UC Irvine

Cohort studies of aging-related diseases often produce event-time data that are subject to left-truncation, since participants enroll after some baseline age while being free of the disease of interest, in addition to right-censoring due to variable follow-up. This dissertation develops a suite of semiparametric methods, organized around a two-part conditional modeling framework, for drawing valid inference from such data and for quantifying healthspan (the portion of life spent in good health) as an endpoint in aging research. The first project proposes a semiparametric sieve likelihood approach for fitting a linear regression model to a response subject to both left-truncation and right-censoring; the resulting estimators are shown to be consistent, asymptotically normal, and semi parametrically efficient, and the method is illustrated on data from the Canadian Study of Health and Aging and The 90+ Study. The second project introduces the Golden Health Index (GHI), the expected proportion of post-baseline life lived disease-free; we estimate the GHI and its covariate associations via a two-part model that accounts for left-truncation, and establish its large-sample properties, and applied to the National Alzheimer’s Coordinating Center data, the framework shows that APOE ε4 allele status is negatively associated with the GHI, whereas lifespan is positively associated with the GHI after an initial dip. The third project closes a theoretical gap by establishing the asymptotic normality of sieve estimates in a broad range of non-and semi-parametric regression models in which a smooth unknown function is among the parameters of interest, thereby justifying likelihood-based inference in the settings of the first two projects.