Reirradiation of Recurrent Pediatric Brain Tumors after Initial Proton Therapy.

Purpose
The use of reirradiation for recurrent pediatric brain tumors has been increasing, but the effect of repeat radiation on critical cranial structures is unknown.


Methods and Materials
Between July 2009 and May 2013, the records of 12 pediatric patients initially treated with proton therapy and then with reirradiation for recurrent brain tumors were retrospectively reviewed for toxicity and outcomes. Initial and repeat radiation dose distributions were deformed and merged to determine the maximum dose to 0.03 cm3 of the optic chiasm, optic nerves, spinal cord, brainstem, cochleae, pituitary, and uninvolved brain, and to 1 cm3 of the brainstem and brain on individual and composite plans. These dosimetric results were compared with auditory, neurocognitive, ophthalmologic, and endocrine outcomes to identify radiation-associated toxicities.


Results
Median follow-up was 3.5 years from diagnosis. Median ages at initial and repeat radiation were 4.6 and 6.7 years, respectively. All patients initially received proton radiotherapy to a median tumor dose of 55.8 Gy relative biological effectiveness (RBE) (range, 45 to 60 Gy [RBE]). At progression, patients completed a second course of radiation to local fields (n = 7) or the craniospinal axis (n = 5) with a median tumor dose of 40 Gy (RBE) (range, 20 to 54 Gy [RBE]). Median progression-free survival was 22.7 months from the last day of the second radiation course. No patient developed central nervous system necrosis requiring treatment. Of evaluable patients, none developed radiation-related high-grade hearing loss (n = 11), visual pathway deficit (n = 10), or significant change in pre- and post-reirradiation full-scale intelligence quotient (n = 4). Of 11 evaluable patients, 4 (36.4%) developed secondary hypothyroidism and 1 (9.1%) developed growth hormone deficiency.


Conclusion
Repeat radiation for recurrent brain tumors after proton therapy may be performed in the pediatric population with acceptable short- and long-term toxicity.

Introduction have typically used the prescribed dose to guide treatment planning, but modern conformal techniques allow for the avoidance of normal structures while providing therapeutic doses to target volumes. Consequently, the dose to critical structures may be less than the prescribed dose.
Proton therapy is increasingly used to treat pediatric brain tumors [6]. Dose-volume constraints based on the photon experience are used to guide treatment planning, but emerging evidence suggests that the effects of photons and protons on the pediatric brain, particularly the brainstem, may not be equivalent [7]. How these differences should be considered in the setting of reirradiation after initial proton therapy remains unknown.
Using conformal doses that consider the high-dose volume treated during the first course of radiation, we have performed reirradiation in select patients. This technique allows for maximal dose to the target while sparing adjacent organs at risk (OARs) as much as possible. Photon, electron, and proton therapy have been used alone or in combination to achieve the optimal dosimetry for each patient.
To explore our institutional outcomes with reirradiation, we retrospectively reviewed the clinical and dosimetric records of 12 pediatric patients initially treated with proton therapy who subsequently underwent a second course of radiation to the brain.

Patient Selection and Data Collection
Almost all patients treated with proton therapy at our institution were enrolled in a prospective registry approved by the institutional review board. Informed consent was obtained for all patients prior to enrollment. This registry was reviewed to identify patients with primary brain tumors who underwent initial radiation with proton therapy and a second course of radiation to the brain with any modality. Patients were required to have detailed dosimetric data for both treatment courses, with the second course of radiation delivered between July 2009 and May 2013. All patients underwent multidisciplinary evaluation prior to reirradiation. Twelve patients met these criteria and were younger than 18 years at the time of their initial radiation.
The records of these patients were reviewed for demographic, tumor-specific, and outcome data. Age was defined at time of initial diagnosis. Extent of surgical resection was obtained from operative reports and imaging studies. Gross total resection was defined as no evidence of remaining tumor by imaging or the surgeon's intraoperative impression. Subtotal resection was defined as anything less. Tumor location was supratentorial or infratentorial. Date and location of progression were defined by imaging studies. Time to progression was the interval from the last day of radiation to the first date of progression.
Acute and late toxicity were assessed using the National Cancer Institute's Common Terminology Criteria for Adverse Events, version 4.0 [8]. Endocrine, audiologic, ophthalmic, and neurocognitive testing were collected with pre-and postreirradiation assessments. For comparison, evaluations were divided among 3 time periods: (1) before initial radiation, (2) the interval between radiation courses, and (3) after reirradiation. If a patient had multiple assessments in a single time period, the mean value was reported for that time period. Ototoxicity was scored according to the Brock ototoxicity grading scale [9,10]. If there was asymmetry in hearing loss between the right and left ear, the score for the worst ear was recorded. Magnetic resonance images of the brain and spine obtained in follow-up were reviewed for possible necrosis, which was scored based on the clinical condition of the patient, timing and nature of imaging changes after radiation, and need for symptomatic intervention.

Radiation Techniques
For the first (RT1) and second (RT2) radiation course, the gross tumor volume (GTV) was based on preoperative and postoperative imaging studies and the surgeon's operative impression of residual disease. Clinical target volume was defined as appropriate for each histology and target volume. For RT2, the clinical target volume was modified at the discretion of the treating radiation oncologist to account for the high dose volume treated in RT1. The planning target volume for patients treated with photon therapy was a 3-to 5-mm geometric expansion on the clinical target volune. Uncertainty margins for proton therapy were calculated as described [11]. Craniospinal radiation with proton therapy was performed with passive scatter proton therapy as described in Giebeler et al [11]. Modified craniospinal radiation with intensity-modulated radiation therapy (IMRT) and electrons was performed as described [12].

Dosimetry
Radiation dose distributions for RT1 and RT2 were independently analyzed. The Eclipse Treatment Planning System (Varian, Salt Lake City, Utah), versions 8.1 and 8.9, was used to calculate all proton therapy plans. Treatment plans for photon and electron therapy were calculated with Pinnacle 3 Treatment Planning (Philips, Andover, Massachusetts), versions 8.0 and 9.0.
A composite dose distribution containing the cumulative dose from RT1 and RT2 was generated. Deformable image registration was used to register the planning computed tomography (CT) dataset from RT1 to that of RT2 [13]. The deformation vector field obtained from the image registration was then used to transfer the dose distribution from RT1 to the planning CT dataset of RT2. The deformed dose from RT1 was added to the dose from RT2 to generate a composite plan (RT1 þ RT2) for analysis.
Dosimetric data were collected for RT1, RT2, and the composite plans, including the maximum dose to 0.03 cm 3 of the optic chiasm, optic nerves, cochleae, pituitary, spinal cord, brainstem, and uninvolved brain. Per Radiation Therapy Oncology Group protocol 0825, 0.03 cm 3 was chosen to represent the volume of the maximum point dose [14]. As drawn on the planning CT dataset for RT2, dose to the uninvolved brain was calculated by subtracting the sum of the dose to the optic chiasm, optic nerves, brainstem, and the GTV2 from the dose to the brain. The maximum dose to 1 cm 3 of the uninvolved brain and brainstem was also calculated.
If both IMRT and proton therapy were used during a single treatment course, the dose distribution for each was calculated and their sum reported. A relative biological effectiveness (RBE) of 1.1 was utilized to convert photon dose in Gray to proton dose in cobalt-60 Gray equivalent (Gy [RBE]) [15].
To compare treatment regimens, the linear-quadratic model was used to calculate the biologically equivalent dose (BED). Accordingly, , and a/b ¼ tissue repair capacity [Gy (RBE)] [16]. An a/b ratio of 2 was used as a measure of late toxicity for the brain [17,18]. If 2 different dose fractionations were used during a single treatment course, the BED value of each was calculated and their sum reported. The total BED (BED total ) was calculated as the sum of the initial radiation BED and the repeat radiation BED, For comparison, BED total values were normalized to 2 Gy (RBE) fractions and reported as normalized total dose (NTD 2 ). The NTD 2 is the ratio of BED to relative effectiveness (RE), where RE ¼ ð1 þ d= a=b ½ Þ; d ¼ 2 Gy and a=b ratio of 2; or NTD 2 ¼ BED=2:

Statistical Analysis
Paired t tests were used to compare differences between the time to initial and second progression and pre-and postreirradiation neurocognitive assessments. Wilcoxon matched-pairs signed-ranks tests were used to compare NTD 2 values with maximum composite dose to 0.03 cm 3 of the brain, brainstem, and GTV2. The Kaplan-Meier method was used to calculate progression-free survival (PFS) and overall survival (OS) [19].

Patient Characteristics
Patient-specific treatment characteristics are shown in Table 1

Progression
Median time to progression was 1.2 years (range, 0.1 to 4 years). Median time to local (n ¼ 8), distant (n ¼ 3), or both local and distant (n ¼ 1) progression was 1.0, 1.4, and 1.2 years, respectively. There were no failures outside the CNS. Six patients underwent gross total resection prior to reirradiation.

Repeat Radiation (RT2)
Median interval from the end of RT1 to the start of RT2 was 14.3 months (range, 1.15 to 47.9 months). Median age at the start of RT2 was 6.7 years (range, 3 to 11.5 years). At progression, patients were treated with local fields (n ¼ 7) or CSI (n ¼ 5) to a Of the 7 patients treated with local fields, 5 were treated with proton therapy and 2 with IMRT. The choice of proton therapy or IMRT was at the discretion of the treating radiation oncologist. Five patients received CSI. In 4 patients, IMRT was used to treat the cranial fields and electrons were used to treat the spine as previously described [12] (Figure 1). One patient (No. 12) received a boost with protons and 3 received a boost with IMRT. The remaining patient (No. 6) had CSI with 6 MV photons due to the palliative intent of treatment because the patient recurred with disseminated disease in the CNS after upfront CSI. In all cases, care was taken to minimize overlap with previous target volumes and OARs (Tables 2 and 3).  Three patients received chemotherapy during reirradiation, with temozolomide alone (n ¼ 1), or in combination with irinotecan and/or vincristine (n ¼ 2).

Survival
Median follow-up was 3.5 years (range, 1.6 to 5.8 years). Median PFS from diagnosis was 19.7 months (range, 8.0 to 49.3 months), and median PFS from the last day of RT1 was 12.7 months (range, 0.6 to 46.6 months) (Figure 2). Median OS was not reached; 1-and 2-year actuarial OS from the last day of RT2 was 66.7% and 58.3%, respectively.
Median PFS was 22.7 months from the last day of RT2 (Figure 2). Seven patients (58%) had a second progression. Three did not receive additional therapy, 1 received surgery (No. 4), and 1 received chemotherapy (No. 1). Two underwent a third course of radiation: patient No. 3 received 20 Gy (RBE) in 10 fractions to disease in the spine, and patient No. 8 received stereotactic radiosurgery to the sphenoid sinus.

Target Volume Dosimetry
Based on prescribed doses, median NTD 2 to the brain on the composite plan was 90.  Table 2). Median maximum composite doses to 0.03 and 1 cm 3 of the uninvolved brain were 72.8 Gy (RBE) (range,   Composite dose-volume histograms for the whole brain and brainstem are shown in Figure 3. One patient (No. 3) developed a 5-mm patch of enhancement in the pons 4 months after RT2 that subsequently resolved without intervention. Because this was asymptomatic and did not require intervention, this was conservatively scored as grade 1 CNS necrosis per National Cancer Institute's Common Terminology Criteria for Adverse Events, version 4.0. This lesion was not biopsied and therefore may have represented a subacute radiographic change after cranial radiation [20] instead of true necrosis. No patient developed CNS necrosis requiring treatment, cranial nerve impairment, or myelopathy after RT2. Four patients (Nos. 2, 7, 9, and 12) had full-scale intelligence quotient (FSIQ) assessments after RT1 and after RT2. Mean pre-and post-reirradiation FSIQ scores were not significantly different, 94.6 6 10.9 versus 88.3 6 11.2 (n ¼ 4; P ¼ .37), and all patients continued to attend school.
Seven patients had normal hearing before RT1 and were evaluable for ototoxicity after RT2. The median interval from the end of RT2 to the last available audiogram was 12 months (range, 6 to 26 months). No patients developed high-grade hearing loss, defined as hearing loss in the frequency range of 0.5 to 2 kHz, the range of audible speech. Three patients (Nos. 1, 9, and 11) developed grade 1 or 2 hearing loss following RT1 after a mean cochlear dose of 23.7 to 38.2 Gy (RBE), all receiving cisplatin ( Table 2). After RT2, these 3 patients had stable hearing. The other 4 patients (Nos. 3, 7, 8, and 12) remained free of hearing loss after RT2, with 3 patients receiving cisplatin.
Ophthalmic testing was available for 10 patients. Two patients (Nos. 6 and 12) developed optic pathway defects related to tumor progression. Eight patients had normal visual field testing following RT2 after a median maximum composite dose to 0.03 cm 3 Table 3). One patient (No. 1), with a composite maximum dose of 39 Gy (RBE) to the pituitary, developed growth hormone deficiency following RT2. Two additional patients, each treated with CSI, developed short stature (No. 9) and linear growth deceleration (No. 3) but had normal insulin-like growth factor 1 levels.

Discussion
Reirradiation is increasingly used to salvage recurrent brain tumors in children [21]. We found that reirradiation after initial proton therapy may be performed with acceptable short-term toxicity by taking into account the previous high dose volume.
The risk of toxicity after reirradiation depends partly on the cumulative dose delivered to the brain. To estimate this dose, many studies of reirradiation have summed the BEDs from the first and second radiation courses to determine the NTD 2 . Most recent pediatric studies demonstrate that NTD 2 values up to 107.5 Gy (RBE) may be well tolerated [1,2,4].
The NTD 2 calculations do not capture the dose distributions possible with modern techniques. In our study, the median NTD 2 value was significantly higher than the median maximum composite dose to 0.03 cm 3 of the brain, brainstem, and GTV2. This suggests that the NTD 2 overestimates dose to the target volume and the brain. For instance, the NTD 2 for patient No. 7 was 111.3 Gy (RBE), but on the composite plan, the maximum dose to 0.03 cm 3 of the GTV was 56.6 Gy (RBE) and the maximum dose to 0.03 cm 3 of the uninvolved brain was 65.2 Gy (RBE) (Tables 1 and 2). Both of these values are well below what would be expected from an NTD 2 of 111.3 Gy (RBE). As illustrated in Figure 4, treatment volumes from the first and second courses of RT for this patient are adjacent but do not overlap. Simple summation of the prescription doses does not reflect the reality of the dose distributions. Volume-based dose measurements such as dose to 0.03 cm 3 and 1 cm 3 of the GTV and uninvolved brain more accurately describe the dose distribution. In future studies, dose-volume measurements will enable more accurate determination of the tolerance of the brain and intracranial structures for reirradiation.
The volume of tissue irradiated, the initial and cumulative radiation doses, and the interval between radiation courses contribute to radiation necrosis [22][23][24]. No cases of necrosis requiring treatment were observed in our cohort of patients, with a maximum composite dose to 0.03 cm 3 of the uninvolved brain of 103.0 Gy (RBE) and a maximum composite dose to 0.03 cm 3 of the brainstem of 76.3 Gy (RBE). Likewise, no radiation-related toxicity of the optic pathway was observed, with a maximum composite dose to 0.03 cm 3 of the optic nerve of 53.2 Gy (RBE) and a maximum composite dose to 0.03 cm 3 of the optic chiasm of 56.9 Gy (RBE). These composite doses were low by design; aggressive sparing of the optic apparatus was a high priority for dose modulation for all patients. No patients were refused treatment because of composite dose to the optic structures.
Patients who developed hearing loss did so after RT1, and their hearing did not worsen after RT2 [25]. The 3 cases of hearing loss were grade 1 or 2; none were high grade. There was not a dose-response relationship between cochlear dose and grade of hearing loss, consistent with previous studies that did not find a dose-response relationship for ototoxicity but instead support a threshold cochlear dose of 36 Gy (RBE) for hearing loss [10,26].
One patient developed growth hormone deficiency and 4 developed secondary hypothyroidism. With longer follow-up, the prevalence of anterior pituitary deficits in this group will likely increase, since radiation doses as low as 18 Gy can cause growth hormone deficiency with increasing prevalence after 40 to 60 Gy [27][28][29].
In this cohort of heavily treated patients, it is difficult to isolate the effect of radiation on neurocognitive function, as surgical interventions, chemotherapy, and tumor can all contribute to neurocognitive decline [30]. Three of four evaluable patients had declines in FSIQ between RT1 and RT2; the oldest patient (7.9 years old at the time of RT1) did not have a decline in FSIQ. It is difficult to draw conclusions from this small number of patients, but this finding agrees with previous findings that older age at radiation protects against neurocognitive decline [31].
Our study emphasizes the importance of using the appropriate radiation modality to individualize treatment for each patient. Depending on the nature of the progression, photons, electrons, or protons can be used alone or in combination to provide the most appropriate treatment that optimally balances risks and benefits. In particular, the increasing use of proton therapy allows new opportunities for sparing previously radiated critical structures. McDonald et al [32] described 2 cases of salvage CSI using 3-dimensional conformal proton therapy with inverse apertures. Hill-Kayser and Kirk [33] described a case of brainstemsparing CSI using scanning beam proton therapy.
One of the largest series of proton reirradiation comes from Eaton et al [34], who described the Massachusetts General Hospital experience with repeat proton therapy for recurrent ependymoma. At a median follow-up of 37.8 months, 3-year OS and PFS were 78.6% and 28.1%, respectively. Of 14 patients who underwent local reirradiation, 3 developed symptomatic radiation-related changes in the brain, brainstem, or spinal cord; all 3 patients received oral dexamethasone with resolution or stabilization of symptoms. Notably, the composite doses to OARs reported in that study are higher than those reported here. The median maximum composite doses to the brainstem, brain, and spinal cord in the Massachusetts General Hospital dataset were 104.41, 105.70, and 70.42 Gy versus 59.4, 72.8, and 43.1 Gy (RBE) in our series, respectively. One reason for the more conservative approach in our series is that the median interval between radiation courses in our series was 14.3 months versus 27.75 months in the Massachusetts General Hospital series. Similarly, in the St. Jude's experience of reirradiation for recurrent ependymoma, 19 of 38 patients (50%) had an interval longer than 24 months between RT1 and RT2 [2]. Histologies such as ependymoma that tend to have late recurrences may be amenable to more aggressive reirradiation, because a longer interval allows for more recovery of normal tissue [22][23][24]. As more centers publish detailed dose, volume, and time interval data for reirradiation of OARs, the field will move toward more evidence-based definitions of dose constraints for these challenging cases.
Limitations of our study include its retrospective nature, our histologically diverse cohort, and limited follow-up. No attempt was made to account for different fraction sizes in initial or final RT, although all patients were treated with conventional fractionation. It is also difficult to draw robust conclusions about the influence of the time interval between radiation courses on the tolerance of intracranial structures because only 3 of 12 patients (25%) in our study underwent RT2 more than 2 years after RT1. This narrow interval contributed to our conservative approach during RT2, as evidenced by the low composite doses to the optic structures and cochleae in our series. As more groups explore reirradiation, additional nuances regarding outcomes and toxicities will emerge.

Conclusion
Reirradiation for recurrent primary brain tumors is feasible and may be performed in the pediatric population with acceptable short-and long-term toxicity. Additional studies are needed to optimize treatment parameters and identify late effects.

ADDITIONAL INFORMATION AND DECLARATIONS
Conflicts of Interest: Dr Yang receives royalties from Varian Medical Systems for licensing of deformable image registration software. The other authors have no conflicts of interest to disclose.