Therapy (SBRT)

Background After stereotactic body radiation therapy (SBRT) for medically inoperable stage I non-small-cell lung cancer (NSCLC), more patients die of comorbidities, particularly severe pulmonary insufficiency, than of tumor progression. The aim of this study was to evaluate correlation between lung biologically effective dose (BED) with an α/β ratio of 3 Gy (BED3) and overall survival (OS) for these patients. Methods From 2012 to 2017, we have developed a prospectively updated institutional database for all first 100 consecutively treated patients with inoperable Stage 1 (T1T2N0M0) NSCLC. All SBRT were conducted on a Novalis Tx® LINAC with two coplanar dynamic conformal arcs (84%) or with coplanar volumetric modulated arc therapy (VMAT) (16%). Mean GTV and PTV were 8.6 cc and 50.8 cc, respectively. The marginal dose prescribed to the PTV was the 80% isodose line (IDL), i.e., 54 Gy in 3 fractions for 76 patients (BED10 = 126 Gy) and 50 Gy in 5 fractions for 24 patients (BED10 = 83.3 Gy). Pulmonary heterogeneity has been taken into account by using Monte Carlo or AAA algorithms. Median follow-up was 25 months. Results At 1, 2, 3 and 5 years, local control (LC) was respectively 100, 98.2, 98.2, and 77.7%, and OS was respectively 83, 71.2, 58.1, and 33.2% (median OS was 49 months). Significant OS prognostic factors in univariate and multivariate analysis were mean lung BED3 (HR = 1.14, p = 0.01) and PTV volume (HR = 1.01, p = 0.004). A mean lung BED3 ≤ 5 Gy was significantly associated with a doubling of median OS from 29 months to more than 60 months (not achieved, p = 0.0068). For patients with a forced expiratory volume in 1 second (FEV1) ≤ 40%, a mean lung BED3 ≤ 4 Gy was significantly associated with a doubling of median OS from 23 to 46 months (p = 0.019). Conclusion Mean lung BED3 is strongly and significantly associated with OS in SBRT for inoperable Stage I NSCLC. For all treated patients, a mean lung BED3 ≤ 5 Gy lead to a doubling of median OS. This threshold value should be reduced to 4 Gy for patients with FEV1 ≤ 40%.

Pre-treatment pulmonary function tests have not been shown to be predictive for RP. Therefore, patients with NSCLC with a poor baseline pulmonary function are not excluded from treatment with SBRT (18). Mean lung dose (MLD) seems to be a strong and reproducible dosimetric parameter of RP, with a significant cut-off at 4-4.7 Gy in three fractions, and is often correlated to the volume of PTV (15,17,(19)(20)(21). Many factors may have confounded the reported MLD because of inadequate heterogeneity correction algorithms, various dose prescriptions and fractionations and probably a lack of lung volume definitions (whole or ipsilateral lung volume minus GTV or PTV) (17,22,23). In this way, the biologically effective dose (BED) determined with adequate heterogeneity correction algorithms may be used for estimating toxicity probabilities. For high fraction doses, the linear-quadratic model with an α/β ratio of 3 Gy is the best method for converting the physical lung dose to predict RP (24).
Our study aimed to evaluate the correlation between lung BED 3 and OS in a prospectively updated institutional cohort of 100 consecutively treated patients with stage I NSCLC.
The secondary objective was to study the impact of lung dosimetric parameters in a population with a poor baseline pulmonary function.

Patients' Selection and Characteristics
From October 2012 to August 2017, we have developed a prospectively updated institutional database for all patients consecutively treated with SBRT for inoperable Stage 1 (T1T2N0M0) NSCLC in our institution. This database has been approved by our local ethics committee and a regional ethics committee (CECIC Rhône-Alpes-Auvergne, Grenoble, IRB 5921), and developed according to the French law regulating clinical research (Loi Huriet). Thus, our study is an observational cohort study from all first 100 patients recorded in this prospectively updated institutional database. All treated patients fulfilled inclusion criteria of those first described by Timmerman et al. in the analysis of RTOG 0236 (25): performance status (PS) ≤ 2, age ≥ 18 years, stages T1T2N0M0, peripherally located NSCLC at least 2 cm from the proximal bronchial tree, and medical inoperability (baseline forced expiratory volume in one second (FEV1) ≤ 40%, predictive postoperative FEV1 ≤ 30%, diffusing capacity for carbon monoxide (DLCO) ≤ 40%, severe cerebral, pulmonary or cardiovascular disease or patient refusal).
All patients were required to have a complete imaging screening performed less than 1 month prior to lung SBRT: a high-resolution contrast-enhanced lung computed tomography (CT), a 18F-fluorodeoxyglucose positron emission tomography (18F-FDG PET) and a cerebral magnetic resonance imaging (MRI) to exclude regional or distant metastases. A flexible bronchoscopy was needed to exclude an endobronchial location or infectious disease such as lung tuberculosis. A cytologically or histologically proven NSCLC was strongly recommended but not mandatory in case of contraindications. If this proof was not obtained, a tumor growth observed with an interval at least 3 months between two CT or a maximum standardized uptake value (SUV max ) in 18F-FDG PET above 2.5-3 was necessary to include patients. Exclusion criteria were small-cell lung cancer, mediastinal location and no meeting of normal tissue dose constraints. Pre-treatment pulmonary function tests were performed for all patients.

SBRT Specifications
The gross tumor volume (GTV) was contoured on 2.5-mmthick lung CT windows. Intrafraction tumor motion due to breathing was limited by an abdominal compression and taken into account by creating an internal target volume (ITV) obtained with a four-dimensional (4D) CT scan at the time of CT simulation. An additional margin of 8 mm for adenocarcinoma and 6 mm for other histologic types was added for microscopic tumor extension to create the clinical target volume (CTV) (26,27). Finally, the planning target volume (PTV) was obtained with a uniform 3 mm CTV expansion, according to our defined geometric stereotactic conditions.
All SBRT treatment characteristics are reported in Table 1. The isocenter prescription dose was 67.5 Gy in 3 fractions and reduced to 62.5 Gy in 5 fractions for central lung tumors to meet normal tissue dose contraints. The marginal isodose line prescribed to the edge of PTV was the 80% isodose: respectively, 54 Gy in 3 fractions and 50 Gy in 5 fractions (BED 10 = 126 Gy and 83.3 Gy). Each fraction was separated by at least 40 h (25).
All treatments used 6-MV photons. Dose distributions were performed with two coplanar dynamic conformal arcs (DCA) in 84% of cases and with volumetric modulated arc therapy (VMAT) in 16%. Treatment planning systems (TPS) were Iplan R v4.1 (Brainlab, Feldkirchen, Germany) for DCA plans and Eclipse R v13.5 (Varian Medical Systems, Palo Alto, CA, United States) for VMAT plans. Pulmonary heterogeneity has been taken into account by using the Monte Carlo algorithm for DCA plans and the AAA algorithm for VMAT plans. Target coverage was adequate when at least 95% of the PTV was covered by 80% of the prescribed isodose. Treatment quality was verified by calculating the conformality index (CI) (a ratio of 80% prescription isodose volume to the PTV), a ratio of 50% prescription isodose volume to the PTV (R50) and the maximum dose 2 cm from the PTV   in any direction (D2cm), as seen in Table 1 (28). Lung SBRT was performed using a Novalis Tx R (Varian Medical Systems, Palo Alto, CA, United States) linear accelator (LINAC) with an integrated Exactrac X-ray 6D system R (Brainlab AG, Feldkirchen, Germany). This system enabled a pretreatment positioning which was then adjusted daily with a Cone Beam CT (CBCT).

Follow-Up
Follow-up included prospective clinical examination and CT scans every 3 months during the first two years post-SBRT and every 6 months afterward. Follow-up PET scans were required only in cases of progressive soft tissue abnormalities observed on CT. Efficacy was assessed with the Response Evaluation Criteria in Solid Tumors (RECIST) (29). A complete response was defined as the disappearance of the target lesion, a partial response as a decrease of at least 30% of the tumor's longest diameter and a progressive disease as an increase of at least 20% of the longest diameter. LC was defined as the absence of local failure. Local failure was characterized as the combination of a RECIST progressive disease and evidence of tumor viability as shown by biopsy or SUV max in 18F-FDG PET above the pre-treatment SUV max or above a value of 5 (30). Progression-free survival (PFS) was defined as the period of time from the end of SBRT to the date of local-regional failure, disseminated (visceral or lymphnode) recurrence or the patient's death. OS was defined as the time between the end of SBRT and the patient's death. Toxicity was evaluated with the National Cancer Institute's Common Toxicity Criteria for Adverse Events version 4.0 (NCI-CTCAE).
Mean and median follow-up were respectively 27 and 25 months (range: 0.6-64). Only one patient was lost to follow-up after 10 months ( Table 1).

Statistical Analysis
No patient included in the prospectively updated institutional database of all first 100 patients consecutively treated with SBRT in our institution was excluded from the study or from the statistical analysis (Figure 1). LC, PFS, and OS were calculated using the Kaplan-Meier method. The Cox proportional hazards model was performed to identify predictive factors of LC and prognostic factors of PFS and OS. A twosided p-value < 0.05 was considered significant. The following factors were included in the univariate analysis for LC: histological type, pre-treatment SUV max , conformality index, R50, GTV, PTV, an 80% prescription isodose volume (%), and maximum and minimum PTV BED 10 (D max , D 98 % , D 2 % , D min ). Concerning PFS and OS, the following factors were included in the univariate analysis in addition to the previously studied LC predictive factors: gender, age, performance status, history of lung surgery, pre-treatment FEV1, pre-treatment DLCO, baseline pulmonary function, and mean lung BED 3 (whole lungs, ipsilateral lung, whole lungs minus PTV, ipsilateral lung minus PTV), The Benjamini-Hochberg method was used to adjust p-values to limit false positives, considering the large number of tests. The Spearman correlation enabled the identification of strongly correlated factors between them that were not included in the multivariate analysis (Spearman's rank correlation coefficient of 0.75). Factors associated with a p-value < 0.25 in the univariate analysis were included in the multivariate analysis if they were also selected by the LASSO method. The Wald test and the Likelihood ratio test were performed to calculate and verify the p-value for each coefficient in multivariate analyses.
Linear correlation between OS and significant prognostic factors was then verified with a Pearson or Spearman correlation coefficient, depending on the cases, especially for patients with a poor pre-treatment pulmonary function (i.e., FEV1 or DLCO ≤ 40%). For each significant linear correlation observed, a ROC (Receiver Operating Characteristics) curve was performed to identify the best threshold. Finally, comparisons of OS curves with obtained thresholds were conducted using the logrank test.

DISCUSSION
To date, the present study is the first to demonstrate a significant correlation between OS and MLD after Stage I NSCLC SBRT. Moreover, this finding leads us to propose reducing published lung constraints with thresholds of 5 and 4 Gy in mean lung BED 3 respectively for patients with FEV1 > 40% and FEV1 ≤ 40%.
Our lung SBRT characteristics are in agreement with international guidelines about Stage I NSCLC, especially with Timmerman et al. in the analysis of RTOG 0236 (25). Marginal prescribed dose to the edge of PTV was 54 Gy in three fractions if it met dose constraints. Concerning target volume delineation, we no longer create a CTV since it is not recommended (31). When we started to treat patients in 2012, guidelines were less clear and CTV creation was debated (26). In our study, LC and OS were respectively over 95% and about 60% at the three-year point, which is consistent with previously published studies (1)(2)(3)25). No significant predictive factor of LC was found, probably because only three local failures were observed and adequate BED 10 was prescribed to the tumor.
Radiation pneumonitis is a known important dose-limiting factor in lung cancer radiation therapy. RP is categorized into two interdependent stages: acute RP and late RP (a chronic injury stage known as pulmonary fibrosis) which can theoretically be fatal, especially if pulmonary function is already impaired. The summarized sequence of classic RP is as follows: cellular injury leads to cytokine release, cytokine recruitment of the inflammatory infiltrate causes acute pneumonitis, and the body's attempt to repair the injury results in pulmonary fibrosis (13). It is accepted that recognition of sporadic RP can be particularly difficult for clinicians because it is rare (≈ 10%) and patients often present with severe dyspnea and/or "out-of-field" radiographic findings that may raise the possibility of other disease process (13). So we can think that RP are probably underestimated in published studies, and especially their exact relationship with the death of patients. This might be one of the explanations of why MLD was a significant prognostic factor of OS in our study. Limitations of the study were in link with the difficulty to recognize RP particularly in patients with a poor baseline pulmonary function and frequent flares of acute pulmonary insufficiency. It was very difficult to know from what patients were dying and the cause of their pulmonary insufficiency: natural and classic outcome of their comorbidities or especially related to RP? Other limitations were that study was mono institutional and not multicentric, only 100 patients were included which could have led to a lack of power for statistical analyses and data were retrospectively analyzed even if they were prospectively updated. Statistical analysis was robust and there was only one patient lost to follow-up after a period of 10 months after SBRT. Pulmonary heterogeneity, which may be a reason for false interpretation, has been taken into account by using the Monte Carlo algorithm for DCA plans and the AAA algorithm for VMAT plans. Various dose prescriptions and fractionations have been studied with the linear-quadratic model with an α/β ratio of 3 Gy which was the best method for converting the physical lung dose to predict RP (24).
Concerning PFS, PFS after lung SBRT was significantly correlated to tumor volume (GTV) in univariate and in multivariate analysis. We did not find mean lung BED 3 as a significant prognostic factor of PFS in univariate analysis. One of the explanations might be a lack of power of our study because the lower threshold value of the 95%CI was very close to 1 (0.988). Interestingly, there was therefore a strong trend for significance when we studied mean ipsilateral lung BED 3 (p-value of 0.063; HR = 2.035, 95%CI = 1.000-1.072).
Concerning OS, it is widely agreed that more patients die of comorbidities than of tumor progression since LC is excellent, over 95% at three years. Therefore, significant published prognostic factors to date mostly relate to comorbidities [age (4), gender (5), performance status (4, 6), platelet-to-lymphocyte ratio (8,10), and pretreatment immune parameters (10)] if the prescribed dose is sufficiently high (11). However, no relationship between toxicity and OS is demonstrated to date. Many patients treated with lung SBRT die from severe pulmonary insufficiency, but it is not easy to distinguish whether the main cause is the natural course of pulmonary or cardiovascular disease or if SBRT lung toxicity might have worsened the situation. Severe RP is probably under-evaluated in literature. It is the main toxicity factor after lung SBRT; furthermore, MLD seems to be the strongest and most reproducible dosimetric parameter of RP (15,17,(19)(20)(21). In addition, tumor volume is a reproducible and frequently reported significant prognostic factor of OS (5-7). We can also surmise that tumor volume is linked to MLD and that patients could die from a higher received MLD than GTV because LC is high. For this reason, we have made the choice to study the correlation between OS and MLD, which we have demonstrated to be significant in multivariate analysis for all patients (HR = 1.14, CI95% = 1.03-1.25, p = 0.01) and in subgroup multivariate analysis, whether for patients with FEV 1 ≤ 40% (HR = 1.55, 95%CI = 1.12-2.14, p = 0.008) or with FEV 1 > 40% (HR = 1.14, 95%CI = 1.02-1.28, p = 0.021) or with GTV < mean GTV, i.e., 8.6 cc (HR = 1.34, 95%CI = 1.06-1.69, p = 0.015). GTV and FEV 1 cannot be modified while the prescribed dose can be adjusted.
Moreover, we showed in our study a significant correlation between OS and MLD with a significant threshold value of 5 Gy for BED 3 , i.e., 3.6 Gy in three fractions. A significant cut-off at 4-4.7 Gy in three fractions is reported in literature concerning the probability of RP after lung SBRT (17,19). So our threshold value of 3.6 Gy in three fractions is not very different from that published for symptomatic RP, which enables us to assume that the threshold value of 5 Gy for OS may be related to RP. Thresholds values may have to be reduced to 5 Gy (i.e., 3.6 Gy in 3 fractions or 4.3 Gy in 5 fractions) and 4 Gy (i.e., 3 Gy in 3 fractions or 3.5 Gy in 5 fractions) in mean lung BED 3 respectively for patients with FEV1 > 40% and FEV1 ≤ 40%, as our study suggests, to have an impact on OS. These two thresholds values are totally in agreement because we may have to be more careful to treat frailty patients and probably we may have to reduce prescribed dose for these patients. We therefore recommend using effective algorithms that take into account pulmonary heterogeneity and limiting PTV irradiated volume to a minimum by reducing margins (not creating a CTV or using gating or tracking techniques). This is particularly crucial for frailty patients with poor baseline pulmonary function (FEV1 ≤ 40%).

CONCLUSION
In summary, our study demonstrates a significant and strong correlation between OS and mean lung BED 3 , confirmed in univariate and multivariate analysis in all patients, in subgroup analysis and in survival curves analysis. Higher mean lung BED 3 is always strongly and significantly associated with a poorer OS. Moreover, significant mean lung BED 3 threshold values have here been shown to correlate with OS: 5 Gy for the entire population (i.e., 3.6 Gy in 3 fractions or 4.3 Gy in 5 fractions) and 4 Gy for patients with FEV1 ≤ 40% (i.e., 3 Gy in 3 fractions or 3.5. Gy in 5 fractions). Stay below these threshold values significantly enabled a doubling of median OS.

DATA AVAILABILITY STATEMENT
The datasets generated for this study are available on request to the corresponding author.

AUTHOR CONTRIBUTIONS
GD and IM analyzed the data. GD wrote and revised the manuscript. JB, VC, VD, ML, and AB-C also revised the manuscript. All authors have actively participated in the data acquisition, commented, and approved the final version of the manuscript.