Markus Diefenhardt, Daniel Martin, Ethan B. Ludmir, Maximilian Fleischmann, Ralf-Dieter Hofheinz, Michael Ghadimi, Rebekka Kosmala, Bülent Polat, Tim Friede, Bruce D. Minsky, Claus Rödel, Emmanouil Fokas

• Background: There is a lack of predictive models to identify patients at risk of high neoadjuvant chemoradiotherapy (CRT)-related acute toxicity in rectal cancer. Patient and Methods: The CAO/ARO/AIO-04 trial was divided into a development (n = 831) and a validation (n = 405) cohort. Using a best subset selection approach, predictive models for grade 3–4 acute toxicity were calculated including clinicopathologic characteristics, pretreatment blood parameters, and baseline results of quality-of-life questionnaires and evaluated using the areaBackground: There is a lack of predictive models to identify patients at risk of high neoadjuvant chemoradiotherapy (CRT)-related acute toxicity in rectal cancer. Patient and Methods: The CAO/ARO/AIO-04 trial was divided into a development (n = 831) and a validation (n = 405) cohort. Using a best subset selection approach, predictive models for grade 3–4 acute toxicity were calculated including clinicopathologic characteristics, pretreatment blood parameters, and baseline results of quality-of-life questionnaires and evaluated using the area under the ROC curve. The final model was internally and externally validated. Results: In the development cohort, 155 patients developed grade 3–4 toxicities due to CRT. In the final evaluation, 15 parameters were included in the logistic regression models using best-subset selection. BMI, gender, and emotional functioning remained significant for predicting toxicity, with a discrimination ability adjusted for overfitting of AUC 0.687. The odds of experiencing high-grade toxicity were 3.8 times higher in the intermediate and 6.4 times higher in the high-risk group (p < 0.001). Rates of toxicity (p = 0.001) and low treatment adherence (p = 0.007) remained significantly different in the validation cohort, whereas discrimination ability was not significantly worse (DeLong test 0.09). Conclusion: We developed and validated a predictive model for toxicity using gender, BMI, and emotional functioning. Such a model could help identify patients at risk for treatment-related high-grade toxicity to assist in treatment guidance and patient participation in shared decision making.…