skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Multivariable modeling of radiotherapy outcomes, including dose-volume and clinical factors

Abstract

Purpose: The probability of a specific radiotherapy outcome is typically a complex, unknown function of dosimetric and clinical factors. Current models are usually oversimplified. We describe alternative methods for building multivariable dose-response models. Methods: Representative data sets of esophagitis and xerostomia are used. We use a logistic regression framework to approximate the treatment-response function. Bootstrap replications are performed to explore variable selection stability. To guard against under/overfitting, we compare several analytical and data-driven methods for model-order estimation. Spearman's coefficient is used to evaluate performance robustness. Novel graphical displays of variable cross correlations and bootstrap selection are demonstrated. Results: Bootstrap variable selection techniques improve model building by reducing sample size effects and unveiling variable cross correlations. Inference by resampling and Bayesian approaches produced generally consistent guidance for model order estimation. The optimal esophagitis model consisted of 5 dosimetric/clinical variables. Although the xerostomia model could be improved by combining clinical and dose-volume factors, the improvement would be small. Conclusions: Prediction of treatment response can be improved by mixing clinical and dose-volume factors. Graphical tools can mitigate the inherent complexity of multivariable modeling. Bootstrap-based variable selection analysis increases the reliability of reported models. Statistical inference methods combined with Spearman's coefficient provide an efficientmore » approach to estimating optimal model order.« less

Authors:
 [1];  [1];  [1];  [1];  [1];  [1];  [2]
  1. Department of Radiation Oncology, Washington University, St. Louis, MO (United States)
  2. Department of Radiation Oncology, Washington University, St. Louis, MO (United States). E-mail: deasy@wustl.edu
Publication Date:
OSTI Identifier:
20793410
Resource Type:
Journal Article
Resource Relation:
Journal Name: International Journal of Radiation Oncology, Biology and Physics; Journal Volume: 64; Journal Issue: 4; Other Information: DOI: 10.1016/j.ijrobp.2005.11.022; PII: S0360-3016(05)02971-8; Copyright (c) 2006 Elsevier Science B.V., Amsterdam, Netherlands, All rights reserved; Country of input: International Atomic Energy Agency (IAEA)
Country of Publication:
United States
Language:
English
Subject:
62 RADIOLOGY AND NUCLEAR MEDICINE; FORECASTING; PERFORMANCE; RADIATION DOSES; RADIOTHERAPY; RELIABILITY; RESPONSE FUNCTIONS; SIMULATION

Citation Formats

El Naqa, Issam, Bradley, Jeffrey, Blanco, Angel I., Lindsay, Patricia E., Vicic, Milos, Hope, Andrew, and Deasy, Joseph O.. Multivariable modeling of radiotherapy outcomes, including dose-volume and clinical factors. United States: N. p., 2006. Web. doi:10.1016/J.IJROBP.2005.1.
El Naqa, Issam, Bradley, Jeffrey, Blanco, Angel I., Lindsay, Patricia E., Vicic, Milos, Hope, Andrew, & Deasy, Joseph O.. Multivariable modeling of radiotherapy outcomes, including dose-volume and clinical factors. United States. doi:10.1016/J.IJROBP.2005.1.
El Naqa, Issam, Bradley, Jeffrey, Blanco, Angel I., Lindsay, Patricia E., Vicic, Milos, Hope, Andrew, and Deasy, Joseph O.. Wed . "Multivariable modeling of radiotherapy outcomes, including dose-volume and clinical factors". United States. doi:10.1016/J.IJROBP.2005.1.
@article{osti_20793410,
title = {Multivariable modeling of radiotherapy outcomes, including dose-volume and clinical factors},
author = {El Naqa, Issam and Bradley, Jeffrey and Blanco, Angel I. and Lindsay, Patricia E. and Vicic, Milos and Hope, Andrew and Deasy, Joseph O.},
abstractNote = {Purpose: The probability of a specific radiotherapy outcome is typically a complex, unknown function of dosimetric and clinical factors. Current models are usually oversimplified. We describe alternative methods for building multivariable dose-response models. Methods: Representative data sets of esophagitis and xerostomia are used. We use a logistic regression framework to approximate the treatment-response function. Bootstrap replications are performed to explore variable selection stability. To guard against under/overfitting, we compare several analytical and data-driven methods for model-order estimation. Spearman's coefficient is used to evaluate performance robustness. Novel graphical displays of variable cross correlations and bootstrap selection are demonstrated. Results: Bootstrap variable selection techniques improve model building by reducing sample size effects and unveiling variable cross correlations. Inference by resampling and Bayesian approaches produced generally consistent guidance for model order estimation. The optimal esophagitis model consisted of 5 dosimetric/clinical variables. Although the xerostomia model could be improved by combining clinical and dose-volume factors, the improvement would be small. Conclusions: Prediction of treatment response can be improved by mixing clinical and dose-volume factors. Graphical tools can mitigate the inherent complexity of multivariable modeling. Bootstrap-based variable selection analysis increases the reliability of reported models. Statistical inference methods combined with Spearman's coefficient provide an efficient approach to estimating optimal model order.},
doi = {10.1016/J.IJROBP.2005.1},
journal = {International Journal of Radiation Oncology, Biology and Physics},
number = 4,
volume = 64,
place = {United States},
year = {Wed Mar 15 00:00:00 EST 2006},
month = {Wed Mar 15 00:00:00 EST 2006}
}
  • Purpose: To analyze the tolerance dose for retention of visual acuity in patients with head-and-neck tumors treated with carbon ion radiotherapy. Methods and Materials: From June 1994 to March 2000, 163 patients with tumors in the head and neck or skull base region were treated with carbon ion radiotherapy. Analysis was performed on 54 optic nerves (ONs) corresponding to 30 patients whose ONs had been included in the irradiated volume. These patients showed no evidence of visual impairment due to other factors and had a follow-up period of >4 years. All patients had been informed of the possibility of visualmore » impairment before treatment. We evaluated the dose-complication probability and the prognostic factors for the retention of visual acuity in carbon ion radiotherapy, using dose-volume histograms and multivariate analysis. Results: The median age of 30 patients (14 men, 16 women) was 57.2 years. Median prescribed total dose was 56.0 gray equivalents (GyE) at 3.0-4.0 GyE per fraction per day (range, 48-64 GyE; 16-18 fractions; 4-6 weeks). Of 54 ONs that were analyzed, 35 had been irradiated with <57 GyE (maximum dose [D{sub max}]) resulting in no visual loss. Conversely, 11 of the 19 ONs (58%) irradiated with >57 GyE (D{sub max}) suffered a decrease of visual acuity. In all of these cases, the ONs had been involved in the tumor before carbon ion radiotherapy. In the multivariate analysis, a dose of 20% of the volume of the ON (D{sub 2}) was significantly associated with visual loss. Conclusions: The occurrence of visual loss seems to be correlated with a delivery of >60 GyE to 20% of the volume of the ON.« less
  • Purpose: To study the impact of clinical predisposing factors on rectal normal tissue complication probability modeling using the updated results of the Dutch prostate dose-escalation trial. Methods and Materials: Toxicity data of 512 patients (conformally treated to 68 Gy [n = 284] and 78 Gy [n = 228]) with complete follow-up at 3 years after radiotherapy were studied. Scored end points were rectal bleeding, high stool frequency, and fecal incontinence. Two traditional dose-based models (Lyman-Kutcher-Burman (LKB) and Relative Seriality (RS) and a logistic model were fitted using a maximum likelihood approach. Furthermore, these model fits were improved by including themore » most significant clinical factors. The area under the receiver operating characteristic curve (AUC) was used to compare the discriminating ability of all fits. Results: Including clinical factors significantly increased the predictive power of the models for all end points. In the optimal LKB, RS, and logistic models for rectal bleeding and fecal incontinence, the first significant (p = 0.011-0.013) clinical factor was 'previous abdominal surgery.' As second significant (p = 0.012-0.016) factor, 'cardiac history' was included in all three rectal bleeding fits, whereas including 'diabetes' was significant (p = 0.039-0.048) in fecal incontinence modeling but only in the LKB and logistic models. High stool frequency fits only benefitted significantly (p = 0.003-0.006) from the inclusion of the baseline toxicity score. For all models rectal bleeding fits had the highest AUC (0.77) where it was 0.63 and 0.68 for high stool frequency and fecal incontinence, respectively. LKB and logistic model fits resulted in similar values for the volume parameter. The steepness parameter was somewhat higher in the logistic model, also resulting in a slightly lower D{sub 50}. Anal wall DVHs were used for fecal incontinence, whereas anorectal wall dose best described the other two endpoints. Conclusions: Comparable prediction models were obtained with LKB, RS, and logistic NTCP models. Including clinical factors improved the predictive power of all models significantly.« less
  • Purpose: To report preliminary clinical outcomes for pediatric patients treated with proton beam radiation for intracranial ependymoma and compare the dose distributions of intensity-modulated radiation therapy with photons (IMRT), three-dimensional conformal proton radiation, and intensity-modulated proton radiation therapy (IMPT) for representative patients. Methods and Materials: All children with intracranial ependymoma confined to the supratentorial or infratentorial brain treated at the Francis H. Burr Proton Facility and Harvard Cyclotron between November 2000 and March 2006 were included in this study. Seventeen patients were treated with protons. Proton, IMRT, and IMPT plans were generated with similar clinical constraints for representative infratentorial andmore » supratentorial ependymoma cases. Tumor and normal tissue dose-volume histograms were calculated and compared. Results: At a median follow-up of 26 months from the start date of radiation therapy, local control, progression-free survival, and overall survival rates were 86%, 80%, and 89%, respectively. Subtotal resection was significantly associated with decreased local control (p = 0.016). Similar tumor volume coverage was achieved with IMPT, proton therapy, and IMRT. Substantial normal tissue sparing was seen with proton therapy compared with IMRT. Use of IMPT will allow for additional sparing of some critical structures. Conclusions: Preliminary disease control with proton therapy compares favorably with the literature. Dosimetric comparisons show the advantage of proton radiation compared with IMRT in the treatment of ependymoma. Further sparing of normal structures appears possible with IMPT. Superior dose distributions were accomplished with fewer beam angles with the use of protons and IMPT.« less
  • Purpose: To compare toxicity profiles and biochemical tumor control outcomes between patients treated with high-dose image-guided radiotherapy (IGRT) and high-dose intensity-modulated radiotherapy (IMRT) for clinically localized prostate cancer. Materials and Methods: Between 2008 and 2009, 186 patients with prostate cancer were treated with IGRT to a dose of 86.4 Gy with daily correction of the target position based on kilovoltage imaging of implanted prostatic fiducial markers. This group of patients was retrospectively compared with a similar cohort of 190 patients who were treated between 2006 and 2007 with IMRT to the same prescription dose without, however, implanted fiducial markers inmore » place (non-IGRT). The median follow-up time was 2.8 years (range, 2-6 years). Results: A significant reduction in late urinary toxicity was observed for IGRT patients compared with the non-IGRT patients. The 3-year likelihood of grade 2 and higher urinary toxicity for the IGRT and non-IGRT cohorts were 10.4% and 20.0%, respectively (p = 0.02). Multivariate analysis identifying predictors for grade 2 or higher late urinary toxicity demonstrated that, in addition to the baseline Internatinoal Prostate Symptom Score, IGRT was associated with significantly less late urinary toxicity compared with non-IGRT. The incidence of grade 2 and higher rectal toxicity was low for both treatment groups (1.0% and 1.6%, respectively; p = 0.81). No differences in prostate-specific antigen relapse-free survival outcomes were observed for low- and intermediate-risk patients when treated with IGRT and non-IGRT. For high-risk patients, a significant improvement was observed at 3 years for patients treated with IGRT compared with non-IGRT. Conclusions: IGRT is associated with an improvement in biochemical tumor control among high-risk patients and a lower rate of late urinary toxicity compared with high-dose IMRT. These data suggest that, for definitive radiotherapy, the placement of fiducial markers and daily tracking of target positioning may represent the preferred mode of external-beam radiotherapy delivery for the treatment of prostate cancer.« less
  • Purpose: To determine in a prospective study, the correlation between radiation dose/volume, clinical toxicities and patient-reported, quality of life (QOL) resulting from lung SBRT. Methods: For 106 non-small cell lung cancer (NSCLC) patients receiving SBRT (12 Gy × 4), symptoms including cough, dyspnea, fatigue and pneumonitis were measured at baseline (before treatment), after treatment and 3, 6, and 12 months post-treatment. Toxicity was graded from zero to five. Dosimetric parameters such as the MLD, D10%, D20%, and lung subvolumes (V10 and V20) were obtained from the treatment plan. Dosimetric parameters and number of patients demonstrating toxicity ≥ grade 2 weremore » tabulated. Linear regression analysis was used to calculate correlations between MLD and D10, D20, V10 and V20. Results: The percentages of patients with > grade 2 pneumonitis, fatigue, cough, and dyspnea over 3 to 12 months increased from 0.0% to 3.5%, 3.2% to 10.5%, 4.3% to 8.3%, and 10.8% to 18.8%, respectively. Computed dose indices D10%, D20% were 7.9±4.8 Gy and 3.0±2.3 Gy, respectively. MLD ranged from 0.34 Gy up to 9.9 Gy with overall average 3.0±1.7 Gy. The averages of the subvolumes V10 and V20 were respectively 8.9±5.3% and 3.0±2.4%. The linear regression analysis showed that V10 and D10 demonstrated the strongest correlation to MLD; R2= 0.92 and 0.87, respectively. V20, and D20 were also strongly correlated with MLD; R2 = 0.81 and 0.84 respectively. A correlation was also found to exist between MLD > 2 Gy and ≥ grade 2 cough and dyspnea. Subvolume values for 2Gy MLD were 5.3% for V10 and 2% for V20. Conclusion: Dosimetric indices: MLD ≥ 2Gy, D10 ≥ 5Gy and V10 ≥ 5% of the total lung volume were predictive of > grade 2 cough and dyspnea QOL data. The QOL results are a novel component of this work. acknowledgement of the Varian grant support.« less