Prostate external beam radiotherapy combined with high-dose-rate brachytherapy: dose-volume parameters from deformably-registered plans correlate with late gastrointestinal complications

This study included 118 prostate cancer patients (tumour T stage ≥ 2a) who were treated with EBRT followed by HDR at Sir Charles Gairdner Hospital in the period 2004 to 2008. These patients were treated as part of the Trans-Tasman Radiation Oncology Group (TROG) 03.04 Randomized Androgen Deprivation and Radiotherapy (RADAR) trial [14, 15]. The patient criteria and treatment methodology for the RADAR trial have previously been detailed [14, 15]. Aspects of the combined EBRT/HDR treatment process have previously been described [16]. The four-field EBRT plans for a prescription dose of 46 Gy in 23 daily fractions were created in the Elekta XiO treatment planning system (Elekta AB, Stockholm, Sweden). The HDR plans for a prescription dose of 19.5 Gy in 3 fractions across 2 days were created in the BrachyVision treatment planning system (Varian Medical Systems, Palo Alto, US) using the TG43 formalism [17]. Additional file 1 (Supplement A) provides additional patient and treatment details.

The external wall of the rectum was manually delineated by treating-clinicians in the HDR CTs using BrachyVision and in the EBRT CTs using the Elekta Focal contouring system (Elekta AB, Stockholm, Sweden). Author with initials MK reviewed rectum outlines for consistency between patients. For rectum contouring, the inferior-superior limits of the rectum were the rectosigmoid flexure and the last slice where the ischial tuberosities were visible. Patients did not commonly require bowel preparation. Examples of the planning CTs and structures for EBRT and HDR TG43 physical dose plans are provided in Additional file 1 (Supplement B, Figures A1 and A2).

Patients were assessed for various gastrointestinal toxicities at baseline (randomisation) and subsequent time points after randomisation. The median of the most recent patient follow-ups was 72 months (range 12–96 months). The Late Effects of Normal Tissue — Subjective, Objective, Management, and Analytic (LENT SOMA) scales were used to assess rectal bleeding, urgency and tenesmus, stool frequency, diarrhoea, anorectal pain and completeness of evacuation [18]. Proctitis was scored by clinicians according to the Common Toxicity Criteria for Adverse Events (CTCAE version 2) [19]. Additional file 1 (Supplement C, Table A2) provides a summary of the grading systems.

Late peak toxicity was calculated for the period from 3 months after radiation therapy and onwards. Figure 1 provides a summary of the late peak toxicity event rates for the follow-up period. Patients were classified to a toxicity group if the late peak toxicity was at least a certain grade (threshold for dichotomisation). In the interest of modelling a moderate severity of toxicity the threshold was grade 2 for rectal bleeding, stool frequency and completeness of evacuation. The threshold was grade 1 for diarrhoea, anorectal pain, proctitis, urgency and tenesmus due to low toxicity rates for grade ≥ 2 toxicity and/or a lack of significance for grade ≥ 2 toxicity. Alternatively, the chosen thresholds for toxicities are indicated in Fig. 1. The analysis was repeated using the prevalence of toxicity at 36 months post-randomisation. This did not reveal any additional trends for dose-histogram effects and so is reported no further.

The HDR CT was registered to the EBRT CT using rigid registration followed by a B-splines multi-pass DIR in Velocity Advanced Imaging 2.8.1 (Varian Medical Systems, Palo Alto, US) [20]. The registration process has been described in detail previously [20, 21]. Visual inspections for the 118 patients were undertaken by authors (initials CRM, VL and CIT) and did not identify any major registration misalignments (e.g. Additional file 1 [Supplement D, Figure A3]). The registrations were quantitatively evaluated for each patient in this study using the overlap of the EBRT rectum and registered HDR rectum (expressed as a percentage of the volume of the registered HDR rectum). As illustrated in Fig. 2, the median overlap is 80.4 % for alignment of EBRT/registered HDR rectum structure volumes. A general structure overlap of 70 % is considered to be the starting point for satisfactory structure-correspondence in the radiotherapy context [22, 23]. The registrations have also previously been extensively evaluated using structure-correspondence metrics, image similarity metrics and qualitative visual inspection by authors (initials CRM, VL and CIT) [21].

The EBRT and registered HDR 3D-doses were imported into MATLAB™ R2010a (The MathWorks Inc., Massachusetts, US) and the Computational Environment for Radiotherapy Research (CERR, version 4.1) [24]. The voxel doses were converted to equieffective doses given in a reference 2 Gy per fraction using the linear-quadratic model [11] with an alpha-beta ratio of 3 Gy for the rectum [2]. The analysis was also performed for an alpha-beta ratio of 5.4 Gy to check the sensitivity of results to the upper limit published for the rectum [25]. The EBRT dose was summed voxel-by-voxel with the registered HDR dose (i.e. accumulated). The rectum dose-volume histograms (DVH) in 1 Gy bins from 1 to 80 Gy were extracted for the total registered dose (with EBRT rectum structure), the unregistered EBRT dose (with EBRT rectum structure) and the unregistered HDR dose (with the HDR rectum structure). The parameters extracted from the rectum DVHs were the V_X (percentage of the rectal volume receiving at least X Gy after applying an alpha-beta ratio) and D_X% (minimum dose to the most irradiated X percent of the rectal volume after applying an alpha-beta ratio). The V_X and D_X% were calculated using the total registered dose and the EBRT rectum structure (‘distribution-adding’). Additionally, the D_X% was alternatively calculated by adding the EBRT D_X% to the unregistered HDR D_X% using the corresponding rectum structures (‘parameter-adding’).

For each type of toxicity, univariate logistic regression was applied at each V_X to obtain an odds ratio (OR) for the increase in toxicity probability per 5 % absolute increase in volume [26]. 95 % confidence intervals (CI) for odds ratios were calculated using bootstrapping with 10,000 resamples from the toxicity and no toxicity groups. Odds ratios were considered significant if the 95 % CIs did not include the value of one. Mann-Whitney U-tests were used to determine whether the median V_X (or D_X%) values for the toxicity and no toxicity groups were significantly different (p-value < 0.05). This analysis was performed in MATLAB™ R2010a (The MathWorks Inc., Massachusetts, US).

Clinical risk factors were not included in dose-response modelling as a previously published analysis determined that clinical covariates did not significantly influence late toxicities for patients in the RADAR trial [27]. An equivalent analysis for the 118 patients in this study confirmed that clinical covariates did not significantly influence late toxicities. The clinical factors considered were age, tumour T stage, Gleason score, initial PSA, risk group, number of HDR catheters, colorectal disorders, hypertension, diabetes, smoking, use of statins, ACE inhibitors and anti-coagulants.

This study is the first to use registration-based distribution-addition to obtain accumulated rectum dose-histogram parameters for combined EBRT/HDR prostate cancer treatment and to then correlate the resulting parameters with gastrointestinal toxicities. Studies have estimated the accumulated rectum dose for combined EBRT/HDR prostate cancer treatment without applying deformable registration [3, 7]. However, Kikuchi et al. [12] acknowledged that deformable image registration should be part of a more accurate method of accumulating the rectum dose. This current study improved upon these studies by applying deformable image registration and then correlating accumulated rectum dose with various gastrointestinal toxicities. This study acknowledges the uncertainties of deformable image registration. Subsequently, it compares the findings for distribution-adding with the findings for parameter-adding. Given the potential uncertainties of deformable registration it is important for registration-based dose-toxicity modelling to be published for a variety of studies. This would allow a multi-institutional comparison of findings to include the confounding factors associated with different registration algorithms, registration circumstances, associated inter-fraction motion constraints and diversity in treatment techniques.

The volume receiving certain doses and the magnitude of dose delivered to volumes have been associated with late gastrointestinal toxicities, typically rectal bleeding, scored after a number of prostate radiotherapy techniques including EBRT only, HDR only, low-dose-rate brachytherapy (LDR) only, combined EBRT/HDR and combined EBRT/LDR [2‐4, 26, 28‐45]. Table 3 summarises the important dose-response findings for the previously mentioned studies and the findings for this study.

The previously mentioned studies commonly suggested that the incidence of late rectal bleeding following prostate radiotherapy can be reduced by constraining the volume of the rectum receiving high doses (e.g. [4, 28, 42, 43, 46]). Additionally, some of the studies have correlated the mid and low-mid dose regions with late rectal bleeding [28, 30, 32, 33] and stool frequency/urgency/tenesmus [26, 30, 32, 37] respectively. Consequently, the mid-high rectum doses in prostate EBRT are typically managed through constraints on the V_{40-75 Gy} [28, 46] whereas treatments involving prostate brachytherapy (HDR or LDR) should consider the high rectum doses via the V_70%-100%, D_1cc, D_2cc and/or near maximum dose [4, 47‐49] due to high-dose hot spots associated with radioactive sources. For treatments involving prostate HDR, the importance of low-dose regions has been explained in terms of considerable inter-patient variation in rectal gas and the distance from the prostate to the anterior rectal wall [45]. The instances where the V_{80 Gy} and V_{90 Gy} have been identified as important for prostate cancer treatments involving EBRT only were related to homogenous irradiation of volumes with hypofractionated doses [45]. The sections to follow will discuss the findings of this study, summarised in Table 3, relative to findings of the previously mentioned dose-response studies, which are also summarised in Table 3.

In agreement with other studies [2, 28‐30] the high-dose metrics for the rectum were significantly correlated with rectal bleeding for both distribution-adding and parameter-adding. The significant correlation between near maximum dose metrics for the rectum and rectal bleeding indicates the dose-volume effects follow a serial response. The confirmation of the expected importance of near maximum doses after registration is important as a previous study without registration did not find any significance for near maximum doses [3] and the GEC/ESTRO recommendation is to limit the D_2cc to 75 Gy [4]. Additionally, the identified seriality is consistent with the suggestion rectal bleeding is associated with epithelial damage and mucositis as a result of exposure of parts of the rectal wall to near maximum doses [31].

Studies have also demonstrated that the mid-dose region (>30 Gy) is important for rectal bleeding [28, 30, 32, 33]. In this study the importance is shifted to relatively higher doses in the mid-high dose range for both distribution-adding and parameter-adding. An influencing factor for the lack of importance of the lower end of the mid-dose range could be that the combined EBRT/HDR treatments were subject to the constraint that the maximum rectum dose from HDR should not exceed 80 % of the 19.5 Gy prescription dose for HDR. Consequently, in the context of the total EBRT/HDR dose this constraint effectively applies more to the lower end of the mid-dose range after adjusting for dose fractionation than it does to the high-dose region. The importance of the high-dose and near maximum dose regions could also be related to the steepness of dose gradients associated with HDR treatments as it has been proposed that a focused high-dose region could aid healing of the vascular sclerosis in high-dose regions via cell migration from the low-dose region [50]. Consequently, it would be useful to determine optimal rectum dose constraints for combined EBRT/HDR based on accumulated dose. A larger sample size containing patients from a variety of institutions would allow for a feasible application of multivariate and cut-point analysis.

The upper end of the mid-high dose range after distribution-adding was important for the non-bleeding toxicities of stool frequency and urgency/tenesmus. This result could support the earlier suggestion that the rectum dose constraint for HDR effectively applies a constraint to the lower end of the mid-dose range when the total EBRT/HDR dose is considered. The dose constraint in one dose region leading to other dose regions becoming important is consistent with a previous study focusing on patients within this trial who received EBRT only [30]. The study indicated the low-mid dose range was important for stool frequency, urgency and tenesmus in the presence of high-dose constraints [30]. More optimised dose constraints for the mid-high dose range based on accumulated dose could be useful for reducing the toxicities associated with these doses. Such constraints could be relatively more important for urgency/tenesmus compared to stool frequency given the higher toxicity rate in this patient sample compared to the other toxicities.

The association of urgency/tenesmus with distribution-adding doses at the lower end of the mid-dose range is consistent with the finding from another study where violation of the V_{40 Gy} dose constraint was important for urgency [26]. Additionally, the results indicate the lower end of the mid-dose range and the low doses may be associated with anorectal pain and rectal bleeding respectively. The correlation of toxicities with low doses and the lower end of the mid-dose range is possible as it is plausible that a low-dose bath to a large volume will be associated with detriment. However, these findings of association should be considered with respect to toxicity event rates, sample size and the potential of random discovery.

Analysis based on contouring and registration is associated with uncertainties. However, the dose regions indicated as being important for toxicity after distribution-adding were in most cases consistent with those indicated as important after parameter-adding. The low-mid dose range for parameter-adding was significantly associated with proctitis and stool frequency. In contrast, these regions after distribution-adding were not identified as important. However, distribution-adding did indicate additional regions as important compared to regions identified by the parameter-adding results. For example, the analysis for parameter-adding did not indicate any significant dose regions for diarrhoea and urgency/tenesmus whereas analysis after distribution-adding indicated the mid-high dose range was important. The alpha-beta ratio is an additional uncertainty for diarrhoea correlations as the mid-high dose range was only important for an alpha-beta ratio of 3 Gy. Further studies for a variety of contouring and registration contexts would be useful for gathering data for the purpose of determining whether registration and distribution-adding reveals correlations which were not identified by parameter-adding.

When considering the distribution-adding findings in isolation it should be noted that errors in contouring and registration accuracy will confound the distribution-adding parameters that have been correlated with toxicity. A median overlap of 80.4 % for the rectum volume correspondence across all patients would indicate the registrations are satisfactory as a general structure overlap of 70 % is considered to be the starting point for satisfactory structure-correspondence in the radiotherapy context [22, 23]. The proximity of the HDR rectum to the HDR catheters makes parameters obtained by registration and distribution-adding sensitive to small localised variations in contouring and/or registration accuracy across the prostate/rectum interface. Given this is the first study to accumulate the rectum dose for combined EBRT/HDR prostate cancer treatments using deformable registration and then correlate the doses with toxicity, we encourage more prostate cancer studies to assess the importance of dose-volume metrics using a variety of registration algorithms. Software developers and treatment planning vendors have the opportunity to greatly improve planning and the reliability of dose-toxicity modelling after registration by improving contouring and registration accuracy for the prostate/rectum interface.

A common uncertainty associated with EBRT, HDR and other radiotherapy techniques is inter-fraction motion of patient anatomy [51‐54]. In response to this uncertainty, it is becoming more common for institutions to adopt repeat imaging over the course of prostate cancer treatment to correct for inter-fraction motions and improve the correspondence between planned dose and delivered dose [53, 55]. However, many studies including this study do not contain repeat imaging due to the retrospective nature of studies where long follow-up is required to correlate dose with late toxicities. Consequently, studies are constrained by treatments performed in the past with the associated resources and protocols at that time.

A consideration for prostate EBRT in this study is rectum motion and variable rectum contents confounding the accuracy of rectum dose distributions obtained from single static planning CTs [52, 56, 57]. A variety of methods have been used to estimate the impact of inter-fraction motion on rectum dose parameters and dose-response modelling for prostate EBRT [8, 52, 58‐61]. To obtain appropriate mean estimates for the difference in EBRT rectum dose between the single CT based values and the motion-corrected values this study analysed the results of another study [52] that used the same registration software and registration algorithm. Consequently, compared to the motion-corrected values the single CT based values may be conservative estimates by 3.9 % for the D_2% and 5.8 % for the equivalent uniform dose [52].

An important consideration for inter-fraction motion during prostate brachytherapy is the movement of the anterior rectal wall relative to the prostate [54, 62]. A variety of methods have been used to estimate inter-fraction motion in prostate HDR [16, 54] and the subsequent impact on rectum dose parameters for prostate brachytherapy [54, 63‐65]. Simnor et at. [54] calculated that the catheter mean caudal displacements of 7.9 mm and 3.8 mm prior to fractions 2 and 3 were associated with mostly systematic increases to the D_2cc of 0.69 Gy (~ 6.6 %) and 0.76 Gy (~ 7.2 %) respectively. For HDR at the institution where patients in this study were treated, the displacement of catheters was checked prior to each of the three fractions using an anterior-posterior radiograph and corrected for using a rigid external holding device as described by Tiong et al. [16]. The catheter mean caudal displacements after this advancement process were reported as 1.7 mm, 1.1 mm and 0.8 mm for fractions 1, 2 and 3 respectively [16]. Consequently, the inter-fraction motion increases to the D_2cc reported by Simnor et al. [54] may be appropriate conservative estimates for the inter-fraction motion of HDR catheters that could be expected for this study.

It is possible that the above mentioned inter-fraction motion could remove significance of dose ranges. However, shifting of dose values identified as significant are likely as the single CT based estimates were mostly identified as being systematically different to the motion-corrected values [52, 54]. The influence of inter-fraction on delivered doses is likely to be important when considering dose constraints recommended by studies where planned dose tends to be less than delivered dose. Consequently, it would be useful to confirm the importance of published dose-volume constraints after registration is applied for repeat daily imaging.