Personalized cancer therapy seeks to tailor treatment to an individual patient's biology. Therefore, a means to characterize radiosensitivity is necessary. In this study, we investigated radiosensitivity in the normal esophagus using an imaging biomarker of radiation-response and esophageal toxicity, esophageal expansion, as a method to quantify radiosensitivity in 134 non-small-cell lung cancer patients, by using K-Means clustering to group patients based on esophageal radiosensitivity. Patients within the cluster of higher response and lower dose were labelled as radiosensitive. This information was used as a variable in toxicity prediction modelling (lasso logistic regression). The resultant model performance was quantified and compared to toxicity prediction modelling without utilizing radiosensitivity information. The esophageal expansion-response was highly variable between patients, even for similar radiation doses. K-Means clustering was able to identify three patient subgroups of radiosensitivity: radiosensitive, radio-normal, and radioresistant groups. Inclusion of the radiosensitive variable improved lasso logistic regression models compared to model performance without radiosensitivity information. Esophageal radiosensitivity can be quantified using esophageal expansion and K-Means clustering to improve toxicity prediction modelling. Finally, this methodology may be applied in clinical trials to validate pre-treatment biomarkers of esophageal toxicity.