Predictive models for 5-year mortality after breast camcer surgery

Abstract

Few studies of breast cancer surgery outcomes have used longitudinal data for more than five years. To validate the use of artificial neural network (ANN) models in predicting 5-year mortality for breast cancer surgery patients and to compare predictive accuracy between an ANN model and a multiple logistic regression (MLR) model. This study compared the performance of ANN and MLR models based on retrospective clinical data of 3,632 breast cancer surgery patients treated during 1996–2010. Global sensitivity score and analysis approach were also employed to assess the relative importance of variables and the relative significance of input parameters in the system model. In the training, testing, and validation groups of breast cancer surgery patients, the ANN model significantly outperformed the MLR model in terms of specificity, sensitivity, negative predictive value (NPV), positive predictive value (PPV), accuracy, and area under the receiver operating characteristic curves. Surgeon volume was the most influential variable affecting 5-year mortality followed by hospital volume, age, and Charlson co-morbidity index (CCI) score. The ANN model achieved higher overall performance indices and was more accurate in predicting 5-year mortality, compared with the conventional MLR model.