- Abstract: Survival Analysis (time-to-event analysis) in the presence of multiple possible adverse events, i.e., competing risks, is a challenging, yet very important problem in medicine, finance, manufacturing, etc. Extending classical survival analysis to competing risks is not trivial since only one event (e.g. one cause of death) is observed and hence, the incidence of an event of interest is often obscured by other related competing events. This leads to the nonidentifiability of the event times’ distribution parameters, which makes the problem significantly more challenging. In this work we introduce Siamese Survival Prognosis Network, a novel Siamese Deep Neural Network architecture that is able to effectively learn from data in the presence of multiple adverse events. The Siamese Survival Network is especially crafted to issue pairwise concordant time-dependent risks, in which longer event times are assigned lower risks. Furthermore, our architecture is able to directly optimize an approximation to the C-discrimination index, rather than relying on well-known metrics of cross-entropy etc., and which are not able to capture the unique requirements of survival analysis with competing risks. Our results show consistent performance improvements on a number of publicly available medical datasets over both statistical and deep learning state-of-the-art methods.
- TL;DR: In this work we introduce a novel Siamese Deep Neural Network architecture that is able to effectively learn from data in the presence of multiple adverse events.
- Keywords: survival analysis, competing risks, siamese neural networks