exDNNsurv: Using Deep Neural Network as An Extension to Traditional Survival Modeling

Abstract

Due to the presence of censoring in survival data, there has been a growing interest in using deep neural networks to deal with the complexities of survival datasets. Though a great number of deep neural network models have been proposed in recent times, these models usually require the development of special loss functions to deal with censoring in the data. In this article, we propose a simpler approach to achieving the same goal by using jackknife pseudo survival probabilities as a quantitative response variable that can be easily handled by a deep neural network model. This approach, thus reduces a complex problem to a traditional regression problem. Further, our model directly outputs the conditional probability of a subject surviving given that the subject earlier on survived. This can be a very desirable feature from a practical point of view. We name our model as exDeepsurv.