Bladder spatial-dose descriptors correlate with acute urinary toxicity after radiation therapy for prostate cancer

Physica Medica xxx (2016) xxx–xxx

Contents lists available at ScienceDirect

Physica Medica journal homepage: http://www.physicamedica.com

Original paper

Bladder spatial-dose descriptors correlate with acute urinary toxicity after radiation therapy for prostate cancer I. Improta a,⇑, F. Palorini a, C. Cozzarini b, T. Rancati c, B. Avuzzi d, P. Franco e, C. Degli Esposti f, E. Del Mastro g, G. Girelli h, C. Iotti i, V. Vavassori j, R. Valdagni d,k, C. Fiorino a a

Medical Physics, San Raffaele Scientific Institute, Milano, Italy Radiotherapy, San Raffaele Scientific Institute, Milano, Italy Prostate Cancer Program, Fondazione IRCCS Istituto Nazionale dei Tumori, Milan, Italy d Radiation Oncology 1, Fondazione IRCCS Istituto Nazionale dei Tumori, Milano, Italy e Radiotherapy, Ospedale Regionale U.Parini – AUSL Valle d’Aosta, Aosta, Italy f Radiotherapy, Ospedale Bellaria, Bologna, Italy g Radiotherapy, IRCCS–Candiolo, Candiolo (TO), Italy h Radiotherapy, Ospedale ASL9, Ivrea, Italy i Radiation Therapy Unit, Department of Oncology and Advanced Technology, ASMN Hospital IRCCS, Reggio Emilia, Italy j Radiotherapy, Cliniche Gavazzeni-Humanitas, Bergamo, Italy k Clinical Sciences and Community Health, University of Milan, Milan, Italy b c

a r t i c l e

i n f o

Article history: Received 16 June 2016 Received in Revised form 16 August 2016 Accepted 17 August 2016 Available online xxxx Keywords: Prostate radiotherapy Acute urinary toxicity Dose-surface maps Bladder spatial-dose descriptors

a b s t r a c t Purpose: To assess bladder spatial-dose parameters predicting acute urinary toxicity after radiotherapy for prostate cancer (PCa) through a pixel-wise method for analysis of bladder dose-surface maps (DSMs). Materials & methods: The final cohort of a multi-institutional study, consisting of 539 patients with PCa treated with conventionally (CONV:1.8–2 Gy/fr) or moderately hypo-fractionated radiotherapy (HYPO:2.2–2.7 Gy/fr) was considered. Urinary toxicity was evaluated through the International Prostate Symptoms Score (IPSS) administered before and after radiotherapy. IPSS increases P10 and 15 points at the end of radiotherapy (DIPSS P 10 and DIPSS P 15) were chosen as endpoints. Average DSMs (corrected into 2 Gy-equivalent doses) of patients with/without toxicity were compared through a pixel-wise method. This allowed the extraction of selected spatial descriptors discriminating between patients with/without toxicity. Previously logistic models based on dose-surface histograms (DSH) were considered and replaced with DSM descriptors. Discrimination power, calibration and log-likelihood were considered to evaluate the impact of the inclusion of spatial descriptors. Results: Data of 375/539 patients were available. DIPSS P 10 was recorded in 76/375 (20%) patients, while 30/375 (8%) experienced DIPSS P 15. The posterior dose at 12 mm from the bladder base (roughly corresponding to the trigone region) resulted significantly associated to toxicity in the whole/HYPO populations. The cranial extension of the 75 Gy isodose along the bladder central axis was the best DSM-based predictor in CONV patients. Multi-variable models including DSM descriptors showed better discrimination (AUC = 0.66–0.77) when compared to DSH-based models (AUC = 0.58–0.71) and higher log-likelihoods. Conclusion: DSMs are correlated with the risk of acute GU toxicity. The incorporation of spatial descriptors improves discrimination and log-likelihood of multi-variable models including dosimetric and clinical parameters. Ó 2016 Associazione Italiana di Fisica Medica. Published by Elsevier Ltd. All rights reserved.

1. Introduction The implementation of intensity modulation radiation therapy (IMRT) and image guidance (IGRT), improved the clinical outcome ⇑ Corresponding author. E-mail address: [email protected] (I. Improta).

after the radical treatment of prostate cancer (PCa), leading to a significant reduction of rectal and gastro-intestinal toxicities [1–4]. By contrast, urinary toxicity remains a relevant side-effect in patients treated with radiation therapy (RT) for PCa, mainly due to the intrinsic difficulty of sparing the bladder neck and the urethra. Acute urinary symptoms occurring during or soon after the treatment were also found to be significantly associated with

http://dx.doi.org/10.1016/j.ejmp.2016.08.013 1120-1797/Ó 2016 Associazione Italiana di Fisica Medica. Published by Elsevier Ltd. All rights reserved.

Please cite this article in press as: Improta I et al. Bladder spatial-dose descriptors correlate with acute urinary toxicity after radiation therapy for prostate cancer. Phys. Med. (2016), http://dx.doi.org/10.1016/j.ejmp.2016.08.013

2

I. Improta et al. / Physica Medica xxx (2016) xxx–xxx

late urinary toxicity [5,6] and their severity considerably affects the quality of life of a still relevant fraction of patients. In order to reduce radiation-induced side effects and improve overall quality of life, the development of predictive models of toxicity after RT for PCa has achieved relevant advances. In this context, the role of medical physics has become essential not only for the detection of the major predictors, but also for the entire decision-making process, the interpretation and application of predictive models in clinical practice [7]. Although the assessment of clinical predictors of urinary toxicity has increased in recent years, dose-volume effect models are still lacking [3,8–14]. In addition, the available knowledge is mainly based on bladder dose–volume/surface histograms (DVH/ DSH) and does not include any information regarding the shape of the dose distribution. Only two published studies [15,16] reported associations between late urinary toxicity and the dose received by specific portions of the bladder, and both suggested that the dose received by the trigone area was correlated with the onset of severe late urinary symptoms. Much research is currently oriented toward the implementation of methods aimed at integrating the spatial information included in the 3D dose distributions into toxicity models. In this context, dose-surface mapping is a suitable method to describe spatial effects of local dose distributions, thus far mostly applied to rectal toxicity [17–20]. An original pixel-wise method aimed at analyzing dose-surface maps (DSMs) of bladder and assessing local dose descriptors of urinary toxicity was recently developed [21] and first applied to a relatively small, homogeneous group of patients treated with moderate hypofractionated RT. In the current study, we exploited this method with the ultimate scope of analyzing the final population of a cohort multi-centric study (DUE01 [9,21–23]), including patients treated for clinically localized PCa with radical intent including both moderate hypofractionated and conventional fractionated regimens. Acute urinary symptoms were evaluated through a patient reported questionnaire (the International Prostate Symptom Score, IPSS) and large worsening of symptoms at the end of RT were here considered as the relevant toxicity endpoint. The aim of the analysis was to assess the best DSMs factors associated with a large worsening of patient-reported acute urinary symptoms and to quantify the possible benefit of the introduction of DSM descriptors into multi-variable predictive models.

2. Materials and methods 2.1. Patient cohort and assessment of GU toxicity Between April 2010 and December 2014, 539 patients were enrolled in the DUE01 trial, a multi-Institutional prospective study aimed at developing predictive models of urinary toxicity and erectile dysfunction after high-dose radical RT for localized PCa. Detailed information on enrolment criteria, contouring/planning procedures, methods for data collection were reported in detail previously [9,23]. Briefly, patients were treated with conventionally (1.8–2 Gy/fr) or moderately hypofractionated RT (2.2–2.7 Gy/ fr) in 5 fractions/week, with 3D conformal RT and IMRT techniques. All patients were scanned supine with empty rectum and full or half/full bladder [24]. Of note, IGRT through CBCT or MVCT (using Tomotherapy) was used for set-up the patients in >80% of patients. Urinary toxicity was evaluated through IPSS, administrated before RT and at its end. IPSS includes 7 questions investigating feeling of incomplete bladder emptying, frequency, intermittency, urgency, weak stream, straining and nocturia: each question scores the severity of each symptom with a scale ranging from 0 (i.e. absence of the

symptom) to 5 (i.e. symptom almost always present). The total score can therefore range from 0 to 35. In the current study we focused on the large worsening of urinary symptoms by considering two previously introduced distinct endpoints [22]: (1) an IPSS increase P10 points at RT end (DIPSS P 10) and (2) DIPSS P 15, expression of a more severe worsening. Analyses were performed for the whole population and for conventionally fractionated/hypofractionated subgroups, separately. In all cases patients with a baseline IPSS P 20 were excluded, in order to consider a population who could in principle experience both worsening endpoints. 2.2. Dose surface maps and pixel-wise analysis An original pixel-wise method for DSMs analysis and assessment of local dose descriptors was previously implemented [21]. In short, DSMs of bladder were generated from the planning CT contours and the calculated dose distributions using a dedicated module of VODCA (MSS Medical Software Solutions, Hagendorn, Switzerland), by unfolding the reconstructed 3D surface doses of the bladder on a 2D plane: the bladder was anteriorly cut at the points intersecting the sagittal plane passing through its center of mass. In this way, the posterior bladder corresponded to the central region of the map, while the anterior one was located laterally. A rectangular map for each patient was then generated: the axial dimension was normalized while the cranial-caudal direction was maintained in absolute units. The laterally normalized DSMs of each patient were superposed and aligned at the central inferior/posterior point and caudally cut at the smallest vertical extension present in the sample (25 mm in our case). This choice of maintaining the absolute vertical extension should allow a better superposition of similar structures at the more stable bladder base, as previously discussed in detail [21]. The population was then divided into two groups (patients with/without toxicity endpoint) and a framework of 4 maps was generated from pixel-by-pixel analysis: the average DSMs of patients with and without toxicity, the dose difference map (DDM) and the t-statistic map (i.e., differences of means divided by the unpooled standard error). The uncorrected p-value contours (60.05) were also added to the t-statistic map to point out the regions that better discriminate between patients with/without toxicity and select reasonable spatial descriptors. The choice of showing uncorrected p-values in DSMs was widely discussed in Palorini et al. [21] and was mainly due to the fact that we are interested in the shape of dose regions that better discriminate the two groups of patients, while we are not looking for significance at the single pixel level. In this context, the strong control of false positives guaranteed by the corrections for multiple testing might imply an unnecessary inflation of false negatives with a loss of interesting information. Note, however, that for such a large dataset and for DSMs that have highly correlated dose regions, we do not expect very different results from maps with corrections for multiple testing, at least in the shape of interesting regions. For comparison, in Supplementary Material (§2) we show DSMs with p-values adjusted by the permutation algorithm described in [25]. In order to separately analyze different fractionation schemes, DSMs of all patients were corrected into 2 Gy equivalent maps using the linear-quadratic model with the inclusion of treatment time correction (a/b = 10 Gy and a time correction factor c = 0.7 Gy/day). Details are reported in the Supplementary Materials (§1). 2.3. Analysis of spatial dose descriptors and their inclusion into multivariable models Based on the analysis of DSMs, several spatial descriptors were extracted, such as local doses, distances between selected isodose

Please cite this article in press as: Improta I et al. Bladder spatial-dose descriptors correlate with acute urinary toxicity after radiation therapy for prostate cancer. Phys. Med. (2016), http://dx.doi.org/10.1016/j.ejmp.2016.08.013

3

I. Improta et al. / Physica Medica xxx (2016) xxx–xxx

levels and their length. A preliminary correlation filter (r < 0.85) was applied to the extracted local dose descriptors to select only independent parameters. For all endpoints and subgroups, DSMbased predictors for patients with/without toxicity were compared through two-sided t-test and ROC analysis in order to identify those exhibiting higher association (lowest p-values, highest AUCs) with toxicity endpoints to be included in further multivariable models. Previously published models of large IPSS worsening trained on the same population were considered [22]; these models included DSH parameters and clinical features that were selected by applying advanced data-mining methods for variable selection [22]. Then, the DSH parameters were replaced with the best spatial dose-descriptor as selected following the above described procedure. The previously introduced interaction term between DSH parameters and drugs (cardiovascular drugs for DIPSS P 10 and anti-hypercholesterolemia drugs for DIPSS P 15 [22]) was unchanged: this last choice was motivated by the idea that the possible role of an impaired healing process of the radiationinduced damage in patients with potential micro-angiopathic disease could exhibit its interaction with the bladder surface irradiated above certain doses and is not depending on the particular spatial location of dose. Concerning DIPSS P 10 in the conventionally treated population, as a ‘‘DSH” model was not available, a model including DSH parameters and clinical features was first developed (Supplementary Material, Table S1) following exactly the same variable selection method previously followed for the whole population and for the hypofractionated group.

2.4. Comparison between models based on DSH and DSM parameters In order to quantify the benefit of replacing DSH with DSM descriptors, the resulting models were compared in terms of discriminative power (Area Under the Receiver Operating Characteristics Curve, AUC), calibration slope, log-likelihood (LLH), sum of squared residuals and sum of squared residuals as calculated in the subpopulation of patients with the toxicity endpoint. This last performance measure was suggested by the observation that a common problem with these toxicity models is that they are mostly failing prediction in patients exhibiting the endpoint, i.e. they are usually estimating a low toxicity probability for patients experiencing toxicity; then, improving the sum of squared residuals on patients with toxicity should be considered to be more relevant. The difference between AUCs of two sets of models was tested with the method of DeLong et al. [26], while for LLH differences we used the Akaike’s Information Criterion (AIC), as proposed by Burnham and Anderson [27]. All analyses were performed with R (www.r-project.org) and Knime software (KNIME GmbH, Germany).

3. Results 3.1. Patient data Of the 539 patients enrolled in the DUE01 trial, 27 were withdrawn before RT and 457/512 had complete planning data; due to technical problems, it was not possible to successfully export DSMs for 20 patients, and of the remaining 437, 387 correctly filled

Table 1 Patient characteristics. All patients (n = 375)

CONV patients (n = 164)

HYPO patients (n = 211)

Age (years, mean ± st.dev.) BMI (mean ± st.dev.) Diabetes (yes) Cardiovascular diseases (yes) Previous abdominal surgeries (yes) Appendectomy (yes) Use of antihypertensives (yes) Use of anti-aggregants (yes) Use of 5alpha reductase inhibitors (yes) Use of drugs for impotence (yes) Use of alpha blockers (yes) Use of drugs for gastrointestinal diseases (yes) Use of drugs for cardiovascular diseases (yes) Use of drugs for hypercholesterolemia (yes) Smoke (yes) Alcohol (yes) Neoadjuvant hormone therapy (yes) Use of antiandrogens (yes) Use of LH-RH analog (yes) Total androgen blockade (yes) Clinical stage = cT1 Clinical stage = cT2 Clinical stage = cT3 Clinical stage = cT4 Clinical stage = cTX

70 ± 6 27 ± 6 47(12%) 98 (26%) 172 (46%) 49 (13%) 206 (55%) 117 (31%) 35 (9%) 27 (7%) 65 (17%) 38 (10%) 128 (34%) 60 (16%) 57 (15%) 180 (48%) 213 (57%) 179 (48%) 86 (23%) 53 (14%) 212 (56%) 109(29%) 37(10%) 4(1%) 12(3%)

70 ± 6 28 ± 8 25 (15%) 50 (31%) 87 (54%) 29 (18%) 98 (60%) 27 (17%) 13 (8%) 13 (8%) 32 (20%) 18 (11%) 69 (43%) 32 (20%) 19 (12%) 78 (48%) 108 (67%) 92 (57%) 28 (17%) 12 (7%) 106(65%) 36(22%) 16(10%) 0 4 (2%)

71 ± 6 27 ± 4 22 (10%) 48 (22%) 85 (40%) 20 (9%) 108 (51%) 34 (16%) 14 (7%) 7 (3%) 33 (15%) 20 (9%) 59 (28%) 28 (13%) 38 (18%) 102 (48%) 105 (49%) 87 (41%) 58 (27%) 41 (19%) 106(50%) 73 (34%) 21(10%) 4(2%) 8(4%)

Dosimetric data Prescription dose (Gy, median, range) Dose/fraction (Gy, median, range) Irradiation of pelvic nodes (yes) Prescribed dose (Gy, mean ± st.dev.) Irradiation of seminal vesicles (yes) CTV (cc, mean ± st.dev.) PTV (cc, mean ± st.dev.) Bladder volume (cc, mean ± st.dev.)

74.2 (59.8–80.0) 2.48 (2.0–2.7) 158 (42%) 51.0 ± 2.0 289 (77%) 55 ± 33 137 ± 63 244 ± 152

78.0 (74.0–80.0) 2.0 66 (41%) 50.0 ± 1.0 127 (78%) 48 ± 20 119 ± 44 244 ± 143

70.2 (59.8–75.2) 2.55 (2.35–2.7) 110 (52%) 52.0 ± 2.0 206 (97%) 61 ± 39 151 ± 71 245 ± 158

HYPO = hypofractionated; CONV = conventionally fractionated; st.dev. = standard deviation; BMI = Body Mass Index; LH-RH = lutinizing hormone-releasing hormone; CTV = Clinical Target Volume; PTV = Planning Target Volume.

Please cite this article in press as: Improta I et al. Bladder spatial-dose descriptors correlate with acute urinary toxicity after radiation therapy for prostate cancer. Phys. Med. (2016), http://dx.doi.org/10.1016/j.ejmp.2016.08.013

4

I. Improta et al. / Physica Medica xxx (2016) xxx–xxx

Fig. 1. Mean dose surface maps (2 Gy equivalent) of patients with/without toxicity at radiotherapy end (left), maps of dose differences between the two groups and corresponding t-statistics maps (right) with overlapped p-value contours (60.05) for (a) DIPSS P 10 and (b) DIPSS P 15.

Please cite this article in press as: Improta I et al. Bladder spatial-dose descriptors correlate with acute urinary toxicity after radiation therapy for prostate cancer. Phys. Med. (2016), http://dx.doi.org/10.1016/j.ejmp.2016.08.013

5

I. Improta et al. / Physica Medica xxx (2016) xxx–xxx

in both baseline and end RT IPSS questionnaires. Twelve patients were excluded because of baseline IPSS > 20; final analyses were thus performed on 375 patients (211 hypofractionated and 164 conventionally fractionated). A summary of the clinical and dose data for the final cohort is presented in Table 1. The median prescribed doses were 78 Gy (2 Gy/fr) and 70.2 Gy (2.35–2.7 Gy/fr) for conventionally fractionated and hypofractionated patients respectively; overall 158/375 (42%) received pelvic limph-nodes irradiation at a mean prescribed dose of 51.0 ± 2 Gy; in 289/375 (77%) the seminal vesicles were irradiated. The very large majority of patients (336/375) was treated with IMRT techniques. In the whole population, 76/375 (20%) and 30/375 (8%) patients exhibited a DIPSS P 10 and DIPSS P 15 respectively. In the hypofractionated subgroup, 49/211 (23%) and 24/211(11%) patients had DIPSS P 10 and DIPSS P 15, respectively, whereas 27/164 (16%, DIPSS P 10) and 6/164 (4%, DIPSS P 15) toxicity events were registered in the conventionally fractionated subgroup. Analyses pertaining to DIPSS P 15 for the conventionally

fractionated subgroup were not performed because of the low number of toxicity events (<10). 3.2. DSMs and extraction of spatial-dose descriptors The average DSMs of patients with/without toxicity, the differences and the t-statistics maps are shown in Fig. 1a, b, for the whole population and hypofractionated/conventionally fractionated subgroups, for DIPSS P 10 and DIPSS P 15 respectively. In the whole population, large dose differences (>5 Gy) were found in the region of high-intermediate doses (50–75 Gy, 2 Gyequivalent) along the whole central region of the DSM. Different patterns were found when separately analyzing hypofractionated and conventionally fractionated cohorts: in the hypofractionated subpopulation DSM analysis highlighted (for both endpoints) the role of the dose received by the posterior region at 5–18 mm from the base of bladder. When considering conventionally fractionated group, the areas that most significantly correlated with DIPSS P 10 were located more cranially and were likely to be associated with

Table 2 Multivariable models with inclusion of dose surface map descriptors for (1) DIPSS P 10 and (2) DIPSS P 15 in (a) whole population, (b) hypofractionated and (c) conventionally fractionated subgroups. Variable

Median coeff

% significant coeff

Median Odds ratio

10–90° percentile ORs

(a) Bootstrap multi-variable regression – whole population (1) Endpoint DIPSS P 10 AUC = 0.70; 95% CI = 0.64–0.76 Calibration slope = 1.10. R2 = 0.78 Neoadjuvant HT (yes vs no) PTV10 (units of 10 cm3) PD12 (Gy) Age (yrs) Hypertension (yes vs no) Cardio drugs  absolute weekly DSH at 8.5 Gy BMI (kg/m2) 5-ARIs (yes vs no) Constant

0.505 0.021 0.077 0.066 0.323 0.005 0.071 0.651 3.28

93.3 75.4 100 99.8 84.2 86.5 98.8 86.9

0.60 1.02 1.08 0.94 1.38 1.005 0.93 0.52

0.40–0.94 0.98–1.06 1.05–1.11 0.91–0.97 0.92–2.19 0.999–1.011 0.88–0.98 0.19–1.08

(2) Endpoint DIPSS P 15 AUC = 0.77, 95% CI = 0.70–0.84 Calibration slope = 0.89; R2 = 0.82 Cardio drugs  absolute weekly DSH at 12 Gy Hypercol drugs  absolute weekly DSH at 12 Gy Neoadjuvant HT (yes vs no) PD12 (Gy) Constant

0.004 0.016 0.461 0.146 14.00

80.2 61.2 84.6 100

1.004 1.02 0.63 1.16

0.980–1.024 0.99–1.04 0.37–1.12 1.10–1.22

93.8 57.5 91.6 100

1.83 0.93 0.962 1.08

1.09–3.03 0.57–1.49 0.925–0.997 1.05–1.12

82.3 100

1.66 1.14

0.81–3.35 1.09–1.21

95.2 100.0

0.46 1.17

0.26–0.85 1.10–1.27

(b) Bootstrap multi-variable regression – hypofractionated population (1) Endpoint DIPSS P 10 AUC = 0.66, 95% CI = 0.58–0.74 Calibration slope = 1.07; R2 = 0.84 Cardio drugs (yes vs no) 0.605 Neoadjuvant HT (yes vs no) 0.073 Age (yrs) 0.039 PD12 (Gy) 0.076 Constant 4.87 (2) Endpoint DIPSS P 15 AUC = 0.72, 95% CI = 0.61–0.82 Calibration slope = 1.00; R2 = 0.79 Anti-hypercholesterolemia drugs (yes vs no) PD12 (Gy) Constant

0.504 0.132 12.99

(c) Bootstrap multi-variable regression – conventionally fractionated population (1) Endpoint DIPSS P 10 AUC = 0.77, 95% CI = 0.68–0.87 Calibration slope = 0.89; R2 = 0.93 Neoadjuvant HT (yes vs no) 0.781 Vert75 (mm) 0.158 Constant 4.20

Abbreviations: Coeff = coefficient; % significant coeff = percentage of coefficients >0 if mean coefficient is >0, percentage of coefficient <0 if mean coefficient <0; ORs = Odds Ratios; IPSS = International Prostate Symptom Score; AUC = Area Under the Curve; HT = hormonal therapy; PD12 = Posterior dose at 12 mm from bladder’s base; cardio = cardiovascular; 5-ARIs = 5-alpha-reductase inhibitors; Vert75 = cranio-caudal extension of 75 Gy isodose; BMI = body mass index; PTV = Planning Target Volume, yrs = years.

Please cite this article in press as: Improta I et al. Bladder spatial-dose descriptors correlate with acute urinary toxicity after radiation therapy for prostate cancer. Phys. Med. (2016), http://dx.doi.org/10.1016/j.ejmp.2016.08.013

6

I. Improta et al. / Physica Medica xxx (2016) xxx–xxx

the cranial extension of the intermediate/high dose isodoses. Similar interesting regions were also found in the permutationcorrected DSMs reported in the Supplementary Material. Then, the following spatial-dose parameters were extracted: – The posterior dose at 10–12–15–18 mm from the base of the bladder (PD10, PD12, PD15, PD18) and the cranio-caudal extension (mm) of the isodose curves from 50 to 70 Gy at 5 Gy interval (Vert50 ? Vert75) for the whole population and for both endpoints – PD10 ? PD18 for DIPSS P 10 and DIPSS P 15 for the hypofractionated population – Vert50 ? Vert75 for DIPSS P 10 for the conventionally fractionated population 3.3. Analysis of local dose descriptors Table S2 (Supplementary Material) summarizes the results for t-tests and discrimination power analysis for the whole population and for hypofractionated/conventionally fractionated subpopulations for the two distinct endpoints. PD12 resulted to be the parameter with the lowest p-value and highest AUC for both endpoint in whole population (DIPSS P 10: AUC = 0.64; DIPSS P 15: AUC = 0.74) and in hypofractionated subset (DIPSS P 10: AUC = 0.65; DIPSS P 15: AUC = 0.70). In the conventionally fractionated subgroup the best discriminating spatial parameter was the Vert75 (DIPSS P 10: AUC = 0.76). The best spatial predictors significantly correlated with toxicities were tested in univariable logistic regression (UVR) and showed a significant increment in AUCs for both endpoints com-

pared to the best DSH predictors (Supplementary Material, Table S3).

3.4. Multi-variable models of urinary toxicity including spatial-dose parameters The resulting multivariable models including DSM parameters were summarized in Table 2. Neoadjuvant hormone therapy (HT) was included as protective factor together with the spatial dose parameters (PD12 and Vert75, continuous variables, OR = 1.08– 1.17 for the risk of DIPSS P 10 and OR = 1.14–1.17 for DIPSS P 15). All models showed moderately high discriminative power (AUC = 0.66–0.77) and good calibration (calibration slopes: 0.89– 1.10). Fig. 2 shows the dose–response relationships for the whole group and the hypofractionated and conventionally fractionated subgroups including the impact of the most relevant clinical risk factors.

3.5. Comparison of models with DSH and DSM based parameters Table 3 reports the results of the comparisons between the two sets of multivariable models (including DSH vs DSM parameters). Models including spatial parameters always showed increased discrimination power (AUC = 0.66–0.77 vs AUC = 0.58–0.72). The largest improvements were observed for both endpoints in the hypofractionated cohort (DAUC = 0.14–0.18; p = 0.06–0.04) and in the whole population for DIPSS P 15 (DAUC = 0.12; p = 0.03). The differences were not statistically significant for DIPSS P 10

Fig. 2. Complication probabilities curves with respect to the main DSM and clinical predictors: whole population, ΔIPSS P 10 and ΔIPSS P 15 (upper panel); hypofractionated and conventionally treated patients, ΔIPSS P 10 (lower panel).

Please cite this article in press as: Improta I et al. Bladder spatial-dose descriptors correlate with acute urinary toxicity after radiation therapy for prostate cancer. Phys. Med. (2016), http://dx.doi.org/10.1016/j.ejmp.2016.08.013

7

I. Improta et al. / Physica Medica xxx (2016) xxx–xxx

Table 3 Comparison of multivariable models based on dose surface map/dose surface histogram parameters and including clinical covariates, for the risk of DIPSS P 10 and DIPSS P 15, in whole, hypofractionated and conventionally fractionated populations. The table summarizes: Areas Under the Receiver Operating Characteristics Curves (AUC) with 95% of confidence interval (CI), calibration slope (CS), log-likelihood (LLH), sum of squared residuals on the whole population (RSS) and sum of squared residuals on the subpopulation exhibiting the considered endpoint (RSSt). P-values for differences of reported performance measures (p-AUC) and delta values of Akaike information criterion (DAIC) are also reported. Endpoint

Population

Variables

Clinical covariates

AUC; 95%CI

CS

R2

RSS

RSSt

p-AUC

DAIC

DIPSS P 10

WHOLE

DSHw at 8.5 Gy PD at 12 mm

(neoadjuvant HT, PTV, age, BMI, hypertension, 5-ARIs, interaction cardio)

0.66; [0.59–0.73] 0.70; [0.64–0.76]

0.98 1.06

0.92 0.77

176.2 171.6

56.5 55.6

43.1 42.2

0.21

9.2

HYPO

DSHw at 12.5 Gy PD at 12 mm

(neoadjuvant HT, cardiovascular drugs, age)

0.58; [0.49–0.67] 0.66; [0.58–0.74]

0.98 1.07

0.65 0.84

110.0 104.3

36.2 34.7

27.0 26.7

0.06

11.4

CONV

DSHw at 6 Gy Vert75 Gy at 5 mm

(neoadjuvant HT)

0.72; [0.60–0.83] 0.77; [0.68–0.87]

1.01 0.89

0.81 0.93

67.2 62.4

20.7 19.1

16.0 14.6

0.39

9.6

WHOLE

DSHw at 12 Gy PD at 12 mm

(neoadjuvant HT, interaction cardio, interaction hypercholesterolemia)

0.69; [0.60–0.78] 0.77; [0.70–0.84]

1.31 0.89

0.84 0.82

98.5 90.7

26.6 25.4

24.9 21.1

0.03

15.6

HYPO

DSHw at 10 Gy PD at 12 mm

(hypercholesterolemia drugs)

0.61; [0.50–0.73] 0.72; [0.61–0.82]

1.07 1.00

0.78 0.79

71.6 66.6

20.6 19.7

18.2 16.3

0.04

10.0

DIPSS P 15

in the whole population and in conventionally treated patients. ROC curves for all models are shown in Fig. 3. The new models also exhibited increased log-likelihood and improved (i.e. decreased) sum of squared residuals on the whole population and in the subpopulation of patients exhibiting worsening of urinary symptoms. The values of DAIC indicate that models with the spatial dose descriptors are better than the models with DSH parameters (DAIC = 9.2–15.6). Calibration slope did not improve when including spatial parameters.

4. Discussion The vast majority of studies on predictive models of urinary toxicity have been focused on dose–volume/surface histograms with the consequent limitation of missing any spatial information related to the dose distribution in the bladder wall. The incorporation of DSM information may partially go beyond this limitation. In this study, we investigated the relationships between bladder DSM and the risk of a large worsening of urinary symptoms at RT end in patients treated for PCa. Two main spatial descriptors were found to be significantly associated with acute urinary toxicity (assessed as worsening of overall urinary symptoms): the posterior dose at 12 mm from the base of bladder, consistently with the bladder trigone (for whole/ hypofractionated subpopulations) and the cranial-caudal extension of the 75 Gy isodose along the central axis of bladder (for conventionally treated patients). The evidence that specific regions of the bladder could be more sensitive than others has also recently been suggested by Heemsbergen et al. [15], who analysed a large population of patients and examined differences in dose maps of patients with and without late urinary obstruction: they demonstrated that the dose received by the trigone was associated with an increased risk of this late severe side effect. The impact of trigone dose on late urinary toxicity was also suggested by Ghadjar et al. [16]. In their study the potential association between the DVHs of specific regions of the bladder and urinary toxicity was analyzed and they found that hot points on the trigone region were correlated with a worsening of late urinary symptoms. Very interestingly, this effect was not visible in current study for conventionally fractionated patients. This fact could depend on the lower 2 Gy equivalent doses delivered in this subgroup, reflecting also in the lower rates of toxicity, especially when considering DIPSS P 15.

LLH

For this group, an interesting relationship between an increased toxicity and a more cranial extension of the high dose isodoses was evident; this effect seems to be masked in the whole population by the dominant ‘‘trigone” dose–effect, driven by the hypofractionated group data. In short, in case of delivered 2 Gy equivalent doses below 80 Gy, the reduction of the cranial extension of the bladder included in the high dose region may translate into a potentially relevant reduction of acute urinary toxicity. Multivariable models for acute urinary toxicity including spatial-dose parameters together with clinical features were also considered by starting from previously developed models using an advanced data-mining method for the selection of the most robustly predictive variables [22]. Overall, the incorporation of DSMs implied a higher discriminative power of the resulting models with respect to models including DSH parameters only [22]: this result is the first showing a benefit of spatial-dose descriptors over DVH/DSH in modeling urinary toxicity. The improvement of the performances was more evident in the hypofractionated population: this may be interpreted again as a prevalent impact of the ‘‘trigone dose” effect, well assessed by the pixel-wise DSM analysis and stronger in this subgroup due to the higher delivered 2 Gy equivalent doses. Given the known correlation between acute and late urinary toxicity [6,28] the application of these models to reduce acute symptoms could likely translate into a reduction of late events. An interesting issue concerns the robustness of DVH/DSH/DSM with respect to variable bladder filling. In a previous study, it was shown that both absolute DVH/DSH and DSM-based parameters are relatively stable in the frame of current study where a full bladder is prescribed to the patients: Palorini et al. [29] showed a very limited systematic error (population averaged value: ±1 Gy; 1SD around 1–3 Gy) for DSM values within the planned high-dose region by analyzing data referred to 472 DSMs of 18 patients based on daily MVCT images collected during IGRT. Focusing on clinical predictors, neoadjuvant HT was originally included as a protective factor; however, the introduction of DSMs translated into a not evidence of any role for the hypofractionated population (Table 2) while it was confirmed as a strong protective factor for the CONV population. The effect of neo-adjuvant HT as a protective factor from urinary toxicity had already been reported [28] and often associated with a reduction of the irradiated volume and a consequent better dosimetry. However, the independent roles of HT and bladder dosimetry parameters found in our population suggest that the protective behaviour of HT might be also related to a possible anti-inflammatory action, as suggested by

Please cite this article in press as: Improta I et al. Bladder spatial-dose descriptors correlate with acute urinary toxicity after radiation therapy for prostate cancer. Phys. Med. (2016), http://dx.doi.org/10.1016/j.ejmp.2016.08.013

8

I. Improta et al. / Physica Medica xxx (2016) xxx–xxx

Fig. 3. ROC curves of multivariable models with DSH or DSM predictors for all subgroups for (a) ΔIPSS P 10 and (b) ΔIPSS P 15. The impact of the inclusion of DSM may be well appreciated.: largest improvement were found for hypofractionated patients and for the ΔIPSS P 15 end point.

several pre-clinical studies [30–32]. There is some indication that, in addition to the known benefit of hormonal therapies on urinary symptoms due to direct inhibitory effects on prostate volume, specific hormonal receptors in the bladder wall as well as transcriptional suppression of genes involved in the inflammatory process, as proinflammatory cytokines and growth factors, may independently act to directly reduce radiation-induced inflammation. The strong confirmation of HT as a protective factor in the conventionally fractionated subgroup suggests that the supposed anti-inflammatory effect of anti-androgens could be less effective when higher equivalent doses are delivered, as in our hypofractionated group. Some specific effect of hypo-fractionation on the intensity of bladder functionality not explained by the LQ model, as previously suggested for late hematuria in the post-operative setting [33,34], could also be claimed, although this is out of the aim of current work. Interactions between dose and the use of cardiovascular drugs and, for DIPSS P 15, with the use of anti-hypercholesterolemia drugs were also found as risk factors, showing that patients who take those drugs have an additional risk which increases with dose. In the hypofractionated population (entailing higher doses) the use of the above mentioned drugs is included as a risk factors without interaction with dose (ORs  1.7–1.8). These drugs are usually prescribed in patients with vasculopaties, thus indirectly suggesting a possible role of an impaired capacity of repairing radiationinduced damage resulting from micro-angiopathic disease. In conclusion, this analysis shows that the inclusion of bladder DSM improves multi-variable models for acute urinary toxicity prediction. The results suggest that a reduction of the dose in the bladder trigone below 80 Gy (2 Gy equivalent) and a reduction of the cranial extension of the bladder surface included in the high dose region (i.e., 75 Gy) may significantly reduce the risk of acute toxicity. More research is warranted to improve our knowledge regarding dose-volume effects of urinary toxicity which still remain largely unknown, especially for late and chronic symptoms [35].

Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.ejmp.2016.08.013. References [1] De Langhe S, De Meerleer G, De Ruyck K, et al. Integrated models for the prediction of late genitourinary complaints after high-dose intensity modulated radiotherapy for prostate cancer: making informed decisions. Radiother Oncol 2014;112:95–9. [2] Zelefsky MJ, Levin EJ, Hunt M, et al. Incidence of late rectal and urinary toxicities after three-dimensional conformal radiotherapy and intensitymodulated radiotherapy for localized prostate cancer. Int J Radiat Oncol Biol Phys 2008;70(4):1124–9. [3] Valdagni R, Vavassori V, Rancati T, et al. Increasing the risk of late rectal bleeding after high-dose radiotherapy for prostate cancer: the case of previous abdominal surgery. Results from a prospective trial. Radiother Oncol 2012;103:252–5. [4] Fiorino C, Fellin G, Rancati T, et al. Clinical and dosimetric predictors of late rectal syndrome after 3d-CRT for localized prostate cancer: preliminary results of a multicenter prospective study. Int J Radiat Oncol Biol Phys 2010;78 (1):19–25. [5] Kim S, Moore DF, Shih W, et al. Severe genitourinary toxicity following radiation therapy for prostate cancer: how long does it last? J Urol 2013;189 (1). [6] Budäus L, Bolla M, Bossi A, Cozzarini C, Crook J, Widmark A, et al. Functional outcomes and complications following radiation therapy for prostate cancer: a critical analysis of the literature. Eur Urol 2012;61(1):112–27. [7] Fiorino C, Seuntjens J. The role of medical physics in prostate cancer radiation therapy. Phys Med 2016;32(3):435–7. [8] Peeters ST, Heemsbergen WD, van Putten WL, et al. Acute and late complications after radiotherapy for prostate cancer: results of a multicenter randomized trial comparing 68 Gy to 78 Gy. Int J Radiat Oncol Biol Phys 2005;61(4):1019–34. [9] Carillo V, Cozzarini C, Rancati T, et al. Relationships between bladder dose volume/surface histograms and acute urinary toxicity after radiotherapy for prostate cancer. Radiother Oncol 2014;111(1):100–5. [10] Cheung MR, Tucker SL, Dong L, et al. Investigation of bladder dose and volume factors influencing late urinary toxicity after external beam radiotherapy for prostate cancer. Int J Radiat Oncol Biol Phys 2007;67(4):1059–65. [11] Carillo V, Cozzarini C, Chietera A, Perna L, Gianolini S, Maggio, et al. Correlation between surrogates of bladder dosimetry and dose-volume histograms of the bladder wall defined on MRI in prostate cancer radiotherapy. Radiother Oncol 2012;105(2):180–3.

Please cite this article in press as: Improta I et al. Bladder spatial-dose descriptors correlate with acute urinary toxicity after radiation therapy for prostate cancer. Phys. Med. (2016), http://dx.doi.org/10.1016/j.ejmp.2016.08.013

I. Improta et al. / Physica Medica xxx (2016) xxx–xxx [12] Yahia N, Ebert MA, Bulsara M, Haworth A, Kennedy A, Joseph DJ, et al. Dosimetry, clinical factors and medication intake influencing urinary symptoms after prostate radiotherapy: an analysis of data from RADAR prostate radiotherapy trial. Radiother Oncol 2015;116:112–8. [13] Pra Dal, Souhami L. Prostate cancer radiation therapy: a physician’s perspective. Phys Med 2016;32(3):438–45. [14] Coates J, El Naqa I. Outcome modeling techniques for prostate cancer radiotherapy: data, models, and validation. Phys Med 2016;32(3):512–20. [15] Heemsbergen WD, Al-Mamgani A, Witte MG, et al. Urinary obstruction in prostate cancer patients from the Dutch trial (68 Gy vs. 78 Gy): relationships with local dose, acute effects, and baseline characteristics. Int J Radiat Oncol Biol Phys 2010;78(1):19–25. [16] Ghadjar P, Zelefsky MJ, Spratt DE, et al. Impact of dose to the bladder trigone on long-term urinary function after high-dose intensity modulated radiation therapy for localized prostate cancer. Int J Radiat Oncol Biol Phys 2014;88 (2):339–44. [17] Acosta O, Drean G, Ospina JD, et al. Voxel-based population analysis for correlating local dose and rectal toxicity in prostate cancer radiotherapy. Phys Med Biol 2013;58:2581–95. [18] Buettner F, Gulliford SL, Webb S, Partridge M. Modeling late rectal toxicities based on a parameterized representation of the 3D dose distribution. Phys Med Biol 2011;56:2103–18. [19] Buettner F, Gulliford SL, Webb S, Partridge M. Using dose-surface maps to predict radiation-induced rectal bleeding: a neural network approach. Phys Med Biol 2009;54:5139–53. [20] Lee R, Chan EK, Kosztyla R, et al. Dose-distance metric that predicts late rectal bleeding in patients receiving radical prostate external-beam radiotherapy. Phys Med Biol 2012;57:8297–307. [21] Palorini F, Cozzarini C, Gianolini S, et al. First application of a pixel-wise analysis on bladder dose-surface maps in prostate cancer radiotherapy. Radiother Oncol 2016;119:123–8. [22] Palorini F, Rancati T, Cozzarini C. Multi-variable models of large IPSS worsening at the end of therapy in prostate cancer radiotherapy. Radiother Oncol 2016;118:92–8. [23] Carillo V, Cozzarini C, Perna L, et al. Contouring variability of the penile bulb on CT images: quantitative assessment using a generalized concordance index. Int J Radiat Oncol Biol Phys 2012;83:841–6.

9

[24] Cozzarini C, Rancati T, Carillo V, et al. Multi-variable models predicting specific patient-reported acute urinary symptoms after radiotherapy for prostate cancer: results of a cohort study. Radiother Oncol 2015;116(2):185–91. [25] Chen C, Witte M, Heemsbergen W, van Herk M. Multiple comparisons permutation tests for image based data mining in radiotherapy. Radiat Oncol 2013;8:293. [26] DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics 1988;44:837–45. [27] Burnham KP, Anderson DR. Multimodel inference: understanding AIC and BIC in Model Selection. Sociol Methods Res 2004;33:261–304. [28] Fiorino C, Valdagni R, Rancati T, Sanguineti G. Dose–volume effects for normal tissues in external radiotherapy: pelvis. Radiother Oncol 2009;93:153–67. [29] Palorini F, Botti A, Carillo V, et al. Bladder dose-surface maps and urinary toxicity: robustness with respect to motion in assessing local dose effects. Phys Med 2016;32:506–11. [30] Cui Y, Zong H, Yan H, Li N, Zhang Y. Degarelix versus goserelin plus bicalutamide therapy for lower urinary tract symptom relief, prostate volume reduction and quality of life improvement in men with prostate cancer: a systematic review and meta-analysis. Urol Int 2014;93(2):152–9. [31] Russo A, Castiglione F, Salonia A, Benigni F, Rigatti P, Montorsi F, et al. Effects of the gonadotropin-releasing hormone antagonist ganirelix on normal micturition and prostaglandin E(2)-induced detrusor overactivity in conscious female rats. Eur Urol 2010;59(5):868–74. [32] Rick FG, Schally AV, Block NL, Halmos G, Perez R, Fernandez JB, et al. LHRH antagonist cetrorelix reduces prostate size and gene expression of proinflammatory cytokines and growth factors in a rat model of benign prostatic hyperplasia. Prostate 2011;71(7):736–47. [33] Fiorino C, Cozzarini C, Rancati T, et al. Modelling the impact of fractionation on late urinary toxicity after postprostatectomy radiation therapy. Int J Radiat Oncol Biol Phys 2014;90:1250–7. [34] Fiorino C, Cozzarini C, et al. Image guided hypofractionated postprostatectomy intensity modulated radiation therapy for prostate cancer. In regard to Lewis et al.. Int J Radiat Oncol Biol Phys 2016;94:859–60. [35] Landoni V, Fiorino C, Cozzarini C, et al. Predicting toxicity in radiotherapy for prostate cancer. Phys Med 2016;32:521–32.

Please cite this article in press as: Improta I et al. Bladder spatial-dose descriptors correlate with acute urinary toxicity after radiation therapy for prostate cancer. Phys. Med. (2016), http://dx.doi.org/10.1016/j.ejmp.2016.08.013