Non-ignorable missing covariate data in parametric survival analysis
Within any epidemiological study missing data is almost inevitable. This missing data is often ignored; however, unless we can assume quite restrictive mechanisms, this will lead to biased estimates. Our motivation are data collected to study the long-term effect of severity of disability upon survi...
Saved in:
| Main Author: | |
|---|---|
| Format: | Dissertation |
| Language: | English |
| Published: |
ProQuest Dissertations & Theses
01-01-2007
|
| Subjects: | |
| Online Access: | Get full text |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Within any epidemiological study missing data is almost inevitable. This missing data is often ignored; however, unless we can assume quite restrictive mechanisms, this will lead to biased estimates. Our motivation are data collected to study the long-term effect of severity of disability upon survival in children with cerebral palsy (henceforth CP). The analysis of such an old data set brings to light statistical difficulties. The main issue in this data is the amount of missing covariate data. We raise concerns about the mechanism causing data to be missing. We present a flexible class of joint models for the survival times and the missing data mechanism which allows us to vary the mechanism causing the missing data. Simulation studies prove this model to be both precise and reliable in estimating survival with missing data. We show that long term survival in the moderately disabled is high and, therefore, a large proportion will be surviving to times when they require care specifically for elderly CP sufferers. In particular, our models suggest that survival from diagnosis is considerably higher than has been previously estimated from this data. This thesis contributes to the discussion of possible methods for dealing with NMAR data. |
|---|