Treatment plans for head and neck (H&N) cancer are highly complex due to multiple dose prescription levels and numerous organs at risk (OARs) close to the target. The plan quality is inter-planner dependent since it is dependent on the skills and experience of the dosimetrist. This study presents a blinded prospective clinical comparison of automatic (AU) and manually (MA) generated H&N VMAT plans made for clinical use. MA and AU plans were generated for 30 consecutive patients in Pinnacle3 using the IMRT optimisation module and the new Autoplan module, respectively. The plan quality was blindedly compared by three senior oncologists and the best plan was selected for treatment of the patient. Planning time was measured as the active operator time used. The plan quality was analysed with DVH metrics and the dose delivery accuracy validated on the ArcCheck phantom. For twenty-nine out of the thirty patients the AU plan was chosen for treatment. Target doses were more homogenous with the AU plans and the OAR doses were significantly reduced, between 0.5 and 6.5 Gy. The average operator time spent on creating a manual plan was 64 min which was halved by Autoplan. The AU plans were more modulated as illustrated by an increase in MUs, which might cause the slightly lower pass rate of 97.7% in the ArcCheck measurements. Target doses were similar between MA and AU plan, while AU plans spared all OAR considerably better than the MA plans.
Federated learning has the potential to perfrom analysis on decentralised data; however, there are some obstacles to survival analyses as there is a risk of data leakage. This study demonstrates how to perform a stratified Cox regression survival analysis specifically designed to avoid data leakage using federated learning on larynx cancer patients from centres in three different countries. Data were obtained from 1821 larynx cancer patients treated with radiotherapy in three centres. Tumour volume was available for all 786 of the included patients. Parameter selection among eleven clinical and radiotherapy parameters were performed using best subset selection and cross-validation through the federated learning system, AusCAT. After parameter selection, β regression coefficients were estimated using bootstrap. Calibration plots were generated at 2 and 5-years survival, and inner and outer risk groups' Kaplan-Meier curves were compared to the Cox model prediction. The best performing Cox model included log(GTV), performance status, age, smoking, haemoglobin and N-classification; however, the simplest model with similar statistical prediction power included log(GTV) and performance status only. The Harrell C-indices for the simplest model were for Odense, Christie and Liverpool 0.75[0.71-0.78], 0.65[0.59-0.71], and 0.69[0.59-0.77], respectively. The values are slightly higher for the full model with C-index 0.77[0.74-0.80], 0.67[0.62-0.73] and 0.71[0.61-0.80], respectively. Smoking during treatment has the same hazard as a ten-years older nonsmoking patient. Without any patient-specific data leaving the hospitals, a stratified Cox regression model based on data from centres in three countries was developed without data leakage risks. The overall survival model is primarily driven by tumour volume and performance status.
Introduction: Prediction models using logistic regression may perform poorly in external patient cohorts. However, there is a need to standardize and validate models for clinical use. The purpose of this project was to describe a method for validation of external NTCP models used for patient selection in the randomized trial of protons versus photons in head and neck cancer radiotherapy, DAHANCA 35. Material and methods: Organs at risk of 588 patients treated primarily with IMRT in the randomized controlled DAHANCA19 trial were retrospectively contoured according to recent international recommendations. Dose metrics were extracted using MatLab and all clinical parameters were retrieved from the DAHANCA database. The model proposed by Christianen et al. to predict physician-rated dysphagia was validated through the closed testing, where change of the model intercept, slope and individual beta’s were tested for significant prediction improvements. Results: Six months prevalence of dysphagia in the validation cohort was 33%. The closed testing procedure for physician-rated dysphagia showed that the Christianen et al. model needed an intercept refitting for the best match for the Danish patients. The intercept update increased the risk of dysphagia for the validation cohort by 7.9 ± 2.5% point. For the raw model performance, the Brier score (mean squared residual) was 0.467, which improved significantly with a new intercept to 0.415. Conclusions: The previously published Dutch dysphagia model needed an intercept update to match the Danish patient cohort. The implementation of a closed testing procedure on the current validation cohort allows quick and efficient validation of external NTCP models for patient selection in the future.
3 March 2022 in Journal of Medical Imaging and Radiation Oncology
Background: The quality of radiotherapy planning has improved substantially in the last decade with the introduction of intensity modulated radiotherapy. The purpose of this study was to analyze the plan quality and efficacy of automatically (AU) generated VMAT plans for inoperable esophageal cancer patients. Material and Methods: Thirty-two consecutive inoperable patients with esophageal cancer originally treated with manually (MA) generated volumetric modulated arc therapy (VMAT) plans were retrospectively replanned using an auto-planning engine. All plans were optimized with one full 6MV VMAT arc giving 60 Gy to the primary target and 50 Gy to the elective target. The planning techniques were blinded before clinical evaluation by three specialized oncologists. To supplement the clinical evaluation, the optimization time for the AU plan was recorded along with DVH parameters for all plans. Results: Upon clinical evaluation, the AU plan was preferred for 31/32 patients, and for one patient, there was no difference in the plans. In terms of DVH parameters, similar target coverage was obtained between the two planning methods. The mean dose for the spinal cord increased by 1.8 Gy using AU (p = .002), whereas the mean lung dose decreased by 1.9 Gy (p < .001). The AU plans were more modulated as seen by the increase of 12% in mean MUs (p = .001). The median optimization time for AU plans was 117 min. Conclusions: The AU plans were in general preferred and showed a lower mean dose to the lungs. The automation of the planning process generated esophageal cancer treatment plans quickly and with high quality.