Modeling dose to normal brain following hypofractionated stereotactic radiotherapy to a single brain metastasis

Brain metastases (BM) occur in at least 20% of people with an advanced solid malignancy and can lead to morbidity or mortality if not controlled. Hypofractionated stereotactic radiotherapy (HF-SRT) may be preferred to single fraction stereotactic radiosurgery (SRS) in certain clinical situations to improve the therapeutic ratio. Certain dose-volume metrics can predict the risk of symptomatic radionecrosis or other toxicities after HF-SRT but dosimetric predictors of more complex outcomes such as cognitive function are less well described. For this reason, clinicians aim to limit the dose to uninvolved brain so that it is as low as reasonably achievable rather than simply aiming for a threshold goal. The number of treatment arcs and arc length, floor angles and volumetric modulated arc therapy (VMAT) optimization strategies impact low dose distribution outside the target volume, and various combinations can be trialed and modified during the plan optimization process. On the other hand, variation in patient-specific factors such as target size and location cannot be changed. This combination of modifiable and fixed factors is expected to influence dose to uninvolved brain, but their relative impacts are not well understood.

This study aims to investigate both modifiable and fixed factors influencing low dose distribution to the normal brain and to develop predictive models for multiple normal brain dose metrics used for developing single BM VMAT HF-SRT plans.

Consecutive patients receiving HF-SRT for a single BM or cavity at a single institution between July 2017 and December 2019 were reviewed. Patient-specific, dosimetric and treatment technique data was collected retrospectively and used to develop statistical models to predict planned dose to the normal brain. Exploratory analyses and relevant parametric and non-parametric bivariate testing were performed to identify significant variables for inclusion in model development. A manual backward stepwise approach of multiple linear regression (MLR) was used to model achievable low dose in normal brain tissue for three outcomes. The three model outcomes produced in this study include predicting normal brain mean in Gray (Gy), the volume of normal brain receiving 50% of the prescribed dose and the volume of normal brain receiving 25% of the prescribed dose.

Planning data from a total of 90 patients was included in the analysis. Over 60 patient-specific, dosimetric and treatment technique variables were examined from each patient, with several variables showing significance for all three outcomes through bivariate testing. The nine most significant variables were included in the model development. MLR modeling produced three or four statistically significant variables of planning target volume (PTV) location, PTV shape, PTV volume and arc combinations. Each model explained between 68.6% and 78.3% of the variation in normal brain dose, and the overall significance of each model is p < 0.001.

This study has produced three predictive models which estimate an achievable amount of low dose to normal brain for use in the development of single BM VMAT HF-SRT 6MV plans. This is one of the first studies to investigate the impact of a comprehensive range of variables on achievable low dose distribution.