Ascertainment correction in frailty models for recurrent events data. (18th April 2016)

• Record Type: Journal Article
• Authors: Balan, Theodor A.; Jonker, Marianne A.; Johannesma, Paul C.; Putter, Hein
• Is Part Of: Statistics in medicine. Volume 35:Number 23(2016)
• Page Start: 4183
• Page End: 4201
• Publication Date: 2016-04-18
• Subjects: frailty -- selection bias -- recurrent events; Medical statistics -- Periodicals; Statistique médicale -- Périodiques; Statistiques médicales -- Périodiques; 610.727
• DOI: 10.1002/sim.6968
• Languages: English
• ISSNs: 0277-6715

Abstract

Abstract : In retrospective studies involving recurrent events, it is common to select individuals based on their event history up to the time of selection. In this case, the ascertained subjects might not be representative for the target population, and the analysis should take the selection mechanism into account. The purpose of this paper is two‐fold. First, to study what happens when the data analysis is not adjusted for the selection and second, to propose a corrected analysis. Under the Andersen–Gill and shared frailty regression models, we show that the estimators of covariate effects, incidence, and frailty variance can be biased if the ascertainment is ignored, and we show that with a simple adjustment of the likelihood, unbiased and consistent estimators are obtained. The proposed method is assessed by a simulation study and is illustrated on a data set comprising recurrent pneumothoraces. Copyright © 2016 John Wiley & Sons, Ltd.