Study of The Predictors for Radiation Pneumonitis in Patient With Non-Small Cell Lung Cancer Received Radiotherapy After Pneumonectomy

Background: To identify the valuable predictors of grade ≥ 2 radiation pneumonitis (RP) in patient treated with radiotherapy after pneumonectomy for non-small cell lung cancer (NSCLC); and to construct a nomogram predicting the incidence of grade ≥ 2 RP in such patients. Methods: We reviewed 74 patients with NSCLC received radiotherapy after pneumonectomy from 2008 to 2018. The endpoint was grade ≥ 2 RP. Univariate and multivariate regression analysis were conducted to evaluate signicant factors of grade ≥ 2 RP. Receiver operating characteristic (ROC) curve was used to establish optimal cutoff values and the nomogram was built to make the predictive model visualized. Descriptive analysis was performed on 5 patients with grade 3 RP. Results: A total of 25(33.8%) patients developed grade ≥ 2 RP and 5(6.8%) patients were grade 3 RP. V5, V10, V20, V30, MLD, PTV, and PTV/TLV were associated with the occurrence of grade ≥ 2 RP in univariate analysis, while none of the clinical factors was signicant; V5(OR,1.195;95%CI,1.082-1.319; P<0.001) and V20(OR,1.412; 95%CI,1.161-1.717; P=0.001) were the independent signicant predictors by multivariate analysis and were included in the nomogram. The ROC analysis for the cutoff values for predicting grade ≥ 2 RP were V5>22%(AUC=0.816, sensitivity:0.712, specicity:0.837) and V20>8%(AUC=0.822, sensitivity:0.681, specicity:0.828). Additionally, grade ≥ 3 RP did not occur when V5<30%, V20<13% and MLD<751.2cGy, respectively. Conclusions: Our study showed that V5 and V20 were independent predictors for grade ≥ 2 RP in NSCLC patients receiving radiotherapy after pneumonectomy. of the lung received more xGy, respectively; MLD, PTV, TLV.

Conclusions: Our study showed that V5 and V20 were independent predictors for grade≥2 RP in NSCLC patients receiving radiotherapy after pneumonectomy. Grade 3 RP did not occur whenV5<30%, V20<13% and MLD<751.2cGy, respectively. In addition, patient underwent right pneumonectomy may have a lower tolerance to radiation compared to left pneumonectomy. Background Radiotherapy given after surgery is an effective treatment for patients with non-small cell lung cancer (NSCLC) to improve local control and survival of the disease. It is not only applied for patients with positive surgical margins or N2 nodal involvement, but also for recurrence and second tumor in NSCLC after surgical resection [1][2][3][4]. However, patient receiving radiotherapy following previous surgery has a higher risk of developing radiation pneumonitis (RP) [5,6]. RP as one of the most common dose-limited toxicities following thoracic radiotherapy can severely affect life quality and uncommonly result in death [7]. Inoue et al. reported the 3-year survival rates of patients, who experienced no, mild, and severe RP, were33.4%, 38.2%, and0%, respectively(P = 0.0028) [8].
Previous studies have tried to correlate the incidence of RP following lung cancer surgery with several clinical and dosimetric variables [9,10].However, most of them were conducted on the effect of RP after pulmonary lobectomy and rarely covered pneumonectomy. Although pneumonectomy is a high-risk procedure associated with greater morbidity and mortality, it is still necessary for a tumor located near the center of the lung invading the large vascular or main bronchus when lobectomy cannot achieve safe margin [11][12][13]. Compared to receive radiotherapy after lobectomy, patients underwent pneumonectomy suffered a signi cant decrease in overall survival [14,15]. To date, still very few related investigations have been reported, so that the speci c independent factors for RP in patients with NSCLC after pneumonectomy were unclear.
In this study, we collected data of baseline patient factors, dose-related variables and rates of RP, aiming to identify the valuable predictors of grade ≥ 2 RP in NSCLC patient received radiotherapy after pneumonectomy, and to construct a nomogram predicting the incidence of grade ≥ 2 RP in such patients.

Patients
From December 2008 to September 2018, patients with NSCLC received radiotherapy after pneumonectomy in Shanghai Chest Hospital were retrospectively reviewed. Clinical, dose-volume histogram (DVH) factors, and radiographic ndings were collected from 92 patients who met the following criteria: (1) Pathologically con rmed non-small cell lung cancer. (2) Underwent pneumonectomy (left or right lung). (3) Received thoracic radiotherapy after pneumonectomy with or without chemotherapy. (4) Having available dosimetric data for radiotherapy and clinical records for at least 6 months of follow-up. The exclusion criteria included receiving stereotactic body radiotherapy (SBRT), neoadjuvant chemotherapy or radiotherapy before surgery. Finally, 74 patients were selected as the study sample.

Treatment
All patients received pneumonectomy and thoracic radiotherapy with or without chemotherapy.
Radiotherapy was delivered by 3D-conformal radiotherapy (3D-CRT) or intensity-modulated radiotherapy (IMRT) once a day through Monday to Friday. The median total radiation dose was 56Gy ranging from 50 to 64Gy, receiving 1.8 to 2.5Gy per fraction. Gross tumor volume (GTV) was de ned as the visible recurrence tumor or lymph node and positive resection margins con rmed by postoperative pathology; Clinical tumor volume (CTV) was included the GTV plus an 0.5-0.8cm margin, or the involved mediastinal lymphatic drainage regions for multiple N2 with R0 resection. Considering respiratory movement and setup uncertainty, planning target volume (PTV) was the GTV and CTV plus 0.5-1cm margins to generate PTV-G and PTV-C, respectively. The dose-volume constraints for normal tissues were set as follows: to the spinal cord <50Gy; to the heart, V40<50%, to the esophagus, V55<50%. The chemotherapy regimens were chosen by the attending medical oncologists.
Clinical and dosimetric factors de nition Clinical factors included gender, age, smoking status, performance status, surgical laterality, tumor pathology, nodal status, tumor status, TNM stage, surgical margin, radiation dose, radiation technique, chemotherapy and the interval between chemotherapy and radiotherapy. Dosimetric factors evaluated were total lung V5/10/20/30(Vx was referred to the percentage of contralateral lung volume receiving xGy), mean lung dose (MLD), planning target volume (PTV), total lung volume (TLV) (de ned as the total lung volume minus the GTV of lung tumor) and PTV/TLV (referred to the ratio of PTV to TLV).

Endpoint and Statistics
The endpoint of this study was grade≥2 RP de ned in the Common Terminology Criteria for Adverse Events Version 5.0(CTCAE) [16].A diagnosis of RP was con rmed by at least two experienced radiation oncologists on the basis of clinical symptoms with radiographic in ltration within the radiation eld during the rst 6 months after completion of radiotherapy. This information was retrospectively extracted from the medical records by carefully reviewing all available chest radiographic images and clinical records until the last follow-up or death of the patient. All patients included in this study had follow-up CTs for at least 6 months.
Data analyses were performed using SPSS (version 23.0) and R software (version3.4.3). Univariate analysis was conducted to evaluate the in uence of both clinical and dosimetric factors on grade≥2 RP risk. Due to the close correlation among dosimetric variables, we chose to generate a set multivariate model that included factors with P<0.25 on univariate analysis, for consistent comparisons of endpoints by dosimetric parameters being reviewed. Receiver operating characteristic (ROC) curve was used to establish optimal cutoffs which was identi ed its in uence of RP in multivariate models, and the nomogram was built to make the predictive model visualized. To evaluate the usability of the nomogram, ROC curve and calibration plot of predictive model were performed. Descriptive analysis was performed on 5 patients with grade 3 RP. Welch's t-test was applied to analyze the subgroup strati ed by surgical laterality in grade≥2 RP patients. All statistical tests were 2-sided and P<0.05 was considered statistical signi cance.
In the univariate analysis, dosimetric factors including V5, V10, V20, V30, MLD, PTV and PTV/TLV were associated with the incidence of grade ≥ 2 RP, while none of the clinical factors was signi cant (Tables 1  and 2). In addition, a trend was observed for nodal status (P = 0.099) and chemotherapy with docetaxel (P = 0.080). Patient gender, smoking status, nodal status, radiation dose, chemotherapy schedule and chemotherapy with docetaxel with P < 0.25 in the univariate analysis were thus included in the multivariate analysis. To avoid multicollinearity, dosimetric factors were not included in the multivariate analysis simultaneously. ROC curves were displayed for the optimal thresholds and found V5 and V20 had the better area under the ROC curve (AUC) value ( Fig. 1     Abbreviations: DVH, dose-volume histogram; IQR, interquartile range; V5/10/20/30, Vx the percentage of the lung volume that received more than xGy, respectively; MLD, mean lung dose; PTV, planning target volume; TLV, total lung volume; PTV/TLV, the ratio of PTV to TLV. P = 0.001) were statistically signi cantly associated with the risk of grade ≥ 2 RP (Table 3). The optimal cutoffs for V5 and V20 in ROC analysis revealed that V5 > 22% (sensitivity:0.712, speci city:0.837) andV20 > 8% (sensitivity:0.681, speci city:0.828) were the reasonable predictors. Then, a nomogram was built to make the predictive model of grade ≥ 2 RP visualized (Fig. 2a). It showed a satisfying AUC value was 0.862 in this patient prediction cohort (Fig. 2b). Also, the calibration curve illustrated favorable consistency between the predicted RP and the actual observation (Fig. 2c). Table 4 showed the domestic values of the 5 patients who developed grade 3 RP.  Abbreviations: V5/10/20/30, Vx the percentage of the lung volume that received more than xGy, respectively; MLD, mean lung dose; PTV, planning target volume; TLV, total lung volume; PTV/TLV, the ratio of PTV to TLV.
Since the different surgical laterality affected received radiation dose of the residual lung and heart, we carried out the subgroup analysis strati ed by surgical side (Table 5). Our results showed that while there was no difference in the incidence of RP between two groups, the mean values of DVH thresholds (V5, V20, V30, MLD) in right pneumonectomy were signi cantly lower than left pneumonectomy in grade ≥ 2 RP patients. Abbreviations: RP, radiation pneumonitis; V5/10/20/30, Vx the percentage of the lung volume that received more than xGy, respectively; MLD, mean lung dose; PTV, planning target volume; PTV/TLV, the ratio of PTV to TLV.
Data were presented as mean ± standard deviation (SD).

Discussion
RP is the most common complication in radiation therapy for lung cancer [5][6][7]. Compared to lobectomy, it could be more threatening for patient underwent pneumonectomy [14]. However, there are (to our knowledge) no published studies expressly investigating the correlation between clinical or dosimetric factors and RP in patients underwent pneumonectomy. It is clinically important to explore possible predictors to mitigate the RP incidence in this special population who had already been in high risks. Hence, we conducted a retrospective study and veri ed V5 and V20 as predictors for grade ≥ 2 RP. To our knowledge, this is the rst study to explore the predictors of RP in patients who once received pneumonectomy.
The dosimetric limitations of radiotherapy for NSCLC patients underwent pneumonectomy mainly drew on the experience of adjuvant hemithoracic radiotherapy for patients with malignant pleural mesothelioma (MPM) following extrapleural pneumonectomy (EPP) for a long time. While considering the difference in radiation volumes between the two malignancies, the experience directly exported to NSCLC patients seemed not conformable [17]. Interestingly, the conclusions drawn from our study were consistent with published literatures on mesothelioma patients. In a cohort of 63 patients treated with IMRT after EPP at the MD Anderson Cancer Center, Rice et al. found that V20 was the sole signi cant factor related to pulmonary-related deaths (PRDs). Strikingly, the results showed that patient with V20 > 7% increased a 42-fold risk of PRDs and V5 was almost more than 80% [18]. Similarly, Bece et al.
described 55 patients receiving hemithoracic radiotherapy, none of whom developed RP under restriction on lung V20 < 10% and V5 < 60% [19].In a review of the literature, Chi et al. reported that high incidence of severe RP was observed when V20 and V5 were higher than 7% and 60%, respectively [20]. Moreover, our ndings were also correlated with several studies exploring the predictors of RP in NSCLC patients underwent surgery. For instance, in a cohort of 109 patients with NSCLC received postoperative radiotherapy (PORT), Xin Tang et al. [21]reported that ipsilateral lung V5 (ilV5 > 64.9%) was signi cantly associated with severe acute radiation pneumonitis (SARP, grade ≥ 3) incidence. In another retrospective report of 199 NSCLC patients treated with PORT, including 23 patients who underwent pneumonectomy, V20 > 20% was identi ed as one of cutoff levels for predicting grade ≥ 2 RP [22]. Analogously, Gong Y and co-workers [23] found that both ipsilateral lung V5(iV5 > 71.2%) and total lung V20(V20 > 21.4%) showed a strong statistically association with acute severe RP in lung cancer patients after surgery.
In general, our ndings suggested that V5 and V20 could keep as predictors of RP for patients with NSCLC following pneumonectomy. However, the cutoff values of our study seemed lower than the above researches, and we reasoned this discrepancy on two aspects as follows. On one hand, we discreetly set lung V5 < 30% and V20 < 10% in more than 85% of radiotherapy treatment plans in consideration of the high mortality and morbidity rate of pneumonectomy. Such strict dose limits have rarely been seen in previous studies that almost excluded patients received pneumonectomy. On the other hand, prior reports mainly set grade ≥ 3 RP as the endpoint leading to a much stricter limitation. Additionally, compared to grade 2 RP, grade ≥ 3 RP with more severe clinical symptoms and poorer survival needed to be treated with steroids and oxygen. In a study of 99 patients, the grade 3 pneumonitis was in 8% patients, and all of them died within 8 months after treatment [24]. Ascribed to the smaller case number and lower incidence, we had di culty obtaining effective thresholds of grade 3 RP from ROC curve, therefore only descriptive results were showed. We found no incidence of grade ≥ 3 RP was observed whenV5 < 30%, V20 < 13% and MLD < 751.2 cGy, respectively. A recent publication, under restrictions on contralateral lung V20 < 10% in 32 patients with stage pIII-N2 NSCLC after pneumonectomy, there were only 2 patients developed RP [25]. This result was veri ed by our study with a larger population, and restrictions of V5 and V20 should be more strictly de ned especially for patients with post-pneumonectomy.
In our research, 25(33.8%) patients developed grade ≥ 2 RP and 5(6.8%) patients were grade 3 RP, consistent with previously reported RP rate ranged from 15-40% in grade 2 and 2-9% in grade 3 RP [7,26]. It seems reasonable to postulate that receiving chemotherapy could increase the risk of RP [27]. While neither chemotherapy agent nor timing was signi cantly associated with grade ≥ 2 RP, only a trend for chemotherapy with docetaxel (P = 0.080) was observed in the current study. That may be confounded by other factors related to the decision of whether using chemotherapy. Additionally, the heterogeneity of chemotherapy regimens also in uenced the effect of timing and interval between chemotherapy and radiotherapy with respect to RP. In the researches by Boonyawan et al. and Kim et al., the same results were shown that chemotherapy did not increase RP risk [22,28]. Like chemotherapy, poor pulmonary function tests (especially FEV1% predicted, DLCO) have also been classically described to increase risk of RP [29]. Because patients who undergoing pneumonectomy were routinely received pulmonary function tests before, they rarely had another test after surgery. We can almost ensure there was no signi cant heterogeneity in the basic lung function of these patients, as they can endure pneumonectomy after being evaluated by professional surgeons. In fact, the evaluation of pulmonary function was seldom considered in other published data if patients underwent surgery [5,19,20,22,23]. In a study involved 260 patients from two centers, poor pulmonary function at baseline was found to have no correlation with the risk of pneumonitis [30].
Right pneumonectomy was known to be associated with a higher mortality and morbidity compared with left pneumonectomy [31,32]. Furthermore, Martin and co-workers observed that risks of perioperative death and complications were signi cantly higher for right pneumonectomy after preoperative radiation in patients with NSCLC [33]. Our results showed that mean DVH thresholds of developing grade ≥ 2 RP in right pneumonectomy group were signi cantly lower than that in left ones, but no difference in incidence between the two groups was observed. Consistent with the series of studies, it suggested that patient underwent right pneumonectomy not only suffered a worse prognosis but also had a lower tolerance to radiation. Therefore, receiving radiotherapy following right pneumonectomy should be performed only in selected patients.
As the nomogram quanti es risk by illustrating and combing the relative importance of various factors, it has been widely used in clinical assessment [34]. Thus, we developed a nomogram predicting the incidence of grade ≥ 2 RP and performed ROC curve and calibration plot to validate its prediction e ciency. The results showed that combing the values of V5 and V20 in our nomogram could be used in practical work with less bias and better accuracy. While there may be other factors in uencing the occurrence of RP, further research should be carried to improve the nomogram.
This study had several limitations. Firstly, it was limited by its retrospective nature, with inevitably uncontrolled and unobserved selection bias in patient enrollment. Secondly, there were only 5 patients with grade 3 RP due to the low incidence, leading to di culties in making further analyses. Finally, as a single-center research, the number of cases in this study was relatively insu cient. However, to be precise, they were rare but precious. Even as one of the lung cancer centers with the largest surgery quantity in China, there were only a few dozen patients with this procedure each year, and less than 10% of them received radiotherapy afterward. Notwithstanding its limitation, this is the rst study exploring the optimal predictors for RP in patient with NSCLC underwent pneumonectomy, making a certain contribution to reduce the risk of RP and improve patients' quality of life and survival.

Conclusion
Our study showed that V5 and V20 were independent risk factors for grade ≥ 2 RP in patient with NSCLC receiving radiotherapy after pneumonectomy. Grade 3 RP did not occur whenV5 < 30%, V20 < 13% and MLD < 751.2 cGy, respectively. In addition, patient underwent right pneumonectomy may have a lower tolerance to radiation compared to left pneumonectomy. Availability of data and materials: The data sets used and/or analyzed during the current study are available from the corresponding author on reasonable request.
Ethics approval and consent to participate: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This retrospective study was approved by the Shanghai Chest Hospital Research Ethics Board. Figure 1 ROC curves of V5, V10, V20, V30, MLD, PTV, and PTV/TLV for predicting the incidence of grade≥2 radiation pneumonitis.