Recurrent Event Models

Models for items that can have the same event many times – repairable systems that fail and are repaired, patients with repeated hospitalisations. The data are each item’s event times, measured from the start of its life, with a right-censored (c=1) row marking the end of its observation. Import everything on these pages from surpyval.recurrent. The theory is in Recurrent Event Analysis and Recurrent Event Regression Analysis; worked examples are in Recurrent Event Modelling with SurPyval and Recurrent Event Regression Modelling with SurPyval.

Each fitter below returns one of three fitted-model classes: a ParametricRecurrenceModel (the Poisson processes), a RenewalModel (the renewal and imperfect-repair models) or a ProportionalIntensityModel (the regression models). The non-parametric and cause-specific fitters return a fitted instance of their own class.

Recurrent Event (Poisson Process) Models

Processes whose events arrive at a rate that depends only on the system’s age (or not at all, for the HPP), so a repair leaves the system as it was just before the failure (“as bad as old”).

Non-Parametric Models

The mean cumulative function estimated without assuming a process.

Non-Parametric Counting (MCF)

Renewal Models

Processes in which a repair restores the system part of the way to new (or all the way, for an ordinary renewal process).

Competing Risks (Marked) Models

Recurrent processes with several kinds of event, each with its own intensity.

Recurrent Event Regression Models

Intensities that depend on covariates.

Trend Tests and Diagnostics

Tests of whether the event rate is changing, and the result classes of the tests and of the models’ goodness-of-fit methods.

Trend Tests and Diagnostics