Patients were simulated in a supine position with their hands on their chest using a leg immobilizer for reproducibility. Bladder filling instructions were provided by the radiation therapists (400–600 mL of water) to ensure a reasonably full bladder. In addition, patients were required to empty their bowels before CT stimulation and daily treatment. CT scans were acquired with a 2.5-mm-slice thickness (120 kVp, GE LightSpeed RT 16). Patients underwent mpMRI scans either on the same day as the CT simulation or the day after. mpMRI datasets consisting of T1-weighted (T1W), T2W, and diffusion-weighted imaging (DWI) scan sequences were acquired using the MAGNETOM Avanto (Siemens AG, Munich, Germany) 1.5 T scanner.

DILs were identified and reported as per Prostate Imaging Reporting and Data System (PI-RADS) version 2.0 criteria (16) by the radiologist. Additional annotations were made with regard to the location and extension of the DILs on the radiological scans to facilitate radiation oncologists (RO) contouring. An experienced RO delineated the DILs on the planned CT scan after rigid registration (focusing on the prostate gland) was performed between the scans. DILs defined as the gross tumor volume were delineated with reference to the low intensity on T2W and hypointense on apparent diffusion coefficient (ADC) maps (17) ( Figure 1 ). PTV_dil was generated from the DIL contour with a 2-mm expansion margin. No expansion margin was applied for DILs that were close to the urethra or rectum. The prostate and urethra were also contoured using the T2W MRI.

The targets were defined as follows: clinical target volume with proximal seminal vesicles (CTVpsv): prostate plus proximal 2 cm or entire sv, depending on the level of sv involvement based on the MRI staging report (18). Planning target volume (PTV) psv: uniform 5 mm expansion margin from CTVpsv. CTVpln: L5-S1 junction or mid L5, encompassing bilateral common iliac, external iliac, internal iliac, presacral, and obturator nodes. PTVpln: uniform expansion of 7 mm from CTVpln. All CTVs and OARs were defined in accordance with the Radiation Therapy Oncology Group (RTOG) consensus guideline ( Figure 2 ) (19). Target optimization for VMAT was based on PTV, while for IMPT, CTV was used. The prescribed dose was planned to deliver 60 Gy in 20 fractions to the CTVpsv/PTVpsv and 46 Gy to the CTVpln/PTVpln using a SIB technique. DV constraints were applied as per departmental guidelines ( Supplementary Table S1 ), as well as an additional mandatory constraint for the urethra. As a constant relative biological effectiveness (RBE) of 1.1 was applied for the proton dose prescription, the term GyRBE was used in proton dose reporting.

To maintain plan consistency, two plans with DIL boost using IMPT and VMAT were generated for each patient by the same dosimetrist. A total of twenty plans (10 IMPT and 10 VMAT) were available for analysis. Both IMPT and VMAT were created using Raystation planning system 10A (RaySearch Laboratories AB, Stockholm, Sweden), and the Monte Carlo algorithm was used for dose calculation. All treatment plans were optimized based on the stated clinical goals in Supplementary Table S1 (12, 20). For VMAT planning, two full arcs (clockwise to anti-clockwise rotations) using 10 MV energy with the collimator tilt at 10°–20° were utilized. The target coverage for VMAT was to have 95% of the PTV covering 95% of the prescription dose (PD).

IMPT using a multifield optimization (MFO) technique was utilized to facilitate dose escalation to the DILs. A proton beam arrangement consisting of two lateral beams and one posterior beam was employed. A central blocking structure was created to prevent spot placement in the contralateral PLN. A posterior block was included to minimize spot placement at the rectal region. IMPT was robustly optimized using a 5-mm setup and a 3.5% range of uncertainties to achieve the target coverage of 98% of the CTV, encompassing 98% of the PD. For proton planning in general, the target coverage requirement was more stringent than VMAT as the target assessment is based on CTV. The 5-mm setup uncertainty used in proton planning coincides with the CTV-to-PTV setup and organ motion expansion margins utilized in the photon environment. The additional 3.5% range uncertainty is necessary for proton planning as the proton beam is sensitive to the heterogeneity of the tissue transverse path (21). From a biological perspective, a robustly optimized treatment plan ensures that the CTV receives a consistent dose and mitigates the impact of uncertainties on the OARs in the simulated scenarios, thereby balancing the achieved TCP and NTCP. DV metrics were extracted based on the planning constraints for the target and OARs.

A linear-quadratic (LQ)-Poisson model was used in this study ( Supplementary Material , Eqs. A.1–4). The TCP parameters were adopted from a study conducted by Sachpazidis et al. (22) on 129 intermediate-HR-PCa patients with a median follow-up of 81.4 months and a biochemical relapse rate of 18%. TCP values were generated separately for the prostate minus DIL (prostate-DIL) and DIL due to the use of different model parameters, assuming the difference in tumor cells’ aggressiveness (6, 23). The low and very low α/β of 3 and 1.5 Gy were used to evaluate the different tumor sensitivity, as the exact value has yet to be determined ( Supplementary Table S2 ) (24). A Lyman–Kutcher–Burman normal tissue complication probability (LKB-NTCP) model with an equivalent uniform dose (EUD) was employed for OAR estimation. Details on the generation of D _A and D _P and model formulation have previously been described (25, 26). LKB-NTCP parameters for the rectum and bladder were adopted from our previous work using D _A and D _P to fit the prospectively collated toxicity data for grade ≥ 2 based on 3-year post-RT follow-up time points. The maximum likelihood estimation method was used to estimate the best-fit parameters for LKB-NTCP (27). Additional rectum and bladder parameters obtained from published work (27–30) were also included for comparison ( Supplementary Table S3 ). Similar to TCP calculations, extracted DV metrics were converted to equivalent doses at 2 Gy per fraction using the LQ model prior to NTCP analysis.

Data normality was tested using the Shapiro–Wilk test. A paired t-test was used for DV analysis between IMPT and VMAT and reported as mean and standard deviations (SDs) as the data were normally distributed. A Wilcoxon signed-rank test was used for analyzing TCP and NTCP variables and documented as median and interquartile ranges (IQR) as the data were not normally distributed. Descriptive statistics were presented as mean and SDs as appropriate. Statistical analysis was conducted using Matlab (MathWorks, Inc., Natick, MA, USA, version 8.0) and IBM SPSS statistics (version 26.0, SPSS, Inc., Chicago, IL, USA). Two-sided p-values of < 0.05 were considered statistically significant.

For DIL coverage, IMPT achieved on average a Dmean of 68.1 Gy, while VMAT obtained a slightly lower Dmean of 66.6 Gy (p < 0.05). The difference in mean dose between IMPT and VMAT was 1.4 Gy (p < 0.05). The maximum level of dose escalation achievable for the two modalities was equivalent, as the constraints for the dose-limiting structures for the urethra PRV 2 mm at D0.5 cm³ and rectum V60 Gy were not statistically significant ( Table 2 ). For TCP modeling, IMPT generated a higher average median TCP with a narrower percentage range for prostate-DIL and DIL compared to VMAT ( Table 3 ; Figures 3A, B ).

For the rectum, dose ranges from V24 to V56 Gy was lower in IMPT compared to VMAT (p < 0.05). At the very high dose region, the dose to the rectum was not statistically significant between the two modalities (V60 Gy, p = 0.08). A substantial dose reduction at the low–intermediate-dose range in IMPT, with differences in mean dose at V24 Gy (IMPT minus VMAT: 36.4%, 9.3 Gy) and at V32 Gy (IMPT minus VMAT: 20%, 6.4 Gy), was observed ( Table 2 ). In NTCP modeling, the rectal n value reported in Quantec (QT) and the derived n values from D _P and D _A ranging from 0.2 to 0.9 suggested an increased risk of high dose to a small rectal volume on GI toxicity ( Table 3 ; Figure 3C ). Rectal D _A-NTCP displayed the highest grade ≥ 2 toxicity risk (IMPT: 5.6% vs. VMAT: 5.5%), followed by QT-NTCP (IMPT: 4.2% vs. VMAT: 4.4%), with D _P-NTCP having the lowest risk estimation (IMPT and VMAT: 2.7%). For the bladder, apart from V60 Gy, a dose reduction was observed in the V40–48 Gy region in IMPT compared to VMAT (p < 0.05). The average bladder Dmean for IMPT was 14.9% (3 Gy) lower compared to the VMAT (p < 0.05) ( Table 2 ). Bladder NTCP was not significant from Burman-generated parameters for both modalities. D _P-NTCP for IMPT was lower compared to VMAT (IMPT: 10.4% vs. VMAT: 12.5%, p < 0.05) ( Table 3 ; Figure 3D ).

Dose to the hottest 0.03 cm³ of the urethra and D0.5 cm³ of the urethra PRV 2 mm for IMPT and VMAT were comparable (p = 0.33) as urethra was considered hard constraints during plan optimization. Similarly, NTCP for IMPT and VMAT were low and not statistically significant. An average mean dose to the femoral head for IMPT and VMAT had no statistically significant differences (p = 0.39) ( Table 2 ). No result was generated for the NTCP for the femoral heads as the overall irradiated dose to the femoral heads was low.

IMPT met the required CTV coverage of D95% > 95% at worst-case scenarios (CTVpsv; median: 59.46 GyRBE, CTVpln: 45.3 GyRBE). Thus, the plans were robust enough to be treated clinically ( Supplementary Table S4 ). This study was designed to perform a safe dose escalation using IMPT and VMAT by setting a mandatory constraint on the surrounding OARs. Notably, urethra PRV 2 mm (D0.5 cm³ < 64 Gy) and rectum (V60 Gy < 5%) were the main dose-limiting constraints. A similar strategy was adopted in the PIVOTALboost (ISRCTN80146950) trial, whereby the dose of the DIL was limited to 70 Gy over 20 fractions (20). Both IMPT and VMAT were able to achieve high TCP results for prostate-DIL and DILs, with IMPT obtaining a moderately higher TCP compared to VMAT (p < 0.05) ( Table 3 ). As the DILs are regarded as high-risk targets, correlating to the likelihood of disease relapse, escalating the dose to an estimated 11.7% above the prescription dose could potentially attain a higher disease control rate of 97% (α/β: 3 Gy) and 99.3% (α/β: 1.5 Gy) for IMPT.

The higher average median TCP achieved for IMPT (p < 0.05) remained consistently high for all the plans compared to VMAT for the α/β ratio of 1.5 Gy and 3 Gy ( Figure 3 ). This could be attributed to the optimal dosimetric properties of the proton beam, which have the ability to create a sharp dose gradient at the distal end of the beam (32). Similar approaches have been demonstrated by Uzan et al. (33) on 11 HR-PCa patients using conventional fractions (74 Gy in 37 fractions), treating only the prostate gland with intensity-modulated radiotherapy. By limiting the DILs to 86 Gy (BED: 219 Gy), while maintaining the rectal NTCP to 5%–6%, an increase in TCP from 71% to 83.6% was observed, with no grade 3 events reported at 36 months post-RT. Conversely, some studies adopted a “maximum achievable dose” escalation strategy by using a step-by-step dose-incremental approach to enhance the dose to the DILs while keeping the NTCP at an acceptable level (32). However, the use of this approach must be carefully balanced between potential increase in TCP and toxicity risk, as the urethra was not routinely contoured and considered a dose-limiting organ in some studies (13, 20).

For the rectal DV metrics, V60 Gy for both modalities was similar due to the close proximity of this organ to the target. However, as the proton beam deposited virtually no dose beyond the target, significant sparing of the rectum was observed in the low-intermediate dose region (8). The increase in linear energy transfer (LET) at the distal edge of the proton beam has the potential to cause an increase in toxicity (34). The use of two laterally opposed beams and the modulation of beam energy and intensity during plan optimization were strategies used to mitigate the risk of increased toxicity at the distal edge (35).

It is still unclear if the extensive reduction in the low-intermediate dose will correspond to a lower risk of GI toxicity. However, a multi-institutional prospective trial database review of a subset of PCa patients with PLN irradiation treated using proton therapy reported low rates of acute (grade 2: 2%, grade 3: 0) and late (grade 2: 2%, grade 3: 1%) GI toxicity (36). The reduction in dose received by the rectum at the high dose range in IMPT compared to VMAT (V56 Gy and V52 Gy, p < 0.05) could potentially be translated into a lower risk of patients having grade ≥2 GI toxicity (27).

For rectal NTCP modeling, Cicchetti et al. (37) reported on late rectal bleeding for 1,633 prostate cancer patients with a minimum of 3-year follow-up using various EUD-based NTCP parameters and concluded that higher n values of 0.18 and 0.24 better predict late grades ≥2 and 3 GI toxicity. The study demonstrated that the rectum might not behave as predominantly serial-like but has some parallel-like properties, whereby the intermediate–high-dose region might increase the risk of GI toxicity. The use of IMPT in this study demonstrates a reduction in rectal dose for the intermediate–high-dose range (V24–56 Gy, p < 0.05) and thus has the potential to minimize the risk of GI toxicity. The risk of having grade ≥2 GI toxicity is generally low, whereby both IMPT and VMAT-derived D _A-NTCP values achieved comparable toxicity risk estimations as a mandatory rectal V60 Gy < 5% constraint was applied. In considering rectal D _A-NTCP, there was a 2.9% and 2.8% increase in the risk of predicting grade ≥2 GI toxicity compared to D _P-NTCP for IMPT and VMAT, respectively. This provides a good reference guide for clinicians when devising dose-escalation protocols and selecting treatment modalities, as interfraction rectal motion should be considered because the impact on the risk of toxicity could be more pronounced (27).

For the bladder DV analysis, a significant reduction in the average mean dose was observed in IMPT, which could potentially be translated into a reduction in acute and late GU toxicity (36). Choo et al. (12) reported acceptable late GU toxicity in their prospective trial, with none of the 56 patients experiencing grade 3 GU toxicity with PLN treated with proton therapy. For NTCP modeling, given the similar D ₅₀ and m values from D _P- and D _A-generated biological parameters, statistical significance was observed between IMPT and VMAT. This could imply that the n value for the bladder could be closer to 0.14, indicating the serial-parallel-like behavior of the bladder to radiation (27).

Due to the high incidences of reported late GU toxicity as well as urethra strictures, recent randomized trials have started to contour and include urethra constraints to minimize toxicity (32). The availability of the mpMRI scans at a similar time point as the planned CT scans in this study allows for accurate contouring of the urethra. An additional urethra 2 mm PRV margin with a mandatory D0.5 cm³ < 64 Gy dose limit further reduced the high-dose spillage to the urethra, especially for DILs that were located close to the urethra (20). In line with the objective of safe-dose escalation, after applying the mandatory urethra dose limit, both IMPT and VMAT achieved a low NTCP of 4.3%. A moderately higher DIL mean dose was associated with the risk of urethra stricture with IMPT compared to VMAT (IMPT: 68.1 GyRBE vs. VMAT: 66.6 Gy, p < 0.05).

Despite the use of two lateral-opposed beams in IMPT, the average Dmean for the femoral heads was low and comparable to VMAT (p = 0.39). Conversely, a study reported by Whitaker et al. (38), comparing the dosimetric difference between IMPT and VMAT on PCa patients with PLN treatment, obtained a higher femoral head Dmean for IMPT (IMPT: 18.2 GyRBE vs. VMAT: 14.8 Gy, p < 0.05). The lower femoral head Dmean achieved in this study could be due to the use of an additional posterior proton beam. The concern of introducing a posterior beam due to the presence of rectal uncertainties was addressed by adding a blocking structure to prevent spot placement around the rectal region. This technique has effectively reduced the femoral head doses even with DIL boosting (32). NTCP for a femoral head was not significant as the dose received was very low.

Firstly, this study only included 10 patients. This study explored our initial experience by comparing photon and proton planning techniques, including the application of biological indices to predict potential treatment outcomes and toxicity risks for patients with HR-PCa with PLN irradiation. The generalizability of these results needs to be confirmed in a larger cohort study. Other clinical aspects such as disease-free survival, overall survival, and toxicity rates based on patient-reported outcomes can be incorporated into the design of the risk-benefit profile in a prospective manner as an extension of this study. Secondly, although biological modeling provides practical metrics for plan comparison and optimization, the results might not be a true representation of actual clinical outcomes (39). The accuracy of TCP/NTCP models is highly dependent on the selection of models, the derived parameters, and the associated clinical endpoints (40). To note, the majority of the parameters (Quantec and Burman) used in NTCP modeling were mainly based on three-dimensional conformal techniques. To minimize these uncertainties, well-established TCP and NTCP models and parameters were selected in this study. Moreover, the addition of previously developed NTCP parameters for the rectum and bladder based on similar patient cohorts and dose distribution improves the reliability of GI and GU toxicity estimates (27). Furthermore, taking into consideration the impact of bladder and rectal motion in these parameters increases the accuracy of toxicity predictions, which has not been reported in the literature. Lastly, the reliability of using TCP/NTCP models on proton therapy has been under ongoing investigation due to the increased application of proton therapy to more disease subsites and the availability of more mature published clinical data (41). Currently, the clinical application of proton therapy assumes a constant RBE of 1.1, which will have minimal impact on the α/β, and hence the dose–response curve remains comparable for both photons and protons (42). Further research is still required in this aspect to fully tap into the benefits of incorporating biological modeling in clinical decision-making.

As half of the study cohort had two DILs, this analysis was able to address the challenging aspects of attempting to achieve an escalated dose of the DILs while meeting the mandatory constraints for the OARs. As IMPT is able to offer a highly conformal dose distribution with excellent dose fall-off properties, all the planned cases have achieved a higher minimum TCP with comparable NTCP to VMAT. Lastly, the incorporation of motion-inclusive NTCP parameters derived from similar patient cohorts further improves the toxicity predictions. Moving forward, apart from increasing the sample size to enhance the reliability of the results, more work is still needed to investigate the utilization of biological models in determining clinical outcomes, especially in the field of proton therapy.

In conclusion, this preliminary study has demonstrated that dose escalation to the DILs using IMPT is able to achieve a higher TCP while keeping the NTCP comparable to the VMAT. Rectal D _A-NTCP displayed a greater predicted GI toxicity compared to D _P- and QT-derived parameters. In contrast, bladder D _A-NTCP has a slightly lower predicted toxicity compared to D _P-NTCP. The inclusion of interfraction organ motion improves the reliability of the predicted toxicity.