Exploratory study of the association of volumetric modulated arc therapy (VMAT) plan robustness with local failure in head and neck cancer

Abstract This work is to show which is more relevant to cause local failures (LFs) due to patient setup uncertainty between the planning target volume (PTV) underdosage and the potential target underdosage subject to patient setup uncertainties in head and neck (H&N) cancer treated with volumetric‐modulated arc therapy (VMAT). Thirteen LFs in 10 H&N patients treated by VMAT were analyzed. Measures have been taken to minimize the chances of insufficient target delineation for these patients and the patients were clinically determined to have LF based on the PET/CT scan results by an experienced radiologist and then reviewed by a second experienced radiation oncologist. Two methods were used to identify the possible locations of LF due to underdosage: (a) examining the standard VMAT plan, in which the underdosed volume in the nominal dose distribution (UVN) was generated by subtracting the volumes receiving the prescription doses from PTVs, and (b) plan robustness analysis, in which in addition to the nominal dose distribution, six perturbed dose distributions were created by translating the CT iso‐center in three cardinal directions by the PTV margin. The coldest dose distribution was represented by the minimum of the seven doses in each voxel. The underdosed volume in the coldest dose distribution (UVC) was generated by subtracting the volumes receiving the prescription doses in the coldest dose distribution from the volumes receiving the prescription doses in the nominal dose distribution. UVN and UVC were subsequently examined for spatial association with the locations of LF. The association was tested using the binominal distribution and the Fisher's exact test of independence. We found that of 13 LFs, 11 were associated with UVCs (P = 0.011), while three were associated with UVNs (P = 0.99). We concluded that the possible target underdosage due to patient setup uncertainties appeared to be a more relevant factor associated with LF in VMAT for H&N cancer than the compromised PTV coverage at least for the patients included in this study.


| INTRODUCTION
Volumetric-modulated arc therapy (VMAT) is a novel form of intensity-modulated radiotherapy (IMRT) that can deliver complex, 3-dimensional dose distributions by using single or multiple arcs.
Compared with conventional static-field IMRT (hereafter termed IMRT), VMAT has the advantages of improved dose distributions, faster treatment delivery, and decreased monitor unit requirements. 1 Therefore, VMAT has been rapidly adopted as the preferred treatment technique for head and neck (H&N) cancers. 2 Despite improved technologies in radiotherapy for H&N cancer, local failure (LF) remains the most important cause of patient morbidity and mortality. [3][4][5][6] LF causes significant morbidity and often leads to death. Many factors are associated with LF after radiotherapy. In addition to biological factors inherent in the disease, technical aspects of radiotherapy may play a role in LF. To minimize severe complications such as brainstem necrosis, xerostomia, and dysphagia, radiation oncologists often reduce target margins, compromise the target coverage, or both, to avoid critical structures, resulting in a region of low dose, which in turn may cause LF. 4 Highly conformal treatment techniques such as IMRT and VMAT are capable of generating sharp dose gradients between tumors and nearby critical structures. However, uncertainties introduced by variations in patient setup (hereafter termed patient-setup uncertainties) and organ motion may lead to target underdosage, thus contributing to LF. 7 In planning for external beam therapy, setup uncertainties and organ motion are addressed by uniform geometric expansion of the clinical target volume (CTV) to form the planning target volume (PTV). 8 In our clinics, the PTV coverage is evaluated to assess the probability of LF due to patient setup uncertainties. 4,9 This approach assumes the static dose cloud approximation, i.e., the approximation that the dose cloud is static relative to the room-coordinate system. 10 During the past decade, researchers have extensively studied the sensitivity of IMRT plans to uncertainties and organ motion in H&N cancer radiotherapy. 7,[11][12][13][14][15][16][17] However, few studies have investigated the sensitivity of VMAT plans to these uncertainties. 18 More importantly, almost all the reported studies 7,11-17 are computer-based dosimetric investigations. To our knowledge, no studies have examined the association of LF observed in the clinic with VMAT plan robustness. Therefore, we evaluated data of 10 patients with H&N cancer presenting with a total of 13 LFs after being treated with VMAT in order to determine whether potential target underdosage subject to patient setup uncertainties may be a more relevant factor in predicting LF than the conventional PTV method.

2.A | Patient data and treatment planning
The scope of this work is limited to show which is more relevant to cause the LF due to patient setup uncertainty between the compromised PTV coverage and the potential target underdosage subject to patient setup uncertainties. Other complicated technical aspects and biological aspects including surgical seeding are not considered in this exploratory study.
In total 396 patients with H&N cancer were treated using VMAT at our institution from January 19, 2012 to August 31, 2015. All these patients had positron emission tomography/computed tomography (PET/CT) (General Electric Discovery* PET/CT 610) and/or magnetic resonance imaging (MRI) (GE Discovery MR750w) before radiation therapy to help diagnostics of the disease and facilitate the delineation of the targets. The postoperative patients were not required to have MRI. The PET/CT has the in-plane spatial resolution of 3.6 mm with the slice thickness of 3.3 mm. The MRI has the in-plane spatial resolution of 1.0 mm with the slice thickness of 3.0 mm. All patients had radiographic staging studies using PET/CT. Pathologic staging was used for patients treated postoperatively, whereas clinical staging was used for patients who underwent definitive treatment with radiation therapy.
10 patients with LF were identified by an experienced radiation oncologist. Characteristics of the patients, tumors, and treatment are shown in Table 1. The subsites of the tumors were oral cavity/ oropharynx (seven patients), nasopharynx (two patients), and supraglottic larynx (one patient). Tumors were classified into two classes: patients with T > 2 or N ≥ 2 were considered to be locally advanced (seven patients), otherwise they were considered as early stage (three patients). None of the patients had distant metastatic disease at presentation. Nine patients had histologically verified squamous cell carcinoma (SCC) and 1 patient had adenocarcinoma. All these patients had follow-up with PET/CT in at least 3 months after radiotherapy. Different from the planning CT simulation, the PET/CT scans are for the purpose of diagnosis and thus patients do not have any immobilization during the PET/CT. The patients were clinically determined to have LF based on the PET/CT scan results by an experienced radiologist and then reviewed by a second experienced radiation oncologist. 6 Each patient was treated using VMAT with 2 or 3 arcs. Two dose levels were prescribed and administered using a simultaneous integrated boost technique. The target region, which received a higher prescribed dose, was referred to as CTV High and the region that received a lower prescribed dose was referred to as CTV Low .
CTVs were delineated by a physician, with CTV High defined as the volume with gross disease or high risk microscopic disease (gross LIU ET AL.  (Table 1).
Doses to critical normal structures were constrained to meet acceptable tolerance dose values whenever possible as defined in the departmental H&N cancer treatment protocol (Table S1). All VMAT plans were generated by experienced dosimetrists or physicists using the treatment planning software Eclipse TM (Varian Medical Systems, Palo Alto, CA, USA) and were approved by the treating physician. A second review by a radiation oncologist was performed to verify that all plans met departmental criteria (Table S1).

2.B | Underdosed volume because of cold spots in
the PTVs in the nominal dose distribution (standard method to assess cold spots) In photon radiation therapy, the dose distributions of PTVs are usually used to evaluate the impact of patient setup uncertainties. Two dose-volume histogram (DVH) indices were used to assess the PTV coverage: D 95% , the dose covering 95% of the PTV and V 95%, the subvolume of the PTV receiving 95% of the prescription dose.

2.D | Association of underdosed volumes with LF
All patients had PET/CT scans performed at least 3 months after radiotherapy. Each patient had LF within and around CTVs (both CTV High and CTV Low ), and three patients (patient # 4, 6 and 7, To facilitate spatial association, the PET/CT scan was registered to the planning CT using landmark-based rigid registration in Eclipse TM . 19 The registration uncertainty study was not performed and will be the research topic of the future study. However, the image registration was reviewed and approved by the treating physician. The UVN (standard method) and UVC (robustness method) were then examined for spatial association with LF. If there was overlap of at least 0.5 cm 3 between the underdosed volume and the region of LF, the variable indicating association between the underdosed volumes and LF was set to be TRUE; otherwise it was set to be FALSE ( Table 3). The value of 0.5 cm 3 is chosen to have enough voxels to minimize the possible random errors. The association was further reviewed and approved by an experienced radiation oncologist.

2.E | Visualization of dose variation in the PTVs
The

2.F | Statistical analyses
The null hypothesis was that the underdosed volume was independent of LF. Therefore, the binomial distribution with a probability of 0.5 was used to calculate the p value for the statistical significance of the association between the underdosed volume and LF. We further tested the relationship between the aforementioned association and tumor histology, location, stage, CTV High volume, and treatment modality by using the Fisher's exact test of independence. The 2 9 2 exact contingency table was used.

| 79
For the analysis of the Fisher's exact test of independence, the patients were divided into two groups according to tumor location (oral cavity/oropharynx or not), stage (advanced stage or not), CTV High volume (> 100 cc or not), and treatment modality (with surgery or not).  Table 2. Of these 13 LFs, nine were in-field, three were marginal, and one was out-of-field LFs ( Table 2).

| RESULTS
The D 95% and V 95% of PTV High and PTV Low in the nominal dose distribution are listed in Table 2 (Table S2).
To show dose variation in the PTVs, the RMSD distributions in three planes calculated from the perturbed doses corresponding to different scenarios of patient-setup uncertainty are shown in Fig. 3. One of the major causes for LFs is the suboptimal target delineation. 4,9 The scope of this work is limited to show which is more relevant to cause the LF due to patient setup uncertainty between the compromised PTV coverage and the potential target underdosage subject to patient setup uncertainties. In this study all the targets were contoured by an experienced radiation oncologist and then reviewed by an independent experienced radiation oncologist.
And an experienced radiation oncologist carefully reviewed all the patients with LF and chose patients, who were least likely to have insufficient target delineation. That is also the reason why none of the patients included in this study had the radiation target volumes cropped due to the protection of the nearby organs at risks. Some PTVs were cropped at least 3 mm from the skin if needed due to the imperfection in the dose calculation of our treatment planning system. Fortunately, none of the LFs of the patients included in this study took place in these cropped volumes. | 81 association between LF and UVN was not (P = 0.99) (  Table S2.
Based on this small patient population, it seems that the conclusion is independent of tumor location, stage, volume, and treatment modality.
However, a larger patient population is warranted to verify the clinical relevance of our brute force method, which will be included in our future study. Our patient cohort was also limited in disease subsites.
All of the patients had the disease below the eye level. Fried et al.,4 reported that in case of sinonasal malignancies, with the disease located at or above the eye level, target coverage was often compromised to protect the critical normal tissue adjacent to the orbit and base of skull, thus increasing the risk of LF in this area. Sinonasal cancer is an uncommon malignancy and we did not have any LFs in the recent past using VMAT that could be included for this study.
In summary, our work alerts radiotherapy practitioners not to rely solely on the standard method for assessment of VMAT PTV coverage to assess target control. The plan robustness analysis appears to be a significantly more important factor associated with LF than the conventional PTV underdosage measures for patients with H&N cancer treated with VMAT. A larger patient population study with a control group would be helpful for further investigation.

CONFLI CTS OF INTEREST
None.

R E F E R E N C E S SUPPORTING INFORMATION
Additional Supporting Information may be found online in the supporting information tab for this article.
Table S1 Dose volume constrains for H&N radiation treatment from our institution protocol.