Radiation-induced lung injury (RILI) is one of the most common clinically challenging toxicities induced after breast radiation therapy (RT). RILI encompasses two phases: an acute phase termed Radiation Pneumonitis, occurring within the first 6 months after RT, consisting in a lung inflammation phase responsible for dyspnea, cough, sometimes pleural effusion and a late phase called radiation lung fibrosis, leading to irreversible tissue damage (1, 2). In both acute and late phases and for all grades, RILI translates in radiologic alterations.

The guidelines parameters for measuring and reporting radiation toxicity and prevent the risk of RILI include the volume of lungs receiving a certain dose of radiation (Vx Gy) and the mean radiation dose (Dmean) to the lungs (3). RILI is one of the main toxicity risk factors that influence the technical RT planning of breast cancer treatment. RILI incidence can vary from 1-3% in retrospective studies to 40% in prospective studies depending on the definition used (clinical and/or radiological). Many monocentric studies determined clinical and dosimetric risk factors associated with RILI and recommended dosimetric thresholds. Clinical risk factors reported include age >50 years, history of smoking, history of pulmonary disease, chronic infection/immune disorder, adjuvant chemotherapy, tamoxifen concomitant with RT and structural deformity of the thoracic wall (4–7). Regarding the dosimetric thresholds, there are no unanimously recognized recommendations for the ipsilateral lung in the context of breast cancer treatment (8–13). The meta-analysis of Gokula et al. (14), recommended thresholds of V20 Gy not exceeding 30% and Dmean not exceeding 15 Gy. Other studies found that locoregional lymph nodes irradiation is a strong clinical risk factor since it results in higher doses delivered to the lung adjacent to the lymph node areas (8).

The CANTO RT cohort provides a unique opportunity to assess long-term outcomes and toxicities of breast cancer RT over time (15). CANTO-RT is a subset of CANTO (NCT01993498) a prospective longitudinal multicentric cohort that enables clinical and dosimetric risk factor assessment for long term RT toxicities (16, 17).

The primary aim of our study was to assess the incidence of RILI after RT treatment in CANTO-RT and to identify clinical and dosimetric risk factors associated with RILI occurrence. The secondary aim was to compare former dosimetric thresholds proposed in the literature, and to select the most appropriate in our multicentric dataset to apply for clinical practice.

This work is an etiological cross-sectional study based on the prospective CANTO-RT cohort, a subset of the CANTO cohort (NCT01993498).

CANTO Inclusion and exclusion criteria were previously published (18). Briefly, patients were women aged 18 years and over, treated for invasive cT0-3 cN0-3 M0 breast cancer, and completed 5-year follow-up for side effects, based on questionnaires and medical visits. All patients included in our study where part the CANTO-RT, which colligated full DICOM data on RT modalities and dosimetric analysis as previously described (15). Data from the patients recruited in the three larger CANTO-RT centers were included in our study.

RILI identification was initially based on Case Report Form (CRF) analysis, to identify patients that presented dyspnea, cough, or both dyspnea and cough at either M0 (3 to 6 months after completion of RT), M12 (12 months after the completion of RT), M36 or M60. Events were graded according to the Radiation Therapy Oncology Group (RTOG) classification: Grade I: asymptomatic or mild symptoms, Grade II: moderate symptoms, Grade III: severe symptoms requiring oxygen therapy, Grade IV: symptoms requiring assisted ventilation, and Grade V: death. The medical files of symptomatic patients were then retrospectively analyzed, and patients were considered having RILI when a CT scan or a chest X-Ray identified compatible pulmonary lesions during the follow-up. RILI was considered acute when the symptoms were declared at M0 and chronic at later time points. An event was defined as RILI when both radiological and clinical symptoms were recorded in a medical consultation report; however, when mentioned only in the medical imaging report, often performed for other causes years after the end of treatment, RILI was not considered as diagnosed by the physician. RILI events were then retrospectively graded according to the RTOG criteria. Absence of pre-existing fibrosis, which could mislead to RILI diagnosis, was confirmed by reviewing the dosimetric CT scans ( Figure 1A ).

RT treatments were performed with LINAC accelerators, using photons with or without electrons beams. RT was mainly performed with 3D conformational RT, however intensity-modulated radiation therapy (IMRT) was allowed. Treatment plans were performed with various Treatment Planning Systems that could use type A (e.g. Clarkson) or type B (e.g. Collapsed Cone Convolution) algorithms. For patients without nodal irradiation, whole breast irradiation was delivered in dorsal or lateral decubitus (19). Dose constraints on the ipsilateral lung and field arrangement were at the discretion of the treating centers ( Supplementary Table S1 ). DIBH technique was allowed to minimize the dose to the heart and lungs. Fractionation scheme were normofractionnated (NF RT) with 50 Gy in 25 fractions over 5 weeks or moderately hypofractionated (HF RT) with 40 Gy in 15 fractions over 3 weeks.

Statistical analysis was performed using R software, version 4.0.4, with the “pROC” package for ROC curve analysis and the “arsenal” package for descriptive table analyses.

RILI+ (patients with RILI) and RILI- (patients without RILI) groups were compared by the Wilcoxon test for continuous characteristics and a Chi2 or Fisher test was used to compare proportions of categorical variables. Correlations between dosimetric parameters were tested using Spearman’s correlation analysis. Statistical analysis did not include missing data. Dosimetric threshold analysis was performed by building contingency tables and calculating the positive and negative predictive values and multivariable analysis were performed using a logistic regression with backwards elimination selection. Models were compared using the Akaike information criterion (AIC) and the Bayesian information criterion (BIC). Some publications outlined that dosimetric threshold definition using ROC curves and univariable models may be less accurate than multivariable model-based threshold assessment and proposed iterative multivariable-model-based threshold definition (20). Such an analysis was thus also performed.

All patients provided a written informed consent to participate in the CANTO cohort. The study protocol was approved by a central ethical committee for human subjects (n° ID RCB: 2011-A01095-36).

Our analysis on 1565 patients within the multicentric prospective longitudinal CANTO-RT cohort estimated a RILI incidence of 2.4% after early breast cancer treatment, with only half of the events prospectively identified by the physicians. The respiratory adverse events were mainly grade I-II (95%) with no grade V, confirming treatment safety regarding RILI. The main clinical predictor was respiratory disease history, especially COPD or medical history of acute pneumonia before treatment. All dosimectric parameters analyzed were higher in RILI+ patients and highly correlated to each other. ROC curves analysis showed a strong prediction for all Vx in ipsilateral lung with V30 Gy and V35 Gy being the most robust factors. Those factors and respiratory disease history remained significantly associated with RILI occurrence in multivariable analysis adjusted for chemotherapy and nodal RT. We found V30 Gy > 15% was strongly associated with RILI occurrence in CANTO RT dataset.

We noted close predictive values for all the dosimetric parameters analysed. This is in accordance with the formerly published papers showing different dosimetric parameters as the most predictive of RILI occurrence. However, our results substantiate data from the monocentric study conducted in an Asian population by Lee et al. (13). These results are important since the Asian population can have different pulmonary responses to exogenous agents compared to the Caucasian population (21). Concerning the V30 Gy parameter, V30 Gy > 10%, published by Lee et al. (13) is close to our threshold of V30 Gy >15%. In the study of Lee et al. (13), RT was performed with IMRT in 44% of cases, mainly with a moderately hypofractionnated regimen with doses reported in EQD2. In our series, IMRT use was limited and hypofractionnation could not be analysed due to missing data.

Other studies recommended V20 Gy as the best predictor for RILI occurrence. Few studies compared V20 Gy and V30 Gy concluding in favor of V20 Gy (9, 22), some analyzed V20 Gy and V30 Gy independently (8). Other studies recommended the V20 Gy parameter in comparison to V10 Gy (4) or V5 Gy (23), but not to V30 Gy. In a study performing prospective pulmonary function test and CT scans, V20 Gy was more predictive than V30 Gy for RILI occurrence in a multivariable analysis with an AUC of 0.69 for V20 Gy, when it was 0.67 for V30 Gy and 0.68 for V30 Gy in our study (22). One study dichotomized dosimetric results between associated and non-associated nodal RT; V20 Gy not exceeding 38% was optimal with nodal RT and V20 Gy not exceeding 20% without nodal RT. This study analyzed also V30 Gy in the subgroup with nodal RT only and showed a slightly higher AUC and a threshold of V30 Gy not exceeding 25.7% compared to V20 Gy (11).

Our backwards multivariable analysis showed that regional nodes irradiation, including IMC, tumor size, nodal status, chemotherapy use and the type of surgery association to RILI were eventually all related to the dose delivered to the lung, as among them the dose is the only significant factor in the multivariable model. Clinical parameters such as age and smoking status (22, 23) were not predictive of RILI appearance in our dataset. In addition, tamoxifen use after RT was not predictive of RILI appearance, showing the safety of its use after RT completion. No concomitant tamoxifen use was noted in the cohort. We did not find any association between the type of chemotherapy regimen used and RILI, in accordance with former studies that have not find any difference in respiratory function decline after breast chemotherapy followed by RT, regardless of chemotherapy drug types (24). Cardiovascular history was less frequent in RILI+ than RILI- patients, which might be explained by the use of more stringent dose constraints to the thorax in such patients, specific volume delineation or specific technique (such as lateral decubitus).

We believe the methodology used for event detection was reliable due to the extensive and accurate two-step cross analysis based on the CRF and the individual medical record verification of the symptomatic events. All the events had imaging proof of new lung injury after RT treatment, limiting the risk of false positive. The incidence of RILI obtained by our two-step analysis is compatible with those found in retrospective studies (25–28).

Four patients presented RILI with surprisingly low doses to the ipsilateral lung. However, RILI occurrence was proven for at least one of them. Indeed, this patient underwent multiple chest CT scans, demonstrating the absence of radiographic event shortly after treatment completion. The patient had a medical history of previous chemotherapy and mediastinal RT for a Hodgkin lymphoma. It is therefore plausible that, for this patient, the low doses of RT received by the lung during the breast cancer treatment, plus the previous doses received during the treatment of Hodgkin lymphoma, resulted in a cumulative dose high enough to induce RILI. We cannot exclude that other patients with low doses to the ipsilateral lung presented unknown risk factors for RILI, suggesting that dosimetric parameters might incompletely predict the risk of RILI.

The main pitfalls of our study are the retrospective analysis of medical records to identify RILI events, although the use of the prospective CANTO CRF allowed us to identify twice as many events as we would have found with medical file analysis alone. Indeed, only half of the events were diagnosed by the physician and therefore reported as such in the clinical file. RILI assessment with systematic imaging usually results in higher proportion of events, the majority of which, however, are asymptomatic with up to 58% grade I and 16% grade II (14, 25, 29).

As RILI detection was not initially planned in CANTO, no systematic imaging was done. We added a retrospective data collection from patient clinical files and imaging from three of the top recruiting centers of CANTO-RT. We thus acknowledge several possible biases introduced by the definition of RILI events that we had to set.

First, our definition of RILI was based on cough and dyspnea presence on the CRF. Those symptoms are pathognomonic of RILI but not diagnosis-defining. We could have missed RILI events with a clinical presentation of fatigue, low-grade fever, chest pain without cough and dyspnea (2). Then, in some patients with underlying pulmonary disease, a modest increase in symptom severity might be present without it being classified as new cough or dyspnea. Hence, it can be suspected that a minority of patients presenting RILI (most likely not clinically meaningful) were not recognized by this method. To better capture these events, monitoring of patient reported outcome measures (PROMs) could have been used.

There was no systematic chest CT among the study population, which might introduce a confounding bias. Patients with underlying pulmonary condition could have been more prone to undergo a CT scan during the follow-up and more RILI could have been detected in those patients. The reason for radiological imaging in those patients was not always recorded but when mentioned in the medical files, it was usually the occurrence of cough, dyspnea or chest pain, or the suspicion of a metastatic evolution.

As only 20% of patients with reported respiratory symptoms on the CRF performed a pulmonary imaging, we may have underestimated the true incidence of RILI grade I events in our population. We also acknowledge that our study did not perform respiratory function tests to identify infra-radiological events, a method that could possibly results in the detection of additional RILI cases (30, 31). In addition, our study design does not enable to analyze RILI occurring more than 5 years after treatment.

Data on fractionation schedule were only available in 54% of the study population, and there was an incomplete record in some centers, resulting in an uneven distribution of the missing data among the three centers. Consequently, we could not analyse the dosimetric data taking into account the fractionation scheme. This was mostly due to the lack of contouring guidelines in CANTO, resulting in the absence of contouring of the target volume in some centers at the time of CANTO beginning. Our study remained possible since lung contours were lacking in only 12 patients out of 1577 in the three centers studied (as shown in Figure 1A ).

To try to assess the impact of fractionation information absence, taking into account the 3 treating centers local guidelines for the indication of hypofractionated radiation therapy, we approximated the fractionation schedule received by the patients based on age and nodal status. Patients were considered having an hypofractionated regimen (40 Gy in 15 fractions) when under 50 years old and having no nodal involvement leading to a total number of 825 patients treated with normofractionated radiation therapy and 733 treated with hypofractionated radiation therapy. Assuming an alpha/beta around 3 we considered the EQD2 (dose equivalent to a 2 Gy by fraction scheme) for lung depicted in Supplementary Table S11 (32). In this analysis, V30 stayed a better dosimetric predictor of RILI than V5 Gy or V20 Gy on the ROC curve analysis ( Supplementary Table S11 ) and it was still associated with RILI occurrence in multivariable analysis ( Supplementary Table S12 ).

RILI events were rare in respect to the general population (38/1565), which limited statistical analysis. Most events were low grade but subgroup analysis of grade III-IV RILI (n=2) or RILI requiring treatment (n=8) was not performed due to small sample size.

We could not directly compare 3D RT to IMRT or VMAT since treatment was performed by conformational 3D RT for 96% of patients. In rotational techniques, the larger lung volumes receiving low-doses may imply different dosimetric thresholds for RILI occurrence (33, 34). Type A or type B calculation algorithms could be used for treatment planning. Type B algorithms are known for higher dose heterogeneity and more precise dose assessment, with statistically larger high-dose volumes (35). In our studies, we cannot evaluate the impact of the Treatment Planning Systems or algorithm types used on the dosimetric results as they were not recorded.

In conclusion, this study is to our knowledge one of the largest multicentric study to analyze the dosimetric parameters associated with RILI occurrence after early breast cancer treatment including 3D RT. Mild respiratory symptoms were frequently reported by CANTO patients. Lung medical imaging was not systematic. Radiologically confirmed RILI were observed in 2.4% of patients (n=38). To further improve lung protection during 3D conformational breast RT, we recommend V30 Gy not exceeding 15% threshold, especially in patients presenting a medical history of respiratory disease such as COPD or acute pneumonia before treatment.

Further studies are needed to provide multicentric data with IMRT including tomotherapy and VMAT and hypofractionated regimens to assess dosimetric risk factors for RILI with modern breast RT.

The data analyzed in this study is subject to the following licenses/restrictions: CANTO-RT dataset published in Sarrade and al, Cancers 2023. Requests to access these datasets should be directed to sofia.rivera@gustaveroussy.fr.

AG: Data collection, statistical analysis, manuscript redaction TS: CANTO-RT database completion. YK, BD, FD, GA, AM, SE, NM, CB, AB, TL, DP, SeR, AM, FP, FA, ED: CANTO data collection and manuscript editing. BD: Advise regarding clinical RILI description. RA: Statistical analysis and manuscript redaction supervision. SR: Study design and manuscript redaction supervision. All authors contributed to the article and approved the submitted version.