Using adaptive magnetic resonance image‐guided radiation therapy for treatment of inoperable pancreatic cancer

Abstract Background Adaptive magnetic resonance imaging‐guided radiation therapy (MRgRT) can escalate dose to tumors while minimizing dose to normal tissue. We evaluated outcomes of inoperable pancreatic cancer patients treated using MRgRT with and without dose escalation. Methods We reviewed 44 patients with inoperable pancreatic cancer treated with MRgRT. Treatments included conventional fractionation, hypofractionation, and stereotactic body radiation therapy. Patients were stratified into high‐dose (biologically effective dose [BED10] >70) and standard‐dose groups (BED10 ≤70). Overall survival (OS), freedom from local failure (FFLF) and freedom from distant failure (FFDF) were evaluated using Kaplan‐Meier method. Cox regression was performed to identify predictors of OS. Acute gastrointestinal (GI) toxicity was assessed for 6 weeks after completion of RT. Results Median follow‐up was 17 months. High‐dose patients (n = 24, 55%) had statistically significant improvement in 2‐year OS (49% vs 30%, P = 0.03) and trended towards significance for 2‐year FFLF (77% vs 57%, P = 0.15) compared to standard‐dose patients (n = 20, 45%). FFDF at 18 months in high‐dose vs standard‐dose groups was 24% vs 48%, respectively (P = 0.92). High‐dose radiation (HR: 0.44; 95% confidence interval [CI]: 0.21‐0.94; P = 0.03) and duration of induction chemotherapy (HR: 0.84; 95% CI: 0.72‐0.98; P = 0.03) were significantly correlated with OS on univariate analysis but neither factor was independently predictive on multivariate analysis. Grade 3+ GI toxicity occurred in three patients in the standard‐dose group and did not occur in the high‐dose group. Conclusions Patients treated with dose‐escalated MRgRT demonstrated improved OS. Prospective evaluation of high‐dose RT regimens with standardized treatment parameters in inoperable pancreatic cancer patients is warranted.


| INTRODUCTION
Radiation therapy (RT) is controversial in the treatment of inoperable pancreatic cancer. The results of the phase III randomized LAP07 trial illustrate that conventional doses of 3D conformal radiation (3DCRT) do not confer an overall survival (OS) advantage over chemotherapy alone in locally advanced pancreatic cancer (LAPC). 1 Investigators have evaluated RT techniques other than 3DCRT to improve outcomes. Retrospective and phase I/II studies of stereotactic body radiation therapy (SBRT) have shown local control outcomes generally exceeding 80% but no change in OS as compared to historical data and some concern with unfavorable toxicity. [2][3][4][5][6][7][8] Multiple institutions have demonstrated favorable survival results with intensity-modulated radiation therapy (IMRT) after induction chemotherapy. 9, 10 Krishnan et al have demonstrated that dose escalation using simultaneous integrated boost (SIB) with IMRT improved survival and local control in LAPC compared to standard radiotherapy doses in a single-institution retrospective study. 11 These techniques suggest opportunities to potentially improve outcomes with dose escalation with RT.
Adaptive magnetic resonance imaging (MRI)-guided radiation therapy (MRgRT) is a novel modality that potentially allows for dose escalation while minimizing excessive radiation dose to the organs at risk (OAR). MRgRT offers excellent visualization of the stomach, duodenum, small, and large intestines in the setting of abdominal RT and thus can account for interfractional variability of these organs. 12,13 The ability to adapt treatment plans daily allows for improved target volume coverage while meeting OAR constraints that is necessitated by daily gastrointestinal (GI) organ motion and deformation. [14][15][16] While dose escalation with MRgRT appears technically feasible and can spare OARs from high doses of radiation, it is not clear whether this leads to changes in clinical outcomes. This multi-institutional study evaluated clinical outcomes of inoperable pancreatic cancer patients treated with MRgRT and compared patients treated with and without adaptive dose-escalated regimens.

| METHODS
This study was a multi-institutional, retrospective, cohort study based on data from 5 institutions. Eligible patients had biopsy-proven, inoperable, pancreatic cancer treated with MRgRT from 2014 through 2016. All patients were evaluated with diagnostic computed tomography (CT) imaging or MRI. Patients deemed medically inoperable were included in our study as well. The study excluded patients with prior pancreas-directed RT, pancreatic surgery or any clinicalradiographic evidence of distant metastasis prior to initiation of RT. Research conformed to the Helsinki Declaration and satisfied retrospective review requirements for each institution.
Systemic therapy was determined by the local medical oncologist or institutional study protocol. Induction chemotherapy regimens included nab-paclitaxel plus gemcitabine, FOLFIRINOX, gemcitabine alone, FOLFOX, and combined regimens. Concurrent chemotherapy was delivered on the day of radiation treatments with conventional fractionated radiation treatments and hypofractionated radiation treatments. Concurrent regimens consisted of nab-paclitaxel plus gemcitabine, gemcitabine alone, and capecitabine alone. After RT, patients received systemic therapy according to institutional preferences.
MRgRT (ViewRay MRIdian System, Oakwood Village, OH) was used to treat all patients included in this study, but trended towards significance for 2-year FFLF (77% vs 57%, P = 0.15) compared to standard-dose patients (n = 20, 45% simulation and planning techniques varied between institutions. The Supplemental methods and Table S1 offer a detailed overview of simulation and planning information for each institution. In general, patients were simulated with a planning CT and planning 0.35T MRI scan in the supine position. Gross tumor volume (GTV) included the primary tumor with or without inclusion of any enlarged regional lymph nodes (>1 cm in short axis) and was delineated using MRI simulation, CT simulation and diagnostic imaging. Some institutions contoured regional lymph nodes into the clinical target volume (CTV) while others did not treat enlarged regional lymph nodes and considered GTV equal to CTV. Planning target volume (PTV) was generated with a 5 mm isotropic margin from GTV or CTV.
Adaptive MRgRT treatments were delivered for patients receiving 15 or fewer fractions and required patients to receive a volumetric MRI prior to each fraction. Patients were aligned daily to the GTV based on their volumetric MRI, and their radiation plan was evaluated on their current anatomy. If the plan did not meet OAR constraints or the target was not covered adequately, a new plan was created using the same beam angles. If the new plan did not resolve constraint violations and/or improve coverage, then the prior plan was utilized for treatment of that fraction. Adaptive MRgRT plans prioritized strict OAR constraints even at the expense of PTV coverage. However, if favorable OAR anatomy was noted during adaptive treatment, dose escalation to improve PTV coverage could be performed while maintaining OAR constraints. [14][15][16][17] In some cases, an adaptive plan could be generated to treat above the prescription dose. For patients receiving >15 fractions of treatment, nonadaptive MRgRT was delivered using conventional treatment planning with no adjustment for daily internal anatomy.
In order to compare the effects of various dose and fractionation schedules on outcomes, the biologically effective dose (BED 10 ) was calculated for each patient using the radiation prescription dose and alpha/beta ratio of 10 as previously reported in the literature. 11 Patients receiving a radiation prescription with BED 10 >70 Gy were grouped in the high-dose cohort while those receiving a radiation prescription with BED 10 ≤70 Gy were grouped in the standard-dose cohort. Table 1 provides complete RT regimen details for the study patients.

| Outcomes and statistical analysis
Endpoints included OS, freedom from local failure (FFLF), freedom from distant failure (FFDF) and rate of acute grade 3 or higher GI toxicities. All time to endpoint calculations were performed from start date of RT. Surviving patients were censored at last follow-up. Both local and distant failure events were determined using routine clinical-radiographic studies. Local failure was defined as radiographic progression noted at the primary tumor site or regional nodes. Distant failure was defined as radiographic progression at any other site. Kaplan-Meier curves were generated and differences between groups were determined using the log-rank method. Acute GI toxicity was graded based on the Common Terminology Criteria for Adverse Events version 4 and recorded from start of RT until 6 weeks after completion of RT. Baseline patient characteristics were evaluated by Fisher's exact testing for categorical variables and Mann-Whitney U testing for continuous variables. Statistical significance was defined as a P < 0.05 and all testing was two-sided. Univariate (UVA) and multivariate (MVA) analyses were performed for OS using Cox regression. CA 19-9 was analyzed as a binary categorical variable (cutoff value of 37 U/mL based on prior studies 18 ). As the number of events was limited, only factors with univariate P < 0.05 were included in the MVA. The MVA defined significance as a P < 0.05. Statistical analyses were performed using IBM SPSS Statistics, version 23 (IBM Corp., Armonk, NY) and SAS (Version 9.4, SAS Institute Inc, Cary, NC).

| RESULTS
A total of 44 patients were included in the study (high-dose, n = 24; standard-dose, n = 20). Concurrent chemotherapy was administered with all the conventional fractionated patients and all but two hypofractionated radiation patients. As expected, treatment fraction adaptation was more common in the high-dose group (83%) vs the standard-dose group (15%). After completion of RT, six patients (two in high-dose group and four in the standard-dose group) underwent surgical resection with five patients (two in high-dose, three in standard-dose) demonstrating microscopically negative margins. Further patient and treatment characteristics are listed in Table 2.
Severe acute GI toxicity (grade 3 or higher) occurred in three patients (7%). All three patients were in the standard-dose cohort and received concurrent gemcitabine-based chemotherapy. Two patients developed abdominal infections requiring hospitalization, and one patient experienced a grade 4 duodenal ulcer requiring intensive care unit admission. No patient died of their acute toxicity.

| DISCUSSION
This study suggests that adaptive MRgRT is a promising approach to advance the treatment of inoperable pancreatic cancer. MaxBED 10 of treatment was shown in our study to be associated with improved OS and FFLF. By using online adaptive treatment planning to spare OARs and optimize RT dose delivery to target volumes, MRgRT appears to be a safe technique for ablative dose escalation.
Dose escalation in pancreatic cancer is an area of keen interest in the radiation oncology community. Early approaches included intraoperative radiation therapy (IORT) where visible tumor was irradiated with a single electron portal. This procedure was limited to those fit for surgical exploration with constraints on size of tumors. Modern reports of this technique show median OS of 25 months from diagnosis for patients receiving IORT without resection. 19 With the advent of 3DCRT and IMRT approaches, moderate dose escalation to BED from 60-70 Gy became possible. A recent review of 134 patients by Huguet et al noted a median survival from diagnosis of 23 months for patients treated with 56 Gy/28 fractions (BED 10 = 67.2 Gy) with chemoradiation. 9 Stereotactic body radiation therapy led to attempts at dose escalation but with a concomitant increase in toxicity. Schellenberg et al evaluated concurrent gemcitabine with high-BED (25 Gy/1 fraction, BED 10 = 87.5 Gy) SBRT for LAPC reported local recurrences in only 19% of patients. They reported 15% grade 2 ulcers and one patient with a duodenal perforation. 20 Moreover, an early SBRT regimen prior to routine onboard image guidance of 45 Gy/3 fractions (BED 10   More recently, data by Krishnan et al found that high-BED treatments (BED 10 : 70.4-100 Gy) using SIB in selected patients lead to improved OS (36% at 2 years) and loco-regional recurrence free survival, but patients only received dose escalation if their tumors were more than 1 cm away from GI mucosal structures. 11 This study did not select specific patients for dose escalation but used MRI-guidance to allow clinicians to make the decision regarding plan adaptation and feasibility based on daily evaluation of internal anatomy. Dose escalation was evaluated F I G U R E 1 Overall survival from start of radiation therapy stratified by biologically effective dose (BED 10 ). Standard error bars displayed at each 6-mo timepoint F I G U R E 2 Freedom from local failure from start of radiation therapy stratified by biologically effective dose (BED 10 ). Standard error bars displayed at each 6-mo timepoint with BED 10 to compare the various radiation treatment regimens. In addition to the survival benefit, high-dose regimens demonstrated a trend toward significance for improved local control. The median OS of 10.8 months in our standard-dose cohort emulated historical controls receiving conventional chemoradiation therapy demonstrating that standard radiation doses may not improve outcomes even if visualization of internal anatomical changes leads to improved localization and motion management. Malik  of induction chemotherapy whereas we calculated OS from start of RT. This accounts for the numerical difference in survival between our standard-dose cohort and the LAP07 cohort. More impressively, patients in the high-dose group of this study demonstrated a sizeable survival advantage compared to the standard-dose group despite this cohort having older age and higher CA 19-9 levels, which are known negative prognostic factors. 23 Table 4 lists survival outcomes from previous studies compared to this study cohorts. FFDF rates were similar between the two groups despite the large difference in OS. We postulate that effective doses of radiation prevent preclinical local failure symptoms such as anorexia, cachexia, and nausea that often preclude effective systemic therapy at progression. In addition, post-RT pancreatectomy rates did not change with the high-dose regimens, and the overall rate of resection of 14% was low in this study as compared to historical series. 9,10 Determining resectability radiographically is difficult after neoadjuvant therapy due to tissue fibrosis complicating imaging interpretation. 24 Use of high-dose regimens may increase the risk of fibrosis even further. However, even without the benefit of surgical resection, high-dose patients had favorable OS outcomes.

Characteristics Overall (N = 44) High-dose (N = 24) Standard-dose (N = 20) P-value
Fear of treatment-related toxicity is a hurdle for using ablative RT doses in pancreatic cancer. Patients in this study experienced an overall acute grade 3+ GI toxicity rate of 7% with one patient experiencing duodenal toxicity. No patient in the high-dose cohort experienced severe toxicity suggesting that adaptive radiation and decreased use of concurrent chemotherapy diminished this risk. Plan adaptation was frequently used in the high-dose group to limit dose to OARs when significant internal anatomical changes occurred. Further follow-up will be required to assess for late toxicities, but it is reassuring that high-dose adaptive MRgRT appears to be well tolerated acutely.
This study highlights a novel approach of dose escalation for treatment of inoperable pancreatic cancer, but the study design poses limitations which render the findings as hypothesis-generating. Medically inoperable patients with resectable disease based on imaging criteria were included in this study, but bias from these patients is unlikely given that only two patients met this criterion. Though this was a multiinstitutional study, the sample size was relatively small, and heterogeneity of treatment approaches existed between institutions. We also note that radiographic assessment for tumor response is challenging after delivery of RT and local evaluation of imaging studies can introduce bias. Furthermore, high-dose RT did not predict for OS on MVA. A larger sample size would be needed to demonstrate significance of highdose RT when grouped by BED 10 , yet, it is still important to note that high-dose RT remained significant on univariate regression and Kaplan-Meier analyses. Another limitation was that a number of patients had the same prescription dose but heterogeneous delivered dose due to adaptive treatments. Although stratifying patients by BED 10 of prescription dose did not account for this dose heterogeneity, the swimmer's plot stratified patients by MaxBED 10 to explore dose heterogeneity and its impact on OS. These study data support the evaluation of dose-escalated regimen in a larger, prospective setting.

| CONCLUSIONS
High-dose adaptive MRgRT in the treatment of inoperable pancreatic cancer resulted in improved OS in this multi-institutional study. Adaptive MRgRT provides an innovative technique to administer higher doses of radiation to patients without increasing the risk for acute toxicity. A prospective, phase II multi-institutional study (NCT03621644) prescribing 50 Gy in 5 fractions (100 Gy BED 10 ) using adaptive dose escalation with MRgRT is open, accruing, and will clarify the role of MRgRT for inoperable pancreatic cancer.

CONFLICT OF INTEREST
The authors listed below report the following financial relationships: NJ reports honoraria from ViewRay, outside the submitted work. LP reports honoraria and stock from ViewRay along with advisory role, speakers' bureau, and research funding from BTG, outside the submitted work. EAM reports travel accommodations from ViewRay, outside the submitted work. AB and FL report honoraria and travel accommodations from ViewRay, outside the submitted work. MCR reports travel expenses from Varian and BTG, outside the submitted work. MFB reports travel accommodations from ViewRay along with research funding from AstraZeneca outside the submitted work, outside the submitted work. PJP reports research funding from ViewRay, outside the submitted work. PL reports honoraria, consulting/ advisory role, and travel accommodations from ViewRay, AstraZeneca, and Varian along with research funding from AstraZeneca, outside the submitted work. All other authors have no conflicts to declare.

AUTHOR CONTRIBUTIONS
Soumon Rudra: data curation, investigation, writing original draft, writing review, and editing. Naomi Jiang: data curation, writing original draft, writing review, and editing. Stephen A. Rosenberg: data curation, writing original draft, writing review, and editing. Jeffrey R. Olsen: investigation, methodology, writing review, and editing. Michael C. Roach: investigation, methodology, writing review, and editing. Leping