2026-09-20T16:36:38Zhttps://digital.lib.washington.edu/server/oai/request

oai:digital.lib.washington.edu:1773/355412026-02-16T02:29:51Zcom_1773_4888col_1773_4900

Heagerty, Patrick Kosel, Alison 2016-04-06T16:30:10Z 2016-04-06T16:30:10Z 2016-04-06 2016-03 Kosel_washington_0250E_15594.pdf http://hdl.handle.net/1773/35541 Thesis (Ph.D.)--University of Washington, 2016-03 Statistical methods that can provide patients and their healthcare providers with individual predictions are needed so that informed medical decisions can be made. Ideally an individual prediction would display the full range of possible outcomes (full predictive distribution), would be obtained with a specified level of precision, and would be minimally reliant on statistical model assumptions. We propose a novel method that satisfies each of these criteria via the semi-supervised creation of an axis-parallel covariate neighborhood constructed around a given point of interest. We then provide non-parametric estimates of the outcome distribution for subjects in this neighborhood, which we refer to as a localized prediction. We implement the local prediction method using dynamic graphical methods that allow the user to vary key options such as the choice of neighborhood variables and the size of the neighborhood. Furthermore, we expand our method to handle multiple treatment groups and longitudinal data. application/pdf en_US local prediction; non-parametric; semi-supervised learning Biostatistics biostatistics Local Estimation of Patient Prognosis Thesis Open Access open.access