ESTRO 2024 - Abstract Book

S1215

Clinical - Head & neck

ESTRO 2024

Material/Methods:

We reviewed data of patients with oral cavity cancer who underwent surgery and adjuvant (chemo)radiotherapy during 2010 – 2020 at our institution. The changes in skeletal muscle index (SMI) were measured using pre- and post radiotherapy computed tomography scans at the C3 vertebral level. Overall survival was defined as the time from the date of surgery to the date of any-cause death or the last follow-up. The cohort was randomly assigned to training and test sets in accordance with a ratio of 6:4. The eXtreme Gradient Boosting (XGBoost) machine learning method was applied to establish a prediction model, and the performances were evaluated using the area under the receiver operating characteristic curve (AUC). Demographic and disease-specific data were used as input variables. The Shapley Additive Explanation (SHAP) method was applied to calculate the SHAP value and interpret the ML models. The SHAP values measure the contribution of each feature on the prediction of the models, with values greater than 0 equivalent to a positive effect where an event is more likely, and vice versa. The SHAP dependence plot was developed to identify cut-off value of muscle loss, focusing on where the SHAP values cross 0. Cox regression was used to validate the cut-off value. A total of 427 patients, including 378 (88.5%) male and 49 (11.5%) female, with a median age of 55 years (IQR: 48-62) were included. Median follow-up time was 4.7 years (IQR: 2.6 – 7.4), and 136 deaths (31.9%) were observed. The mean SMI decreased by 1.7% during adjuvant radiotherapy (52.0 cm 2 /m 2 vs. 51.1 cm 2 /m 2 ; 95% confidence interval: −1.2 to −0.6; p <0.001). The XGBoost model achieved an AUC of 0.864 (95% confidence interval: 0.860-0.866) in the training set and 0.890 in the test set. The SHAP summary plot shows that SMI change was the most important feature to the higher likelihood of death (Figure 1A). The SHAP dependence plot shows that patients with decreasing SMI during radiotherapy (decrease in the x-axis value) were associated with a higher SHAP value, indicating a higher likelihood of death (increase in the y-axis value) (Figure 1B). The XGBoost model suggested that SMI loss of 4.2% during radiotherapy for predicting survival. The 5-year overall survival rates were 20.0% and 86.0% in patients with SMI with or without loss ≥4.2%, respectively ( p <0.001, Figure 2). On multivariable Cox regression analysis, SMI loss ≥4.2% was independently associated with poorer overall survival (hazard ratio: 9.15, 95% confidence interval: 6.05– 13.82, p <0.001; Table 1). Figure 1 (A) SHAP summary plot showing the distribution of the SHAP values of each feature. Features are ranked according to the importance of features on the prediction. (B) SHAP dependence plot of SMI change. Feature values are plotted on the x-axis, and the SHAP values are plotted on the y-axis. Each dot represents a SHAP value for a feature per patient, and color from red to blue represents the feature’s value from high to low. The red arrow indicates that the SHAP values cross 0 at SMI loss of 4.2%. Results: