Simplifying Data Structuring for Complex Survival Analysis with pammtools
Regular talk, 2:00 - 3:00 PM
Survival analysis is a powerful tool for clinical and population health research, but standard survival models (e.g., Cox Proportional Hazard models) are limited by semiparametric model assumptions. These assumptions may be violated with dynamic health data when events are time dependent, can recur, and when hazards are not proportional. Piece-wise exponential Additive Mixed Models (PAMMs) are a flexible tool for modeling complex time-to-event data, including recurrent events (Ramjith, Bender, Roes & Jonker, 2022). Moving from Cox models to PAMM survival models requires transforming data into a piecewise exponential data (PED) format. This talk will provide an accessible introduction to complex survival modelling with PAMMs and will demonstrate an example workflow of formatting a PED with the pammtools package in R (Bender & Scheipl, 2018). This workflow will illustrate an applied example with de-identified perinatal acute health utilization data from Oregon Health Authority and will demonstrate how to combine multiple datasets into a PED style format for both single- and recurrent-event survival analysis. This presentation will also illustrate the decision-points along the PED data structuring process, including selecting gap-time vs. calendar time, specifying event time intervals, and incorporating time-dependent covariates. The goal of this talk is to increase awareness of complex survival modelling approaches and demystify the data configuration process.
 | Pronouns: she/herEugene, OR, USA |