In this dissertation we develop a framework that combines data

mining, statistics and operations research methods for improving

real-time decision support systems in healthcare. Our approach

consists of three main concepts: data gathering and preprocessing,

modeling, and deployment. We introduce the notion of offline and

semi-offline modeling to differentiate between models that are based on known baseline behavior and those based on a baseline with missing information. We apply and illustrate the framework in the context of two important healthcare contexts: biosurveillance and

kidney allocation. In the biosurveillance context, we address the

problem of early detection of disease outbreaks. We discuss integer

programming-based univariate monitoring and statistical and

operations research-based multivariate monitoring approaches. We

assess method performance on authentic biosurveillance data. In the kidney allocation context, we present a two-phase model that

combines an integer programming-based learning phase and a

data-analytical based real-time phase. We examine and evaluate our

method on the current Organ Procurement and Transplantation Network (OPTN) waiting list. In both contexts, we show that our framework produces significant improvements over existing methods.