Predicting outcomes in anal cancer patients using multi-centre data and distributed learning - A proof-of-concept study.

摘要: Abstract Background and purpose Predicting outcomes is challenging in rare cancers. Single-institutional datasets are often small multi-institutional data sharing complex. Distributed learning allows machine models to use from multiple institutions without exchanging individual patient-level data. We demonstrate this technique a proof-of-concept study of anal cancer patients treated with chemoradiotherapy across European countries. Materials methods atomCAT three-centre collaboration between Leeds Cancer Centre (UK), MAASTRO Clinic (The Netherlands) Oslo University Hospital (Norway). trained validated Cox proportional hazards regression model distributed fashion using 281 radical, conformal for three institutions. Our primary endpoint was overall survival. selected disease stage, sex, age, tumour size, planned radiotherapy dose (in EQD2) priori as predictor variables. Results The all centres found worse survival high risk stage (HR = 2.02), male sex (HR = 3.06), older age (HR = 1.33 per 10 years), larger volume (HR = 1.05 10 cm3) lower (HR = 1.20 5 Gy). A mean concordance index 0.72 achieved during validation, limited variation (Leeds = 0.72, MAASTRO = 0.74, Oslo = 0.70). global performed well stratification two out centres. Conclusions Using learning, we accessed analysed one the largest available cohorts modern techniques. This demonstrates value outcome modelling