The proposed studies are applicable to all newly diagnosed H&N cancer patients consecutively treated at Mayo Clinic Rochester and Arizona between April 2013 and August 2019 (>2 years’ follow-up) with curative intent definitive chemoradiation therapy (VMAT or PBSPT) regardless of gender, age, minority status, vulnerable population status, and weight with a confirmed histologic diagnosis. The data was limited to (1): the patients treated with fractionation sizes of 120 cGy[RBE] to 220 cGy[RBE] per fraction (2); the patients treated with the prescription dose of at least 60 Gy[RBE] to the high-risk tumor target for both modalities; and (3) the patients with re-treatment only if the dose to the mandible was negligible from the re-treatment plan or if patients had already developed ORN before re-treatment. In total we got 1,266 patients. Among them, 931 were treated with VMAT and 335 were treated with PBSPT. Patients were non-intentionally selected by any clinical factors for either treatment modality. Note that we used Gy[RBE] to present doses for both VMAT and PBSPT if the RBE value was not specified in the bracket. For PBSPT, a fixed value of RBE=1.1 was assumed following current clinical practice (i.e. 60 Gy[RBE=1.1]). For VMAT, Gy[RBE] represented the physical dose (Gy) (i.e. 60 Gy[RBE=1.0]).

Most of the ORN cases occurred in patients with a primary tumor arising within the oral cavity or oropharynx or with unknown primary metastatic to nodes, in which case the oropharynx was treated (VMAT: 530 vs. PBSPT: 161). All patient data were stored in our institutional patient outcomes database (26). Demographic (age, gender) and related clinical information (prescription dose, tumor stage, concurrent chemotherapy, hypertension, diabetes, dental extraction, smoking history (former or current smoker) and current smoker) were extracted (Table 1). This study was approved by our institutional review board (IRB).

Patients with ORN were identified by experienced physicians clinically (bone exposure on physical examination), radiographically (Panorex, CT, MR, PET), and/or pathologically via debridement and resection/mandibulectomy. Patients with ORN were staged using the Marx system (27) based on their treatments with Trental/vitamin E, hyperbaric oxygen therapy (HBO), debridement, and mandibulectomy (Supplemental Materials Section 1 for details of Marx staging). Details of demographic and graded Marx staging information for all ORN patients are listed in Supplemental Table 1.

Both VMAT and PBSPT plans were generated using a commercial treatment planning system, Eclipse™ (Varian medical system, Palo Alto, CA) based on patients’ simulation CTs (Supplemental Materials Section 2 for details of treatment planning). All plans were evaluated to ensure that institutional dose volume constraints (DVCs) were met, if possible. The LET distributions of PBSPT plans were computed using an in-house fast Monte Carlo dose/LET calculation engine with a minimum electron energy cutoff of 50 keV (28, 29).

To minimize the impact of the possible imbalance in clinical factors between VMAT and PBSPT patients in the dosimetric comparison (dose and LET) between VMAT and PBSPT and the resulting RBE quantification, a subset of case-matched cohort of 335 patients from each modality was selected to carry out the analysis. The VMAT-PBSPT matched cohort was based on balancing clinical factors including gender, tumor stage, concurrent chemotherapy, hypertension, diabetes, dental extraction, current smoker, and smoking history using the greedy nearest neighbor matching of propensity scores. The propensity score represented the probability of one patient being treated with VMAT or PBSPT based on the observed clinical factors. In the greedy nearest neighbor matching, the VMAT patient whose propensity score was the closest to that of the PBSPT patient was selected as the match without replacement (30). The matched cohort was generated using “MatchIt” package of R (version 4.1.2).

Dose volume histograms (DVHs) were calculated within Eclipse™. DVH indices of absolute volumes receiving doses of at least 40 Gy[RBE] (V40Gy[RBE](cc)), 50 Gy[RBE] (V50Gy[RBE](cc)), 60 Gy[RBE] (V60Gy[RBE](cc)), 70 Gy[RBE] (V70Gy[RBE](cc)), and 75 Gy[RBE] (V75Gy[RBE](cc)) to the mandible, the minimum dose irradiated to the hottest 0.01 cc of mandible (D0.01cc), and mean dose (Dmean) of mandible were extracted for analysis.

The distribution of DVH indices were visualized using box plots. The bottom and top of each box were 25 and 75 percentiles from the population, and the middle line indicated the median value. Minimum and maximum values of whiskers were set as half of the interquartile range below or above the 25 or 75 percentiles.

The performance of all DVH indices was evaluated by calculating the area under curve (AUC) of the corresponding receiver operating characteristic (ROC) curve. Critical volumes were determined as the thresholds that performed at the optimal operating point of ROCs by minimizing the distance of the optimal operating point to the ideal case [true positive rate (TPR) = 1, false positive rate (FPR) = 0]. To estimate the uncertainties in the derived critical volumes and the ROC curve, we performed a bootstrapping of 1,000 times. 95% bootstrap confidence intervals of AUCs and critical volumes were computed. The true positive rates and false positive rates for the critical volumes were also calculated. Using this data we created volume tolerance curves of the mandible for both VMAT and PBSPT. This represented the possible tolerance volumes at different doses from both modalities.

We derived empirical proton RBEs by comparing volume tolerance curves between VMAT and PBSPT. According to the definition of RBE, it is the ratio of doses to reach the same clinical endpoint when a novel RT modality, such as PBSPT, is compared to photon irradiation (in this case VMAT). In this study, we considered the critical volumes of VMAT and PBSPT as the endpoints. We sought the equivalent constraint doses based on the VMAT volume tolerance curve via linear interpolation so the critical volumes of VMAT were equal to those of PBSPT at different physical doses. Empirical RBEs of PBSPT for mandible ORN were obtained by calculating the ratio between the derived equivalent constraint doses and physical doses (RBE=1.0) of PBSPT. 95% confidence intervals of RBEs were obtained using a bootstrapping of 1,000 times.

DLVH is a recently proposed cumulative volume histogram tool following the similar statistical study concept of DVH and aimed to evaluate the impact of LET by bypassing the uncertainties in the existing RBE models (25). It presents dosimetric variables including dose, LET, and normalized volume, all of which can be calculated relatively accurately. Figure 1 A illustrates a typical DLVH plot, in which the X axis is the RBE=1.1 dose and the Y axis is LET. Details of DLVH are included in Supplemental Materials Section 3.

All DVH indices were tested using Mann-Whitney U test. Categorical factors (gender, stage, chemotherapy, hypertension, diabetes, dental extraction, smoking history, and current smoker) of patients were tested using χ² test and numerical factors (age and prescription doses) were tested using two-sided two-sample t test. 95% confidence intervals of critical volumes, AUCs, equivalent constraint doses, and RBEs were derived based on a bootstrapping of 1000 times. All the statistical analyses were scripted using Matlab 2019a (MathWorks, Inc., Natick, Massachusetts, United States). P-value smaller than 0.05 was considered statistically significant.

Of 1,266 patients included in this study, the median age when patients finished the treatment was 62 (interquartile range: 15) years, and 327 (25.8%) were women. Of the VMAT group, 26 (2.8%) patients developed ORN of the mandible (grade≥2, Common Terminology Criteria for Adverse Events, CTCAE v4.0), while 9 (2.7%) patients experienced ORN of the mandible (grade≥2, CTCAE v4.0) in the PBSPT group. These 35 patients constituted the ORN group, while all the others constituted the control group. The ORN incidence rates were 4.91% (VMAT) and 4.97% (PBSPT) respectively among those oral cavity and oropharynx patients (Table 1).

No statistically significant differences of age (p=0.468), gender (p=0.424), tumor stage (p=0.797), prescription dose (p=0.349), concurrent chemotherapy (p=0.175), conditions of hypertension (p=0.987), diabetes (p=0.227), dental extraction (p=0.582), smoking history (p=0.611), or current smoker (p=0.245) between the ORN and control group were observed in the overall patient cohort (Supplemental Table 2 column 2).

For the clinical factor comparison of the ORN vs. control group in the patients either treated with VMAT or PBSPT, no statistically significant associations of age, gender, tumor stage, prescription dose, chemotherapy, hypertension, diabetes, dental extraction, or current smoker to ORN were observed (Supplemental Table 2 column 3 and 4). Tests showed that smoking history was significantly associated with the ORN occurrence in patients treated with PBSPT only (p=0.015), but not in patients treated with VMAT (p=0.527). Detailed analysis regarding these risk factors to the ORN incidence will be reported in a separate study (31).

For the clinical factor comparison of the patients treated with PBSPT vs. VMAT in either the ORN group or the control group, no aforementioned clinical factors were found to be significantly different in the ORN group (Supplemental Table 3 column 3). Tumor stage, hypertension, and smoking history were observed to be significantly different between VMAT and PBSPT in the control group (p<0.05) (Supplemental Table 3 column 4) and thus the overall patient cohort (Supplemental Table 3 column 2).

For the case-matched patient subset cohort between the two modalities, all of 335 PBSPT patients were included in this subset cohort, including 9 ORN patients. In the case-matched photon group of 335 patients, 8 patients experienced osteoradionecrosis. After the case matching, all clinical factors were balanced between VMAT and PBSPT in both the ORN and control groups (p>0.05, Supplemental Table 4). Similar to the results based on the entire patient cohort, no clinical factors showed significant difference between the ORN and control group for both modalities, except smoking history between the ORN and control group in PBSPT patients (p=0.046) (Supplemental Table 5). All the dosimetric analysis below were based on this case-matched patient subset cohort.

For all patients, and patients in each individual modality, V40Gy[RBE], V50Gy[RBE], and V60Gy[RBE] were statistically significantly different between the ORN group and control group in all comparisons (V40Gy[RBE]: p=0.007 and p=0.004; V50Gy[RBE]: p=0.004 and p=0.002; V60Gy[RBE]: p=0.014 and p=0.003 [VMAT and PBSPT], Table 2). The DVH indices of V70Gy[RBE], V75Gy[RBE], Dmean, and D0.01cc are not considered in the following analysis since they are not statistically significantly different in some or all comparisons.

Figure 2 A shows the axial (top rows) and sagittal view (bottom rows) of the dose distributions from typical patients with (1^st and 3^rd column) and without ORN (2^nd and 4^th column) from VMAT (left two columns) and PBSPT (right two columns) (one patient for each modality). The ORN region is contoured in pink. Note that these injury regions of patients treated with PBSPT were typically located distal to the edge of the high-risk CTV (CTV_High, cyan) in the beam direction.

Figure 3 shows the box plots of V40Gy[RBE], V50Gy[RBE], V60Gy[RBE], V70Gy[RBE] and V75Gy[RBE] between the patients treated with VMAT and PBSPT overall (Figure 3 A) and between the ORN and control group in VMAT and PBSPT, respectively (Figure 3 B). PBSPT showed better sparing of the mandible than VMAT as suggested by lower dose volumes of V40Gy[RBE], V50Gy[RBE], and V60Gy[RBE] (V40Gy[RBE]: 30.94 cc vs. 15.25 cc; V50Gy[RBE]: 20.44 cc vs. 9.25 cc; V60Gy[RBE]: 8.68 cc vs. 2.12 cc [VMAT vs PBSPT, medians]) (Figure 3 A). Despite having improved sparing of the mandible in PBSPT compared to VMAT, the proton group did not show obvious reduction in the incidence rate of ORN. The critical volumes of the derived mandible DVCs in PBSPT were lower than those of the derived mandible DVCs in VMAT for V40Gy[RBE], V50Gy[RBE], and V60Gy[RBE] (VMAT: V40Gy[RBE]: 38.96 cc (95% CI: 32.74-52.49 cc); V50Gy[RBE]: 31.85 cc (95% CI: 25.49-43.39 cc); V60Gy[RBE]: 17.61 cc (95% CI: 9.49-19.35 cc); PBSPT: V40Gy[RBE]: 21.26 cc (95% CI: 19.75-29.44 cc); V50Gy[RBE]: 16.42 cc (95% CI: 13.95-19.63 cc); V60Gy[RBE]: 3.95 cc (95% CI: 2.19-10.82 cc)) (Figure 3 C and Table 3). High AUCs, high TPRs, and low FPRs of V40Gy[RBE], V50Gy[RBE], and V60Gy[RBE] from the derived DVCs were observed (Table 3).

Figure 2 B shows the axial (top rows) and sagittal view (bottom rows) of the LET distributions from typical patients with (left) and without ORN (right) treated with PBSPT. Elevated LET distribution appears to correlate with the injury sites with moderate doses (above 40 Gy[RBE=1.1]).

Empirical RBEs were calculated by comparing the volume tolerance curves between PBSPT and VMAT (Figure 3 C) via equivalent constraint dose analysis. Empirical RBEs of 1.58 (95% CI: 1.34-1.64), 1.34 (95% CI: 1.23-1.40), and 1.24 (95% CI: 1.15-1.26) were obtained at a proton dose of 40 Gy[RBE=1.1], 50 Gy[RBE=1.1], and 60 Gy[RBE=1.1], respectively (Table 3). The empirical RBEs decreased with the increase of the physical dose in PBSPT.

Figures 1A, B show DLVHs of typical PBSPT patients with and without mandible ORN, respectively. More voxels of high LETs above 5 keV/µm were observed in the ORN patients (above a dose of 40 Gy[RBE=1.1]) than those in the control patients (comparing Figure 1 A to Figure 1 B). Both DLVHs showed a decrease of the LET top edges at each dose bin when the physical dose increased from 40 to 60 Gy[RBE=1.1]. However, a generally higher LET distribution was observed in the ORN patients than that in the control patients (~8 keV/µm to ~4 keV/µm in the ORN group and ~6 keV/µm to ~3 keV/µm in the control group when the physical dose was increased from 40 to 60 Gy[RBE=1.1], as shown by comparing red ovals in Figures 1A, B).