Joint Models for rare events

Sophie Potts, University of Goettingen

Co-authors: Elisabeth Bergherr, University of Goettingen

Abstract: Joint Models for longitudinal and time-to-event data are increasingly used in various fields of application.

Regarding the time-to-event part of the model, which is usually fitted via a Cox proportional hazards model, one borrows all limitations from the latter, such as low numbers of events (overall or in individual categories of covariates) leading to monotone likelihoods. The Firth correction has been used successfully for Cox models to prevent from those issues by introducing a correction term in the score function.

In order to make joint models estimable in cases of rare events and/or highly imbalanced covariates, we implement the Firth correction for joint models.

With the corrected model we analyse the influence of personality traits on marriage dissolution.