Prioritizing Toxicities in Automated Outcome-Directed Treatment Plan Optimization

We recently reported on quality of life (QOL)-guided radiotherapy including NTCP-based treatment plan optimization. This method employs NTCP objective functions, prioritized based on their impact on QOL, instead of conventional dose-based objective functions for individual organs at risk (OAR). We recently expanded this method to include simple models for tumor control probability (TCP) and general plan quality (e.g., hotspot reduction, target dose homogeneity), allowing for fully automated planning. The aim of this study was to test the feasibility of this method and to study predicted toxicity profiles of various plans by prioritizing toxicities differently during treatment plan optimization.Thirty patients, representative of our population receiving definitive (chemo)radiotherapy for head and neck cancer, were selected for this study. Fully automated VMAT treatment plans were created using a mix of NTCP and TCP objective functions. TCP objective functions were based on a PTV D98 ≥ 95% of the prescribed dose. Automated plans were created for each patient using different priorities for 20 toxicities including salivary, swallowing and speech problems, pain and general symptoms. Toxicities were prioritized in different ways: 1) minimizing the impact on QOL; 2) minimizing salivary toxicity; 3) minimizing swallowing related toxicity; 4) equal weights for each toxicity. Manually created clinical VMAT plans were used as reference.Toxicity priorities were easily selected prior to optimization and resulting plans did not need any manual post-processing. Fully automated NTCP-TCP-based optimization consistently resulted in superior plans when compared to the clinical reference plans and always in lower NTCPs for swallowing, speech, pain and general toxicities. Dose reductions in OARs depended on which toxicities were prioritized during plan optimization, e.g., the mean dose in the parotid glands was on average 13 Gy lower in salivary sparing plans compared with plans focused on swallowing sparing, and the average NTCP of salivary toxicities in those plans was 7% lower. However, these reductions were at the cost of higher NTCPs for dysphagia and moderate-to-severe general symptoms such as fatigue. Plans optimized based on the QOL model had the highest QOL scores.Automated outcome-directed treatment plan optimization is feasible and consistently outperforms manually generated VMAT plans in terms of NTCP and QOL scores. Predicted toxicity profiles depend on toxicity priority settings prior to planning. The highest QOL score was obtained when toxicities were prioritized according to relative impact on QOL.