Development of Prediction Model for Mean Parotid Dose of HNC Undergoing Radiotherapy - A Single Institutional Study

The aim of the study was to develop a simple prediction model based on previous treatment plans for head-and-neck cancer (HNC).

This study was conducted on 95 patients who underwent volumetric-modulated arc therapy (VMAT) with curative intent for HNC at our institute between January 2016 and December 2022 with intact bilateral parotid glands. Two simple prediction models were used: one linear regression model and one exponential model. Both models use fractional overlapping parotid volume with planning target volume (PTV) as a predictor of mean parotid dose. The fractional overlapping volume was calculated as the difference between the volume of the parotid gland minus the volume of the parotid gland outside the PTV plus a 2 mm margin, divided by the volume of the parotid gland. Statistical calculations were done using data analysis tools and Solver in Microsoft Excel (Microsoft Office 2013, Redmond, WA, USA). To enhance the accuracy of the results, outliers were excluded with residuals >2 standard deviations below and above the residuals. R² and root-mean-square error were calculated for both models to evaluate the quality of the predictions. The normality of both models' residuals was validated using the Shapiro-Wilk test.

Both linear and exponential prediction models exhibited strong correlation statistics, with r² = 0.85 and 0.82, respectively. The authors found a fractional overlap of 16.4% and 18.9% in linear and exponential models that predict parotid mean dose 26 Gy. The implementation was carried out on a cohort of 12 prospective patients, demonstrating a remarkable improvement in minimizing the dose to the parotid glands.

In this single-institutional study, the authors successfully developed a prediction model for mean parotid dose in HNC patients undergoing radiotherapy. The model showed promising accuracy and has the potential to assist planners in optimizing treatment plans and minimizing radiation-related toxicity. It is possible to avoid under sparing the organs at risks in some cases and wasting time or effort on physically impossible goals in others using this prediction model. As a result, planning resources can be used much more efficiently. Future studies should focus on validating the model's performance using external datasets and exploring its integration into clinical practice.