Doubly robust estimation, optimally truncated inverse-intensity weighting and increment-based methods for the analysis of irregularly observed longitudinal data
Article first published online: 10 OCT 2012
Copyright © 2012 John Wiley & Sons, Ltd.
Statistics in Medicine
Volume 32, Issue 6, pages 1054–1072, 15 March 2013
How to Cite
Pullenayegum, E. M. and Feldman, B. M. (2013), Doubly robust estimation, optimally truncated inverse-intensity weighting and increment-based methods for the analysis of irregularly observed longitudinal data. Statist. Med., 32: 1054–1072. doi: 10.1002/sim.5640
- Issue published online: 14 FEB 2013
- Article first published online: 10 OCT 2012
- Manuscript Accepted: 10 SEP 2012
- Manuscript Received: 21 FEB 2012
- longitudinal data;
- inverse-probability weighting;
- doubly robust estimation
Longitudinal data arising from routine follow-up of patients will often have irregular measurement times. Existing methods for analysis include joint modelling of the outcome and measurement processes, and inverse-intensity weighting (IIW). This work extends previously proposed analysis of increments to the case of irregular follow-up, yielding a model for the increments that can be used as a stand-alone method. Furthermore, we propose two ways of combining the increments and IIW estimators. First, we use the increment model to select the truncation point for the inverse-intensity weights that minimises the mean squared error of the IIW estimator. Second, we use the increment model to augment the usual IIW estimating equations to form a doubly robust estimator. We evaluate the methods through simulation and apply these to a recent study of juvenile dermatomyositis. Copyright © 2012 John Wiley & Sons, Ltd.