LAUSR

For patients with oesophageal cancer, producing accurate prediction models for survival after oesophagectomy has proved challenging. We investigated whether Random Survival Forests (RSF), a novel machine learning method, could produce an accurate prognostic model for overall survival after oesophagectomy. The study used data from the National Oesophago-Gastric Cancer Audit and included patients diagnosed with oesophageal adenocarcinoma or squamous cell carcinoma between 2012 and 2018 in England and Wales and who underwent a curative oesophagectomy with adequate lymphadenectomy (15 LN) and survived to discharge (n=6198). Missing data was handled using multiple imputation and the data was split into training and validation cohorts. 13 variables were selected for inclusion using Random Forest variable importance and used to train the final model. The same variables were used to develop a traditional Cox regression model. Median survival was 53 months in both cohorts. The final RSF model had good discrimination in the validation cohort with a C-index of 0.757(0.755-0.759), exceeding the Cox model; 0.748(0.746-0.750). At 3 years post-surgery, overall survival was 56.2%. The RSF yielded a mean predicted survival of 55.8%(IQR 29.5%-81.7%) compared to 55.4%(40.0%-77.7%) for the Cox model. The most important variables were lymph node involvement and pT/ypT stage, however other variables including neoadjuvant treatment completion and surgical complications were also found to be important. A Random Forest survival model provided better performance in predicting survival after curative oesophagectomy. This will allow more personalised predictions to be delivered clinicians and patients. Random Forest survival models can accurately predict post-operative prognosis after oesophagectomy.