Description
Extend Cox regression to people who can experience the same event repeatedly, choosing how risk sets and time are defined to match the question.
Standard Cox regression stops at the first event, discarding later ones. Three extensions keep them, differing mainly in who is at risk for each event. The Andersen-Gill model treats each person as a counting process with one common baseline hazard, using (start, stop] intervals, and asks how covariates affect the overall event rate. Prentice-Williams-Peterson models stratify by event number: only people who have had s-1 events are at risk for the s-th, with time measured from entry (total time) or from the previous event (gap time), so first and later events can have different baselines. The Wei-Lin-Weissfeld model fits a marginal Cox model for each event number with everyone at risk from time zero and robust variances for within-person correlation. The models answer different questions, so the choice must precede the fit.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Unmigrated
- Stage
- Evaluation, Post-launch
- Data
- Quantitative
- Effort
- Days
Steps
- Decide whether the question concerns the overall rate (Andersen-Gill), the ordered sequence of events (PWP) or each event number separately (WLW), and choose total or gap time.
- Arrange the data: one record per person per at-risk interval for Andersen-Gill and PWP, or one record per person per event number with time from zero for WLW.
- Fit the chosen Cox model, stratified by event number for PWP and WLW, with robust (sandwich) standard errors clustered by person.
- Check proportional hazards for key covariates within strata.
- Compare covariate effects across event numbers or across models, and report hazard ratios with their intervals.
- Complement the regression with a mean cumulative function plot of the raw recurrence pattern.
Sources
- Statistical Methods for Survival Data Analysis (Lee & Wang, 2003)
Also known as
- Andersen-Gill model
- AG model
- PWP model
- Prentice-Williams-Peterson model
- WLW marginal model
- Counting process Cox model for repeated events
- Recurrent-event proportional hazards regression
Categories
- Analytics & Behaviour
- Statistics & Modelling