|
BACKGROUND/RATIONALE:
The goal of this project is to develop a principal stratification framework that can correctly address the issue of truncation by death with possible dropouts in causal inferences of interventions on health related quality of life (HRQOL). Interest in HRQOL is soaring because veterans are living longer, want to stay healthy and active for as long as possible, and many interventions do not have a large effect on mortality, yet are intended to improve HRQOL. Commonly used HRQOL measures do not include death and as a result, deceased subjects have undefined HRQOL. Existing methods for dealing with truncation by death have some serious limitations.
OBJECTIVE(S):
Health services research methodology is an emphasis area in VA HSR&D and statistical methods and trial design are essential to a wide variety of VA studies. HRQOL has become a standard outcome in VA studies. Due to the effects of an older population and advanced diseases in many VA funded HRQOL studies, some persons die before the end of the study, resulting in "missing data" since the longitudinal quality of life data of those subjects are truncated. The most commonly used analytic methods on HRQOL in these studies generally ignore subjects who die and can lead to biased results. The newly designed methods will: estimate causal effects, not just associational effects, of interventions on HRQOL measures; perform well even when the sample size is not large since they do not resort to asymptotic properties of estimators as classical statistical inference procedures do; and handle both missing data and longitudinal structures to make them more useful in practice.
METHODS:
The statistical problems to be addressed are how to make causal inferences about the efficacy of the treatment using randomized and observational data truncated by death with possible missing values. Bayesian methods will be utilized to solve these statistical problems. The data will come from three studies: the VA Adult Day Health Care (ADHC) Evaluation study (Hedrick et al, 1993), a randomized trial of a computer reminder for managing veterans with chronic heart failure (Subramanian et al, 2004), and the VA Ambulatory Care Quality Improvement Project (ACQUIP) (Dominitz et al, 2001).
FINDINGS/RESULTS:
Paper: Chi, Y. Y. and Zhou, X. H. Using principal stratification in longitudinal causal inference with truncation to death (submitted)
Conclusion: In this paper, we extended the use of principal stratifcation to facilitate causal analysis of longitudinal data. Observations truncated by death are philosophically different from observations missing during the lifetime, with the former not well-defined for existence. The periodically observed death truncation and its counterfactual play central roles in the construction of principal strata to form patient groups that are independent of their assignment
of treatment. Causal relationship is thus established within each group of patients. Even though inference for total survivors is almost of the greatest interest, the causal effects of treatment on responses that would and would have been observed for partial survivors may be helpful in practice. Missing data with ignorable missing data mechanism are considered in the model that is speci c to each principal stratum, and the model could naturally be extended to allow for non-ignorable missing data mechanism.
IMPACT:
There has been none to date.
PUBLICATIONS:
None at this time.
DRA:
Aging and Age-Related Changes
DRE:
Quality of Care
Keywords:
Quality of life, Research measure, Research method
MeSH Terms:
none
|
