Abstract:
|
Neurodegenerative diseases last decades, and knowledge of an individual's disease progression can be invaluable to identify interventions. Unfortunately, predicting progressions is not easy because clinical data do not cover the full disease course. One option is to align the pieces of short-term clinical data along an event time on the disease path. This variable for alignment may be not observed for everyone, but is censored. Towards solving this problem, we focus on a simple regression model with censored covariates. The goal in regression is to produce unbiased and efficient parameter estimates, which is not straightforward when covariates are censored. The commonly used complete case analysis, multiple imputation and inverse probability weighting (IPW) methods all have disadvantages. We propose an augmented IPW method, which only requires one of two models to be correctly specified for guaranteed consistency. Moreso, efficiency is improved by making better use of the information available from censored observations. We tested our method in simulation studies and applied it to a data example, evaluating how motor onset impacts the mental health of Huntington disease patients.
|