Treatment planning and outcomes effects of reducing the preferred mean esophagus dose for conventionally fractionated non‐small cell lung cancer radiotherapy

Abstract Based on an analysis of published literature, our department recently lowered the preferred mean esophagus dose (MED) constraint for conventionally fractionated (2 Gy/fraction in approximately 30 fractions) treatment of locally advanced non‐small cell lung cancer (LA‐NSCLC) with the goal of reducing the incidence of symptomatic acute esophagitis (AE). The goal of the change was to encourage treatment planners to achieve a MED close to 21 Gy while still permitting MED to go up to the previous guideline of 34 Gy in difficult cases. We compared all our suitable LA‐NSCLC patients treated with plans from one year before through one year after the constraint change. The primary endpoint for this study was achievability of the new constraint by the planners; the secondary endpoint was reduction in symptomatic AE. Planners were able to achieve the new constraint in statistically significantly more cases during the year following its explicit implementation than in the year before (P = 0.0025). Furthermore, 38% of patients treated after the new constraint developed symptomatic AE during their treatment as opposed to 48% of the patients treated before. This is a clinically desirable endpoint although the observed difference was not statistically significant. A subsequent power calculation suggests that this is due to the relatively small number of patients in the study.

To limit normal tissue complications, our institution sets organ at risk (OAR) treatment planning dose constraints for each type of treatment in a departmental spreadsheet and updates them in response to new published data or internal analyses of our own results. The constraints fall into two categories: 'limits' which can only be violated after consultation with the prescribing physician and possible peer review by another attending physician, and 'guidelines' which, while desirable, can be violated in the interest of target coverage or sparing of other higher priority OARs. Guideline violation is always discussed with the prescribing physician. In all cases, planners aim to both meet diseasespecific target coverage metrics and to minimize normal tissue doses.
For LA-NSCLC patients, a physician contours the esophagus from the cricoid cartilage to the gastro-esophageal junction on the planning CT scan. For several years, the esophageal planning guidelines for these patients have been mean esophageal dose (MED) ≤ 34 Gy and V60 Gy ≤ 17%; the latter is conditional on not compromising target coverage without explicit physician approval. However, there are several published dose-volume models of symptomatic AE and an analysis based on four such models was performed at our institution as part of a long-range effort to integrate such a model directly into planning. 3 The model predictions were compared with the outcomes of Stage IIIA-IIIB LA-NSCLC patients treated at our institution with CCRT and IMRT between 2004 and 2014. 3 The best model was found to be that of Huang et al. 4,5 The model predicted that AE would be notably reduced if MED could be kept in the low 20 Gy range. In September 2018 (before the publication of Reference 3), we changed our clinical guidelines to state that MED ≤ 21 Gy was preferred. Planners were instructed to aim for a MED below 21 Gy. If that prevented acceptable target coverage, they were asked to aim for a MED between 21 Gy and 34 Gy. The V60 Gy guideline remained in place and no 'limits' were added. The guidelines and general plan requirements are the same regardless of whether the patient receives CCRT or not.
Each patient has a weekly on-treatment visit (OTV) with an attending radiation oncologist. Complications noted at the OTV are recorded in the electronic medical record (EMR) per the Common Terminology Criteria for Adverse Events (CTCAE) v4.0.
When the new esophagus planning guideline was instituted, we had two questions: 1. Was the lower MED sufficiently easy to achieve that simply sensitizing planners to the desirability of a lower MED would lead to a significant reduction in the planned MED?
2. If the answer to the first question was 'yes', would an accompanying decrease in incidence of symptomatic AE be observed: specifically would the incidence of Grade 2 or higher AE (AE2+) reported in OTVs during the course of treatment be reduced? 2 | ME TH ODS All patients were planned in Eclipse (v.15; Varian Medical Systems, Palo Alto, CA) for 6 MV photon treatment on a variety of Varian linacs; doses were calculated with the Analytical Anisotropic Algorithm (AAA) algorithm. At simulation, the patient was immobilized in an alpha-cradle (Smithers Medical Products, Inc.) mounted on a device that was indexed to the couch. Isocenter and associated skin marks (tattoos) were placed at simulation, where both a free-breathing scan and a 4DCT were acquired. The physician delineated the GTV based on the free-breathing scan and a fused FDG-PET scan, determined the internal target volume (ITV) using the 4DCT, formed the clinical target volume (CTV) by adding a 5-7 mm 3-dimensional margin to the ITV and formed the planning target volume (PTV) by a further 5 mm margin around the CTV. The esophagus, heart, and other OARs that might receive significant dose (e.g. brachial plexus, liver) were delineated by a physician while the planner delineated lungs and the spinal canal. Patients were set up by tattoos followed by daily orthogonal images matched to the spine on an orthogonal pair of digitally reconstructed radiographs (DRRs); a cone-beam CT was acquired weekly, primarily to follow tumor changes.
We reviewed all consecutive patients treated definitively for LA-NSCLC with a prescription of 2 Gy/fraction for one year before and one year after the new guideline was put in place. No other planning, technique, delivery, or imaging changes were deliberately introduced during that period. For each patient, we recorded the MED and V60 Gy from the treatment plan and whether the patient had CCRT. The clinical charts were retrospectively reviewed to determine the maximum grade of esophagitis noted in at least one weekly OTV. Esophagitis after the treatment course was finished was not recorded because of the sporadic nature of the post-treatment records in the EMR. The analyses comparing clinical and dosimetric factors before and after the guideline change was done with Fisher's test and significance was determined by P < 0.05.
As part of our analysis, we examined agreement of outcome with the model on which the guideline change was based. 3,4 This predicts the percent of patients with CTCAE4 v4.0 Grade 2 or higher AE (AE2+) to be given by: where X = 1.5 CCRT + 0.07 MED −3.1.
The variable CCRT is 1 for patients with concurrent chemotherapy and 0 for those without; MED is the mean esophagus dose in Gy. The model was applied to Groups A and B with no re-fitting performed. Model discrimination was assessed through the area under the receiver-operating characteristics curve (AUC) and calibration between observed and predicted AE2 was judged visually.
This study was completed under an institutional review board approved protocol.

| RESULTS
Sixty-four patients meeting criteria were treated the year before (Group A) the guideline change and coincidentally 64 were treated the year after (Group B). The results for the two groups that is described below are also summarized in Table 1. Hereafter, we focus on the relationship between the incidence of AE2+ and MED, which was the subject of the stated constraint change.
In Group A, the MED was less than 21 Gy for 21 patients (33%) of whom 13 had CCRT (28 % of Group A CCRT patients). In Group T A B L E 1 Groups A (within 1 year before guideline change) and B (within 1 year after) characteristics.  The gross differences between Groups A and B before and after the change in planning guideline are shown graphically in Fig. 4.

Subcategories
Details of the features that were explored and their statistical significance are summarized in Table 2.

| DISCUSSION
Radiation therapy treatments must strike a balance between effective tumor control and avoidance of serious treatment complications.
To do this, treatment planners must avoid 'cold spots' in the target

combination of Groups A and B and in the combination of Group A
and B patients with CCRT, the number of AE2+ cases with MED below 21 Gy was statistically significantly lower than for those with MED above. The incidence of AE2+ in Group B was 38% while in the earlier Group A it was 48%; for patients with CCRT, the AE2+ rates were 41% in Group B versus 57% in the earlier Group A.
Though the changes in this relatively small study are not statistically significant, we are encouraged by both the improvement in outcome and the agreement with the model (Fig. 3).
The small number of patients is a limitation of this study. Therefore, we performed a power calculation to determine the number of patients in the two groups that would be needed to detect, with P ≤ 0.05, the observed change in AE2+ due to the changed MED

| CONCLUSION
To reduce the risk of esophageal toxicity in NSCLC patients receiving conventionally fractionated radiation therapy (2 Gy/fraction for approximately 30 fractions) planners were requested to try to achieve a MED below 21 Gy if possible while still respecting target coverage and other OAR constraints. The requested MED was based on a formal outcomes study of a large group of NSCLC patients (3,4). If an otherwise acceptable treatment plan could not be achieved with such a low MED, planners were instructed to aim for a MED between 21 Gy and the prior planning goal of 34 Gy. Reducing the MED was not a mandatory 'hard constraint', but rather intended to be a 'gentle reminder'. The plans for patients one year before (Group A) and one year after (Group B) this change were analyzed. A statistically significant MED reduction was accomplished.
Although the reported incidence of AE2+ was lower in Group B, this was not statistically significant. Nevertheless, the clinically observed overall incidence of AE2+ was significantly lower in patients with mean doses below 21 Gy and there was good agreement between the observed rates of esophagitis and those predicted by the model both before and after the guideline was implemented. Implementing 'soft' planning guidelines to better control normal tissue doses is feasible and can improve clinical outcomes.

ACKNOWLEDG EMENTS
All the authors acknowledge support from NIH/NCI Cancer Center Support Grant P30 CA008748.

AUTHORS' CONTRIBUTIONS
E. Yorke designed the study, reviewed the patients for relevance, collected the patient-specific data, did some of the data analysis and did the initial drafting of the manuscript. M. Thor