Counting recurrent events when death gets in the way
A new parametric model for the average number of recurrent events over time — built to handle competing risks.
The published work
The challenge
Recurrent events — infections, cancer recurrences, hospitalisations, childbirth — occur many times for the same individual and are everywhere in medical research. But individuals may also face competing events such as death, which prevent further recurrences and complicate estimation. A method that ignores the competing risk gives the wrong answer.
The approach
Led by Red Door Analytics' Joshua Entrop with Michael Crowther, the work develops a new parametric model that handles the recurrent-event and competing-event processes simultaneously, producing a smooth, easily interpretable estimate of the average number of events over time. It was published in Biometrical Journal.
What it enables
The result is a clear summary measure — for example, the average number of hospitalisations after colon-cancer surgery as a window into recovery: patients with more hospitalisations likely face more severe complications and a slower recovery.
Where it fits
By design it is a summary measure. For a more granular view of the recurrent-event process — explicit transitions between states or event occurrences — multi-state models are complementary, which is exactly what RDA's multistate and pendragon provide.