Arjen van der Schaaf

NTCP models for patient-rated xerostomia and sticky saliva after treatment with intensity modulated radiotherapy for head and neck cancer: The role of dosimetric and clinical factors

PURPOSE The purpose of this multicentre prospective study was to develop multivariable logistic regression models to make valid predictions about the risk of moderate-to-severe patient-rated xerostomia (XER(M6)) and sticky saliva 6 months (STIC(M6)) after primary treatment with intensity modulated radiotherapy (IMRT) with or without chemotherapy for head and neck cancer (HNC). METHODS AND MATERIALS The study population was composed of 178 consecutive HNC patients treated with IMRT. All patients were included in a standard follow up programme in which acute and late side effects and quality of life were prospectively assessed, prior to, during and after treatment. The primary endpoints were XER(M6) and STIC(M6) as assessed by the EORTC QLQ-H&N35 after completing IMRT. Organs at risk (OARs) potentially involved in salivary function were delineated on planning-CT, including the parotid, submandibular and sublingual glands and the minor glands in the soft palate, cheeks and lips. Patients with moderate-to-severe xerostomia or sticky saliva, respectively, at baseline were excluded. The optimal number of variables for a multivariate logistic regression model was determined using a bootstrapping method. RESULTS Eventually, 51.6% of the cases suffered from XER(M6). The multivariate analysis showed that the mean contralateral parotid gland dose and baseline xerostomia (none vs. a bit) were the most important predictors for XER(M6). For the multivariate NTCP model, the area under the receiver operating curve (AUC) was 0.68 (95% CI 0.60-0.76) and the discrimination slope was 0.10, respectively. Calibration was good with a calibration slope of 1.0. At 6 months after IMRT, 35.6% of the cases reported STIC(M6). The mean contralateral submandibular gland dose, the mean sublingual dose and the mean dose to the minor salivary glands located in the soft palate were most predictive for STIC(M6). For this model, the AUC was 0.70 (95% CI 0.61-0.78) and the discrimination slope was 0.12. Calibration was good with a calibration slope of 1.0. CONCLUSIONS The multivariable NTCP models presented in this paper can be used to predict patient-rated xerostomia and sticky saliva. The dose volume parameters included in the models can be used to further optimise IMRT treatment.

Sparing the region of the salivary gland containing stem cells preserves saliva production after radiotherapy for head and neck cancer

Avoiding irradiation of the region of the parotid gland containing stem cells reduces the risk of xerostomia (dry mouth). Preserving saliva flow after radiotherapy Radiotherapy for head and neck cancer may damage the salivary glands, resulting in reduced salivation with consequent xerostomia (dry mouth). Xerostomia affects the quality of life of patients with head and neck cancer. van Luijk and co-workers reported the location of salivary (parotid) gland stem cells in the mouse, rat, and human. Next, they showed in rat and human that irradiation of the salivary gland region containing the highest number of stem cells resulted in the greatest loss of saliva production after treatment. Finally, the authors showed that it is possible to avoid irradiation of this specific area during therapy, which may reduce the patient’s risk of developing post-radiotherapy xerostomia. Each year, 500,000 patients are treated with radiotherapy for head and neck cancer, resulting in relatively high survival rates. However, in 40% of patients, quality of life is severely compromised because of radiation-induced impairment of salivary gland function and consequent xerostomia (dry mouth). New radiation treatment technologies enable sparing of parts of the salivary glands. We have determined the parts of the major salivary gland, the parotid gland, that need to be spared to ensure that the gland continues to produce saliva after irradiation treatment. In mice, rats, and humans, we showed that stem and progenitor cells reside in the region of the parotid gland containing the major ducts. We demonstrated in rats that inclusion of the ducts in the radiation field led to loss of regenerative capacity, resulting in long-term gland dysfunction with reduced saliva production. Then we showed in a cohort of patients with head and neck cancer that the radiation dose to the region of the salivary gland containing the stem/progenitor cells predicted the function of the salivary glands one year after radiotherapy. Finally, we showed that this region of the salivary gland could be spared during radiotherapy, thus reducing the risk of post-radiotherapy xerostomia.

The potential of intensity-modulated proton radiotherapy to reduce swallowing dysfunction in the treatment of head and neck cancer: A planning comparative study.

Abstract Background. Predictive models for swallowing dysfunction were developed previously and showed the potential of improved intensity-modulated radiotherapy to reduce the risk of swallowing dysfunction. Still the risk is high. The aim of this study was to determine the potential of swallowing-sparing (SW) intensity-modulated proton therapy (IMPT) in head and neck cancer (HNC) for reducing the risk of swallowing dysfunction relative to currently used photon therapy. Material and methods. Twenty-five patients with oropharyngeal (n = 21) and hypopharyngeal (n = 4) cancer received primary radiotherapy, including bilateral neck irradiation, using standard (ST) intensity-modulated photon therapy (IMRT). Prophylactic (54 Gy) and therapeutic (70 Gy) target volumes were defined. The dose to the parotid and submandibular glands was reduced as much as possible. Four additional radiotherapy plans were created for each patient: SW-IMRT, ST-IMPT, 3-beam SW-IMPT (3B-SW-IMPT) and 7-beam SW-IMPT (7B-SW-IMPT). All plans were optimized similarly, with additional attempts to spare the swallowing organs at risk (SWOARs) in the SW plans. Probabilities of swallowing dysfunction were calculated with recently developed predictive models. Results. All plans complied with standard HNC radiotherapy objectives. The mean parotid gland doses were similar for the ST and SW photon plans, but clearly lower in all IMPT plans (ipsilateral parotid gland ST-IMRT: 46 Gy, 7B-SW-IMPT: 29 Gy). The mean dose in the SWOARs was lowest with SW-IMPT, in particular with 7B-SW-IMPT (supraglottic larynx ST-IMRT: 60 Gy, 7B-SW-IMPT: 40 Gy). The observed dose reductions to the SWOARs translated into substantial overall reductions in normal tissue complication risks for different swallowing dysfunction endpoints. Compared with ST-IMRT, the risk of physician-rated grade 2–4 swallowing dysfunction was reduced on average by 8.8% (95% CI 6.5–11.1%) with SW-IMRT, and by 17.2% (95% CI: 12.7–21.7%) with 7B-SW-IMPT. Conclusion. SWOAR-sparing with proton therapy has the potential to substantially reduce the risk of swallowing dysfunction compared to similar treatment with photons.

Functional Diffusion Tensor Imaging: Measuring Task-Related Fractional Anisotropy Changes in the Human Brain along White Matter Tracts

Background Functional neural networks in the human brain can be studied from correlations between activated gray matter regions measured with fMRI. However, while providing important information on gray matter activation, no information is gathered on the co-activity along white matter tracts in neural networks. Methodology/Principal Findings We report on a functional diffusion tensor imaging (fDTI) method that measures task-related changes in fractional anisotropy (FA) along white matter tracts. We hypothesize that these fractional anisotropy changes relate to morphological changes of glial cells induced by axonal activity although the exact physiological underpinnings of the measured FA changes remain to be elucidated. As expected, these changes are very small as compared to the physiological noise and a reliable detection of the signal change would require a large number of measurements. However, a substantial increase in signal-to-noise ratio was achieved by pooling the signal over the complete fiber tract. Adopting such a tract-based statistics enabled us to measure the signal within a practically feasible time period. Activation in the sensory thalamocortical tract and optic radiation in eight healthy human subjects was found during tactile and visual stimulation, respectively. Conclusions/Significance The results of our experiments indicate that these FA changes may serve as a functional contrast mechanism for white matter. This noninvasive fDTI method may provide a new approach to study functional neural networks in the human brain.

Development of a multivariable normal tissue complication probability (NTCP) model for tube feeding dependence after curative radiotherapy/chemo-radiotherapy in head and neck cancer

BACKGROUND AND PURPOSE Curative radiotherapy/chemo-radiotherapy for head and neck cancer (HNC) may result in severe acute and late side effects, including tube feeding dependence. The purpose of this prospective cohort study was to develop a multivariable normal tissue complication probability (NTCP) model for tube feeding dependence 6 months (TUBEM6) after definitive radiotherapy, radiotherapy plus cetuximab or concurrent chemoradiation based on pre-treatment and treatment characteristics. MATERIALS AND METHODS The study included 355 patients with HNC. TUBEM6 was scored prospectively in a standard follow-up program. To design the prediction model, the penalized learning method LASSO was used, with TUBEM6 as the endpoint. RESULTS The prevalence of TUBEM6 was 10.7%. The multivariable model with the best performance consisted of the variables: advanced T-stage, moderate to severe weight loss at baseline, accelerated radiotherapy, chemoradiation, radiotherapy plus cetuximab, the mean dose to the superior and inferior pharyngeal constrictor muscle, to the contralateral parotid gland and to the cricopharyngeal muscle. CONCLUSIONS We developed a multivariable NTCP model for TUBEM6 to identify patients at risk for tube feeding dependence. The dosimetric variables can be used to optimize radiotherapy treatment planning aiming at prevention of tube feeding dependence and to estimate the benefit of new radiation technologies.

The Quest for Evidence for Proton Therapy: Model-Based Approach and Precision Medicine

PURPOSE Reducing dose to normal tissues is the advantage of protons versus photons. We aimed to describe a method for translating this reduction into a clinically relevant benefit. METHODS AND MATERIALS Dutch scientific and health care governance bodies have recently issued landmark reports regarding generation of relevant evidence for new technologies in health care including proton therapy. An approach based on normal tissue complication probability (NTCP) models has been adopted to select patients who are most likely to experience fewer (serious) adverse events achievable by state-of-the-art proton treatment. RESULTS By analogy with biologically targeted therapies, the technology needs to be tested in enriched cohorts of patients exhibiting the decisive predictive marker: difference in normal tissue dosimetric signatures between proton and photon treatment plans. Expected clinical benefit is then estimated by virtue of multifactorial NTCP models. In this sense, high-tech radiation therapy falls under precision medicine. As a consequence, randomizing nonenriched populations between photons and protons is predictably inefficient and likely to produce confusing results. CONCLUSIONS Validating NTCP models in appropriately composed cohorts treated with protons should be the primary research agenda leading to urgently needed evidence for proton therapy.

Validation and Modification of a Prediction Model for Acute Cardiac Events in Patients With Breast Cancer Treated With Radiotherapy Based on Three-Dimensional Dose Distributions to Cardiac Substructures

Purpose A relationship between mean heart dose (MHD) and acute coronary event (ACE) rate was reported in a study of patients with breast cancer (BC). The main objective of our cohort study was to validate this relationship and investigate if other dose-distribution parameters are better predictors for ACEs than MHD. Patients and Methods The cohort consisted of 910 consecutive female patients with BC treated with radiotherapy (RT) after breast-conserving surgery. The primary end point was cumulative incidence of ACEs within 9 years of follow-up. Both MHD and various dose-distribution parameters of the cardiac substructures were collected from three-dimensional computed tomography planning data. Results The median MHD was 2.37 Gy (range, 0.51 to 15.25 Gy). The median follow-up time was 7.6 years (range, 0.1 to 10.1 years), during which 30 patients experienced an ACE. The cumulative incidence of ACE increased by 16.5% per Gy (95% CI, 0.6 to 35.0; P = .042). Analysis showed that the volume of the left ventricle receiving 5 Gy (LV-V5) was the most important prognostic dose-volume parameter. The most optimal multivariable normal tissue complication probability model for ACEs consisted of LV-V5, age, and weighted ACE risk score per patient (c-statistic, 0.83; 95% CI, 0.75 to 0.91). Conclusion A significant dose-effect relationship was found for ACEs within 9 years after RT. Using MHD, the relative increase per Gy was similar to that reported in the previous study. In addition, LV-V5 seemed to be a better predictor for ACEs than MHD. This study confirms the importance of reducing exposure of the heart to radiation to avoid excess risk of ACEs after radiotherapy for BC.

Multivariate modeling of complications with data driven variable selection: Guarding against overfitting and effects of data set size

PURPOSE Multivariate modeling of complications after radiotherapy is frequently used in conjunction with data driven variable selection. This study quantifies the risk of overfitting in a data driven modeling method using bootstrapping for data with typical clinical characteristics, and estimates the minimum amount of data needed to obtain models with relatively high predictive power. MATERIALS AND METHODS To facilitate repeated modeling and cross-validation with independent datasets for the assessment of true predictive power, a method was developed to generate simulated data with statistical properties similar to real clinical data sets. Characteristics of three clinical data sets from radiotherapy treatment of head and neck cancer patients were used to simulate data with set sizes between 50 and 1000 patients. A logistic regression method using bootstrapping and forward variable selection was used for complication modeling, resulting for each simulated data set in a selected number of variables and an estimated predictive power. The true optimal number of variables and true predictive power were calculated using cross-validation with very large independent data sets. RESULTS For all simulated data set sizes the number of variables selected by the bootstrapping method was on average close to the true optimal number of variables, but showed considerable spread. Bootstrapping is more accurate in selecting the optimal number of variables than the AIC and BIC alternatives, but this did not translate into a significant difference of the true predictive power. The true predictive power asymptotically converged toward a maximum predictive power for large data sets, and the estimated predictive power converged toward the true predictive power. More than half of the potential predictive power is gained after approximately 200 samples. Our simulations demonstrated severe overfitting (a predicative power lower than that of predicting 50% probability) in a number of small data sets, in particular in data sets with a low number of events (median: 7, 95th percentile: 32). Recognizing overfitting from an inverted sign of the estimated model coefficients has a limited discriminative value. CONCLUSIONS Despite considerable spread around the optimal number of selected variables, the bootstrapping method is efficient and accurate for sufficiently large data sets, and guards against overfitting for all simulated cases with the exception of some data sets with a particularly low number of events. An appropriate minimum data set size to obtain a model with high predictive power is approximately 200 patients and more than 32 events. With fewer data samples the true predictive power decreases rapidly, and for larger data set sizes the benefit levels off toward an asymptotic maximum predictive power.

Acute symptoms during the course of head and neck radiotherapy or chemoradiation are strong predictors of late dysphagia

PURPOSE To determine if acute symptoms during definitive radiotherapy (RT) or chemoradiation (CHRT) are prognostic factors for late dysphagia in head and neck cancer (HNC). MATERIAL AND METHODS This prospective cohort study consisted of 260 HNC patients who received definitive RT or CHRT. The primary endpoint was grade 2-4 swallowing dysfunction at 6 months after completing RT (SWALM6). During treatment, acute symptoms, including oral mucositis, xerostomia and dysphagia, were scored, and the scores were accumulated weekly and entered into an existing reference model for SWALM6 that consisted of dose-volume variables only. RESULTS Both acute xerostomia and dysphagia were strong prognostic factors for SWALM6. When acute scores were added as variables to the reference model, model performance increased as the course of treatment progressed: the AUC rose from 0.78 at the baseline to 0.85 in week 6. New models built for weeks 3-6 were significantly better able to identify patients with and without late dysphagia. CONCLUSION Acute xerostomia and dysphagia during the course of RT are strong prognostic factors for late dysphagia. Including accumulated acute symptom scores on a weekly basis in prediction models for late dysphagia significantly improves the identification of high-risk and low-risk patients at an early stage during treatment and might facilitate individualized treatment adaptation.

Impact of statistical learning methods on the predictive power of multivariate normal tissue complication probability models.

PURPOSE To study the impact of different statistical learning methods on the prediction performance of multivariate normal tissue complication probability (NTCP) models. METHODS AND MATERIALS In this study, three learning methods, stepwise selection, least absolute shrinkage and selection operator (LASSO), and Bayesian model averaging (BMA), were used to build NTCP models of xerostomia following radiotherapy treatment for head and neck cancer. Performance of each learning method was evaluated by a repeated cross-validation scheme in order to obtain a fair comparison among methods. RESULTS It was found that the LASSO and BMA methods produced models with significantly better predictive power than that of the stepwise selection method. Furthermore, the LASSO method yields an easily interpretable model as the stepwise method does, in contrast to the less intuitive BMA method. CONCLUSIONS The commonly used stepwise selection method, which is simple to execute, may be insufficient for NTCP modeling. The LASSO method is recommended.