Machine Learning Survival Analysis

The Piece-wise exponential (additive mixed) model (PEM/PAMM) has become a flexible and popular approach for statistical analysis of time-to-event data, especially suitable for estimation of time-varying effects and incorporation of time-dependent covariates. This talk will discuss some current developments in machine learning-based survival analysis, in particular a general machine learning framework for survival analysis based on PEMs and a deep learning extension of PAMMs called DeepPAMM. The latter inherits all the abilities of PAMMs and, among other advantages, enables estimation of models with multiple modes of features, e.g. tabular data and image data.