Validation of a predictive clinical risk model for acute skin toxicity in patients undergoing breast radiotherapy


Session type:


Tim Rattay1,Kerstie Johnson2,R Paul Symonds1,Jenny Chang-Claude3,Petra Seibold3,Gillian Barnett4,Charlotte Coles4,Catharine West5,Christopher Talbot1
1University of Leicester,2Nottingham University Hospitals NHS Trust,3DKFZ German Cancer Research Institute,4Cambridge University Hospitals NHS Trust,5University of Manchester



Clinically significant side-effects from radiotherapy affect around a quarter of breast cancer patients and may have a considerable impact on breast cosmesis and quality of life.  If patients at high risk of radiation toxicity could be identified at breast cancer diagnosis, this could be taken into account when discussing surgical treatment options.  The aim of this study was to validate a predictive model for acute skin toxicity in patients undergoing breast radiotherapy.


Using multivariate logistic regression and backwards elimination, the risk model for acute skin toxicity (≥1 acute desquamation) was first developed in patient cohorts treated by breast-conserving surgery and whole breast radiotherapy in three European centres (Leicester, Heidelberg/Mannheim, Cambridge; total n=2,012) with a biologically effective dose (BED) range from 47.1 to 67.2 Gy.  The model was subsequently updated and externally validated in breast cancer patients enrolled in the multi-centre REQUITE cohort study (n=2,059; BED range 44.6 to 75.4 Gy).


The final updated model with the variables age, breast size, BED + boost dose, use of intensity-modulated radiotherapy (IMRT), and presence/absence of diabetes, smoking, and hypertension or cardiovascular disease proved to give best prediction of acute skin toxicity with a c-statistic (AUC) of 0.77 in the development and 0.74 in the validation cohort and was well calibrated (Hosmer-Lemeshow p=0.43). 


This updated predictive risk model for acute skin toxicity has the potential to give clinicians important information when planning treatment to reduce side-effects and optimise quality of life.  The addition of predictive genetic markers investigated as part of the study is likely to improve model performance and will be presented.