Dependence of the AUC of NTCP models on the observational dose-range highlights cautions in comparison of discriminative performance.

BACKGROUND: Normal tissue complication probability (NTCP) models are probabilistic models that describe the risk of radio-induced toxicity in tissues or organs. In the field of radiotherapy, the area under the ROC curve (AUC) is widely used to estimate the performance in risk prediction of NTCP models. METHODS: In this work, we derived an analytical expression of the AUC for the logistic NTCP model in the case of both symmetrical and asymmetrical dose (to the normal tissue) windows around D50. Using numerical simulations, we studied the behavior of the AUC in general clinical settings, enforcing non-logistic NTCP models (Lyman-Kutcher-Burman and LogEUD) and including risk factors beyond the dose. We validated our findings using real-world radiotherapy data sets of prostate cancer patients. RESULTS: Our analytical expression of the AUC made explicit the dependence on both the steepness of the logistic curve (ß) and the dose window width (w), showing that an increase of w pushes AUC towards higher values. Increasing values of the AUC with increasing values of w were consistently observed across simulated data sets with diverse clinical settings from published studies and real clinical data sets. CONCLUSION: Our results reveal that the AUC of NTCP models inherits intrinsic characteristics from the clinical setting of the data set on which the models are developed, and warn against the use of the AUC to compare the performance of models constructed upon data from trials in which substantially different dose ranges were administered or accounting for different risk factors beyond the dose.

Bladder dose-surface maps and urinary toxicity: Robustness with respect to motion in assessing local dose effects.

The purpose of this study was to quantify the impact of inter-fraction modifications of bladder during RT of prostate cancer on bladder dose surface maps (DSM). Eighteen patients treated with daily image-guided Tomotherapy and moderate hypofractionation (70-72.8Gy at 2.5-2.6Gy/fr in 28 fractions and full bladder) were considered. Bladder contours were delineated on co-registered daily Megavoltage CT (MVCT) by a single observer and copied on the planning CT to generate dose-volume/surface histograms (DVH/DSH) and bladder DSMs. Discrepancies between planned and daily absorbed doses were analyzed through the average of individual systematic errors, the population systematic errors and the population random errors for the DVH/DSHs and DSMs. In total, 477 DVH/DSH and 472 DSM were available. DSH and DVH showed small population systematic errors of absolute surfaces (<3.4cm(2)) and volumes (<8.4cm(3)) at the highest doses. The dose to the posterior bladder base assessed on DSMs showed a mean systematic error below 1Gy, with population systematic and random errors within 4 and 3Gy, respectively. The region surrounding this area shows higher mean systematic errors (1-3Gy), population systematic (8-11Gy) and random (5-7Gy) errors. In conclusion, DVH/DSH and DSMs are quite stable with respect to inter-fraction variations in the high-dose region, within about 2cm from bladder base. Larger systematic variations occur in the anterior portion and cranially 2.5-3.5cm from the base. Results suggest that dose predictors related to the high dose area (including the trigone dose) are likely to be sufficiently reliable with respect to the expected variations due to variable bladder filling.

Modelling late stool frequency and rectal pain after radical radiotherapy in prostate cancer patients: Results from a large pooled population.

AIM: To investigate late gastrointestinal toxicity in a large pooled population of prostate cancer patients treated with radical radiotherapy. Normal tissue complication probability models were developed for late stool frequency and late rectal pain. METHODS AND MATERIALS: Population included 1336 patients, 3-year minimum follow-up, treated with 66-80Gy. Toxicity was scored with LENT-SOMA-scale. Two toxicity endpoints were considered: grade ⩾2 rectal pain and mean grade (average score during follow-up) in stool frequency >1. DVHs of anorectum were reduced to equivalent uniform dose (EUD). The best-value of the volume parameter n was determined through numerical optimization. Association between EUD/clinical factors and the endpoints was investigated by logistic analyses. Likelihood, Brier-score and calibration were used to evaluate models. External calibration was also carried out. RESULTS: 4% of patients (45/1122) reported mean stool frequency grade >1; grade ⩾2 rectal pain was present in the TROG 03.04 RADAR population only (21/677, 3.1%): for this endpoint, the analysis was limited to this population. Analysis of DVHs highlighted the importance of mid-range doses (30-50Gy) for both endpoints. EUDs calculated with n=1 (OR=1.04) and n=0.35 (OR=1.06) were the most suitable dosimetric descriptors for stool frequency and rectal pain respectively. The final models included EUD and cardiovascular diseases (OR=1.78) for stool frequency and EUD and presence of acute gastrointestinal toxicity (OR=4.2) for rectal pain. CONCLUSION: Best predictors of stool frequency and rectal pain are consistent with findings previously reported for late faecal incontinence, indicating an important role in optimization of mid-range dose region to minimize these symptoms highly impacting the quality-of-life of long surviving patients.

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

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:1.8-2Gy/fr) or moderately hypo-fractionated radiotherapy (HYPO:2.2-2.7Gy/fr) was considered. Urinary toxicity was evaluated through the International Prostate Symptoms Score (IPSS) administered before and after radiotherapy. IPSS increases ⩾10 and 15 points at the end of radiotherapy (ΔIPSS⩾10 and ΔIPSS⩾15) were chosen as endpoints. Average DSMs (corrected into 2Gy-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. ΔIPSS⩾10 was recorded in 76/375 (20%) patients, while 30/375 (8%) experienced ΔIPSS⩾15. The posterior dose at 12mm 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 75Gy 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.