A novel approach to predict acute radiation dermatitis in patients with head and neck cancer using a model based on Bayesian probability.

Purpose: In this study, we aimed to establish a method for predicting the probability of each acute radiation dermatitis (ARD) grade during the head and neck Volumetric Modulated Arc Therapy (VMAT) radiotherapy planning phase based on Bayesian probability.

Methods: The skin dose volume >50 Gy (V), calculated using the treatment planning system, was used as a factor related to skin toxicity. The empirical distribution of each ARD grade relative to V was obtained from the ARD grades of 119 patients (55, 50, and 14 patients with G1, G2, and G3, respectively) determined by head and neck cancer specialists. Using Bayes' theorem, the Bayesian probabilities of G1, G2, and G3 for each value of V were calculated with an empirical distribution. Conversely, V was obtained based on the Bayesian probabilities of G1, G2, and G3.

Results: The empirical distribution for each graded patient group demonstrated a normal distribution. The method predicted ARD grades with 92.4 % accuracy and provided a V value for each grade. For example, using the graph, we could predict that V should be ≤24.5 cm to achieve G1 with 70 % probability.

Conclusions: The Bayesian probability-based ARD prediction method could predict the ARD grade at the treatment planning stage using limited patient diagnostic data that demonstrated a normal distribution. If the probability of an ARD grade is high, skin care can be initiated in advance. Furthermore, the V value during treatment planning can provide radiation oncologists with data for strategies to reduce ARD.